text
stringlengths 47
469k
| meta
dict | domain
stringclasses 1
value |
|---|---|---|
---
abstract: 'A debated issue in the physics of the BCS-BEC crossover with trapped Fermi atoms is to identify characteristic properties of the superfluid phase. Recently, a condensate fraction was measured on the BCS side of the crossover by sweeping the system in a fast (nonadiabatic) way from the BCS to the BEC sides, thus “projecting” the initial many-body state onto a molecular condensate. We analyze here the theoretical implications of these projection experiments, by identifying the appropriate quantum-mechanical operator associated with the measured quantities and relating them to the many-body correlations occurring in the BCS-BEC crossover. Calculations are presented over wide temperature and coupling ranges, by including pairing fluctuations on top of mean field.'
author:
- 'A. Perali, P. Pieri, and G.C. Strinati'
title: Extracting the condensate density from projection experiments with Fermi gases
---
The current experimental advances with trapped Fermi atoms have attracted much interest in the physics of the BCS-BEC crossover. In this context, one of the most debated issues is the unambiguous detection of superfluid properties on the BCS side of the crossover. Several attempts have been made in this direction. They include absorption images of the “projected” density profiles for $^{40}$K [@Jin-PRL-2004] and $^{6}$Li [@Ketterle-PRL-2004], rf spectroscopy to detect single-particle excitations [@Grimm-Science-2004], and measurements of collective modes [@Thomas-PRL-2004; @Grimm-PRL-2004]. In addition, a number of schemes to detect superfluid properties on the BCS side of the crossover have been proposed, including Josephson oscillations [@Josephson] and vortices [@vortices].
In particular, the experimental procedure of Refs. [@Jin-PRL-2004; @Ketterle-PRL-2004] pairwise “projects” fermionic atoms onto molecules, by preparing the system of trapped Fermi atoms on the BCS side with a tunable Fano-Feshbach resonance and then rapidly sweeping the magnetic field to the BEC side. In this way, the same two-component fit of density profiles routinely used for Bose gases is exploited to extract from these “projected” density profiles the analog of a condensate fraction, which is now associated with the equilibrium state on the BCS side before the sweep took place.
Purpose of this paper is to provide a theoretical interpretation of the experiments of Refs. [@Jin-PRL-2004; @Ketterle-PRL-2004], by obtaining the “projected” density profiles in terms of the correlation functions of the Fermi gas at equilibrium. This will be based on a number of physical assumptions which we associate with the experimental procedure of Refs. [@Jin-PRL-2004; @Ketterle-PRL-2004]. Our calculation evidences how the projection procedure amplifies the visibility of the emergence of the condensate as the temperature is lowered below the critical temperature $T_{c}$, when compared with the ordinary density profiles of Ref. [@PPPS-PRL-04]. We also attempt an analysis of the “projected” density profiles in terms of a two-component fit, in analogy to what is done with the experimental data [@Jin-PRL-2004; @Ketterle-PRL-2004]. A prediction is further made of a reduced molecular fraction that depends on the initial many-body state, in agreement with a late experimental evidence [@Ketterle-cond-mat-2004].
Inclusion of pairing fluctuations on top of mean field along the lines of Ref. [@PPS-PRB-04] enables us to cover a wide temperature range even in the intermediate- and strong-coupling regimes, in contrast to Refs. [@Ho-condensate-2004; @Bohn-condensate-2004] where only mean field was taken into account.
To account for the “projected” density profiles of Refs. [@Jin-PRL-2004; @Ketterle-PRL-2004], we consider the boson-like field operator: $$\Psi_{B}({\bf r}) = \int \! d\bm{ \rho} \, \phi(\rho)
\psi_{\downarrow}({\bf r}-\bm{ \rho}/2) \psi_{\uparrow}({\bf r}+
\bm{ \rho}/2)\; .
\label{definition-boson-fermion}$$ Here, $\psi_{\sigma}(\mathbf{r})$ is a fermion field operator with spin $\sigma$, and $\phi(\mathbf{\rho})$ a real and normalized function which specifies the probability amplitude for the fermion pair. An operator of the form (\[definition-boson-fermion\]) was considered in Ref. [@APS-2003] to obtain the condensate density for composite bosons.
Our theoretical analysis of the experiments of Refs. [@Jin-PRL-2004; @Ketterle-PRL-2004] is based on the following *physical assumptions*, that we infer from the experimental procedure:
\(i) Atoms of a specific spin state were detected, which originate from the dissociation of molecules after applying an rf pulse. The object of the measurement is thus the bosonic (molecular) density $n_{B}(\mathbf{r})$ at position $\mathbf{r}$ (and not the fermionic (atomic) density $n(\mathbf{r})$).
\(ii) A rather large conversion efficiency into molecules results when rapidly sweeping the magnetic field in the experiments. This suggests that molecules form just past the unitarity limit on the BEC side, where the “final” molecular wave function and the many-body correlations for the “initial”states considered in Refs. [@Jin-PRL-2004; @Ketterle-PRL-2004] have maximum overlap. Correspondingly, we assume that the wave function in the expression (\[definition-boson-fermion\]) refers to this “final” coupling, and represent it by $\phi_{f}(\mathbf{\rho})$. As molecules form in a medium, we take into account the effect of Pauli blocking in analogy with the original Cooper argument [@Cooper-1956] and identify $\phi_{f}(\rho)$ with the bound-state solution of the two-body problem, with the condition that its Fourier transform $\phi_{f}(\mathbf{k})$ vanishes when the magnitude of the wave vector $\mathbf{k}$ is smaller than a characteristic value $k_{\mu_{f}}=\sqrt{2 m \mu_{f}}$ for $\mu_{f} > 0$ ( $m$ being the fermion mass), while no constraint is enforced for $\mu_{f} < 0$. Here, the value of the chemical potential $\mu_{f}$ depends on the “final” coupling at which the molecular state is assumed to form, thus implying that some sort of local equilibrium can be established around a molecule. We shall present our calculations for two “final” couplings that bound the interval where the maximum overlap occurs, and are at the same time representative of the two cases where $\mu_{f}$ is positive or negative.
\(iii) In the experiments, bosonic Thomas-Fermi (TF) profiles for the molecular condensate were extracted from position-dependent density profiles, thus entailing an assumption of thermal equilibrium. We assume that this thermal equilibrium corresponds to the state prepared *before* the rapid sweep of the magnetic field. The validity of this assumption is supported by a recent experimental study of the formation time of a fermion-pair condensate[@Ketterle-cond-mat-2004]. In our calculations, we then use $\langle \cdots \rangle_{i}$ as expressions for the thermal averages, where the suffix $i$ stands for “initial”.
All these assumptions are summarized by stating that the *“projected” bosonic density profile* given by $$n_{B}^{fi}(\mathbf{r}) \, = \, \langle \Psi_{B}^{f}(\mathbf{r})^{\dagger} \, \Psi_{B}^{f}(\mathbf{r}) \rangle_{i}
\label{projected-bosonic-density}$$ represents the “in situ” molecular density which would be measured after the rapid sweep but before the cloud expansion performed in Refs. [@Jin-PRL-2004; @Ketterle-PRL-2004] (connection with the density measured after the expansion will be discussed below). In this expression, the boson-like field operator of Eq. (\[definition-boson-fermion\]) contains the *final* molecular wave function $\phi_{f}(\mathbf{\rho})$ on the BEC side of the crossover, while the thermal average $\langle \cdots \rangle_{i}$ is taken with reference to the state in which the system was *initially* prepared.
Consistently with our previous work [@SPS-2004], we describe the interaction term of the many-fermion Hamiltonian via an effective single-channel model. The parameter $(k_{F} a_{F})^{-1}$ then drives the crossover from the BCS side (identified by $(k_{F} a_{F})^{-1} \lesssim -1$) to the BEC side (identified by $1 \lesssim (k_F a_F)^{-1}$) across the unitarity limit $(k_{F} a_{F})^{-1} = 0$. Here, $a_{F}$ is the two-fermion scattering length and the Fermi wave vector $k_{F}$ results by setting $k_{F}^{2}/(2m)$ equal to the noninteracting Fermi energy.
The calculation proceeds by expressing the four-fermion field operator in Eq. (\[projected-bosonic-density\]) in terms of the two-particle Green’s function $\mathcal{G}_{2}(1,2,1',2') \, = \, \langle T_{\tau}[\Psi(1) \, \Psi(2) \, \Psi^{\dagger}(2') \, \Psi^{\dagger}(1')] \rangle$, where $T_{\tau}$ is the imaginary-time ordering operator. We have introduced the spinor $\Psi(\mathbf{r}) = (\psi_{\uparrow}(\mathbf{r}),\psi_{\downarrow}^{\dagger}(\mathbf{r}))$ as well as the short-hand notation $1=(\mathbf{r}_{1}, \tau_{1}, \ell_{1})$ with imaginary time $\tau$ and spinor component $\ell$, such that $\Psi(1) = \exp\{K \tau_{1}\} \Psi_{\ell_{1}} (\mathbf{r_{1}}) \exp\{- K \tau_{1}\}$. The thermal average contains the grand-canonical Hamiltonian $K = H - \mu N$ with fermionic chemical potential $\mu$ and is taken in the initial state, as specified above.
The two-particle Green’s function $\mathcal{G}_{2}$ is, in turn, expressed in terms of the many-particle T-matrix, by solving formally the Bethe-Salpeter equation as follows: $$\begin{aligned}
& &\mathcal{G}_{2}(1,2,1',2') =\mathcal{G}(1,1') \, \mathcal{G}(2,2') \, - \, \mathcal{G}(1,2') \, \mathcal{G}(2,1')\phantom{1111} \nonumber\\
& & - \int \! d3456 \,\, \mathcal{G}(1,3) \, \mathcal{G}(6,1') \, T(3,5;6,4) \, \mathcal{G}(4,2') \, \mathcal{G}(2,5)\phantom{1111} \label{G-T} \end{aligned}$$ where $\mathcal{G}(1,1') \, = \, - \, \langle T_{\tau}[\Psi(1) \, \Psi^{\dagger}(1')] \rangle$ is the fermionic single-particle Green’s function. Accordingly, the “projected” bosonic density (\[projected-bosonic-density\]) reads: $$n_{B}^{fi}(\mathbf{r}) \, = \, \int \! d\bm{\rho} \, d\bm{\rho}' \,\, \phi_{f}(\mathbf{\rho}) \,\, \phi_{f}(\mathbf{\rho'}) \,\,
\mathcal{G}_{2}^{i}(1,2,1',2') \label{pbd-Green-function}$$ where $1 = (\mathbf{r}-\bm{\rho}/2,\tau + 2 \eta,\ell = 2)$, $ 2 = (\mathbf{r}+\bm{\rho}'/2,\tau,\ell = 1)$, $1' = (\mathbf{r}+\bm{\rho}/2,\tau + 3 \eta,\ell = 1)$, and $2' = (\mathbf{r}-\bm{\rho}'/2,\tau + \eta,\ell = 2)$ ($\eta$ being a positive infinitesimal).
Implementation of the above expressions to the trapped case is readily obtained via a local-density approximation, whereby a local gap parameter $\Delta(\mathbf{r})$ is introduced and the chemical potential $\mu$ is replaced (whenever it occurs for both “initial” and “final” couplings) by the quantity $\mu(\mathbf{r}) = \mu - V(\mathbf{r})$ that accounts for the trapping potential $V(\mathbf{r})$.
The three terms on the right-hand side of Eq. (\[G-T\]) correspond to physically different contributions to the expression (\[pbd-Green-function\]). In particular, the first term can be written as $|\alpha_{fi}|^{2}$, where $\alpha_{fi} = \int \! d\bm{\rho} \, \phi_{f}(\mathbf{\rho}) \mathcal{G}_{12}^{(i)}(\mathbf{\rho},\tau = - \eta)$ represents the *overlap* between the fermionic correlations (embodied in the anomalous single-particle Green’s function $\mathcal{G}_{12}^{(i)}$) and the molecular wave function $\phi_{f}$. This contribution vanishes with the gap parameter $\Delta$ when approaching $T_{c}$, and is identified with the *condensate density* for composite bosons when both the “initial” thermal equilibrium and the “final” molecular wave function are taken at the same coupling deep in the BEC region [@APS-2003]. Only this contribution was considered in Ref. [@Bohn-condensate-2004] in connection with the experiments of Refs. [@Jin-PRL-2004; @Ketterle-PRL-2004].
The second term on the right-hand side of Eq. (\[G-T\]) represents fermionic correlations in the normal state, that are relevant in the presence of an underlying Fermi surface. This term is most sensitive to the “final” molecular wave function $\phi_{f}$ being affected by Pauli blocking when $\mu_{f}$ is positive. This term would be irrelevant if the “initial” thermal equilibrium and the “final” molecular wave function were taken deep in the BEC side. When $\mu_{f}$ is positive, this term can lead to an overestimate of the value of the condensate fraction when the “projected” density profiles are fitted in terms of TF and Gaussian functions, as argued below. Both this and the previous contribution were considered in Ref. [@Ho-condensate-2004] (where the “final” coupling was, however, taken deep in the BEC region).
The third term on the right-hand side of Eq. (\[G-T\]) will be calculated in the following within the off-diagonal BCS-RPA approximation considered in Ref. [@APS-2003]. This contribution is identified with the *noncondensate density* for composite bosons when both the “initial” thermal equilibrium and the “final” molecular wave function are taken at the same coupling deep in the BEC side [@APS-2003]. It is thus of particular importance for increasing temperature when approaching the normal phase.
When $i=f$ deep in the BEC region, the “projected” density profile $n_{B}^{ii}(\mathbf{r})$ coincides with (half) the ordinary density profile $n(\mathbf{r})$ calculated at the same coupling. In the following, the values for the local chemical potential and gap parameter, to be inserted in the expression (\[pbd-Green-function\]) for $n_{B}^{fi}(\mathbf{r})$, are obtained with the theory of Ref. [@PPPS-PRL-04] where pairing fluctuations are included on top of mean field.
Figure 1 shows the axially-integrated “projected” density profiles calculated for the coupling values $(k_{F} a_{F})_{i}^{-1} = (-0.50,-0.25,0.00)$ and for the two representative values $0.40$ (upper panel) and $1.50$ (lower panel) of the “final” coupling $(k_{F} a_{F})_{f}^{-1}$. Two characteristic temperatures (just above the critical temperature and near zero temperature) are considered in each case. Note the marked temperature dependence of the “projected” density profiles when entering the superfluid phase, as signaled by the emergence of a “condensate” component near the center of the trap. This contrasts the milder dependence (especially on the BCS side) of the density profiles without projection [@PPPS-PRL-04]. The “projection” technique introduced in Ref. [@Jin-PRL-2004] is thus demonstrated to *amplify* the effects due to the presence of a condensate on the density profiles, which would otherwise be almost temperature independent on the BCS side of the crossover.
In Fig. 1 the densities are normalized to half the total number $N_{F}$ of fermionic atoms. This number differs, in general, from the total number $N_{mol}$ of molecules obtained by integrating the “projected” density profiles. In particular, $N_{mol}$ can vary significantly when scanning the “initial” coupling $(k_{F} a_{F})_i^{-1}$ on the BCS side of the crossover for given “final” molecular-like state. This effect is shown in Fig. 2 for the same temperatures and “final” couplings of Fig. 1. Our finding that the total number $N_{mol}$ of molecules constitutes only a fraction of the original atom number $N_{F}/2$ for each spin state is supported by the experimental results of Refs. [@Jin-PRL-2004] and [@Ketterle-PRL-2004]. In addition, our prediction that the reduced value of the molecular fraction depends on the “initial” many-body state is in agreement with the experimental evidence recently reported in Ref. [@Ketterle-cond-mat-2004]. Note that, for both values of the “final coupling”, the total number of molecules increases upon lowering the temperature. This result indicates that the conversion efficiency for the condensate fraction is larger than for the thermal component, as also observed in Ref. [@Ketterle-cond-mat-2004].
In our procedure, the [*condensate*]{} and [*noncondensate*]{} components of the “projected” density profiles are calculated separately. By our definition, they correspond to the first term and to the remaining terms on the right-hand side of Eq. (\[G-T\]), respectively. The *condensate fraction* is obtained accordingly from the ratio of the corresponding areas. Yet, the total “projected” density profiles obtained theoretically could be analyzed in terms of a two-component fit with a TF plus a Gaussian function (or, better, a $\mathrm{g}_{3/2}$ function for the Bose gas), in analogy to a standard experimental procedure. This kind of analysis is reported in Fig. 3 for the two low-temperature curves shown on the right panels of Fig. 1. In both cases, a good overall fit is obtained by the $\chi$-square method. Separate comparison is also made in the figure with the theoretical condensate and noncondensate components defined above, which appears rather good for the value $1.50$ of the “final” coupling while for the value $0.40$ an overestimate (of about $50 \%$) of the condensate results from the fit. This discrepancy stems mostly from the second term on the right-hand side of Eq. (\[G-T\]), which contributes to the TF component of the fit owing to a peculiar shape of the corresponding “projected” density profile.
Extention of this analysis to “initial” couplings on the BCS side reveals unconventional forms of the theoretical condensed and noncondensed contributions to the “projected” density profiles, so that the above two-component fit fails. For negative values of the initial coupling when the two-component fit fails, we have verified that the difference $N_{mol}(T=0) - N_{mol}(T=1.10 T_c)$ approximately coincides with $N_0(T=0)$ within a relative error not larger than $10 \%$ when $(k_F a_F)^{-1}_f=0.4$. This suggests a practical prescription to extract $N_0(T=0)$ from the values of $N_{mol}$ at low temperature and slightly above the critical temperature, without relying on a two-component fit.
In this respect, recall that our theoretical “projected” density profiles are calculated when molecules just form on the BEC side near the unitarity limit. In the experiments, however, a further ramp of the magnetic field is performed together with a subsequent cloud expansion. Only at this stage the profile of the cloud is detected. Comparison between theory and experiments is thus meaningful since the further ramp of the magnetic field and the subsequent cloud expansion are expected to have no influence on the values of $N_{mol}$ and $N_{0}/N_{mol}$. This is because, by the further ramp, the molecules shrink following the field adiabatically. They then become tightly bound and weakly interacting among themselves, while their local counting is unaffected. Under these conditions, the condensate fraction, too, should not be modified by the subsequent expansion as it is the case for an ordinary dilute Bose gas, even though the expansion may affect the details of the density profiles.
In Fig. 4 the condensate fraction $N_{0}/N_{mol}$, obtained from our theoretical expressions at the lowest temperature $T/T_{c}=0.05$, is plotted vs $(k_{F} a_{F})^{-1}_{i}$ on the BCS side of the crossover. The data from Refs. [@Jin-PRL-2004] and [@Ketterle-PRL-2004] are also reported in the figure. The agreement between the overall trends of the theoretical and experimental curves appears satisfactory, although quantitative discrepancies result between the two sets of curves. They might be due to an overestimate of the TF component of the fits in Ref. [@Ketterle-PRL-2004] for the reasons discussed in Ref. [@Ketterle-cond-mat-2004], and to a possible underestimate of the condensate component in Ref. [@Jin-PRL-2004] due to a preferential loss of molecules in the condensate itself. We have, finally, verified that a linear dependence occurs between $N_{mol}/(N_{F}/2)$ and $N_{0}/N_{mol}$ (inset of Fig. 4). A similar linear dependence is also reported in Fig. 4(b) of Ref. [@Ketterle-PRL-2004], although with a different definition of $N_{mol}$ that includes also molecular states not directly detected in the experiment.
We are indebted to C. Salomon and H. Stoof for discussions, and to D. Jin for a critical reading of the manuscript. This work was partially supported by the Italian MIUR with contract Cofin-2003 “Complex Systems and Many-Body Problems”.
[99]{}
C.A. Regal, M. Greiner, and D.S. Jin, Phys. Rev. Lett. [**92**]{}, 040403 (2004).
M.W. Zwierlein [*et al.*]{}, Phys. Rev. Lett. [**92**]{}, 120403 (2004).
C. Chin [*et al.*]{}, Science [**305**]{}, 1128 (2004).
J. Kinast [*et al.*]{}, Phys. Rev. Lett. [**92**]{}, 150402 (2004).
M. Bartenstein [*et al.*]{}, Phys. Rev. Lett. [**92**]{}, 203201 (2004).
F. S. Cataliotti [*et al.*]{}, Science [**293**]{}, 843 (2001); G. Orso, L.P. Pitaevskii, and S. Stringari, Phys. Rev. Lett. [**93**]{}, 020404 (2004); M. Wouters, J. Tempere, and J. T. Devreese, Phys. Rev. A [**70**]{}, 013616 (2004). M. Greiner, C.A. Regal, and D.S. Jin, cond-mat/0502539.
A. Perali, P. Pieri, L. Pisani, and G.C. Strinati, Phys. Rev. Lett. [**92**]{}, 220404 (2004). M.W. Zwierlein, [*et al.*]{}, cond-mat/0412675.
P. Pieri, L. Pisani, and G.C. Strinati, Phys. Rev. B [**70**]{}, 094508 (2004).
R.B. Diener and T.L. Ho, cond-mat/0404517.
A.V. Avdeenkov and J.L. Bohn, Phys. Rev. A [**71**]{}, 023609 (2005).
N. Andrenacci, P. Pieri, and G.C. Strinati, Phys. Rev. B [**68**]{}, 144507 (2003).
L.N. Cooper, Phys. Rev. [**104**]{}, 1189 (1956).
S. Simonucci, P. Pieri, and G.C. Strinati, Europhys. Lett. [**69**]{}, 713 (2005).
| {
"pile_set_name": "ArXiv"
} | ArXiv |
---
abstract: 'We introduce an index for symmetry protected topological (SPT) phases of infinite fermionic chains with an on-site symmetry given by a finite group $G$. This index takes values in $\mathbb{Z}_2 \times H^1(G,\mathbb{Z}_2) \times H^2(G, U(1)_{\mathfrak{p}})$ with a generalized Wall group law under stacking. We show that this index is an invariant of the classification of SPT phases. When the ground state is translation invariant and has reduced density matrices with uniformly bounded rank on finite intervals, we derive a fermionic matrix product representative of this state with on-site symmetry.'
address:
- 'WPI-AIMR, Tohoku University, 2-1-1 Katahira, Aoba-ku, Sendai, 980-8577 *and* RIKEN iTHEMS, Wako, Saitama 351-0198, Japan'
- 'Graduate School of Mathematical Sciences, The University of Tokyo, Komaba, Tokyo, 153-8914, Japan'
author:
- 'C. Bourne'
- 'Y. Ogata'
title: 'The classification of symmetry protected topological phases of one-dimensional fermion systems'
---
Introduction {#intro}
============
The notion of symmetry protected topological (SPT) phases was introduced by Gu and Wen [@GuWen2009]. We consider the set of all Hamiltonians with a prescribed symmetry and which have a unique gapped ground state in the bulk. Two Hamiltonians in this set are equivalent if there is a smooth path within the set connecting them. By this equivalence relation, we may classify the Hamiltonians in this family. The equivalence class of a Hamiltonian with only on-site interactions is regarded as a trivial phase. If a phase is non-trivial, it is called a SPT phase.
A basic question is how to show that a given Hamiltonian belongs to a SPT phase. A mathematically natural approach for this problem is to define an invariant of the classification. This approach has been studied in the physics literature using matrix product states (MPS) [@po; @po2; @GuWen2009; @ChenGuWEn2011; @Perez-Garcia2008]. MPS is a powerful framework introduced in [@Fannes:1992vq], after the discovery of the famous AKLT model [@Affleck:1988vr]. Hastings showed that MPS approximates unique gapped ground states of quantum spin chains well [@area]. However, an approximation of gapped ground states by MPS is insufficient to analyze the complete classification problem and is furthermore restricted to [*translationally invariant*]{} systems. If the index is not defined everywhere, there is no way to discuss if it is actually an invariant or not.
In [@OgataTRI; @OgataRF; @OgataSPT], an index for SPT phases with on-site finite group symmetry and global reflection symmetry was defined for infinite quantum spin chains in fully general setting. In these papers, it was proven that the index is actually an invariant of the classification of SPT phases. An important observation for stability of the index is the factorization property of the automorphic equivalence. The key ingredient for the definition of the index is the split property of unique gapped ground states, proven by Matsui [@Matsui13]. The index introduced in [@OgataTRI; @OgataRF] generalizes the indices introduced for MPS in [@po; @po2; @GuWen2009; @ChenGuWEn2011; @Perez-Garcia2008], where a MPS emerges naturally from translation invariant split states with a reduced density matrices whose rank is uniformly bounded on finite intervals [@BJKW; @Matsui13]. In this paper, we are interested in the analogous problem for fermionic chains with on-site finite group symmetries. Fermionic SPT phases for finite systems in one dimension have already been extensively studied in the physics literature [@fk; @FK2; @BWHV; @KT; @KapustinfMPS; @TurzilloYou]. In contrast to quantum spin chains, for parity-symmetric gapped ground states without additional symmetries, there are two distinct phases. A ${{\mathbb Z}}_2$-index to distinguish these phases in infinite systems was introduced in [@BSB] and independently in [@Matsui20]. It was outlined in [@BSB] that this $\mathbb{Z}_2$-index is an invariant of the classification of unique parity-invariant gapped ground state phases using techniques from [@OgataTRI] and [@NSY18]. The aim of this paper is to extend the analysis of fermionic gapped ground states to the case with an on-site symmetry. Namely, a classification of one-dimensional fermionic SPT phases.
Setting and outline {#setting-and-outline .unnumbered}
-------------------
We assume the reader has some familiarity with the basics of operator algebras and their application to quantum statistical mechanical systems, see [@BR1; @BR2]. Throughout this paper, we fix $d\in{{\mathbb N}}$. Let ${\mathfrak h}:=l^2({{\mathbb Z}})\otimes {{\mathbb C}}^d$ and ${{\mathcal A}}$ be the CAR-algebra over ${\mathfrak h}$, i.e. the universal $C^*$-algebra generated by the identity and $\{a(f)\}_{f\in{\mathfrak h}}$ such that $f\mapsto a(f)$ is anti-linear and $$\begin{aligned}
&\{a(f_1),a(f_2)\} = 0, &&\{a(f_1),a(f_2)^*\} = \langle f_1,f_2\rangle.\end{aligned}$$ For each subset $X$ of ${{\mathbb Z}}$, we set ${\mathfrak h}_X:=l^2(X)\otimes{{\mathbb C}}^d$, and denote by ${{\mathcal A}}_X$ the CAR-algebra over ${\mathfrak h}_X$. We naturally regard ${{\mathcal A}}_X$ as a subalgebra of ${{\mathcal A}}$. We also use the notation ${{\mathcal A}}_R:={{\mathcal A}}_{{{\mathbb Z}}_{\ge 0}}$ and ${{\mathcal A}}_L:={{\mathcal A}}_{{{\mathbb Z}}_{<0}}$. We denote the set of all finite subsets in ${{{\mathbb Z}}}$ by ${\mathfrak S}_{{{\mathbb Z}}}$ and set ${{\mathcal A}}_{\rm loc}:=\bigcup_{X\in {\mathfrak S}_{{{\mathbb Z}}}}{{\mathcal A}}_X$. Given a Hilbert space $\mathfrak{K}$, the fermionic Fock space of anti-symmetric tensors is denoted by $\calF(\mathfrak K)$. For a unitary/anti-unitary operator $U$ on ${{\mathbb C}}^d$, we denote the second quantization of $U$ on the Fock space $\calF({\ensuremath{\mathbb{C}}}^d)$ by $\mathfrak{\Gamma}(U)$.
By the universality of the CAR-algebra, for any unitary/anti-unitary $v$ on ${\mathfrak h}$, we may define a linear/anti-linear automorphism $\beta_v$ on ${{\mathcal A}}$ such that $\beta_v(a(f))=a(vf)$, $f\in{\mathfrak h}$. In particular, for $v=-{\mathbb I}$, we obtain the parity operator $\Theta:=\beta_{-{\mathbb I}}$. For each $X\in {\mathfrak S}_{{{\mathbb Z}}}$, ${{\mathcal A}}_X$ is $\Theta$-invariant. We denote the restriction $\Theta\vert_{{{\mathcal A}}_X}$ by $\Theta_X$. For $\sigma=0,1$, set of elements $A$ in ${{\mathcal A}}$ with $\Theta(A)=(-1)^\sigma A$ is denoted by ${{\mathcal A}}^{(\sigma)}$. Elements in ${{\mathcal A}}^{(0)}$ are said to be even and elements in ${{\mathcal A}}^{(1)}$ are said to be odd.
In this paper, we consider an on-site symmetry given by a finite group $G$. We let ${\mathop{\mathrm{M}}\nolimits}_d$ denote the algebra of $d\times d$ matrices with complex entries and consider a projective unitary/anti-unitary representation of $G$ on ${{\mathbb C}}^d$ relative to a group homomorphism ${\mathfrak{p}}:G\to{{\mathbb Z}}_2$.[^1] That is, there is a projective representation $U$ of $G$ on ${{\mathbb C}}^d$ such that $U_g$ is unitary if ${\mathfrak{p}}(g)=0$ and anti-unitary if ${{\mathfrak{p}}(g)}=1$. Because $U$ is projective, there is a $2$-cocycle $\upsilon:G\times G\to U(1)$ such that $U_g U_h = \upsilon(g,h)U_{gh}$ and for all $f,g,h\in G$ $$\begin{aligned}
&\upsilon(e,g)=1=\upsilon(g,e),
&&\frac{{\overline}{\upsilon(g,h)}^{{{\mathfrak p}}(f)} \upsilon(f,gh)}{\upsilon(f,g)\upsilon(fg,h)} = 1,\end{aligned}$$ where ${\overline}{z}^{{{\mathfrak p}}(f)} = z$ if ${{\mathfrak p}}(f)=0$ and ${\overline}{z}^{{{\mathfrak p}}(f)} = {\overline}{z}$ if ${{\mathfrak p}}(f)=1$. For a fixed homomorphism ${\mathfrak{p}}$, equivalence classes of such $2$-cocycles give rise to the cohomology group $H^2(G, U(1)_{\mathfrak{p}})$.
For a fixed projective unitary/anti-unitary representation $U$ of $G$ on ${{\mathbb C}}^d$ relative to ${\mathfrak{p}}:G\to {{\mathbb Z}}_2$, we can extend this representation to an on-site representation $\bigoplus_\mathbb{Z}U$ on $l^2(\mathbb{Z}) \otimes {{\mathbb C}}^d$. We therefore can define the linear/anti-linear automorphism $\alpha$ on ${{\mathcal A}}$, where $$\begin{aligned}
\alpha_g:=\beta_{{\left (}\bigoplus_{{{\mathbb Z}}} U_g {\right )}},\quad g\in G.\end{aligned}$$ If ${{\mathfrak{p}}(g)}=0$, then $\alpha_g$ is an automorphism on ${{\mathcal A}}$ and if ${{\mathfrak{p}}(g)}=1$, then $\alpha_g$ is an anti-linear automorphism on ${{\mathcal A}}$. Note that $\alpha$ satisfies $$\begin{aligned}
\alpha_g\circ\Theta=\Theta\circ\alpha_g, \qquad
\alpha_g({{\mathcal A}}_X)={{\mathcal A}}_X, \quad g\in G, \quad X\in {\mathfrak S}_{{{\mathbb Z}}}.\end{aligned}$$ For each $g\in G$ and a state $\varphi$ on ${{\mathcal A}}$, we define a state $\varphi_g$ by $\varphi_g(A)=\varphi\circ\alpha_g(A)$, $A\in{{\mathcal A}}$ if ${{\mathfrak{p}}(g)}=0$, and by $\varphi_g(A)=\varphi\circ\alpha_g(A^*)$, $A\in{{\mathcal A}}$ if ${{\mathfrak{p}}(g)}=1$. We say $\varphi$ is $\alpha$-invariant if $\varphi_g=\varphi$ for any $g\in G$.
In the latter half of the paper we also consider space translations $\beta_{S_x}$, $x\in{{\mathbb Z}}$. Here the unitary $S_x$ is given by $S_x=s_x\otimes {\mathbb I}_{{{\mathbb C}}^d}$ with $s_x$ the shift by $x$ on $l^2({{\mathbb Z}})$.
Throughout this paper, for a state $\varphi$ on ${{\mathcal A}}_{X}$ (with $X$ a subset of ${{\mathbb Z}}$), $({{\mathcal H}}_\varphi,\pi_\varphi,\Omega_\varphi)$ denotes a GNS triple of $\varphi$. When $\varphi$ is $\Theta_{X}$-invariant, then $\hat \Gamma_\varphi$ denotes the self-adjoint unitary on ${{\mathcal H}}_\varphi$ defined by $\hat\Gamma_\varphi\pi_\varphi(A)\Omega_\varphi=\pi_\varphi\circ\Theta_X(A)\Omega_\varphi$ for $A\in{{\mathcal A}}_X$. If $\varphi$ is $\alpha$-invariant, then we denote by $\hat\alpha_\varphi$ the extension of $\alpha\vert_{{{\mathcal A}}_X}$ to $\pi_{\varphi}({{\mathcal A}}_X)''$.
The mathematical model of a one-dimensional fermionic system is fully specified by the interaction $\Phi$. An interaction is a map $\Phi$ from ${\mathfrak S}_{{{\mathbb Z}}}$ into ${{{\mathcal A}}}_{\rm loc}$ such that $\Phi(X) \in {{{\mathcal A}}}_{X}$ and $\Phi(X) = \Phi(X)^*$ for $X \in {\mathfrak S}_{{{\mathbb Z}}}$. When we have $\Theta(\Phi(X))=\Phi(X)$ for all $X\in {\mathfrak S}_{{{\mathbb Z}}}$, $\Phi$ is said to be even. We say $\Phi$ is $\alpha$-invariant if we have $\alpha_g(\Phi(X))=\Phi(X)$ for all $X\in {\mathfrak S}_{{{\mathbb Z}}}$ and $g\in G$. An interaction $\Phi$ is translation invariant if $\Phi(X+x)=\beta_{S_x}(\Phi(X))$, for all $x\in{\mathbb Z}$ and $X\in {\mathfrak S}_{{{\mathbb Z}}}$. Furthermore, an interaction $\Phi$ is finite range if there exists an $m\in {\mathbb N}$ such that $\Phi(X)=0$ for any $X$ with diameter larger than $m$. We denote by ${{\mathcal B}}_{f}^{e}$, the set of all finite range even interactions $\Phi$ which satisfy $$\begin{aligned}
\label{fi}
\sup_{X\in {\mathfrak S}_{{{\mathbb Z}}}}
{\left \Vert}\Phi{\left (}X{\right )}{\right \Vert}<\infty.\end{aligned}$$
For an interaction $\Phi$ and a finite set $\Lambda\in{\mathfrak S}_{{{\mathbb Z}}}$, we define the local Hamiltonian on $\Lambda$ by $$\label{GenHamiltonian}
H_{\Lambda,\Phi}:=\sum_{X\subset{\Lambda}}\Phi(X).$$ The dynamics given by this local Hamiltonian is denoted by $$\begin{aligned}
\label{taulamdef}
\tau_t^{\Phi,\Lambda}{\left (}A{\right )}:= e^{itH_{\Lambda,\Phi}} Ae^{-itH_{\Lambda,\Phi}},\quad
t\in{{\mathbb R}},\quad A\in{{\mathcal A}}.\end{aligned}$$ If $\Phi$ belongs to ${{\mathcal B}}_{f}^e$, the limit $$\begin{aligned}
\label{taudef}
\tau_t^{\Phi}{\left (}A{\right )}=\lim_{\Lambda\to{{\mathbb Z}}}
\tau_t^{\Phi,\Lambda}{\left (}A{\right )}\end{aligned}$$ exists for each $A\in {{\mathcal A}}$ and $t\in{\mathbb R}$, and defines a strongly continuous one parameter group of automorphisms $\tau^\Phi$ on ${{\mathcal A}}$ (see Appendix \[flr\]). We denote the generator of $\tau^{\Phi}$ by $\delta_{\Phi}$.
For $\Phi\in{{\mathcal B}}_f^{e}$, a state $\varphi$ on ${{\mathcal A}}$ is called a state if the inequality $
-i\,\varphi(A^*{\delta_{\Phi}}(A))\ge 0
$ holds for any element $A$ in the domain ${{\mathcal D}}({\delta_{\Phi}})$ of ${\delta_\Phi}$. If $\varphi$ is a $\tau^\Phi$-ground state with GNS triple $({{\mathcal H}}_\varphi,\pi_\varphi,\Omega_\varphi)$, then there exists a unique positive operator $H_{\varphi,\Phi}$ on ${{\mathcal H}}_\varphi$ such that $e^{itH_{\varphi,\Phi}}\pi_\varphi(A)\Omega_\varphi=\pi_\varphi(\tau_t^\Phi(A))\Omega_\varphi$, for all $A\in{{\mathcal A}}$ and $t\in\mathbb R$. We call this $H_{\varphi,\Phi}$ the bulk Hamiltonian associated with $\varphi$. Note that $\Omega_\varphi$ is an eigenvector of $H_{\varphi,\Phi}$ with eigenvalue $0$. The following definition clarifies what we mean by a model with a unique gapped ground state.
We say that a model with an interaction $\Phi\in{{\mathcal B}}_f^{e}$ has a unique gapped ground state if (i) the $\tau^\Phi$-ground state, which we denote as $\varphi$, is unique, and (ii) there exists a $\gamma>0$ such that $\sigma(H_{\varphi,\Phi})\setminus\{0\}\subset [\gamma,\infty)$, where $\sigma(H_{\varphi,\Phi})$ is the spectrum of $H_{\varphi,\Phi}$.
Note that the uniqueness of $\varphi$ implies that 0 is a non-degenerate eigenvalue of $H_{\varphi,\Phi}$.
If $\varphi$ is a state of $\alpha$-invariant and $\Theta$-invariant interaction $\Phi\in {{{\mathcal B}}}_f^e$, then $\varphi \circ \Theta$ and $\varphi_g$ is also a state for each $g\in G$. In particular, if $\varphi$ is a unique state, it is pure, $\Theta$-invariant and $\alpha$-invariant. We denote by ${{\mathcal G}}_{f}^{e,\alpha}$ the set of all $\alpha$-invariant interactions $\Phi\in{{\mathcal B}}_f^{e}$ with a unique gapped ground state.
Now the classification problem of SPT phases is the classification of ${{\mathcal G}}_{f}^{e,\alpha}$ with respect to the following equivalence relation: $\Phi_0, \Phi_1\in {{\mathcal G}}_{f}^{e,\alpha}$ are equivalent if there is a smooth path in ${{\mathcal G}}_{f}^{e,\alpha}$ connecting them. (See Section \[stability\] for a more precise definition.)
We now outline the main results of the paper. In Section \[indsec\], we introduce an index for $\Theta$-invariant and $\alpha$-invariant gapped ground states in a fully general setting. This index takes value in ${{\mathbb Z}}_2\times H^1(G,{{\mathbb Z}}_2)\times H^2(G, U(1)_{{\mathfrak{p}}})$, which is analogous to the indices introduced in [@KT] in the context of spin-TQFT and [@BWHV; @KapustinfMPS; @TurzilloYou] for the fermionic MPS setting. When $G$ is trivial, the index is ${{\mathbb Z}}_2$-valued and recovers the index studied in [@BSB; @Matsui20]. The key ingredient for the definition is again the split property of unique gapped ground states for fermionic systems proven recently in [@Matsui20]. In Section \[stability\], we show that our defined index is an invariant of the classification, i.e., it is stable along the smooth path in ${{\mathcal G}}_{f}^{e,\alpha}$.
Because our index takes values in a group, it suggests that one may compose fermionic SPT phases to obtain a new phase with index determined from the original systems. In the physics literature, this is achieved by stacking of systems, see [@FK2; @TurzilloYou] for example. Mathematically this operation corresponds to a (graded) tensor product of ground states. In Section \[stacksec\], we show that our index is indeed closed under this tensor product operation. However, despite the notation, the group operation on ${{\mathbb Z}}_2\times H^1(G,{{\mathbb Z}}_2)\times H^2(G, U(1)_{{\mathfrak{p}}})$ is *not* the direct sum, but rather a twisted product that follows a generalized Wall group law, cf. [@Wall]. As an example, we consider the case of an anti-linear ${{\mathbb Z}}_2$-action (e.g. an on-site time-reversal symmetry) and show that our index takes values in ${{\mathbb Z}}_8$. This recovers the ${{\mathbb Z}}_8$-classification of time-reversal symmetric one-dimensional fermionic SPT phases noted in [@fk; @FK2] and extends this classification to infinite systems. In Sections \[transsec\] and \[fmpssec\], we consider the unique ground state of a translation invariant $\Phi\in {{\mathcal G}}_{f}^{e,\alpha}$. For quantum spin systems, it is known that a representation of Cuntz algebra emerges from translation invariant pure split states [@BJKW; @Matsui3]. The generators of this Cuntz algebra representation give an operator product representation of the state and also implement the space translation. We find an analogous object for fermionic systems in Section \[transsec\]. Because odd elements with disjoint support anti-commute in the CAR-algebra, the operator product representation and space translation is more complicated than the quantum spin chain setting. The results of Section \[transsec\] are then applied to the study of fermionic MPS in Section \[fmpssec\]. When the rank of the reduced density matrices of the infinite volume ground state is uniformly bounded, we show that the ground state has a presentation as a fermionic MPS with on-site symmetry.
Basic properties of graded von Neumann algebras are reviewed in Appendix \[sec:GradedvN\]. In Appendix \[flr\], we adapt the Lieb–Robinson bound to the setting of lattice fermions (see also [@BSP; @NSY17]).
The index of fermionic SPT phases {#indsec}
=================================
Graded von Neumann algebras and dynamical systems {#Subsec:graded_prelims}
-------------------------------------------------
In order to introduce the index, we first need to introduce type I central balanced graded $W^*$-$(G,{\mathfrak{p}})$-dynamical systems. Further details on graded von Neumann algebras can be found in Appendix \[sec:GradedvN\].
A graded von Neumann algebra is a pair $({{\mathcal M}},\theta)$ with ${{\mathcal M}}$ a von Neumann algebra $\theta$ an involutive automorphism on ${{\mathcal M}}$, $\theta^2 = \mathrm{Id}$. If ${{\mathcal M}}\subset {{\mathcal B}}({{\mathcal H}})$ and there is a self-adjoint unitary $\Gamma$ on ${{\mathcal H}}$ such that $\mathrm{Ad}_{\Gamma}|_{{\mathcal M}}= \theta$, then we call $({{\mathcal M}},\theta)$ a spatially graded von Neumann algebra acting with grading operator $\Gamma$. If $\theta$ is the identity automorphism, then we say that $({{\mathcal M}}, \theta)$ is trivially graded.
We say that a graded von Neumann algebra $({{\mathcal M}},\theta)$ is of type $\lambda$, $\lambda \in \{\mathrm{I, II, III}\}$, if ${{\mathcal M}}$ is type $\lambda$.
Given a graded von Neumann algebra $({{\mathcal M}},\theta)$, ${{\mathcal M}}$ is a direct sum of two self-adjoint $\sigma\mathrm{\hbox{-}weak}$-closed linear subspaces as ${{\mathcal M}}={{\mathcal M}}^{(0)}\oplus {{\mathcal M}}^{(1)}$, where $$\begin{aligned}
{{\mathcal M}}^{(\sigma)}:=
\left\{
x\in{{\mathcal M}}\mid
\theta (x)=(-1)^\sigma x
\right\},\quad x\in{{\mathcal M}}, \,\, \sigma \in \{0,1\}.\end{aligned}$$ An element of ${{\mathcal M}}^{(\sigma)}$ is said to be homogeneous of degree $\sigma$, or even/odd for $\sigma=0$/$\sigma=1$, respectively. For a homogeneous $x\in{{\mathcal M}}$, its degree is denoted by $\partial x$. For graded von Neumann algebras $({{\mathcal M}}_1,\theta_1)$, $({{\mathcal M}}_2,\theta_2)$, a homomorphism $\phi: {{\mathcal M}}_1\to{{\mathcal M}}_2$ is a graded homomorphism if $\phi\big({{\mathcal M}}_1^{(\sigma)}\big) \subset {{\mathcal M}}_2^{(\sigma)}$ for $\sigma=0,1$.
Let $({{\mathcal M}}, \theta)$ be a graded von Neumann algebra. We say $({{\mathcal M}},\theta)$ is balanced if ${{\mathcal M}}$ contains an odd self-adjoint unitary. If $Z({{\mathcal M}})\cap {{\mathcal M}}^{(0)}={{\mathbb C}}{\mathbb I}$ for the center $Z({{\mathcal M}})$ of ${{\mathcal M}}$, we say $({{\mathcal M}}, \theta)$ is central.
We now consider dynamics on graded von Neumann algebras via a linear/anti-linear group action.
Let $G$ be a finite group and ${\mathfrak{p}}: G\to{{\mathbb Z}}_2$ be a group homomorphism. A graded $W^*$-$(G,{\mathfrak{p}})$-dynamical system $({{\mathcal M}},\theta,{\hat\alpha})$ is a graded von Neumann algebra $({{\mathcal M}}, \theta)$ with an action ${\hat\alpha}$ of $G$ on ${{\mathcal M}}$ such that ${\hat\alpha}_g$ is a linear automorphism if ${{\mathfrak{p}}(g)}=0$ and ${\hat\alpha}_g$ is an anti-linear automorphism if ${{\mathfrak{p}}(g)}=1$, satisfying ${\hat\alpha}_g\circ\theta=\theta\circ {\hat\alpha}_g$.
We consider some key examples that will play an important role in defining our index. We fix a group homomorphism ${\mathfrak{p}}:G \to {{\mathbb Z}}_2$. A projective unitary/anti-unitary representation of $G$ relative to ${\mathfrak{p}}$ is a projective representation $V$ of $G$ such that $V_g$ is unitary if ${\mathfrak{p}}(g)=0$ and anti-unitary if ${\mathfrak{p}}(g)=1$.
\[ichi\] Let ${{\mathcal K}}$ be a Hilbert space and set $\Gamma_{{{\mathcal K}}}:={\mathbb I}_{{{\mathcal K}}}\otimes \sigma_z$, a self-adjoint unitary on ${{\mathcal K}}\otimes {{\mathbb C}}^2$.[^2] We set ${{\mathcal R}}_{0,{{\mathcal K}}}:={{\mathcal B}}({{\mathcal K}})\otimes {\mathop{\mathrm{M}}\nolimits}_2$ and so $({{\mathcal R}}_{0,{{\mathcal K}}}, {\mathop{\mathrm{Ad}}\nolimits}_{\Gamma_{{\mathcal K}}})$ is a spatially graded von Neumann algebra acting on ${{\mathcal K}}\otimes {{\mathbb C}}^2$ with grading operator $\Gamma_{{\mathcal K}}$. Let $V$ be a projective unitary/anti-unitary representation of $G$ on ${{\mathcal K}}\otimes {{\mathbb C}}^2$ relative to ${\mathfrak{p}}$. We also assume that there is a homomorphism ${\mathfrak{q}}:G\to {{\mathbb Z}}_2$ such that ${\mathop{\mathrm{Ad}}\nolimits}_{V_g}(\Gamma_{{\mathcal K}})=(-1)^{{{\mathfrak{q}}(g)}}\Gamma_{{\mathcal K}}$. We then obtain a graded $W^*$-$(G,{\mathfrak{p}})$-dynamical system $({{\mathcal R}}_{0,{{\mathcal K}}},{\mathop{\mathrm{Ad}}\nolimits}_{\Gamma_{{\mathcal K}}},{\mathop{\mathrm{Ad}}\nolimits}_{V_g})$.
We denote the set of all $W^*$-$(G,{\mathfrak{p}})$-dynamical systems of the form in Example \[ichi\] by ${{\mathcal S}}_0$.
\[ni\] Let ${{\mathcal K}}$ be a Hilbert space and set $\Gamma_{{\mathcal K}}:={\mathbb I}_{{{\mathcal K}}}\otimes \sigma_z$. Let ${\mathop{\mathfrak{C}}\nolimits}$ be the subalgebra of ${\mathop{\mathrm{M}}\nolimits}_2$ generated by $\sigma_x$ and set ${{\mathcal R}}_{1,{{\mathcal K}}}:={{\mathcal B}}({{\mathcal K}})\otimes {\mathop{\mathfrak{C}}\nolimits}$.[^3] Then $({{\mathcal R}}_{1,{{\mathcal K}}}, {\mathop{\mathrm{Ad}}\nolimits}_{\Gamma_{{\mathcal K}}})$ is a spatially graded von Neumann algebra acting on ${{\mathcal K}}\otimes {{\mathbb C}}^2$ with grading operator $\Gamma_{{\mathcal K}}$. Let $V$ be a projective unitary/anti-unitary representation of $G$ relative to ${\mathfrak{p}}$ such that ${\mathop{\mathrm{Ad}}\nolimits}_{V_g}({\mathbb I}_{{{\mathcal K}}}\otimes \sigma_x)= (-1)^{{{\mathfrak{q}}(g)}}{\left (}{\mathbb I}_{{\mathcal K}}\otimes \sigma_x{\right )}$ and ${\mathop{\mathrm{Ad}}\nolimits}_{V_g}(\Gamma_{{\mathcal K}})=(-1)^{{{\mathfrak{q}}(g)}}\Gamma_{{\mathcal K}}$ for ${\mathfrak{q}}:G\to {{\mathbb Z}}_2$ a group homomorphism. These assumptions imply that ${\mathop{\mathrm{Ad}}\nolimits}_{V_g}{\left (}{{\mathcal R}}_{1,{{\mathcal K}}}{\right )}={{\mathcal R}}_{1,{{\mathcal K}}}$ and so $({{\mathcal R}}_{1,{{\mathcal K}}},{\mathop{\mathrm{Ad}}\nolimits}_{\Gamma_{{\mathcal K}}},{\mathop{\mathrm{Ad}}\nolimits}_{V_g})$ is a graded $W^*$-$(G,{\mathfrak{p}})$-dynamical system.
We denote the set of all $W^*$-$(G,{\mathfrak{p}})$-dynamical systems of the form of Example \[ni\] by ${{\mathcal S}}_1$. Given a $W^*$-$(G,{\mathfrak{p}})$-dynamical systems in ${{\mathcal S}}_1$, we can construct a projective representation of $G$ on ${{\mathcal K}}$ from the projective representation on ${{\mathcal K}}\otimes {{\mathbb C}}^2$.
We first establish some notation. Let $C$ be the complex conjugation on ${{{\mathbb C}}^2}$ with respect to the standard basis. Given two group homomorphisms ${\mathfrak{q}}_1, \, {\mathfrak{q}}_2 \in {\ensuremath{\mathrm{Hom}}}(G,{{\mathbb Z}}_2) \cong H^1(G, {{\mathbb Z}}_2)$, we can define a group $2$-cocyle, $$\begin{aligned}
\label{enn}
\epsilon({{\mathfrak{q}}}_1,{{\mathfrak{q}}}_2)(g,h)
=(-1)^{{{\mathfrak{q}}_1(g)}{{\mathfrak{q}}}_2(h)},\quad g,h\in G.\end{aligned}$$
\[homo\] Note that $ [\epsilon ({{\mathfrak{q}}_1},{\mathfrak{q}}_2)] = [\epsilon ({{\mathfrak{q}}_2},{\mathfrak{q}}_1)] \in H^2(G,U(1)_{{\mathfrak{q}}_1})$.
\[zv\] For $({{\mathcal R}}_{1,{{\mathcal K}}},{\mathop{\mathrm{Ad}}\nolimits}_{\Gamma_{{\mathcal K}}},{\mathop{\mathrm{Ad}}\nolimits}_{V_g})\in{{\mathcal S}}_1$, there is a unique projective unitary/anti-unitary representation $V^{(0)}$ of $G$ on ${{\mathcal K}}$ relative to ${\mathfrak{p}}$ such that $V_g=V_g^{(0)}\otimes C^{{{\mathfrak{p}}(g)}}\sigma_y^{{\mathfrak{q}}(g)}$. If $[\tilde\upsilon]$ and $[\upsilon]$ are the second cohomology classes associated to $V$ and $V^{(0)}$ respectively, then $[\tilde{\upsilon}] = [\upsilon \, \epsilon({\mathfrak{q}},{\mathfrak{p}})] \in H^2(G, U(1)_{\mathfrak{p}})$.
Because ${\mathop{\mathrm{Ad}}\nolimits}_{V_g} \circ \mathrm{Ad}_{\Gamma_{{\mathcal K}}} = \mathrm{Ad}_{\Gamma_{{\mathcal K}}}\circ {\mathop{\mathrm{Ad}}\nolimits}_{V_g}$, we have ${\mathop{\mathrm{Ad}}\nolimits}_{V_g}({{\mathcal B}}({{\mathcal K}})\otimes{{\mathbb C}}{\mathbb I}_{{{\mathbb C}}^2})
=
{{\mathcal B}}({{\mathcal K}})\otimes{{\mathbb C}}{\mathbb I}_{{{\mathbb C}}^2}
$. Therefore, ${\mathop{\mathrm{Ad}}\nolimits}_{V_g}$ induces a linear/anti-linear $*$-automorphism on ${{{\mathcal B}}({{\mathcal K}})}$. Applying Wigner’s Theorem, there is a unitary/anti-unitary $\tilde V_g^{(0)}$ on ${{\mathcal K}}$ such that $$\begin{aligned}
{\mathop{\mathrm{Ad}}\nolimits}_{V_g}{\left (}x\otimes {\mathbb I}_{{{{\mathbb C}}^2}}{\right )}={\mathop{\mathrm{Ad}}\nolimits}_{\tilde V_g^{(0)}} (x)\otimes {\mathbb I}_{{{\mathbb C}}^2},\quad x\in{{{\mathcal B}}({{\mathcal K}})}.\end{aligned}$$ It is clear that $\tilde V^{(0)}$ gives a unitary/anti-unitary projective representation relative to ${\mathfrak{p}}$. Note that $V_g^*\big(\tilde V_g^{(0)}\otimes C^{{{\mathfrak{p}}(g)}}\sigma_y^{{{\mathfrak{q}}(g)}}\big)$ is a unitary which commutes with ${{{\mathcal B}}({{\mathcal K}})}\otimes{{\mathbb C}}{\mathbb I}_{{{{\mathbb C}}^2}}$, ${\mathbb I}_{{\mathcal K}}\otimes \sigma_x$, ${\mathbb I}_{{\mathcal K}}\otimes \sigma_z$ and therefore commutes with ${{{\mathcal B}}({{\mathcal K}})}\otimes{\mathop{\mathrm{M}}\nolimits}_2$. Therefore, there is a $c(g)\in {{\mathbb T}}$ such that $V_g=c(g){\left (}\tilde V_g^{(0)}\otimes C^{{{\mathfrak{p}}(g)}}\sigma_y^{{{\mathfrak{q}}(g)}}{\right )}$. Setting $V_g^{(0)}:=c(g)\tilde V_g^{(0)}$, we obtain $V_g=V_g^{(0)}\otimes C^{{{\mathfrak{p}}(g)}}\sigma_y^{{\mathfrak{q}}(g)}$. Clearly $V^{(0)}$ satisfies the required conditions. Because $\sigma_y^{{{\mathfrak{q}}(g)}} C^{{{\mathfrak{p}}(h)}}=(-1)^{{{\mathfrak{q}}(g)}{{\mathfrak{p}}(h)}} C^{{{\mathfrak{p}}(h)}}\sigma_y^{{{\mathfrak{q}}(g)}}$, we obtain the last statement.
We introduce the following equivalence relation on graded $W^*$-$(G,{\mathfrak{p}})$-dynamical systems.
Let $G$ be a finite group and ${\mathfrak{p}}: G\to{{\mathbb Z}}_2$ be a group homomorphism. We say that two graded $W^*$-$(G,{\mathfrak{p}})$-dynamical systems $({{\mathcal M}}_1, \theta_1, {\hat\alpha}^{(1)})$, $({{\mathcal M}}_2, \theta_2, {\hat\alpha}^{(2)})$ are equivalent and write $({{\mathcal M}}_1, \theta_1, {\hat\alpha}^{(1)})\sim({{\mathcal M}}_2, \theta_2, {\hat\alpha}^{(2)})$ if there is a $*$-isomorphism $\iota: {{\mathcal M}}_1\to{{\mathcal M}}_2$ such that $$\begin{aligned}
&\iota\circ{\hat\alpha}_g^{(1)}={\hat\alpha}_g^{(2)}\circ\iota,\quad g\in G\label{eone}\\
&\iota\circ\theta_1=\theta_2\circ\iota\label{etwo}.\end{aligned}$$
Clearly, this is an equivalence relation.
Using equivalence of $W^*$-$(G,{\mathfrak{p}})$-dynamical systems, we can reduce all type I balanced central graded $W^*$-$(G,{\mathfrak{p}})$-dynamical systems to the case of either Example \[ichi\] or [\[ni\]]{}.
\[casebycase\] Let $({{\mathcal M}},\theta,{\hat\alpha})$ be a graded $W^*$-$(G,{\mathfrak{p}})$-dynamical systems with $({{\mathcal M}},\theta)$ balanced, central and type I. Then there is a $\kappa\in{{\mathbb Z}}_2$ and $({{\mathcal R}}_{\kappa, {{\mathcal K}}},{\mathop{\mathrm{Ad}}\nolimits}_{\Gamma_{{\mathcal K}}},{\mathop{\mathrm{Ad}}\nolimits}_{V_g})\in{{\mathcal S}}_{\kappa}$ such that $({{\mathcal M}},\theta,{\hat\alpha})\sim ({{\mathcal R}}_{\kappa, {{\mathcal K}}},{\mathop{\mathrm{Ad}}\nolimits}_{\Gamma_{{\mathcal K}}},{\mathop{\mathrm{Ad}}\nolimits}_{V_g})$.
Because $({{\mathcal M}},\theta)$ is central, by Lemma \[jh\] either ${{\mathcal M}}$ is a factor or $Z({{\mathcal M}})$ has an odd self-adjoint unitary $b\in Z({{\mathcal M}})\cap {{\mathcal M}}^{(1)}$ such that $$\begin{aligned}
\label{bb}
Z({{\mathcal M}})\cap {{\mathcal M}}^{(1)}={{\mathbb C}}b.\end{aligned}$$ We set $\kappa=0$ for the former case, and $\kappa=1$ for the latter case.
[**(Case: $\kappa=0$)**]{} Suppose ${{\mathcal M}}$ is a type I factor. Because $({{\mathcal M}},\theta)$ is balanced, there is an odd self-adjoint unitary $U\in {{\mathcal M}}^{(1)}$.
We claim that ${{\mathcal M}}^{(0)}$ is not a factor. If ${{\mathcal M}}^{(0)}$ is a factor, by Lemma \[gradei\], it is of type I. Note then that ${\mathop{\mathrm{Ad}}\nolimits}_U\vert_{{{\mathcal M}}^{(0)}}$ is an automorphism on the type I factor ${{\mathcal M}}^{(0)}$. By Wigner’s Theorem, there is a unitary $u\in {{\mathcal M}}^{(0)}$ such that ${\mathop{\mathrm{Ad}}\nolimits}_U(x)={\mathop{\mathrm{Ad}}\nolimits}_u(x)$, $x\in {{\mathcal M}}^{(0)}$. Therefore, $u^*U\in {\left (}{{{\mathcal M}}^{(0)}}{\right )}'$. At the same time, $u^*U$ commutes with $U$ because ${\mathop{\mathrm{Ad}}\nolimits}_{U}(u^*)={\mathop{\mathrm{Ad}}\nolimits}_u(u^*)=u^*$ for $u\in {{\mathcal M}}^{(0)}$. Hence $u^*U\in {{\mathcal M}}'\cap {{\mathcal M}}={{\mathbb C}}{\mathbb I}$. This is a contradiction because $u^*U$ is non-zero and odd. Hence we conclude that ${{\mathcal M}}^{(0)}$ is not a factor.
Therefore, there is a projection $z$ in $Z({{\mathcal M}}^{(0)})$ which is not $0$ nor ${\mathbb I}$. For such a projection, we have $z+{\mathop{\mathrm{Ad}}\nolimits}_{U}(z)\in{{\mathcal M}}\cap {\left (}{{{\mathcal M}}^{(0)}}{\right )}'\cap\{U\}'=Z({{\mathcal M}})={{\mathbb C}}{\mathbb I}$, which then implies that $z+{\mathop{\mathrm{Ad}}\nolimits}_{U}(z)={\mathbb I}$. (We note that for orthogonal projections $p,q$ satisfying $p+q=t{\mathbb I}$ with $t\in{{\mathbb R}}$, either $p+q={\mathbb I}$ or $p=0,\,{\mathbb I}$ holds, by considering the spectrum of $p=t{\mathbb I}-q$.)
We claim $Z({{\mathcal M}}^{(0)})={{\mathbb C}}z+{{\mathbb C}}{\mathbb I}$. Now, for any projection $s$ in $Z({{\mathcal M}}^{(0)})$, $zs$ is a projection in $Z({{\mathcal M}}^{(0)})$. Therefore either $zs=0$ or $zs+{\mathop{\mathrm{Ad}}\nolimits}_{U}(zs)={\mathbb I}$. The latter is possible only if $zs=z$ because $z+{\mathop{\mathrm{Ad}}\nolimits}_{U}(z)={\mathbb I}$. Similarly, we have $({\mathbb I}-z)s=0$ or $({\mathbb I}-z)s={\mathbb I}-z$. Hence we have $Z({{\mathcal M}}^{(0)})={{\mathbb C}}z+{{\mathbb C}}{\mathbb I}$, proving the claim.
Combining this with ${\mathop{\mathrm{Ad}}\nolimits}_U(z)={\mathbb I}-z$, ${{\mathcal M}}^{(0)}$ is a direct sum of two same-type factors ${{\mathcal M}}^{(0)}z$ and ${{\mathcal M}}^{(0)}({\mathbb I}-z)$. Applying Lemma \[gradei\], we see that ${{\mathcal M}}^{(0)}$ is of type I, and ${{\mathcal M}}^{(0)}z$ and ${{\mathcal M}}^{(0)}({\mathbb I}-z)$ are type I factors.
Set $\Gamma:=z-({\mathbb I}-z)$. Note that ${\mathop{\mathrm{Ad}}\nolimits}_\Gamma$ and $\theta$ are identity on ${{\mathcal M}}^{(0)}$. We also have ${\mathop{\mathrm{Ad}}\nolimits}_U(\Gamma)=({\mathbb I}-z)-z=-\Gamma$, hence ${\mathop{\mathrm{Ad}}\nolimits}_\Gamma(U)=-U=\theta(U)$. Therefore, we get $$\begin{aligned}
\label{tgr}
\theta(x)={\mathop{\mathrm{Ad}}\nolimits}_\Gamma(x),\quad x\in {{\mathcal M}}.\end{aligned}$$
Next we claim that there is a Hilbert space ${{\mathcal K}}$ and a $*$-isomorphism $\iota:{{\mathcal M}}\to {{\mathcal B}}({{\mathcal K}})\otimes {\mathop{\mathrm{M}}\nolimits}_2$ such that $$\begin{aligned}
\label{clclcl}
\iota\circ \theta={\mathop{\mathrm{Ad}}\nolimits}_{\Gamma_{{\mathcal K}}}\circ \iota,\quad\text{and}\quad
\iota(\Gamma)={\mathbb I}_{{\mathcal K}}\otimes \sigma_z=:\Gamma_{{\mathcal K}}.\end{aligned}$$ As ${{\mathcal M}}$ is a type I factor, there is a Hilbert space $\hat {{\mathcal K}}$ and a $*$-isomorphism $\hat\iota:{{\mathcal M}}\to{{\mathcal B}}(\hat{{\mathcal K}})$. Let $\hat\iota(\Gamma)=Q_0-Q_1$ be the spectral decomposition of a self-adjoint unitary $\hat\iota(\Gamma)$, with orthogonal projections $Q_0,Q_1$, corresponding to eigenvalues $1,-1$. As we have ${\mathop{\mathrm{Ad}}\nolimits}_{\hat\iota(\Gamma)}\circ \hat\iota(x)=\hat\iota\circ{\mathop{\mathrm{Ad}}\nolimits}_\Gamma(x)=\hat\iota\circ\theta(x)$ for $x\in{{\mathcal M}}$ by , we have $
\hat\iota({{\mathcal M}}^{(0)})={{\mathcal B}}(Q_0\hat{{\mathcal K}})\oplus{{\mathcal B}}(Q_1\hat{{\mathcal K}})$. Because ${\mathop{\mathrm{Ad}}\nolimits}_\Gamma(U)=-U$, we have ${\mathop{\mathrm{Ad}}\nolimits}_{\hat\iota(U)}{\left (}\hat \iota(\Gamma){\right )}=-\hat\iota(\Gamma)$. From the spectral decomposition, we then have ${\mathop{\mathrm{Ad}}\nolimits}_{\hat\iota(U)}(Q_0)=Q_1$ and ${\mathop{\mathrm{Ad}}\nolimits}_{\hat\iota(U)}(Q_1)=Q_0$. We therefore see that $v:=Q_0 \hat\iota (U)Q_1$ is a unitary from $Q_1\hat{{\mathcal K}}$ onto $Q_0\hat{{\mathcal K}}$. We set ${{\mathcal K}}:=Q_0\hat{{\mathcal K}}$ and define a unitary $W:\hat {{\mathcal K}}\to {{\mathcal K}}\otimes {{\mathbb C}}^2$ by $$\begin{aligned}
W\begin{pmatrix}
\xi_0\\\xi_1
\end{pmatrix}
=\xi_0\otimes e_0+v\xi_1\otimes e_1,\quad\xi_0\in Q_0\hat {{\mathcal K}},\quad\xi_1\in Q_1\hat{{\mathcal K}}.\end{aligned}$$ Here $\{e_0,e_1\}$ is the standard basis of ${{\mathbb C}}^2$. Note that ${\mathop{\mathrm{Ad}}\nolimits}_W\circ\hat\iota(\Gamma)={\mathbb I}_{{\mathcal K}}\otimes\sigma_z=\Gamma_{{{\mathcal K}}}$. Then $\iota:={\mathop{\mathrm{Ad}}\nolimits}_W\circ\hat\iota: {{\mathcal M}}\to {{\mathcal B}}({{\mathcal K}})\otimes {\mathop{\mathrm{M}}\nolimits}_2$ is a $*$-isomorphism satisfying , proving the claim.
Next we consider the action of $G$. Because $Z({{\mathcal M}}^{(0)})={{\mathbb C}}z+{{\mathbb C}}({\mathbb I}-z)$, $\Gamma=z-({\mathbb I}-z)$ and $-\Gamma=-z+({\mathbb I}-z)$ are the only self-adjoint unitaries in $Z({{\mathcal M}}^{(0)})\setminus {{\mathbb C}}{\mathbb I}$. As ${\hat\alpha}_g$ preserves ${{\mathcal M}}^{(0)}$, ${\hat\alpha}_g(\Gamma)$ is a self-adjoint unitary in $Z({{\mathcal M}}^{(0)})\setminus {{\mathbb C}}{\mathbb I}$ and so ${\hat\alpha}_g(\Gamma)=(-1)^{{{\mathfrak{q}}(g)}}\Gamma$ for ${{\mathfrak{q}}(g)}=0$ or ${{\mathfrak{q}}(g)}=1$. Clearly, ${{\mathfrak{q}}}:G\to{{\mathbb Z}}_2$ is a group homomorphism.
Because $\iota\circ{\hat\alpha}_g\circ \iota^{-1}$ is a linear/anti-linear automorphism on ${{\mathcal B}}({{\mathcal K}})\otimes{\mathop{\mathrm{M}}\nolimits}_2$, by Wigner’s Theorem there is a projective representation $V$ satisfying $$\begin{aligned}
\label{vdef}
{\mathop{\mathrm{Ad}}\nolimits}_{V_g} (x)=\iota\circ{\hat\alpha}_g\circ \iota^{-1}(x),\quad x\in {{\mathcal B}}({{\mathcal K}})\otimes {\mathop{\mathrm{M}}\nolimits}_2,\quad
g\in G,\end{aligned}$$ and where $V_g$ is unitary/anti-unitary depending on ${\mathfrak{p}}(g)$. Because ${\hat\alpha}_g(\Gamma)=(-1)^{{{\mathfrak{q}}(g)}}\Gamma$, we have $$\begin{aligned}
{\mathop{\mathrm{Ad}}\nolimits}_{V_g}{\left (}\Gamma_{{\mathcal K}}{\right )}=(-1)^{{{\mathfrak{q}}(g)}} \Gamma_{{\mathcal K}},\quad g\in G.\label{vpar}\end{aligned}$$ Hence we obtain $
({{\mathcal R}}_{0,{{\mathcal K}}},{\mathop{\mathrm{Ad}}\nolimits}_{\Gamma_{{\mathcal K}}},{\mathop{\mathrm{Ad}}\nolimits}_{V_g})\in{{\mathcal S}}_0
$. By and , we also have $({{\mathcal M}},\theta,{\hat\alpha})\sim ({{\mathcal R}}_{0,{{\mathcal K}}},{\mathop{\mathrm{Ad}}\nolimits}_{\Gamma_{{\mathcal K}}},{\mathop{\mathrm{Ad}}\nolimits}_{V_g})$.
[**(Case: $\kappa=1$)**]{} Suppose that ${{\mathcal M}}$ has a self-adjoint unitary $b\in Z({{\mathcal M}})\cap {{\mathcal M}}^{(1)}$ satisfying . Set $P_\pm :=\frac{1\pm b}{2}$, where $P_\pm$ are orthogonal projections in $Z({{\mathcal M}})$ such that $P_++P_-={\mathbb I}$. By , $Z(M)={{\mathbb C}}b +{{\mathbb C}}{\mathbb I}={{\mathbb C}}P_++{{\mathbb C}}P_-$. As ${{\mathcal M}}$ is type I, ${{\mathcal M}}$ is a direct sum of the type I factors, ${{\mathcal M}}P_+$ and ${{\mathcal M}}P_-$.
We claim that ${{\mathcal M}}^{(0)}$ is a type I factor. For any $x\in Z{\left (}{{\mathcal M}}^{(0)}{\right )}$, we have $x\in {{\mathcal M}}^{(0)}\cap {\left (}{{\mathcal M}}^{(0)}{\right )}'\cap\{b \}'
={{\mathcal M}}^{(0)}\cap{{\mathcal M}}'=Z({{\mathcal M}})\cap{{\mathcal M}}^{(0)}={{\mathbb C}}{\mathbb I}$, because $b$ is a self-adjoint unitary in $Z({{\mathcal M}})\cap {{\mathcal M}}^{(1)}$. Hence $Z({{\mathcal M}}^{(0)})={{\mathbb C}}{\mathbb I}$ and by Lemma \[gradei\], ${{\mathcal M}}^{(0)}$ is a type I factor.
Next we claim that there is a Hilbert space ${{\mathcal K}}$ and a $*$-isomorphism $\iota:{{\mathcal M}}\to {{\mathcal B}}({{\mathcal K}})\otimes {\mathop{\mathfrak{C}}\nolimits}$ such that $$\begin{aligned}
\label{clc}
\iota\circ\theta={\mathop{\mathrm{Ad}}\nolimits}_{\Gamma_{{\mathcal K}}}\circ\iota,
\quad
\iota(b)={\mathbb I}_{{\mathcal K}}\otimes \sigma_x,\end{aligned}$$ for $\Gamma_{{\mathcal K}}={\mathbb I}_{{\mathcal K}}\otimes \sigma_z$. (Recall Example \[ni\] for ${\mathop{\mathfrak{C}}\nolimits}$.) Because ${{\mathcal M}}^{(0)}$ is a type I factor, there is a Hilbert space ${{\mathcal K}}$ and a $*$-isomorphism $\iota_0: {{\mathcal M}}^{(0)}\to {{\mathcal B}}({{\mathcal K}})$. As ${{\mathcal M}}={{\mathcal M}}^{(0)}\oplus {{\mathcal M}}^{(0)}b$, we may define a linear map $\iota: {{\mathcal M}}\to {{{\mathcal B}}({{\mathcal K}})}\otimes {\mathop{\mathfrak{C}}\nolimits}$ by $$\begin{aligned}
\iota(x+yb):=\iota_0(x)\otimes {\mathbb I}+\iota_0(y)\otimes \sigma_x,\quad
x,y\in{{\mathcal M}}^{(0)}.\end{aligned}$$ It can be easily checked that $\iota$ is a $*$-isomorphism satisfying .
Now we consider the group action. Because $Z({{\mathcal M}})\cap {{\mathcal M}}^{(1)}={{\mathbb C}}b$, $b$ and $-b$ are the only self-adjoint unitaries in $Z({{\mathcal M}})\cap{{\mathcal M}}^{(1)}$. As ${\hat\alpha}_g$ commutes with the grading automorphism, ${\hat\alpha}_g(b)$ is a self-adjoint unitary in $Z({{\mathcal M}})\cap{{\mathcal M}}^{(1)}$. Therefore, ${\hat\alpha}_g(b)=(-1)^{{{\mathfrak{q}}(g)}}b$ with ${{\mathfrak{q}}}:G\to{{\mathbb Z}}_2$ a group homomorphism.
Because ${\hat\alpha}_g({{\mathcal M}}^{(0)})={{\mathcal M}}^{(0)}$ and $\iota({{\mathcal M}}^{(0)})={{{\mathcal B}}({{\mathcal K}})}\otimes{{\mathbb C}}{\mathbb I}$ by , $\iota\circ{\hat\alpha}_g\circ\iota^{-1} $ induces a linear/anti-linear automorphism on ${{{\mathcal B}}({{\mathcal K}})}$ which is implemented by a unitary/antiunitary $V_g^{(0)}$ on ${{\mathcal K}}$ by Wigner’s Theorem. That is, $$\begin{aligned}
\iota\circ{\hat\alpha}_g\circ\iota^{-1}(a\otimes {\mathbb I}_{{{{\mathbb C}}^2}})
={\mathop{\mathrm{Ad}}\nolimits}_{V_g^{(0)}}( a)\otimes {\mathbb I}_{{{\mathbb C}}^2},\quad
a\in {{{\mathcal B}}({{\mathcal K}})},\quad g\in G.
\end{aligned}$$ with $V^{(0)}$ a projective unitary/anti-unitary representation of $G$ on ${{\mathcal K}}$ relative to ${\mathfrak{p}}$. Set $ V_g:= V_g^{(0)}\otimes C^{{{\mathfrak{p}}(g)}}\sigma_y^{{{\mathfrak{q}}(g)}}$, with the complex conjugation $C$ on ${{\mathbb C}}^2$ with respect to the standard basis. Clearly $V$ is also a projective unitary/anti-unitary representation of $G$ on ${{\mathcal K}}\otimes{{\mathbb C}}^2$ relative to ${\mathfrak{p}}$. We then have $$\begin{aligned}
{\mathop{\mathrm{Ad}}\nolimits}_{V_g} (a\otimes {{\mathbb I}_{{{\mathbb C}}^2}})
&=\iota\circ{\hat\alpha}_g\circ \iota^{-1}(a\otimes{{\mathbb I}_{{{\mathbb C}}^2}}),\quad a\in {{\mathcal B}}({{\mathcal K}}),\\
{\mathop{\mathrm{Ad}}\nolimits}_{V_g}({\mathbb I}_{{\mathcal K}}\otimes \sigma_x)
&=(-1)^{{{\mathfrak{q}}(g)}}({\mathbb I}_{{\mathcal K}}\otimes \sigma_x)
=\iota\circ{\hat\alpha}_g (b)
=\iota\circ{\hat\alpha}_g\circ \iota^{-1}({\mathbb I}_{{\mathcal K}}\otimes \sigma_x).
\end{aligned}$$ Combining these identities, we obtain $$\begin{aligned}
\label{vvve}
{\mathop{\mathrm{Ad}}\nolimits}_{V_g}\circ\iota(x)=\iota\circ{\hat\alpha}_g(x),\quad x\in{{\mathcal M}}.
\end{aligned}$$ We also have $$\begin{aligned}
{\mathop{\mathrm{Ad}}\nolimits}_{V_g}(\Gamma_{{\mathcal K}})
=(-1)^{{{\mathfrak{q}}(g)}}\Gamma_{{\mathcal K}}.
\end{aligned}$$ Hence we obtain $({{\mathcal R}}_{1,{{\mathcal K}}},{\mathop{\mathrm{Ad}}\nolimits}_{\Gamma_{{\mathcal K}}},{\mathop{\mathrm{Ad}}\nolimits}_{V_g})\in{{\mathcal S}}_1$ such that $({{\mathcal M}},\theta,{\hat\alpha})\sim ({{\mathcal R}}_{1,{{\mathcal K}}},{\mathop{\mathrm{Ad}}\nolimits}_{\Gamma_{{\mathcal K}}},{\mathop{\mathrm{Ad}}\nolimits}_{V_g})$.
\[index\] Let $({{\mathcal M}},\theta,{\hat\alpha})$ a graded $W^*$-$(G,{\mathfrak{p}})$-dynamical system with $({{\mathcal M}},\theta)$ balanced, central and type I. By Proposition \[casebycase\], there is a $\kappa\in{{\mathbb Z}}_2$ and $({{\mathcal R}}_{\kappa, {{\mathcal K}}},{\mathop{\mathrm{Ad}}\nolimits}_{\Gamma_{{\mathcal K}}},{\mathop{\mathrm{Ad}}\nolimits}_{V_g})\in{{\mathcal S}}_{\kappa}$ such that $({{\mathcal M}},\theta,{\hat\alpha})\sim ({{\mathcal R}}_{\kappa, {{\mathcal K}}},{\mathop{\mathrm{Ad}}\nolimits}_{\Gamma_{{\mathcal K}}},{\mathop{\mathrm{Ad}}\nolimits}_{V_g})$. Let ${{\mathfrak{q}}}:G\to {{\mathbb Z}}_2$ be a group homomorphism such that ${\mathop{\mathrm{Ad}}\nolimits}_{V_g}(\Gamma_{{\mathcal K}})=(-1)^{{{\mathfrak{q}}(g)}}\Gamma_{{\mathcal K}}$ and $[\upsilon]$ the second cohomology class associated to the projective representation $V_g$ if $\kappa=0$, and $V_g^{(0)}$ (from Lemma \[zv\]) if $\kappa=1$. We define an index of $({{\mathcal M}},\theta,{\hat\alpha})$ by $$\begin{aligned}
{\mathop{\mathrm{Ind}}\nolimits}({{\mathcal M}},\theta,{\hat\alpha}):=(\kappa, {{\mathfrak{q}}}, [\upsilon])\in{{\mathbb Z}}_2\times H^1(G,{{\mathbb Z}}_2)\times H^2(G, U(1)_{{\mathfrak{p}}}).\end{aligned}$$
\[lem:index\_indep\_of\_R\_iso\] The quantity ${\mathop{\mathrm{Ind}}\nolimits}({{\mathcal M}},\theta,{\hat\alpha})$ is well-defined and independent of the choice of $({{\mathcal R}}_{\kappa, {{\mathcal K}}},{\mathop{\mathrm{Ad}}\nolimits}_{\Gamma_{{\mathcal K}}},{\mathop{\mathrm{Ad}}\nolimits}_{V_g})\in{{\mathcal S}}_{\kappa}$ such that $({{\mathcal M}},\theta,{\hat\alpha})\sim ({{\mathcal R}}_{\kappa, {{\mathcal K}}},{\mathop{\mathrm{Ad}}\nolimits}_{\Gamma_{{\mathcal K}}},{\mathop{\mathrm{Ad}}\nolimits}_{V_g})$.
Suppose that both $({{\mathcal R}}_{\kappa_1,{{\mathcal K}}_1},{\mathop{\mathrm{Ad}}\nolimits}_{\Gamma_{{{\mathcal K}}_1}},{\mathop{\mathrm{Ad}}\nolimits}_{V_g^{(1)}})\in{{\mathcal S}}_{\kappa_1}$ and $({{\mathcal R}}_{\kappa_2, {{\mathcal K}}_2},{\mathop{\mathrm{Ad}}\nolimits}_{\Gamma_{{{\mathcal K}}_2}},{\mathop{\mathrm{Ad}}\nolimits}_{V_g^{(2)}})\in{{\mathcal S}}_{\kappa_2}$ are equivalent to $({{\mathcal M}},\theta,{\hat\alpha})$, via $*$-isomorphisms $\iota_i: {{\mathcal M}}\to {{\mathcal R}}_{\kappa_i, {{\mathcal K}}_i}$, $i=1,2$, respectively. Then $\iota_2\circ\iota_1^{-1}: {{\mathcal R}}_{\kappa_1,{{\mathcal K}}_1}\to {{\mathcal R}}_{\kappa_2, {{\mathcal K}}_2}$ is a $*$-isomorphism such that for all $g\in G$, $$\begin{aligned}
\label{iiin}
&\iota_2\circ\iota_1^{-1}\circ{\mathop{\mathrm{Ad}}\nolimits}_{V_g^{(1)}}={\mathop{\mathrm{Ad}}\nolimits}_{V_g^{(2)}}\circ \iota_2\circ\iota_1^{-1},
&&\iota_2\circ\iota_1^{-1}\circ{\mathop{\mathrm{Ad}}\nolimits}_{\Gamma_{{{\mathcal K}}_1}}=
{\mathop{\mathrm{Ad}}\nolimits}_{\Gamma_{{{\mathcal K}}_2}}\circ \iota_2\circ\iota_1^{-1}.\end{aligned}$$ Let $(\kappa_i, {{\mathfrak{q}}}_i, [\upsilon_i])$ be indices obtained from $({{\mathcal R}}_{\kappa_i,{{\mathcal K}}_i},{\mathop{\mathrm{Ad}}\nolimits}_{\Gamma_{{{\mathcal K}}_i}},{\mathop{\mathrm{Ad}}\nolimits}_{V_g^{(i)}})$, for $i=1,2$. Because of the $*$-isomorphism $\iota_2\circ\iota_1^{-1}$, we clearly have $\kappa_1=\kappa_2$. If $\kappa_1=\kappa_2=0$, then both of $\iota_i^{-1}({\mathbb I}_{{{\mathcal K}}_i}\otimes\sigma_z)$, $i=1,2$, are self-adjoint unitaries in $Z({{\mathcal M}}^{(0)})\setminus{{\mathbb C}}{\mathbb I}$. From the proof of Proposition \[casebycase\], this means that $\iota_2\circ\iota_1^{-1}({\mathbb I}_{{{\mathcal K}}_1}\otimes\sigma_z)=\pm
({\mathbb I}_{{{\mathcal K}}_2}\otimes\sigma_z)$. Hence we get $$\begin{aligned}
&(-1)^{{{\mathfrak{q}}_1(g)}}\iota_2\circ\iota_1^{-1}({\mathbb I}_{{{\mathcal K}}_1}\otimes\sigma_z)
=\iota_2\circ\iota_1^{-1}\circ{\mathop{\mathrm{Ad}}\nolimits}_{V_g^{(1)}}({\mathbb I}_{{{\mathcal K}}_1}\otimes\sigma_z)
={\mathop{\mathrm{Ad}}\nolimits}_{V_g^{(2)}}\circ \iota_2\circ\iota_1^{-1}({\mathbb I}_{{{\mathcal K}}_1}\otimes\sigma_z)\nonumber\\
&\hspace{1.5cm} =\pm {\mathop{\mathrm{Ad}}\nolimits}_{V_g^{(2)}}({\mathbb I}_{{{\mathcal K}}_2}\otimes\sigma_z)
=\pm (-1)^{{{\mathfrak{q}}}_2(g)}
({\mathbb I}_{{{\mathcal K}}_2}\otimes\sigma_z)
=(-1)^{{{\mathfrak{q}}}_2(g)}\iota_2\circ\iota_1^{-1}({\mathbb I}_{{{\mathcal K}}_1}\otimes\sigma_z)\end{aligned}$$ We therefore obtain that ${{\mathfrak{q}}_1(g)}={{\mathfrak{q}}}_2(g)$. When $\kappa_1=\kappa_2=1$, an analogous argument for $\iota_i^{-1}({\mathbb I}_{{{\mathcal K}}_i}\otimes\sigma_x)\in Z({{\mathcal M}})\cap{{\mathcal M}}^{(1)}$, $i=1,2$ implies ${{\mathfrak{q}}_1(g)}={{\mathfrak{q}}}_2(g)$.
If $\kappa_1=\kappa_2=0$, the $*$-isomorphism $\iota_2\circ\iota_1^{-1}:{{\mathcal B}}({{\mathcal K}}_1)\otimes{\mathop{\mathrm{M}}\nolimits}_2\to{{\mathcal B}}({{\mathcal K}}_2)\otimes{\mathop{\mathrm{M}}\nolimits}_2$ is implemented by a unitary $W:{{\mathcal K}}_1\otimes {{\mathbb C}}^2\to{{\mathcal K}}_2\otimes {{\mathbb C}}^2$. Hence we see from that ${\mathop{\mathrm{Ad}}\nolimits}_{WV_g^{(1)}W^*}(x)={\mathop{\mathrm{Ad}}\nolimits}_{V_g^{(2)}}(x)$ for all $x\in {{\mathcal B}}({{\mathcal K}}_1)\otimes{\mathop{\mathrm{M}}\nolimits}_2$. This means that $[\upsilon_1]=[\upsilon_2]$. If $\kappa_1=\kappa_2=1$, the restriction of the $*$-isomorphism $\iota_2\circ\iota_1^{-1}$ onto ${{{\mathcal B}}({{\mathcal K}}_1)}$ induces a $*$-isomorphism from ${{\mathcal B}}({{\mathcal K}}_1)$ to ${{\mathcal B}}({{\mathcal K}}_2)$. Therefore, there is a unitary $W:{{\mathcal K}}_1\to{{\mathcal K}}_2$ such that $
\iota_2\circ\iota_1^{-1}(x\otimes {\mathbb I})
={\mathop{\mathrm{Ad}}\nolimits}_W(x)\otimes {\mathbb I}$, for all $ x\in{{\mathcal B}}({{\mathcal K}}_1)$. Therefore, from we have ${\mathop{\mathrm{Ad}}\nolimits}_{W(V_g^{(1)})^{(0)}W^*}(x)={\mathop{\mathrm{Ad}}\nolimits}_{(V_g^{(2)})^{(0)}}(x)$ for all $x\in{{\mathcal B}}({{\mathcal K}}_1)$. This means that $[\upsilon_1]=[\upsilon_2]$.
Proposition \[casebycase\], Lemma \[lem:index\_indep\_of\_R\_iso\] and the fact that equivalence of $W^*$-$(G,{\mathfrak{p}})$-dynamical systems is an equivalence relation gives us the following.
\[wsta\] Let $({{\mathcal M}}_1,\theta_1,{\hat\alpha}_1)$, $({{\mathcal M}}_2,\theta_2,{\hat\alpha}_2)$ be graded $W^*$-$(G,{\mathfrak{p}})$-dynamical systems of balanced, central and type I graded von Neumann algebras. If $({{\mathcal M}}_1,\theta_1,{\hat\alpha}_1)\sim({{\mathcal M}}_2,\theta_2,{\hat\alpha}_2)$, then ${\mathop{\mathrm{Ind}}\nolimits}({{\mathcal M}}_1,\theta_1,{\hat\alpha}_1)={\mathop{\mathrm{Ind}}\nolimits}({{\mathcal M}}_2,\theta_2,{\hat\alpha}_2)$.
The index for pure split states
-------------------------------
We now define an index to fermionic SPT phases. For each $\Theta$-invariant and $\alpha$-invariant state ${{\mathcal A}}$, $(\pi_{\varphi}({{\mathcal A}}_R)'', {\mathop{\mathrm{Ad}}\nolimits}_{\hat \Gamma_\varphi}, \hat\alpha_\varphi)$ is a graded $W^*$-$(G,{\mathfrak{p}})$-dynamical system.
We first review the split property and recent results of Matsui [@Matsui20] that relate the split property to unique gapped ground states of the CAR-algebra. Given a state $\varphi$ on ${{\mathcal A}}$, $\varphi\vert_{{{\mathcal A}}_R}$ denotes the restriction of $\varphi$ to ${{\mathcal A}}_R$ and $\pi_{\varphi\vert_{{{\mathcal A}}_R}}$ is the GNS representation of ${{\mathcal A}}_R$ from this restricted state.
Let $\varphi$ be a pure $\Theta$-invariant state on ${{\mathcal A}}$. We say that $\varphi$ satisfies the split property if $\pi_{\varphi\vert_{{{\mathcal A}}_R}}({{\mathcal A}}_R)''$ is a type I von Neumann algebra.
Recall the notation ${{\mathcal B}}_f^e$ which denotes the set of all finite-range even interactions that satisfy the bound . Similarly, ${{\mathcal G}}_{f}^{e,\alpha}$ denotes the set of all $\alpha$-invariant interactions $\Phi\in{{\mathcal B}}_f^{e}$, with a unique gapped ground state.
Let $\varphi$ be a unique gapped $\tau^\Phi$-ground state of an interaction $\Phi\in{{{\mathcal B}}}_f^e$. Then $\varphi$ satisfies the split property.
To apply Matsui’s result to graded $W^*$-$(G,{\mathfrak{p}})$-dynamical systems, we must first relate the split ground state of an interaction $\Phi\in {{{\mathcal G}}}_f^{e,\alpha}$ to balanced and central graded type I von Neumann algebras. To show this, we first note the following.
\[hs\] Let $\varphi$ be a $\Theta$-invariant pure state on ${{\mathcal A}}$. Then
1. $Z(\pi_\varphi({{\mathcal A}}_R)'')\cap {\left (}\pi_\varphi({{\mathcal A}}_R)''{\right )}^{(0)}={{\mathbb C}}{\mathbb I}$,
2. The representation $\pi_{\varphi\vert_{{{\mathcal A}}_R}}$ and $(\pi_\varphi)\vert_{{{\mathcal A}}_R}$, the restriction of $\pi_{\varphi}$ to ${{\mathcal A}}_R$, are quasi-equivalent.
\(i) We have that $$\begin{aligned}
Z(\pi_\varphi({{\mathcal A}}_R)'')\cap {\left (}\pi_\varphi({{\mathcal A}}_R)''{\right )}^{(0)}
\subset
\pi_\varphi({{\mathcal A}}_L)'\cap\pi_\varphi({{\mathcal A}}_R)'=\pi_\varphi({{\mathcal A}})'={{\mathbb C}}{\mathbb I},\end{aligned}$$ where the last equality is because $\varphi$ is pure.
\(ii) Let ${\hat \Gamma_\varphi}$ be a self-adjoint unitary on ${{\mathcal H}}_\varphi$ given by ${\hat \Gamma_\varphi}\pi_\varphi(A)\Omega_\varphi=\pi_\varphi\circ\Theta(A)\Omega_\varphi$, $A\in{{\mathcal A}}$. Let $p$ denote the the orthogonal projection onto $\overline{\pi_\varphi({{\mathcal A}}_R)\Omega_\varphi}$. Then $(p{{\mathcal H}}_\varphi, \pi_\varphi(\cdot )\vert_{{{\mathcal A}}_R}p,\Omega_\varphi)$ is a GNS triple of $\varphi\vert_{{{\mathcal A}}_R}$. To show (ii), it suffices to show that $\tau: \pi_\varphi({{\mathcal A}}_R)''\to (\pi_\varphi({{\mathcal A}}_R)p)''$ defined by $\tau(x)=xp$ is a $*$-isomorphism. It is standard to see that $\tau$ is a surjective $*$-homomorphism. To see that $\tau$ is injective, note that from (i) and Lemma \[jh\], either $\pi_\varphi({{\mathcal A}}_R)''$ is factor or $Z(\pi_\varphi({{\mathcal A}}_R)'')={{\mathbb C}}{\mathbb I}+{{\mathbb C}}b$ with some self-adjoint unitary $b\in Z(\pi_\varphi({{\mathcal A}}_R)'')\cap (\pi_\varphi({{\mathcal A}}_R)'')^{(1)}$. For the former case, $\tau$ is clearly injective. For the latter case, let $b=P_+-P_-$ be the spectral decomposition. Because $b$ is odd, we have ${\mathop{\mathrm{Ad}}\nolimits}_{\hat \Gamma_\varphi}(P_\pm)=P_{\mp}$. If $\tau$ is not injective, the kernel of $\tau$ is either $\pi_\varphi({{\mathcal A}}_R)''P_+$ or $\pi_\varphi({{\mathcal A}}_R)''P_-$. If $\tau(P_+)=0$, then we have $P_+\Omega_\varphi$=0. We then have $$\begin{aligned}
P_-\Omega_\varphi={\hat \Gamma_\varphi} P_+{\hat \Gamma_\varphi}\Omega_\varphi={\hat \Gamma_\varphi} P_+\Omega_\varphi=0.\end{aligned}$$ Hence we obtain $\Omega_\varphi=(P_++P_-)\Omega_\varphi=0$, which is a contradiction. Similarly, we have $\tau(P_-)\neq 0$. Therefore, $\tau$ is injective.
\[deru\] Let $\varphi$ be a split pure $\Theta$-invariant and $\alpha$-invariant state on ${{\mathcal A}}$. Then $\pi_{\varphi}({{\mathcal A}}_R)''$ is balanced, central with respect to the grading given by $\hat\Gamma_\varphi$, and it is type I. The triple $(\pi_{\varphi}({{\mathcal A}}_R)'', {\mathop{\mathrm{Ad}}\nolimits}_{\hat \Gamma_\varphi}, \hat\alpha_\varphi)$ is a graded $W^*$-$(G,{\mathfrak{p}})$-dynamical system.
Because $\varphi$ is pure and $\Theta$-invariant, $\pi_{\varphi}({{\mathcal A}}_R)''$ is central by part (i) of Lemma \[hs\]. Because $\varphi$ is split, $\pi_{\varphi\vert_{{{\mathcal A}}_R}}({{\mathcal A}}_R)''$ is type I by definition. Because $(\pi_{\varphi})\vert_{{{\mathcal A}}_R}$ is quasi-equivalent to $\pi_{\varphi\vert_{{{\mathcal A}}_R}}$ by part (ii) of Lemma \[hs\], $\pi_{\varphi}({{\mathcal A}}_R)''$ is also type I. It is also balanced because ${{\mathcal A}}_R$ has an odd self-adjoint unitary. Because $\alpha_g\circ\Theta=\Theta\circ\alpha_g$, for all $g\in G$ then we have ${\left (}\hat\alpha_\varphi{\right )}_g\circ{\mathop{\mathrm{Ad}}\nolimits}_{\hat \Gamma_\varphi}={\mathop{\mathrm{Ad}}\nolimits}_{\hat \Gamma_\varphi}\circ{\left (}\hat\alpha_\varphi{\right )}_g$.
\[sds\] Consider the setting of Lemma \[deru\]. Let $\varphi_R:=\varphi\vert_{{{\mathcal A}}_R}$. Then $(\pi_{\varphi_R}({{\mathcal A}}_R)'', {\mathop{\mathrm{Ad}}\nolimits}_{\hat \Gamma_{\varphi_R}}, \hat\alpha_{\varphi_R})$ is also a graded $W^*$-$(G,{\mathfrak{p}})$-dynamical system of a balanced, central and type I graded von Neumann algebra with $$\begin{aligned}
(\pi_{\varphi}({{\mathcal A}}_R)'', {\mathop{\mathrm{Ad}}\nolimits}_{\hat \Gamma_\varphi}, \hat\alpha_\varphi)
\sim
(\pi_{\varphi_R}({{\mathcal A}}_R)'', {\mathop{\mathrm{Ad}}\nolimits}_{\hat \Gamma_{\varphi_R}}, \hat\alpha_{\varphi_R}).\end{aligned}$$
From Lemma \[deru\], we see that our index of $W^*$-$(G,{\mathfrak{p}})$-dynamical systems can be applied to split, pure, $\Theta$-invariant and $\alpha$-invariant states on ${{\mathcal A}}$. In particular, we may define an index for $\Phi\in {{{\mathcal G}}}_f^{e,\alpha}$.
Let $\varphi$ be a $\Theta$-invariant, $\alpha$-invariant, split and pure state on ${{\mathcal A}}$ with $\varphi_R:=\varphi\vert_{{{\mathcal A}}_R}$. We set $$\begin{aligned}
{\mathop{\mathrm{ind}}\nolimits}\varphi:={\mathop{\mathrm{Ind}}\nolimits}(\pi_{\varphi}({{\mathcal A}}_R)'', {\mathop{\mathrm{Ad}}\nolimits}_{\hat \Gamma_\varphi}, \hat\alpha_\varphi)
= {\mathop{\mathrm{Ind}}\nolimits}(\pi_{\varphi_R}({{\mathcal A}}_R)'', {\mathop{\mathrm{Ad}}\nolimits}_{\hat \Gamma_{\varphi_R}}, \hat\alpha_{\varphi_R}).\end{aligned}$$ For interactions $\Phi\in {{{\mathcal G}}}_f^{e,\alpha}$, we define the index of $\Phi$ by ${\mathop{\mathrm{ind}}\nolimits}(\Phi):={\mathop{\mathrm{ind}}\nolimits}(\varphi_\Phi)$, with $\varphi_\Phi$ the unique ground state of $\Phi$.
The stability of the index {#stability}
==========================
In this section we prove that ${\mathop{\mathrm{ind}}\nolimits}(\Phi)$ is an invariant of the classification of SPT phases. That is, for a path of interactions $\{\Phi(s)\}_{s\in[0,1]}$ satisfying Assumption \[assump\] below, we show that ${\mathop{\mathrm{ind}}\nolimits}(\Phi(0)) = {\mathop{\mathrm{ind}}\nolimits}(\Phi(1))$.
For each $N\in{{\mathbb N}}$, we denote $[-N,N]\cap {{\mathbb Z}}$ by $\Lambda_N$. Let ${{\mathbb E}}_{N}:{{\mathcal A}}\to {{\mathcal A}}_{\Lambda_N}$ be the conditional expectation with respect to the trace state, see [@am]. We consider the following subset of ${{\mathcal A}}$.
Let $f:(0,\infty)\to (0,\infty)$ be a continuous decreasing function with $\lim_{t\to\infty}f(t)=0$. For each $A\in{{\mathcal A}}$, let $$\begin{aligned}
{\left \Vert}A{\right \Vert}_f:={\left \Vert}A{\right \Vert}+ \sup_{N\in {\mathbb{N}}}{\left (}\frac{{\left \Vert}A-{{\mathbb E}}_{N}(A)
{\right \Vert}}
{f(N)}
{\right )}.\end{aligned}$$ We denote by ${{\mathcal D}}_f$ the set of all $A\in{{\mathcal A}}$ such that ${\left \Vert}A{\right \Vert}_f<\infty$.
We consider a path in ${{\mathcal G}}_f^{e, \alpha}$ satisfying the following conditions.
\[assump\] Let $[0,1]\ni s\mapsto \Phi ( s) \in {{\mathcal B}}_f^e$ be a path of interactions on ${{\mathcal A}}$. We assume the following:
1. For each $X\in{\mathfrak S}_{{{\mathbb Z}}}$, the map $[0,1]\ni s\mapsto \Phi(X;s)\in{{\mathcal A}}_{X}$ is continuous and piecewise $C^1$. We denote by $\dot{\Phi}(X;s)$ the corresponding derivatives. The interaction obtained by differentiation is denoted by $\dot\Phi(s)$, for each $s\in[0,1]$.
2. There is a number $R\in{\mathbb{N}}$ such that $X \in {\mathfrak S}_{{{\mathbb Z}}}$ and ${\text{diam}(X)}\ge R$ implies $\Phi(X;s)=0$ for all $s\in[0,1]$.
3. For each $s\in[0,1]$, $\Phi(s)\in {{\mathcal G}}^{e, \alpha}_f$. We denote the unique $\tau^{\Phi(s)}$-ground state by $\varphi_s$.
4. Interactions are bounded as follows $$\begin{aligned}
\sup_{s\in[0,1]}\sup_{X\in {\mathfrak S}_{{{\mathbb Z}}}}
{\left (}{\left \Vert}\Phi{\left (}X;s{\right )}{\right \Vert}+|X|{\left \Vert}\dot{\Phi} {\left (}X;s{\right )}{\right \Vert}{\right )}<\infty.\end{aligned}$$
5. Setting $$\begin{aligned}
b(\varepsilon):=\sup_{Z\in{\mathfrak S}_{{{\mathbb Z}}}}
\sup_{\substack{s,s_0 \in[0,1], \\ 0<| s-s_0|<\varepsilon}}
{\left \Vert}\frac{\Phi(Z;s)-\Phi(Z;s_0)}{s-s_0}-\dot{\Phi}(Z;s_0)
{\right \Vert}\end{aligned}$$ for each $\varepsilon>0$, we have $\lim_{\varepsilon\to 0} b(\varepsilon)=0$.
6. There exists a $\gamma>0$ such that $\sigma(H_{\varphi_s,\Phi(s)})\setminus\{0\}\subset [\gamma,\infty)$ for all $s\in[0,1]$, where $\sigma(H_{\varphi_s,\Phi(s)})$ is the spectrum of $H_{\varphi_s,\Phi(s)}$.
7. There exists $0<\beta<1$ satisfying the following: Set $\zeta(t):=e^{-t^{ \beta}}$. Then for each $A\in D_\zeta$, $\varphi_s(A)$ is differentiable with respect to $s$, and there is a constant $C_\zeta$ such that $$\begin{aligned}
\label{dcon}
{\left \vert}\dot{\varphi_s}(A)
{\right \vert}\le C_\zeta{\left \Vert}A{\right \Vert}_\zeta,\end{aligned}$$ for any $A\in D_\zeta$.
The main result of this section is the following.
\[c1t\] Let $[0,1]\ni s\mapsto \Phi ( s) \in {{\mathcal B}}_f^e$ be a path of interactions on ${{\mathcal A}}$ satisfying Assumption \[assump\]. Then ${\mathop{\mathrm{ind}}\nolimits}(\Phi(0))={\mathop{\mathrm{ind}}\nolimits}(\Phi(1))$.
The proof relies on the idea introduced in [@OgataTRI], that is, using the factorization property of automorphic equivalence. Namely, we note the following.
\[aep\] Let $[0,1]\ni s\mapsto \Phi ( s) \in {{\mathcal B}}_f^e$ be a path of interactions on ${{\mathcal A}}$ satisfying Assumption \[assump\]. Let $\varphi_s$ be the unique $\tau^{\Phi(s)}$-ground state, for each $s\in[0,1]$. Then there is an automorphism $\Xi$ on ${{\mathcal A}}$ and a unitary $u\in{{\mathcal A}}$ such that for all $g\in G$, $$\begin{aligned}
&\Xi({{\mathcal A}}_L)={{\mathcal A}}_L, &&\Xi({{\mathcal A}}_R)={{\mathcal A}}_R, &&\Xi\circ\Theta=\Theta\circ\Xi,
&&\Xi\circ\alpha_g=\alpha_g\circ\Xi, \\
&\Theta(u)=u, &&\alpha_g(u)=u,
&&\varphi_1=\varphi_0\circ{\mathop{\mathrm{Ad}}\nolimits}_u\circ\Xi.\end{aligned}$$
In Appendix \[flr\], we prove the Lieb-Robinson bound and a locality estimate for lattice fermion systems. Having them, the proof of Proposition \[aep\] is the same as that of [@mo Theorem 1.3] and [@OgataTRI Proposition 3.5].
To prove Theorem \[c1t\], we first prove a preperatory lemma.
\[hi\] Let $\varphi_1,\varphi_2$ be pure $\Theta$-invariant states on ${{\mathcal A}}$. If $\varphi_1$ and $\varphi_2$ are quasi-equivalent, then $\varphi_1\vert_{{{\mathcal A}}_R}$ and $\varphi_2\vert_{{{\mathcal A}}_R}$ are quasi-equivalent.
Let $\pi_i$, $\pi_{i,R}$ be GNS representations of $\varphi_i$ and $\varphi_i\vert_{{{\mathcal A}}_R}$ respectively for $i=1,2$. By Lemma \[hs\], there are $*$-isomorphisms $\tau_{i}:\pi_i({{\mathcal A}}_R)''\to \pi_{i,R}({{\mathcal A}}_R)''$, for $i=1,2$ such that $\tau_{i}\circ\pi_i(A)=\pi_{i,R}(A)$, $A\in{{\mathcal A}}_R$. Because $\varphi_1$ and $\varphi_2$ are quasi-equivalent, there is a $*$-isomorphism $\tau: \pi_1({{\mathcal A}})''\to\pi_2({{\mathcal A}})''$ such that $\tau\circ\pi_1(A)=\pi_2(A)$, for $A\in{{\mathcal A}}$. The restriction of $\tau$ to $\pi_1({{\mathcal A}}_R)''$ gives a $*$-isomorphism $\tau_R:\pi_1({{\mathcal A}}_R)''\to \pi_2({{\mathcal A}}_R)''$. Hence we obtain a $*$-isomorphism $\hat\tau:=\tau_2\circ\tau_R\circ\tau_1^{-1}: \pi_{1,R}({{\mathcal A}}_R)''\to \pi_{2,R}({{\mathcal A}}_R)''$ such that $\hat\tau\circ\pi_{1,R}(A)=\pi_{2,R}(A)$, $A\in{{\mathcal A}}_R$. Therefore, $\varphi_1\vert_{{{\mathcal A}}_R}$ and $\varphi_2\vert_{{{\mathcal A}}_R}$ are quasi-equivalent.
Now we are ready to prove the Theorem.
Let $({{\mathcal H}}_i,\pi_i,\Omega_i)$ be the GNS triple of the states $\varphi_i\vert_{{{\mathcal A}}_R}$ for $i=0,1$. Let $\Gamma_i$ be a self-adjoint unitary given by $\Gamma_i\pi_i(A)\Omega_i=\pi_i\circ\Theta(A)\Omega_i$, $A\in{{\mathcal A}}_R$. Let $\hat\alpha_i$ be the extension of $\alpha\vert_{{{\mathcal A}}_R}$ to $\pi_i({{\mathcal A}}_R)''$. From Proposition \[wsta\] and Remark \[sds\], it suffices to show that $(\pi_0({{\mathcal A}}_R)'', {\mathop{\mathrm{Ad}}\nolimits}_{\Gamma_0}, \hat\alpha_0)\sim
(\pi_1({{\mathcal A}}_R)'', {\mathop{\mathrm{Ad}}\nolimits}_{\Gamma_1}, \hat\alpha_1)$. Recalling the $\ast$-automorphism $\Xi$ from Proposition \[aep\], $\Xi({{\mathcal A}}_R)={{\mathcal A}}_R$ and so $\Xi_R:=\Xi\vert_{{{\mathcal A}}_R}$ defines a $*$-automorphism on ${{\mathcal A}}_R$. Note that $({{\mathcal H}}_0,\pi_0\circ\Xi_R,\Omega_0)$ is a GNS triple of $\varphi_0\vert_{{{\mathcal A}}_R}\circ\Xi_R$. The state $\varphi_1=\varphi_0\circ{\mathop{\mathrm{Ad}}\nolimits}_u\circ\Xi$ is quasi-equivalent to $\varphi_0\circ\Xi$. Because $\Xi\circ\Theta=\Theta\circ\Xi$, both $\varphi_0\circ\Xi$ and $\varphi_1$ are $\Theta$-invariant pure states. Applying Lemma \[hi\], $\varphi_1\vert_{{{\mathcal A}}_R}$ and $\varphi_0\circ\Xi\vert_{{{\mathcal A}}_R}=\varphi_0\vert_{{{\mathcal A}}_R}\circ\Xi_R$ are quasi-equivalent. Hence there is a $*$-isomorphism $$\begin{aligned}
\tau: \pi_0\circ\Xi_R({{\mathcal A}}_R)''=\pi_0({{\mathcal A}}_R)''\to \pi_1({{\mathcal A}}_R)'', \qquad
\tau\circ\pi_0\circ\Xi_R(A)=\pi_1(A), \quad A \in {{\mathcal A}}_R.\end{aligned}$$ Using properties of the quasi-equivalence $\tau$ and automorphism $\Xi_R$, we see that $$\begin{aligned}
\tau\circ\hat\alpha_{0,g}\circ \pi_0\circ\Xi_R(A)
&=\tau\circ\pi_0\circ\alpha_g\circ \Xi_R(A)
=\tau\circ\pi_0\circ \Xi_R\circ\alpha_g(A) \nonumber \\
&=\pi_1\circ\alpha_g(A)
=\hat\alpha_{1,g}\circ \pi_1(A)
=\hat\alpha_{1,g}\circ \tau\circ\pi_0\circ\Xi_R(A),\\
\tau\circ{\mathop{\mathrm{Ad}}\nolimits}_{\Gamma_0}\circ\pi_0\circ\Xi_R(A)
&=\tau\circ\pi_0\circ\Theta\circ\Xi_R(A)
=\tau\circ\pi_0\circ\Xi_R\circ\Theta(A)\nonumber\\
&=\pi_1\circ\Theta(A)
={\mathop{\mathrm{Ad}}\nolimits}_{\Gamma_1}\circ\pi_1(A)
={\mathop{\mathrm{Ad}}\nolimits}_{\Gamma_1}\circ\tau\circ\pi_0\circ\Xi_R(A)\end{aligned}$$ for all $A\in{{\mathcal A}}_R$. Hence we obtain $$\begin{aligned}
\tau\circ\hat\alpha_{0,g}(x)=\hat\alpha_{1,g}\circ \tau(x),\quad
\tau\circ{\mathop{\mathrm{Ad}}\nolimits}_{\Gamma_0}(x)=
{\mathop{\mathrm{Ad}}\nolimits}_{\Gamma_1}\circ\tau(x),\quad
x\in \pi_0({{\mathcal A}}_R)''.\end{aligned}$$ This completes the proof.
Stacking and group law of fermionic SPT phases {#stacksec}
==============================================
The graded tensor product {#subsec:graded_product_def}
-------------------------
Let $({{\mathcal M}}_1,{\mathop{\mathrm{Ad}}\nolimits}_{\Gamma_1})$ and $({{\mathcal M}}_2,{\mathop{\mathrm{Ad}}\nolimits}_{\Gamma_2})$ be spatially graded von Neumann algebras acting on on ${{\mathcal H}}_1$, ${{\mathcal H}}_2$ with grading operators $\Gamma_1$, $\Gamma_2$. We define a product and involution on the algebraic tensor product ${{\mathcal M}}_1\odot {{\mathcal M}}_2$ by $$\begin{aligned}
(a_1{\,\hat{\otimes}\,}b_1)(a_2{\,\hat{\otimes}\,}b_2)
&=(-1)^{\partial b_1\partial a_2}(a_1a_2{\,\hat{\otimes}\,}b_1b_2), \nonumber \\
(a{\,\hat{\otimes}\,}b)^* &=(-1)^{\partial a\partial b} a^*{\,\hat{\otimes}\,}b^*.
$$ for homogeneous elementary tensors. The algebraic tensor product with this multiplication and involution is a $*$-algebra, denoted ${{\mathcal M}}_1\,\hat{\odot}\,{{\mathcal M}}_2$. On the Hilbert space ${{\mathcal H}}_1\otimes{{\mathcal H}}_2$, $$\begin{aligned}
\label{pprep}
\pi (a{\,\hat{\otimes}\,}b)
:=a\Gamma_1^{\partial b}\otimes b\end{aligned}$$ for homogeneous $a\in {{\mathcal M}}_1$, $b\in{{\mathcal M}}_2$ defines a faithful $*$-representation of ${{\mathcal M}}_1\,\hat{\odot}\,{{\mathcal M}}_2$. We call the von Neumann algebra generated by $\pi({{\mathcal M}}_1\,\hat{\odot}\,{{\mathcal M}}_2)$ the graded tensor product of $({{\mathcal M}}_1,{{\mathcal H}}_1,\Gamma_1)$ and $({{\mathcal M}}_2,{{\mathcal H}}_2,\Gamma_2)$ and denote it by $\calM_1 {\,\hat{\otimes}\,}\calM_2$. It is simple to check that $\calM_1 {\,\hat{\otimes}\,}\calM_2$ is a spatially graded von Neumann algebra with a grading operator $\Gamma_1\otimes \Gamma_2$.
For $a\in {{\mathcal M}}_1$ and homogeneous $b\in{{\mathcal M}}_2$, we denote $\pi(a{\,\hat{\otimes}\,}b)$ by $a{{\,\hat{\otimes}\,}} b$, embedding ${{\mathcal M}}_1\,\hat{\odot}\,{{\mathcal M}}_2$ in ${{\mathcal M}}_1{\,\hat{\otimes}\,}{{\mathcal M}}_2$. Note that $\partial(a{\,\hat{\otimes}\,}b) = \partial(a) + \partial(b)$ for homogeneous $a\in {{\mathcal M}}_1$ and $b\in {{\mathcal M}}_2$.
Fix a finite group $G$ and a homomorphism ${\mathfrak{p}}: G\to{{\mathbb Z}}_2$. Let $({{\mathcal M}}_1,{\mathop{\mathrm{Ad}}\nolimits}_{\Gamma_1},\alpha_1)$ and $({{\mathcal M}}_2,{\mathop{\mathrm{Ad}}\nolimits}_{\Gamma_2},\alpha_2)$ be graded $W^*$-$(G,{\mathfrak{p}})$-dynamical systems, where $({{\mathcal M}}_1, \mathrm{Ad}_{\Gamma_1})$ and $({{\mathcal M}}_2,\mathrm{Ad}_{\Gamma_2})$ are spatially graded, balanced, central and type I. We may define an action $\alpha_1{\,\hat{\otimes}\,}\alpha_2$ of $G$ on ${{\mathcal M}}_1{\,\hat{\otimes}\,}{{\mathcal M}}_2$ by $$\begin{aligned}
{\left (}\alpha_1{\,\hat{\otimes}\,}\alpha_2{\right )}_g
(a{\,\hat{\otimes}\,}b)
=\alpha_{1,g}(a){{\,\hat{\otimes}\,}} \alpha_{2,g}(b),\quad g\in G\end{aligned}$$ for all homogeneous $a\in {{\mathcal M}}_1$ and $b\in{{\mathcal M}}_2$, see Lemma \[ga8\].
Stacking and the group law
--------------------------
In this section, we show that $W^*$-$(G,{\mathfrak{p}})$-dynamical systems of balanced, central, type I and spatially graded von Neumann algebras are closed under graded tensor products. Furthermore, our index from Definition \[index\] obeys a twisted group law (a generalized Wall group law) under this operation.
\[stlaw\] Let $({{\mathcal M}}_1,{\mathop{\mathrm{Ad}}\nolimits}_{\Gamma_1},\alpha_1)$, $({{\mathcal M}}_2,{\mathop{\mathrm{Ad}}\nolimits}_{\Gamma_2},\alpha_2)$ be graded $W^*$-$(G,{\mathfrak{p}})$-dynamical systems with balanced, central and spatially graded type I von Neumann algebras. Then the triple $({{\mathcal M}}_1{\,\hat{\otimes}\,}{{\mathcal M}}_2, {\mathop{\mathrm{Ad}}\nolimits}_{\Gamma_1\otimes\Gamma_2}, \alpha_1{\,\hat{\otimes}\,}\alpha_2)$ is a graded $W^*$-$(G,{\mathfrak{p}})$-dynamical system with a balanced, central and spatially graded type I von Neumann algebra. If ${\mathop{\mathrm{Ind}}\nolimits}({{\mathcal M}}_i,{\mathop{\mathrm{Ad}}\nolimits}_{\Gamma_i},\alpha_i)=(\kappa_i,{\mathfrak{q}}_i,[\upsilon_i])$, $i=1,2$, then $$\begin{aligned}
\label{totind}
{\mathop{\mathrm{Ind}}\nolimits}({{\mathcal M}}_1{\,\hat{\otimes}\,}{{\mathcal M}}_2, {\mathop{\mathrm{Ad}}\nolimits}_{\Gamma_1\otimes\Gamma_2}, \alpha_1{\,\hat{\otimes}\,}\alpha_2)
=\big(\kappa_1+\kappa_2,\, {\mathfrak{q}}_1+{\mathfrak{q}}_2+\kappa_1\kappa_2{\mathfrak{p}}, \, [\upsilon_1\, \upsilon_2 \, \epsilon_{\mathfrak{p}}(\kappa_1,{\mathfrak{q}}_1,\kappa_2,{\mathfrak{q}}_2)] \big),\end{aligned}$$ where $\epsilon_{\mathfrak{p}}(\kappa_1,{\mathfrak{q}}_1,\kappa_2,{\mathfrak{q}}_2)$ is a group $2$-cocycle defined by $$\begin{aligned}
\epsilon_{\mathfrak{p}}(\kappa_1,{\mathfrak{q}}_1,\kappa_2,{\mathfrak{q}}_2)(g,h)
=(-1)^{{{\mathfrak{q}}_1(g)}{\mathfrak{q}}_2(h)+(\kappa_1-\kappa_2)(\kappa_1{\mathfrak{q}}_2(g)+\kappa_2{{\mathfrak{q}}_1(g)})\cdot{{\mathfrak{p}}(h)}},\quad
g,h\in G.\end{aligned}$$
1. One can check that gives an abelian group law, which is not surprising because of the corresponding properties of the graded tensor product.
2. The group law is a little cumbersome in full generality, but simplifies in many examples of interest. For example, if $\alpha$ is a linear group action only, ${\mathfrak{p}}(g)=0$ for all $g\in G$, we recover the more familiar twisted sum formula, $$\begin{aligned}
(\kappa_1,\, {\mathfrak{q}}_1,\, [\upsilon_1]) \cdot (\kappa_2,\, {\mathfrak{q}}_2,\, [\upsilon_2])
= (\kappa_1 +\kappa_2,\, {\mathfrak{q}}_1+{\mathfrak{q}}_2,\, [\upsilon_1\, \upsilon_2 \, \epsilon({\mathfrak{q}}_1,{\mathfrak{q}}_2)] ).
\end{aligned}$$
By Lemma \[oisoiso\] and Lemma \[wsta\], we may assume that $$\begin{aligned}
({{\mathcal M}}_i,{\mathop{\mathrm{Ad}}\nolimits}_{\Gamma_i},\alpha_i)=({{\mathcal R}}_{\kappa_i, {{\mathcal K}}_i},{\mathop{\mathrm{Ad}}\nolimits}_{\Gamma_{{{\mathcal K}}_i}},{\mathop{\mathrm{Ad}}\nolimits}_{V_{ i}})\in{{\mathcal S}}_{\kappa_i}.\end{aligned}$$ Let $$\begin{aligned}
{\mathop{\mathrm{Ind}}\nolimits}\big( {{\mathcal R}}_{\kappa_i,{{\mathcal K}}_i},{\mathop{\mathrm{Ad}}\nolimits}_{\Gamma_{{{\mathcal K}}_i}},{\mathop{\mathrm{Ad}}\nolimits}_{V_{ i}} \big)
=(\kappa_i,{\mathfrak{q}}_i, [\upsilon_i]),\quad i=1,2.\end{aligned}$$ We would like to show that $$\begin{aligned}
\label{eqvrr}
{\left (}{{\mathcal R}}_{\kappa_1,{{\mathcal K}}_1}{\,\hat{\otimes}\,}{{\mathcal R}}_{\kappa_2, {{\mathcal K}}_2},
{\mathop{\mathrm{Ad}}\nolimits}_{\Gamma_{{{\mathcal K}}_1}}{{\,\hat{\otimes}\,}}{\mathop{\mathrm{Ad}}\nolimits}_{\Gamma_{{{\mathcal K}}_1}},
{\mathop{\mathrm{Ad}}\nolimits}_{ V_{1}}{\,\hat{\otimes}\,}{\mathop{\mathrm{Ad}}\nolimits}_{V_{2}}
{\right )}\sim
({{\mathcal R}}_{\kappa, {{\mathcal K}}},{\mathop{\mathrm{Ad}}\nolimits}_{\Gamma_{{{\mathcal K}}}},{\mathop{\mathrm{Ad}}\nolimits}_{V})\in{{\mathcal S}}_{\kappa},\end{aligned}$$ for suitably chosen $\kappa=0,1$, Hilbert space ${{\mathcal K}}$ and projective representation $V$ on ${{\mathcal K}}\otimes{{{\mathbb C}}^2}$, satisfying $$\begin{aligned}
\label{tind}
{\mathop{\mathrm{Ind}}\nolimits}({{\mathcal R}}_{\kappa, {{\mathcal K}}},{\mathop{\mathrm{Ad}}\nolimits}_{\Gamma_{{{\mathcal K}}}},{\mathop{\mathrm{Ad}}\nolimits}_{V_{g}})
=\big(\kappa_1+\kappa_2,\, {\mathfrak{q}}_1+{\mathfrak{q}}_2+\kappa_1\kappa_2{\mathfrak{p}}, \, [\upsilon_1\, \upsilon_2 \,
\epsilon_{\mathfrak{p}}(\kappa_1,{\mathfrak{q}}_1,\kappa_2,{\mathfrak{q}}_2)] \big).\end{aligned}$$
[**(Case: $\kappa_1=0$ or $\kappa_2=0$)** ]{}
We set the following notation, $$\begin{aligned}
&{{\mathcal K}}:= {{\mathcal K}}_1\otimes{{\mathcal K}}_2\otimes{{{\mathbb C}}^2},
&&\lambda = \begin{cases} 1, &\text{if}\,\, \kappa_1=\kappa_2 = 0, \\ 2, &\text{if}\,\, \kappa_1=1, \,\, \kappa_2 = 0, \\
3, & \text{if}\,\,\kappa_1 = 0, \,\, \kappa_2 = 1, \end{cases}\end{aligned}$$ and define the unitary $v:{{\mathbb C}}^2\otimes {{\mathcal K}}_2\to {{\mathcal K}}_2\otimes {{{\mathbb C}}^2}$, $$\begin{aligned}
\label{swap}
v(\xi\otimes \eta)=\eta\otimes \xi,\quad \xi\in {{{\mathbb C}}^2},\quad \eta\in{{\mathcal K}}_2.\end{aligned}$$
Using the standard basis $\{e_0,e_1\}$ of ${{{\mathbb C}}^2}$, we define the unitaries $w_1,w_2,w_3$ on ${{{\mathbb C}}^2}\otimes{{{\mathbb C}}^2}$ by $$\begin{aligned}
&w_1(e_0\otimes e_0)=e_0\otimes e_0,
&&w_1(e_1\otimes e_1)=e_1\otimes e_0,
&&w_1(e_1\otimes e_0)=e_0\otimes e_1,
&&w_1(e_0\otimes e_1)=e_1\otimes e_1,\\
&w_2(e_0\otimes e_0)=e_0\otimes e_0,
&&w_2(e_1\otimes e_1)=e_1\otimes e_0,
&&w_2(e_1\otimes e_0)=e_0\otimes e_1,
&&w_2(e_0\otimes e_1)=-e_1\otimes e_1,\\
&w_3(e_0\otimes e_0)=e_0\otimes e_0,
&&w_3(e_1\otimes e_1)=e_1\otimes e_0,
&&w_3(e_1\otimes e_0)=e_1\otimes e_1,
&&w_3(e_0\otimes e_1)=e_0\otimes e_1.\end{aligned}$$ By direct calculation, we may check $$\begin{aligned}
&{\mathop{\mathrm{Ad}}\nolimits}_{w_\lambda}{\left (}\sigma_z\otimes\sigma_z{\right )}={\mathbb I}_{{{{\mathbb C}}^2}}\otimes\sigma_z,\quad \lambda=1,2,3,\label{ttz}\\
&{\mathop{\mathrm{Ad}}\nolimits}_{w_2}{\left (}\sigma_x\otimes\sigma_z{\right )}={\mathbb I}_{{{{\mathbb C}}^2}}\otimes\sigma_x,\quad
{\mathop{\mathrm{Ad}}\nolimits}_{w_3}{\left (}{\mathbb I}_{{{{\mathbb C}}^2}}\otimes\sigma_x{\right )}={\mathbb I}_{{{{\mathbb C}}^2}}\otimes\sigma_x.\label{ttx}\end{aligned}$$
We now define unitary $U_\lambda:{{\mathcal K}}_1\otimes{{{\mathbb C}}^2}\otimes{{\mathcal K}}_2\otimes {{{\mathbb C}}^2}\to
{{\mathcal K}}\otimes {{{\mathbb C}}^2}$ such that $$\begin{aligned}
U_\lambda:=({\mathbb I}_{{{\mathcal K}}_1}\otimes {\mathbb I}_{{{\mathcal K}}_2}\otimes w_\lambda)({\mathbb I}_{{{\mathcal K}}_1}\otimes v\otimes {\mathbb I}_{{{{\mathbb C}}^2}}),\quad\lambda=1,2,3.\end{aligned}$$ By , we have $$\begin{aligned}
\label{uttz}
{\mathop{\mathrm{Ad}}\nolimits}_{U_\lambda}{\left (}\Gamma_{{{\mathcal K}}_1}\otimes \Gamma_{{{\mathcal K}}_2}{\right )}=\Gamma_{{{\mathcal K}}},\quad\lambda=1,2,3,\end{aligned}$$ hence $$\begin{aligned}
\label{tu}
{\mathop{\mathrm{Ad}}\nolimits}_{U_\lambda}\circ{\left (}{\mathop{\mathrm{Ad}}\nolimits}_{\Gamma_{{{\mathcal K}}_1}}{\,\hat{\otimes}\,}{\mathop{\mathrm{Ad}}\nolimits}_{\Gamma_{{{\mathcal K}}_2}}{\right )}(x)
={\mathop{\mathrm{Ad}}\nolimits}_{\Gamma_{{{\mathcal K}}}}\circ{\mathop{\mathrm{Ad}}\nolimits}_{U_\lambda}(x),
\quad x\in {{\mathcal R}}_{\kappa_1, {{\mathcal K}}_1}{\,\hat{\otimes}\,}{{\mathcal R}}_{\kappa_2,{{\mathcal K}}_2},\quad
\lambda=1,2,3.\end{aligned}$$ By , when $\lambda=2$, for ${\left (}{\mathbb I}_{{{\mathcal K}}_1}\otimes \sigma_x{\right )}{{\,\hat{\otimes}\,}}
{\left (}{\mathbb I}_{{{\mathcal K}}_2}\otimes \sigma_z{\right )}\in
{{\mathcal R}}_{\kappa_1,{{\mathcal K}}_1}{{\,\hat{\otimes}\,}} {{\mathcal R}}_{\kappa_2,{{\mathcal K}}_2}
$, we have $$\begin{aligned}
\label{uttx2}
{\mathop{\mathrm{Ad}}\nolimits}_{U_2}{\left (}{\left (}{\mathbb I}_{{{\mathcal K}}_1}\otimes \sigma_x{\right )}{{\,\hat{\otimes}\,}}
{\left (}{\mathbb I}_{{{\mathcal K}}_2}\otimes \sigma_z{\right )}{\right )}={\mathbb I}_{{{\mathcal K}}}\otimes \sigma_x.\end{aligned}$$ Similarly, when $\lambda=3$, for $
{\left (}{\mathbb I}_{{{\mathcal K}}_1}\otimes \sigma_z{\right )}{{\,\hat{\otimes}\,}}
{\left (}{\mathbb I}_{{{\mathcal K}}_2}\otimes \sigma_x{\right )}\in{{\mathcal R}}_{\kappa_1, {{\mathcal K}}_1}{{\,\hat{\otimes}\,}} {{\mathcal R}}_{\kappa_2, {{\mathcal K}}_2}
$, $$\begin{aligned}
\label{uttx3}
\quad
{\mathop{\mathrm{Ad}}\nolimits}_{U_3}{\left (}{\left (}{\mathbb I}_{{{\mathcal K}}_1}\otimes \sigma_z{\right )}{{\,\hat{\otimes}\,}}
{\left (}{\mathbb I}_{{{\mathcal K}}_2}\otimes \sigma_x{\right )}{\right )}={\mathbb I}_{{{\mathcal K}}}\otimes \sigma_x.\end{aligned}$$
Let $[\tilde \upsilon_i]$ be the second cohomology class associated to the projective representation $V_{i}$, $i=1,2$. We set $$\begin{aligned}
V_g:={\mathop{\mathrm{Ad}}\nolimits}_{U_\lambda}{\left (}V_{1,g}\otimes V_{2,g}\Gamma_{{{\mathcal K}}_2}^{{{\mathfrak{q}}_1(g)}}{\right )},\quad g\in G,\quad\lambda=1,2,3.\end{aligned}$$ This gives a projective unitary/anti-unitary representation $V$ of $G$ on ${{\mathcal K}}\otimes{{{\mathbb C}}^2}$ relative to ${\mathfrak{p}}$. Using that ${\mathop{\mathrm{Ad}}\nolimits}_{V_{2,g}}{\left (}\Gamma_{{{\mathcal K}}_2}{\right )}=(-1)^{{\mathfrak{q}}_2(g)}\Gamma_{{{\mathcal K}}_2}$ for $g\in G$, the second cohomology class associated to $V$ is equal to $[\tilde \upsilon_1 \,\tilde \upsilon_2 \, \epsilon({\mathfrak{q}}_1,{\mathfrak{q}}_2)] \in H^2(G, U(1)_{\mathfrak{p}})$, where $\epsilon({\mathfrak{q}}_1,{\mathfrak{q}}_2)$ is given in . By Lemma \[isoiso\] and \[aisoaiso\], we have that for $x\in {{\mathcal R}}_{\kappa_1, {{\mathcal K}}_1}{{\,\hat{\otimes}\,}} {{\mathcal R}}_{\kappa_2, {{\mathcal K}}_2}$, $g\in G$ and any $\lambda=1,2,3$, $$\begin{aligned}
\label{uv}
{\mathop{\mathrm{Ad}}\nolimits}_{V_g}\circ{\mathop{\mathrm{Ad}}\nolimits}_{U_\lambda}(x)={\mathop{\mathrm{Ad}}\nolimits}_{U_\lambda}\circ{\mathop{\mathrm{Ad}}\nolimits}_{ V_{1,g}\otimes V_{2,g}\Gamma_{{{\mathcal K}}_2}^{{{\mathfrak{q}}_1(g)}}}(x)
={\mathop{\mathrm{Ad}}\nolimits}_{U_\lambda}\circ{\left (}\alpha_{1,g}{{\,\hat{\otimes}\,}} \alpha_{2,g}{\right )}{\left (}x
{\right )}.
$$ In particular, for $\lambda=2,3$, we also have $$\begin{aligned}
\label{uttxl}
{\mathop{\mathrm{Ad}}\nolimits}_{V_g} ({\mathbb I}_{{{\mathcal K}}}\otimes \sigma_x)=
(-1)^{{{\mathfrak{q}}_1(g)}+{\mathfrak{q}}_2(g)}
{\left (}{\mathbb I}_{{{\mathcal K}}}\otimes \sigma_x{\right )},\quad g\in G,\end{aligned}$$ from and .
By , we have $$\begin{aligned}
\label{vzz}
{\mathop{\mathrm{Ad}}\nolimits}_{V_g}{\left (}\Gamma_{{\mathcal K}}{\right )}={\mathop{\mathrm{Ad}}\nolimits}_{U_\lambda}\circ{\mathop{\mathrm{Ad}}\nolimits}_{ V_{1,g}\otimes V_{2,g}\Gamma_{{{\mathcal K}}_2}^{{{\mathfrak{q}}_1(g)}}}{\left (}\Gamma_{{{\mathcal K}}_1}\otimes\Gamma_{{{\mathcal K}}_2}
{\right )}=(-1)^{{{\mathfrak{q}}_1(g)}+{\mathfrak{q}}_2(g)}\Gamma_{{\mathcal K}},\quad g\in G.\end{aligned}$$
Having set up the required preliminaries, we now consider the $W^*$-$(G,{\mathfrak{p}})$-dynamical system $({{\mathcal R}}_{\kappa, {{\mathcal K}}},{\mathop{\mathrm{Ad}}\nolimits}_{\Gamma_{{{\mathcal K}}}},{\mathop{\mathrm{Ad}}\nolimits}_{V_{g}})\in {{\mathcal S}}_{\kappa}$ and show equivalence with the graded tensor product in the three cases where $\kappa_1$ or $\kappa_2=0$.
[(i)-1]{} For $\lambda=1$ (i.e. $\kappa_1=\kappa_2=0$), we set $\kappa=0$ and note from that $({{\mathcal R}}_{\kappa, {{\mathcal K}}},{\mathop{\mathrm{Ad}}\nolimits}_{\Gamma_{{{\mathcal K}}}},{\mathop{\mathrm{Ad}}\nolimits}_{V_{g}})\in {{\mathcal S}}_{\kappa}$. In this case, $[\tilde \upsilon_i]= [\upsilon_i]$ and $\epsilon_{\mathfrak{p}}(0,{\mathfrak{q}}_1,0,{\mathfrak{q}}_2)=\epsilon({\mathfrak{q}}_1,{\mathfrak{q}}_2)$. Hence the second cohomology class of $V$ is $[\upsilon_1\upsilon_2 \epsilon_{\mathfrak{p}}(0,{\mathfrak{q}}_1,0,{\mathfrak{q}}_2)]$. With this and (\[vzz\]), the index of $({{\mathcal R}}_{\kappa, {{\mathcal K}}},{\mathop{\mathrm{Ad}}\nolimits}_{\Gamma_{{{\mathcal K}}}},{\mathop{\mathrm{Ad}}\nolimits}_{V_{g}})$ is given by . So we just need to show equivalence of the $W^*$-$(G,{\mathfrak{p}})$-dynamical system with the graded tensor product. The equivalence is given by a $*$-isomorphism $$\begin{aligned}
\iota:={\mathop{\mathrm{Ad}}\nolimits}_{U_1}: {{\mathcal B}}({{\mathcal K}}_1\otimes{{{\mathbb C}}^2}\otimes{{\mathcal K}}_2\otimes{{{\mathbb C}}^2})
={{\mathcal R}}_{\kappa_1, {{\mathcal K}}_1}{{\,\hat{\otimes}\,}} {{\mathcal R}}_{\kappa_2, {{\mathcal K}}_2}\to
{{\mathcal B}}({{\mathcal K}}\otimes{{{\mathbb C}}^2})={{\mathcal R}}_{0,{{\mathcal K}}}.\end{aligned}$$ By and , $\iota$ satisfies the required conditions and for equivalence of $W^*$-$(G,{\mathfrak{p}})$-dynamical systems.
[(i)-2]{} For $\lambda=2$ (i.e. $\kappa_1=1,\, \kappa_2=0$), set $\kappa=1$. By and , we see that $({{\mathcal R}}_{\kappa, {{\mathcal K}}},{\mathop{\mathrm{Ad}}\nolimits}_{\Gamma_{{{\mathcal K}}}},{\mathop{\mathrm{Ad}}\nolimits}_{V_{g}})\in {{\mathcal S}}_{\kappa}$. Note that $[\tilde \upsilon_1]= [\upsilon_1\, \epsilon ({\mathfrak{q}}_1,{\mathfrak{p}})] \in {H^2(G,U(1)_{\mathfrak{p}})}$, see Lemma \[zv\] and Definition \[index\], with $\tilde \upsilon_2=\upsilon_2$. Hence the second cohomology associated to our projective representation $V$ is $$\begin{aligned}
\big[ \tilde \upsilon_1 \, \tilde \upsilon_2 \, \epsilon({\mathfrak{q}}_1,{\mathfrak{q}}_2)\big] =
\big[ \upsilon_1 \,\upsilon_2 \,\epsilon ({\mathfrak{q}}_1,{\mathfrak{p}}) \, \epsilon({\mathfrak{q}}_1,{\mathfrak{q}}_2)\big].\end{aligned}$$ Combining this and , the second cohomology associated to the projective representation $V^{(0)}$ (cf. Lemma \[zv\] and Definition \[index\]) is $$\begin{aligned}
\big[\tilde \upsilon_1 \,\tilde \upsilon_2 \, \epsilon({\mathfrak{q}}_1,{\mathfrak{q}}_2) \,\epsilon({\mathfrak{q}}_1+{\mathfrak{q}}_2,{\mathfrak{p}}) \big]
= \big[ \upsilon_1 \,\upsilon_2 \,\epsilon ({\mathfrak{q}}_1,{\mathfrak{p}}) \, \epsilon({\mathfrak{q}}_1,{\mathfrak{q}}_2)
\, \epsilon({\mathfrak{q}}_1+{\mathfrak{q}}_2,{\mathfrak{p}}) \big]
= \big[\upsilon_1 \,\upsilon_2 \,\epsilon_{\mathfrak{p}}(1,{\mathfrak{q}}_1,0,{\mathfrak{q}}_2) \big].\end{aligned}$$ From this and , we see the index of $({{\mathcal R}}_{\kappa, {{\mathcal K}}},{\mathop{\mathrm{Ad}}\nolimits}_{\Gamma_{{{\mathcal K}}}},{\mathop{\mathrm{Ad}}\nolimits}_{V_{g}})\in {{\mathcal S}}_{\kappa}$ is given by .
Now we show $({{\mathcal R}}_{\kappa, {{\mathcal K}}},{\mathop{\mathrm{Ad}}\nolimits}_{\Gamma_{{{\mathcal K}}}},{\mathop{\mathrm{Ad}}\nolimits}_{V_{g}})$ is equivalent to the graded tensor product (\[eqvrr\]). From Lemma \[comgra\], the commutant of ${{\mathcal R}}_{\kappa_1, {{\mathcal K}}_1}{{\,\hat{\otimes}\,}} {{\mathcal R}}_{\kappa_2,{{\mathcal K}}_2}$ is ${{\mathbb C}}{\mathbb I}_{{{\mathcal K}}_1\otimes{{{\mathbb C}}^2}\otimes{{\mathcal K}}_2\otimes{{{\mathbb C}}^2}}+{{\mathbb C}}{\mathbb I}_{{{\mathcal K}}_1}\otimes\sigma_x\otimes{\mathbb I}_{{{\mathcal K}}_2}\otimes\sigma_z$. Note that by , ${\mathop{\mathrm{Ad}}\nolimits}_{U_2}$ maps the commutant to ${{\mathbb C}}{\mathbb I}_{{{\mathcal K}}}\otimes {\mathbb I}_{{{{\mathbb C}}^2}}+{{\mathbb C}}{\mathbb I}_{{{\mathcal K}}}\otimes \sigma_x=({{\mathcal R}}_{\kappa,{{\mathcal K}}})'$. Therefore we have ${\mathop{\mathrm{Ad}}\nolimits}_{U_2}({{\mathcal R}}_{\kappa_1, {{\mathcal K}}_1}{{\,\hat{\otimes}\,}}{{\mathcal R}}_{\kappa_2, {{\mathcal K}}_2})
={{\mathcal R}}_{\kappa, {{\mathcal K}}}$. Hence $\iota:={\mathop{\mathrm{Ad}}\nolimits}_{U_2}\vert_{{{\mathcal R}}_{\kappa_1, {{\mathcal K}}_1}{{\,\hat{\otimes}\,}}{{\mathcal R}}_{\kappa_2,{{\mathcal K}}_2}}$ defines a $*$-isomorphism $\iota: {{\mathcal R}}_{\kappa_1, {{\mathcal K}}_1}{{\,\hat{\otimes}\,}} {{\mathcal R}}_{\kappa_2,{{\mathcal K}}_2}\to {{\mathcal R}}_{\kappa,{{\mathcal K}}}$. By and , $\iota$ satisfies the required conditions of an equivalence of $W^*$-$(G,{\mathfrak{p}})$-dynamical systems.
[(i)-3]{} For $\lambda=3$ (i.e. $\kappa_1=0,\, \kappa_2=1$), we set $\kappa=1$. By and , we see that $({{\mathcal R}}_{\kappa, {{\mathcal K}}},{\mathop{\mathrm{Ad}}\nolimits}_{\Gamma_{{{\mathcal K}}}},{\mathop{\mathrm{Ad}}\nolimits}_{V_{g}})\in {{\mathcal S}}_{\kappa}$. We also have that $[\tilde \upsilon_1]=[\upsilon_1]$ and $[\tilde \upsilon_2]=[\upsilon_2\,\epsilon ({\mathfrak{q}}_2,{\mathfrak{p}})]$. Hence the second cohomology class associated to $V$ is $$\begin{aligned}
\big[ \tilde \upsilon_1 \,\tilde \upsilon_2 \,\epsilon({\mathfrak{q}}_1,{\mathfrak{q}}_2)\big] =
\big[ \upsilon_1 \,\upsilon_2 \, \epsilon ({\mathfrak{q}}_2,{\mathfrak{p}}) \, \epsilon({\mathfrak{q}}_1,{\mathfrak{q}}_2)\big]
.\end{aligned}$$ Hence, from the cohomology class associated to $V^{(0)}$ is $$\begin{aligned}
\big[ \tilde \upsilon_1\, \tilde \upsilon_2 \, \epsilon({\mathfrak{q}}_1,{\mathfrak{q}}_2)
\, \epsilon({\mathfrak{q}}_1+{\mathfrak{q}}_2,{\mathfrak{p}})\big]
= \big[ \upsilon_1 \, \upsilon_2 \, \epsilon_{\mathfrak{p}}(0,{\mathfrak{q}}_1,1,{\mathfrak{q}}_2)\big]\end{aligned}$$ and the index of $({{\mathcal R}}_{\kappa, {{\mathcal K}}},{\mathop{\mathrm{Ad}}\nolimits}_{\Gamma_{{{\mathcal K}}}},{\mathop{\mathrm{Ad}}\nolimits}_{V_{g}})\in {{\mathcal S}}_{\kappa}$ is given by .
We now show that $({{\mathcal R}}_{\kappa, {{\mathcal K}}},{\mathop{\mathrm{Ad}}\nolimits}_{\Gamma_{{{\mathcal K}}}},{\mathop{\mathrm{Ad}}\nolimits}_{V_{g}})$ is equivalent to the graded tensor product. From Lemma \[comgra\], the commutant of ${{\mathcal R}}_{\kappa_1,{{\mathcal K}}_1}{{\,\hat{\otimes}\,}}{{\mathcal R}}_{\kappa_2,{{\mathcal K}}_2}$ is ${{\mathbb C}}{\mathbb I}_{{{\mathcal K}}_1\otimes{{{\mathbb C}}^2}\otimes{{\mathcal K}}_2\otimes{{{\mathbb C}}^2}}+
{{\mathbb C}}{\mathbb I}_{{{\mathcal K}}_1\otimes{{{\mathbb C}}^2}\otimes{{\mathcal K}}_2}
\otimes\sigma_x$, which by is mapped to ${{\mathbb C}}{\mathbb I}_{{{\mathcal K}}}\otimes {\mathbb I}_{{{{\mathbb C}}^2}}+{{\mathbb C}}{\mathbb I}_{{{\mathcal K}}}\otimes \sigma_x=({{\mathcal R}}_{\kappa,{{\mathcal K}}})'$ by ${\mathop{\mathrm{Ad}}\nolimits}_{U_3}$. Therefore, ${\mathop{\mathrm{Ad}}\nolimits}_{U_3}({{\mathcal R}}_{\kappa_1, {{\mathcal K}}_1}{{\,\hat{\otimes}\,}}{{\mathcal R}}_{\kappa_2, {{\mathcal K}}_2})
={{\mathcal R}}_{\kappa,{{\mathcal K}}}$ and $\iota:={\mathop{\mathrm{Ad}}\nolimits}_{U_3}\vert_{{{\mathcal R}}_{\kappa_1,{{\mathcal K}}_1}{{\,\hat{\otimes}\,}}{{\mathcal R}}_{\kappa_2,{{\mathcal K}}_2}}$ defines a $*$-isomorphism $\iota: {{\mathcal R}}_{\kappa_1,{{\mathcal K}}_1}{{\,\hat{\otimes}\,}}{{\mathcal R}}_{\kappa_2,{{\mathcal K}}_2}\to {{\mathcal R}}_{\kappa,{{\mathcal K}}}$ and implements an equivalence of $W^*$-$(G,{\mathfrak{p}})$-dynamical systems.
[**(Case: $\kappa_1=\kappa_2=1$)** ]{}
Set $\kappa:=0$ and ${{\mathcal K}}:={{\mathcal K}}_1\otimes{{\mathcal K}}_2$. We define a projective representation $V$ of $G$ on ${{\mathcal K}}\otimes{{{\mathbb C}}^2}$ relative to ${\mathfrak{p}}$ by $$\begin{aligned}
V_g:=V_{1, g}^{(0)}\otimes V_{2, g}^{(0)}\otimes C^{{{\mathfrak{p}}(g)}}\sigma_y^{{{\mathfrak{q}}_1(g)}}\sigma_x^{{\mathfrak{q}}_2(g)+{{\mathfrak{p}}(g)}},\quad
g\in G.\end{aligned}$$ Here $V_{i}^{(0)}$ is the projective representation on ${{\mathcal K}}_1$ such that $V_{i,g}=V_{i,g}^{(0)}\otimes C^{{{\mathfrak{p}}(g)}}\sigma_y^{{\mathfrak{q}}_i(g)}$ for $i=1,2$ (see Lemma \[zv\]). Then we have $$\begin{aligned}
{\mathop{\mathrm{Ad}}\nolimits}_{V_g}{\left (}{\Gamma_{{\mathcal K}}}{\right )}= (-1)^{{{\mathfrak{q}}_1(g)}+{\mathfrak{q}}_2(g)+{{\mathfrak{p}}(g)}}{\Gamma_{{\mathcal K}}},\quad
g\in G.\end{aligned}$$Hence $({{\mathcal R}}_{\kappa, {{\mathcal K}}},{\mathop{\mathrm{Ad}}\nolimits}_{\Gamma_{{{\mathcal K}}}},{\mathop{\mathrm{Ad}}\nolimits}_{V_{g}})\in{{\mathcal S}}_\kappa$. Because $\sigma_y$ anti-commutes with $\sigma_x$ and $C$, while $\sigma_x$ commutes with $C$, the second cohomology class associate to the projective representation $V$ is $$\begin{aligned}
\big[ \upsilon_1 \, \upsilon_2 \, \epsilon ({\mathfrak{q}}_1,{\mathfrak{q}}_2)\big]
= \big[ \upsilon_1 \, \upsilon_2 \, \epsilon_{\mathfrak{p}}(1,{\mathfrak{q}}_1,1,{\mathfrak{q}}_2)\big],\end{aligned}$$ where we recall that $[\epsilon ({\mathfrak{q}}_1,{\mathfrak{q}}_2)] = [\epsilon ({\mathfrak{q}}_2,{\mathfrak{q}}_1)]$. Hence the triple $({{\mathcal R}}_{\kappa, {{\mathcal K}}},{\mathop{\mathrm{Ad}}\nolimits}_{\Gamma_{{{\mathcal K}}}},{\mathop{\mathrm{Ad}}\nolimits}_{V_{g}})\in {{\mathcal S}}_{\kappa}$ has index given by .
Now we show for the constructed $({{\mathcal R}}_{\kappa, {{\mathcal K}}},{\mathop{\mathrm{Ad}}\nolimits}_{\Gamma_{{{\mathcal K}}}},{\mathop{\mathrm{Ad}}\nolimits}_{V_{g}})$. Regarding ${\mathop{\mathfrak{C}}\nolimits}$ as a graded von Neumann algebra $({\mathop{\mathfrak{C}}\nolimits},{\mathop{\mathrm{Ad}}\nolimits}_{\sigma_z})\subset {\mathop{\mathrm{M}}\nolimits}_2$, there is a $\ast$-isomorphism $\iota_0:{\mathop{\mathfrak{C}}\nolimits}{{\,\hat{\otimes}\,}}{\mathop{\mathfrak{C}}\nolimits}\to{\mathop{\mathrm{M}}\nolimits}_2$ such that $$\begin{aligned}
\label{ioz}
\iota_0({\mathbb I}{{\,\hat{\otimes}\,}}{\mathbb I})={\mathbb I},\quad
\iota_0(\sigma_x{{\,\hat{\otimes}\,}}{\mathbb I}):=\sigma_x,\quad
\iota_0({\mathbb I}{{\,\hat{\otimes}\,}}\sigma_x):=\sigma_y,\quad
\iota_0(\sigma_x{{\,\hat{\otimes}\,}}\sigma_x):=i\sigma_z.\end{aligned}$$ Noting ${\mathop{\mathrm{Ad}}\nolimits}_{{\mathbb I}_{{{\mathcal K}}_1}\otimes v\otimes {\mathbb I}_{{{{\mathbb C}}^2}}}
\big({{\mathcal R}}_{\kappa_1,{{\mathcal K}}_1}{{\,\hat{\otimes}\,}}{{\mathcal R}}_{\kappa_2,{{\mathcal K}}_2}\big)
={{\mathcal B}}({{\mathcal K}}_1)\otimes {{\mathcal B}}({{\mathcal K}}_2)\otimes {\left (}{\mathop{\mathfrak{C}}\nolimits}{{\,\hat{\otimes}\,}}{\mathop{\mathfrak{C}}\nolimits}{\right )}$ with $v$ in , we obtain a $*$-isomorphism $\iota: {{\mathcal R}}_{\kappa_1,{{\mathcal K}}_1}{{\,\hat{\otimes}\,}}{{\mathcal R}}_{\kappa_2,{{\mathcal K}}_2}\to
{{{\mathcal B}}({{\mathcal K}})}\otimes {\mathop{\mathrm{M}}\nolimits}_2={{\mathcal R}}_{\kappa,{{\mathcal K}}}$ given by $$\begin{aligned}
\iota(x):={\left (}{\mathop{\mathrm{id}}\nolimits}_{{{\mathcal K}}}\otimes\iota_0{\right )}\circ{\mathop{\mathrm{Ad}}\nolimits}_{{\mathbb I}_{{{\mathcal K}}_1}\otimes v\otimes {\mathbb I}_{{{{\mathbb C}}^2}}}(x),\quad
x\in {{\mathcal R}}_{\kappa_1,{{\mathcal K}}_1}{{\,\hat{\otimes}\,}}{{\mathcal R}}_{\kappa_2,{{\mathcal K}}_2}.\end{aligned}$$ We then have $$\begin{aligned}
&{\mathop{\mathrm{Ad}}\nolimits}_{V_g}\circ\iota{\left (}{\left (}a\otimes\sigma_x{\right )}{{\,\hat{\otimes}\,}} {\left (}b\otimes{\mathbb I}_{{{{\mathbb C}}^2}}{\right )}{\right )}={\mathop{\mathrm{Ad}}\nolimits}_{V_g}{\left (}a\otimes b\otimes \sigma_x
{\right )}={\mathop{\mathrm{Ad}}\nolimits}_{V_{1, g}^{(0)}}(a)\otimes {\mathop{\mathrm{Ad}}\nolimits}_{V_{2, g}^{(0)}}(b)
\otimes (-1)^{{{\mathfrak{q}}_1(g)}}\sigma_x \\
&\hspace{1.5cm} =\iota{\left (}{\mathop{\mathrm{Ad}}\nolimits}_{V_{1, g}}{\left (}a\otimes \sigma_x{\right )}\hat
\otimes {\left (}{\mathop{\mathrm{Ad}}\nolimits}_{V_{2, g}}{\left (}b\otimes {\mathbb I}_{{{{\mathbb C}}^2}}{\right )}{\right )}{\right )}=\iota\circ{\left (}\alpha_{1,g}{{\,\hat{\otimes}\,}} \alpha_{2,g}{\right )}{\left (}{\left (}a\otimes\sigma_x{\right )}{{\,\hat{\otimes}\,}} {\left (}b\otimes{\mathbb I}_{{{{\mathbb C}}^2}}{\right )}{\right )},\end{aligned}$$ and $$\begin{aligned}
&{\mathop{\mathrm{Ad}}\nolimits}_{V_g}\circ\iota{\left (}{\left (}a\otimes{\mathbb I}_{{{{\mathbb C}}^2}}
{\right )}{{\,\hat{\otimes}\,}} {\left (}b\otimes\sigma_x{\right )}{\right )}={\mathop{\mathrm{Ad}}\nolimits}_{V_g}{\left (}a\otimes b\otimes \sigma_y
{\right )}={\mathop{\mathrm{Ad}}\nolimits}_{V_{1, g}^{(0)}}(a)\otimes {\mathop{\mathrm{Ad}}\nolimits}_{V_{2, g}^{(0)}}(b)
\otimes (-1)^{{\mathfrak{q}}_2(g)}\sigma_y \\
&\hspace{1.5cm} =\iota{\left (}{\mathop{\mathrm{Ad}}\nolimits}_{V_{1, g}}{\left (}a\otimes {\mathbb I}_{{{{\mathbb C}}^2}}{\right )}\hat
\otimes {\left (}{\mathop{\mathrm{Ad}}\nolimits}_{V_{2, g}}{\left (}b\otimes \sigma_x{\right )}{\right )}{\right )}=\iota\circ{\left (}\alpha_{1,g}{{\,\hat{\otimes}\,}} \alpha_{2,g}{\right )}{\left (}{\left (}a\otimes{\mathbb I}_{{{{\mathbb C}}^2}}{\right )}{{\,\hat{\otimes}\,}} {\left (}b\otimes\sigma_x{\right )}{\right )}\end{aligned}$$ for all $a\in{{\mathcal B}}({{\mathcal K}}_1)$, $b\in {{\mathcal B}}({{\mathcal K}}_2)$. Because the elements $(a\otimes\sigma_x){{\,\hat{\otimes}\,}} (b\otimes{\mathbb I}_{{{{\mathbb C}}^2}})$ and $(a\otimes{\mathbb I}_{{{{\mathbb C}}^2}}) {\,\hat{\otimes}\,}(b \otimes \sigma_x)$ generate ${{\mathcal R}}_{\kappa_1,{{\mathcal K}}_1}{{\,\hat{\otimes}\,}}{{\mathcal R}}_{\kappa_2,{{\mathcal K}}_2}$, we see that ${\mathop{\mathrm{Ad}}\nolimits}_{V_g}\circ\iota(x)=\iota\circ{\left (}\alpha_{1,g}{{\,\hat{\otimes}\,}} \alpha_{2,g}{\right )}(x)$ for $x\in {{\mathcal R}}_{\kappa_1,{{\mathcal K}}_1}{{\,\hat{\otimes}\,}}{{\mathcal R}}_{\kappa_2,{{\mathcal K}}_2}$. We also see from that ${\mathop{\mathrm{Ad}}\nolimits}_{\Gamma_{{\mathcal K}}}\circ\iota(x)=\iota\circ{\left (}{\mathop{\mathrm{Ad}}\nolimits}_{\Gamma_1}{{\,\hat{\otimes}\,}} {\mathop{\mathrm{Ad}}\nolimits}_{\Gamma_2}{\right )}(x)$ for $x\in {{\mathcal R}}_{\kappa_1,{{\mathcal K}}_1}{{\,\hat{\otimes}\,}}{{\mathcal R}}_{\kappa_2,{{\mathcal K}}_2}$. Hence we obtain .
As a simple example, let us consider fermionic SPT phases with time-reversal symmetry. That is, we take $G={{\mathbb Z}}_2 =\{0,1\}$ with with ${{\mathfrak{p}}(1)}=1$. We let $\alpha= \alpha_1$ be the anti-linear $\ast$-automorphism of order $2$ from the non-trivial element. Therefore, if $G$ acts on a balanced, central and type I von Neumann algebra, then $\alpha$ is implemented on a graded Hilbert space ${{\mathcal K}}$ by $\mathrm{Ad}_R$ with $R$ anti-unitary. Following [@OgataTRI], we can ensure that $R^2 = \pm {\mathbb I}_{{\mathcal K}}$ and so the group $2$-cocycle is determined by the sign of $R^2$.
The data ${\ensuremath{\mathbb{Z}}}_2 \times H^1({{\mathbb Z}}_2,{{\mathbb Z}}_2)\times H^2({{\mathbb Z}}_2,U(1)_{\mathfrak{p}})$ from Theorem \[stlaw\] is wholly determined by the triple $[\kappa; \varepsilon, \pm]$, where $\varepsilon={\mathfrak{q}}(1) \in {{\mathbb Z}}_2$ and $\pm$ is the sign of $R^2$. Our choice of notation is so that our results can easily be compared with [@MoutuouBrauer Appendix A] and [@Wall]. Following , the triple has the (abelian) composition law under stacking $$\begin{aligned}
[0;\varepsilon_1, \xi_1] [0, \varepsilon_2,\xi_2] &= [0; \varepsilon_1+\varepsilon_2, (-)^{\varepsilon_1\varepsilon_2} \xi_1 \xi_2] \\
[0;\varepsilon_1, \xi_1] [1, \varepsilon_2,\xi_2] &= [1; \varepsilon_1+\varepsilon_2, (-)^{\varepsilon_1+\varepsilon_1\varepsilon_2} \xi_1 \xi_2] \\
[1;\varepsilon_1, \xi_1] [1, \varepsilon_2,\xi_2] &= [0; \varepsilon_1+\varepsilon_2+1, (-)^{\varepsilon_1\varepsilon_2} \xi_1 \xi_2].\end{aligned}$$ One therefore sees that ${\ensuremath{\mathbb{Z}}}_2 \times H^1({{\mathbb Z}}_2,{{\mathbb Z}}_2)\times H^2({{\mathbb Z}}_2,U(1)_{\mathfrak{p}}) \cong {{\mathbb Z}}_8$ with generator $[1;0,+]$. Hence we recover and extend the ${{\mathbb Z}}_8$-classification of time-reversal symmetric fermionic SPT phases in one dimension considered for finite systems in [@fk; @FK2; @BWHV].
Translation invariant states {#transsec}
============================
In this section, we derive a representation of pure, split, translation invariant and $\alpha$-invariant states in terms of a finite set of operators on Hilbert spaces. The idea of the proof is the same as quantum spin case, cf. [@bjp; @Matsui3], although anti-commutativity results in richer structures.
Recall the integer shift $S_x$ on $l^2({{\mathbb Z}})\otimes{{\mathbb C}}^d$, $x\in {{\mathbb Z}}$, which defines the $\ast$-automorphism $\beta_{S_x} \in \operatorname{Aut}({{\mathcal A}})$. Let $\omega$ be a pure, split, $\alpha$-invariant and translation invariant state on ${{{\mathcal A}}}$. In particular, such states are $\Theta$-invariant (see [@BR2 Example 5.2.21]). By Proposition \[casebycase\] and Lemma \[deru\] the graded $W^*$-$(G,{\mathfrak{p}})$-dynamical system $(\pi_\omega({{\mathcal A}}_R)'', {\mathop{\mathrm{Ad}}\nolimits}_{\Gamma_\omega}, \hat\alpha_\omega)$ associated to $\omega$ is equivalent to some $({{\mathcal R}}_{\kappa, {{\mathcal K}}},{\mathop{\mathrm{Ad}}\nolimits}_{\Gamma_{{\mathcal K}}},{\mathop{\mathrm{Ad}}\nolimits}_{V_g})\in{{\mathcal S}}_{\kappa}$. We denote this $\kappa$ by $\kappa_\omega$. The space translation lifts to an endomorphism on $\pi_\omega({{\mathcal A}}_R)''$.
\[iorho\] Let $\omega$ be a pure, split, $\alpha$-invariant and translation invariant state on ${{\mathcal A}}$. Suppose that the graded $W^*$-$(G,{\mathfrak{p}})$-dynamical system $({\pi_\omega}({{\mathcal A}}_R)'', {\mathop{\mathrm{Ad}}\nolimits}_{\Gamma_\omega}, \hat\alpha_\omega)$ associated to $\omega$ is equivalent to $({{\mathcal R}}_{\kappa, {{\mathcal K}}},{\mathop{\mathrm{Ad}}\nolimits}_{\Gamma_{{\mathcal K}}},{\mathop{\mathrm{Ad}}\nolimits}_{V_g})\in{{\mathcal S}}_{\kappa}$, via a $*$-isomorphism $\iota: {\pi_\omega}({{\mathcal A}}_R)''\to {{\mathcal R}}_{\kappa, {{\mathcal K}}}$. Then there is an injective $*$-endomorphism $\rho$ on ${{\mathcal R}}_{\kappa, {{\mathcal K}}}$, such that $$\begin{aligned}
\label{ipr}
\iota\circ{\pi_\omega}\circ\beta_{S_1}(A)=\rho\circ\iota\circ{\pi_\omega}(A),\quad
A\in{{\mathcal A}}_R.\end{aligned}$$ Furthermore, we have $$\begin{aligned}
\label{ababy}
a\rho(b)-(-1)^{\partial a\partial b}\rho(b)a=0,\end{aligned}$$ for homogeneous $a\in\iota\circ{\pi_\omega}\big( {{\mathcal A}}_{\{0\}}\big)$ and $b\in {{\mathcal R}}_{\kappa, {{\mathcal K}}}$.
By the translation invariance of $\omega$, the space translation $\beta_{S_1}$ is lifted to an automorphism $\hat\beta_{S_1}$ on ${\pi_\omega}({{\mathcal A}})''$. Restricting $\hat\beta_{S_1}$ to ${\pi_\omega}({{\mathcal A}}_R)''$, we obtain an injective $*$-endomorphism $\tilde \beta$ on ${\pi_\omega}({{\mathcal A}}_R)''$. We then see that $\rho:=\iota\circ \tilde \beta \circ\iota^{-1}: {{\mathcal R}}_{\kappa, {{\mathcal K}}}\to {{\mathcal R}}_{\kappa, {{\mathcal K}}}$ is an injective endomorphism on ${{\mathcal R}}_{\kappa, {{\mathcal K}}}$ satisfying . Because $\beta_{S_1}({\mathcal{A}}_R) \subset {\mathcal{A}}_{{{\mathbb Z}}\geq 1}$, we see that $a_0 \beta_{S_1}(a_1) -(-1)^{\partial a_0\partial a_1}\beta_{S_1}(a_1)a_0=0$ for homogeneous $a_0 \in {{\mathcal A}}_{\{0\}}$ and $a_1 \in {{\mathcal A}}_R$. Then, because $\rho{\left (}{{\mathcal R}}_{\kappa, {{\mathcal K}}}{\right )}={\left (}\iota\circ{\pi_\omega}\circ\beta_{S_1}{\left (}{{\mathcal A}}_{R}{\right )}{\right )}''$, equation follows.
Let ${{\mathcal P}}$ be the power set $\calP = \calP(\{1,\ldots,d\}) = 2^{\{1,\ldots,d\}}$ of $\{1,\ldots,d\}$. We denote the parity of the number of the elements in $\mu\in{{\mathcal P}}$ by $|\mu| = \# \mu \,\mathrm{mod} \, 2$. We denote by $\{\psi_\mu\}_{\mu\in{{\mathcal P}}}$ the standard basis of $\calF({{\mathbb C}}^d)$. Namely, with the Fock vacuum $\Omega_d$ of $\calF({{\mathbb C}}^d)$ and the standard basis $\{e_{i}\}_{i=1}^d$ of ${{\mathbb C}}^d$, $\psi_\mu$ for $\mu\neq\emptyset$ is given by $\psi_\mu=C_\mu a^*(e_{\mu_1})a^*(e_{\mu_2})\cdots a^*(e_{\mu_l})\Omega_d$ with $l=\#\mu$, $\mu=\{\mu_1,\mu_2,\ldots,\mu_l\}$ with $\mu_1<\mu_2\cdots<\mu_l$ and a suitable normalization factor $C_\mu\in{{\mathbb C}}\setminus \{0\}$. For the empty set $\mu=\emptyset$, we set $\psi_\emptyset:=\Omega_d$.
We denote the matrix units of ${{\mathcal A}}_{\{0\}}\simeq {{\mathcal B}}({{\mathcal F}}({{\mathbb C}}^d)) \simeq {\mathop{\mathrm{M}}\nolimits}_{2^d}$ associated to the standard basis $\{\psi_\mu\}_{\mu\in{{\mathcal P}}}$ by $\{ E_{\mu,\nu}^{(0)}\}$, $\mu,\nu\in\calP$. Because $\Theta$ is implemented by the second quantization of $-{\mathbb I}_{{{\mathbb C}}^d}$, $$\begin{aligned}
\mathfrak\Gamma(-{\mathbb I})=\sum_{\mu\in {{\mathcal P}}}(-1)^{|\mu|} E_{\mu\mu}^{(0)}\in{{\mathcal A}}_{\{0\}},\end{aligned}$$ we see that $$\begin{aligned}
\Theta(E_{\mu,\nu}^{(0)})=(-1)^{|\mu|+|\nu|} E_{\mu,\nu}^{(0)},\qquad
\mu,\nu\in {{\mathcal P}}.\end{aligned}$$ We set $E_{\mu,\nu}^{(x)}:=\beta_{S_x}\big( E_{\mu,\nu}^{(0)}\big)$ for general $x\in{{\mathbb Z}}$. Clearly, $\{ E_{\mu,\nu}^{(x)}\}_{\mu,\nu\in{{\mathcal P}}}$ are matrix units of ${{\mathcal A}}_{\{x\}}$.
\[lem:trans\_weak\_conv\_to\_state\] Let $\omega$ be a pure, split and translation invariant state on ${{{\mathcal A}}}$ and $\hat\beta_{S_n}$ be the extension of $\beta_{S_n}$ to $\pi_\omega({{\mathcal A}})''$, i.e. $\hat\beta_{S_n}\circ\pi_\omega(A)=\pi_\omega\circ\beta_{S_n}(A)$, $A\in{{\mathcal A}}$.
1. If $x\in{\left (}\pi_\omega({{\mathcal A}}_R)''{\right )}^{(0)}$, then $\sigma\mathrm{\hbox{-}weak}\lim_{n\to\infty}\hat\beta_{S_n}(x)
={\langle\Omega_\omega,x\Omega_\omega\rangle}{\mathbb I}_{{{{{\mathcal H}}_\omega}}}$.
2. If $x\in{\left (}\pi_\omega({{\mathcal A}}_R)''{\right )}^{(1)}$ and $\pi_\omega({{\mathcal A}}_R)''$ is a factor, then $$\begin{aligned}
\label{app}
\sigma\mathrm{\hbox{-}weak}\lim_{n\to\infty}
\pi_\omega{\left (}{\mathfrak\Gamma}(-{\mathbb I})
\beta_{S_1}
{\left (}{\mathfrak\Gamma}(-{\mathbb I}){\right )}\cdots
\beta_{S_{n-1}}
{\left (}{\mathfrak\Gamma}(-{\mathbb I}){\right )}{\right )}\hat \beta_{S_n}(x)
=0={\langle\Omega_\omega,x\Omega_\omega\rangle}{\mathbb I}_{{{\mathcal H}}_\omega}.\end{aligned}$$
3. If $x\in{\left (}\pi_\omega({{\mathcal A}}_R)''{\right )}^{(1)}$ and $Z{\left (}\pi_\omega({{\mathcal A}}_R)''{\right )}\cap {\left (}\pi_\omega({{\mathcal A}}_R)''{\right )}^{(1)}\neq \{0\}$, then $\sigma\mathrm{\hbox{-}weak}\lim_{n\to\infty}\hat\beta_{S_n}(x)=0
={\langle\Omega_\omega,x\Omega_\omega\rangle}$.
First we note from the $\sigma$-weak continuity of $\hat\beta_{S_n}$ that $$\begin{aligned}
\label{marui}
\hat\beta_{S_n}{\left (}{\left (}\pi_\omega({{\mathcal A}}_R)''{\right )}^{(\sigma)}{\right )}\subset
{\left (}{\left (}\pi_\omega\circ\beta_{S_n}({{\mathcal A}}_R){\right )}''{\right )}^{(\sigma)},\quad n\in{{\mathbb N}},\quad \sigma=0,1.\end{aligned}$$ (i) By , we have $\hat\beta_{S_n}{\left (}{\left (}\pi_\omega({{\mathcal A}}_R)''{\right )}^{(0)}{\right )}\subset \pi_\omega({{\mathcal A}}_{[0,n-1]})'$. Therefore for any $x\in{\left (}\pi_\omega({{\mathcal A}}_R)''{\right )}^{(0)}$, any $\sigma$-weak accumulation point $z$ of $\{\hat\beta_{S_n}(x)\}$ belongs to $\pi_\omega({{\mathcal A}}_R)'\cap {\left (}\pi_\omega({{\mathcal A}}_R)''{\right )}^{(0)}$. But $\pi_\omega({{\mathcal A}}_R)'\cap {\left (}\pi_\omega({{\mathcal A}}_R)''{\right )}^{(0)}={{\mathbb C}}{\mathbb I}_{{{\mathcal H}}_\omega}$ by Lemma \[hs\]. Hence we have $z\in {{\mathbb C}}{\mathbb I}_{{{\mathcal H}}_\omega}$. Because ${\langle\Omega_\omega,\hat\beta_{S_n}(x){{\Omega_\omega}}\rangle}={\langle{{\Omega_\omega}}, x{{\Omega_\omega}}\rangle}$, this means $z={\langle{{\Omega_\omega}}, x{{\Omega_\omega}}\rangle}{\mathbb I}_{{{{{\mathcal H}}_\omega}}}$. As this holds for any accumulation point, we obtain $\sigma\mathrm{\hbox{-}weak}\lim_{n\to\infty}\hat\beta_{S_n}(x)
={\langle\Omega_\omega,x\Omega_\omega\rangle}{\mathbb I}_{{{{{\mathcal H}}_\omega}}}$.
\(ii) Suppose that $\pi_\omega({{\mathcal A}}_R)''$ is a factor and set $Y_n:={\mathfrak\Gamma}(-{\mathbb I})
\beta_{S_1}({\mathfrak\Gamma}(-{\mathbb I}))\cdots \beta_{S_{n-1}}({\mathfrak\Gamma}(-{\mathbb I}))$. Note that ${\mathop{\mathrm{Ad}}\nolimits}_{Y_n}(B)=\Theta(B)$ for any $B\in{{\mathcal A}}_{[0,n-1]}$. Therefore, by , we have ${{\pi_{\omega}}}(Y_n) \hat\beta_{S_n}\big( ({{\pi_{\omega}}}({{\mathcal A}}_R)'')^{(1)}\big)
\subset {{\pi_{\omega}}}({{\mathcal A}}_{[0,n-1]})'$. Hence, for any $x\in{\left (}\pi_\omega({{\mathcal A}}_R)''{\right )}^{(1)}$, any $\sigma$-weak accumulation point $z$ of the set $\{{{\pi_{\omega}}}{\left (}Y_n{\right )}\hat\beta_{S_n}(x)\}$ belongs to $\pi_\omega({{\mathcal A}}_R)'\cap {\left (}\pi_\omega({{\mathcal A}}_R)''{\right )}^{(1)}=\{0\}$. As such, $z=0$. As this holds for any accumulation point, we obtain (ii).
\(iii) Suppose that $Z{\left (}\pi_\omega({{\mathcal A}}_R)''{\right )}\cap {\left (}\pi_\omega({{\mathcal A}}_R)''{\right )}^{(1)}\neq \{0\}$. By , we have that $$\begin{aligned}
\hat\beta_{S_n}\big( (\pi_\omega({{\mathcal A}}_R)'')^{(1)}\big)
\subset {{\pi_{\omega}}}({{\mathcal A}}_R)''\cap \pi_\omega({{\mathcal A}}_{[0,n-1]})'\Gamma_\omega.
\end{aligned}$$ Therefore for any $x\in{\left (}\pi_\omega({{\mathcal A}}_R)''{\right )}^{(1)}$, any $\sigma$-weak accumulation point $z$ of $\{\hat\beta_{S_n}(x)\}$ belongs to ${\left (}\pi_\omega({{\mathcal A}}_R)'\Gamma_\omega{\right )}\cap {\left (}\pi_\omega({{\mathcal A}}_R)''{\right )}^{(1)}$. Because $Z{\left (}\pi_\omega({{\mathcal A}}_R)''{\right )}$ has an odd element, $\pi_\omega({{\mathcal A}}_R)''$ is not a factor. Lemma \[jh2\] then implies that $\pi_\omega({{\mathcal A}}_R)'\Gamma_\omega \cap \pi_\omega({{\mathcal A}}_R)''=\{0\}$. Hence we have $z=0$. As this holds for any accumulation point, we obtain (iii).
Before stating the result, we fix some notation. Given the operators $\{W_\mu\}_{\mu\in\calP}$ we define the CP map $T_{\bf W}$ by $$T_{\bf W}(x) = \sum_{\mu\in\calP} W_\mu x W_\mu^*.$$ Because the algebraic structure of the von Neumann algebra of interest changes depending on whether $\kappa_\omega=0,1$, we treat each case separately, though the general strategy of proof is the same.
Case: $\kappa_\omega=0$
-----------------------
Recall that $\mathfrak{\Gamma}(U_g)$ denotes the second quantization of $U_g$ on $\calF({\ensuremath{\mathbb{C}}}^d)$. In this subsection we prove the following.
\[cuntzthm0\] Let $\omega$ be a pure $\alpha$-invariant and translation invariant split state on ${{\mathcal A}}$. Suppose that the graded $W^*$-$(G,{\mathfrak{p}})$-dynamical system $(\pi_\omega({{\mathcal A}}_R)'', {\mathop{\mathrm{Ad}}\nolimits}_{\Gamma_\omega}, \hat\alpha_\omega)$ associated to $\omega$ is equivalent to $({{\mathcal R}}_{0,{{\mathcal K}}},{\mathop{\mathrm{Ad}}\nolimits}_{\Gamma_{{\mathcal K}}},{\mathop{\mathrm{Ad}}\nolimits}_{V_g})\in{{\mathcal S}}_{0}$, via a $*$-isomorphism $\iota: \pi_\omega({{\mathcal A}}_R)''\to {{\mathcal B}}({{\mathcal K}}\otimes{{{\mathbb C}}^2})$. Let $\rho$ be the $*$-endomorphism on ${{\mathcal R}}_{0,{{\mathcal K}}}$ given in Lemma \[iorho\]. Then there is a set of isometries $\{B_\mu\}_{\mu \in \calP}$ on ${{\mathcal K}}\otimes{{{\mathbb C}}^2}$ such that $B_\nu^* B_\mu = \delta_{\mu,\nu} {\mathbb I}$, $$\begin{aligned}
\label{bimp}
\rho\circ\iota\circ\pi_\omega(A)
= \sum_{\mu\in\calP} \mathrm{Ad}_{B_\mu\Gamma_{{{\mathcal K}}}^{|\mu|}}\circ\iota\circ\pi_\omega (A)
= \sum_{\mu\in\calP} \mathrm{Ad}_{\Gamma_{{{\mathcal K}}}^{|\mu|}B_\mu}\circ\iota\circ\pi_\omega (A),\quad
A\in{{\mathcal A}}_R,\end{aligned}$$ and $$\begin{aligned}
\label{bbbo}
\iota\circ\pi_\omega{\left (}E_{\mu_0,\nu_0}^{(0)} E_{\mu_1,\nu_1}^{(1)}
\cdots E_{\mu_N,\mu_N}^{(N)}{\right )}=
(-1)^{\sum\limits_{k=1}^N (|\mu_k|+|\nu_k|) \sum\limits_{j=0}^{k-1} |\nu_j|}
B_{\mu_0} \cdots B_{\mu_N} B_{\nu_N}^* \cdots B_{\nu_0}^*\end{aligned}$$ for all $N\in{{\mathbb N}}\cup\{0\}$ and $\mu_0,\ldots\mu_N,\nu_0,\ldots,\nu_N\in{{\mathcal P}}$. The operators $B_\mu$ have homogeneous parity and are such that ${\mathop{\mathrm{Ad}}\nolimits}_{\Gamma_{{\mathcal K}}}{\left (}B_\mu{\right )}= (-1)^{|\mu|+\sigma_0}B_\mu$, with some uniform $\sigma_0\in\{0,1\}$. Furthermore, $$\begin{aligned}
\label{sat}
\sigma\mathrm{\hbox{-}weak}\lim_{N\to\infty}T_{\bf B}^N\circ\iota(x)
={\langle{{\Omega_\omega}}, x{{\Omega_\omega}}\rangle}{\mathbb I}_{{{\mathcal K}}\otimes{{{\mathbb C}}^2}},\quad x\in{{\pi_{\omega}}}({{\mathcal A}}_R)''\end{aligned}$$ and for each $g\in G$, there is some $c_g \in \mathbb{T}$ such that $$\begin{aligned}
\label{gvb}
\sum_{\mu\in\calP} \langle \psi_\mu, \, \mathfrak{\Gamma}(U_g) \psi_\nu \rangle B_\mu =
c_g V_g B_\nu V_g^*.
\end{aligned}$$
We will prove this result in several steps. First we note some properties of endomorphisms of operators on graded Hilbert spaces and the Cuntz algebra.
\[cuntzpropi\] Let ${{\mathcal H}}$ be a Hilbert space with a self-adjoint unitary $\Gamma$ that gives a grading for $\calB(\calH)$. Let ${{\mathcal M}}$ be a finite type I von Neumann subalgebra of ${{\mathcal B}}({{\mathcal H}})$ with matrix units $\{E_{\mu,\nu}\}_{\mu,\nu\in\calP} \subset {{\mathcal M}}$ spanning ${{\mathcal M}}$. Assume that $$\begin{aligned}
\label{grae}
{\mathop{\mathrm{Ad}}\nolimits}_{\Gamma}(E_{\mu,\nu})=(-1)^{|\mu|+|\nu|} E_{\mu,\nu},\qquad
\mu,\nu\in {{\mathcal P}}\end{aligned}$$ and set $\Gamma_0:=\sum_{\mu\in{{\mathcal P}}} (-1)^{|\mu|}E_{\mu\mu}$. Let $\rho: \calB(\calH) \to \calB(\calH)$ be a graded, unital $\ast$-endomorphism such that $\rho(a)b- (-1)^{\partial a \partial b} b \rho(a) = 0$ for $a \in \calB(\calH)$ $b\in {{\mathcal M}}$ with homogeneous grading. Suppose further that $\calB(\calH) = \rho(\calB(\calH)) \vee \calM$. Then there exist isometries $\{S_\mu\}_{\mu \in \calP}$ on ${{\mathcal H}}$ with the property that $$\begin{aligned}
S_\nu^* S_\mu = \delta_{\mu,\nu} {\mathbb I},
\qquad
\rho(x) = \sum_\mu S_\mu x S_\mu^* \end{aligned}$$ for all $\mu,\nu \in\calP$ and $x \in \calB(\calH)$. The operators $S_\mu$ have homogeneous parity and are such that ${\mathop{\mathrm{Ad}}\nolimits}_\Gamma{\left (}S_\mu{\right )}= (-1)^{|\mu|+\sigma_0}S_\mu$ with some uniform $\sigma_0\in\{0,1\}$. Furthermore, setting $B_\mu:=(\Gamma_0\Gamma)^{|\mu|} S_\mu$, for $\mu\in{{\mathcal P}}$, we have $B_\nu^* B_\mu = \delta_{\mu,\nu} {\mathbb I}$, $$\begin{aligned}
\label{rhotrans}
\rho(x)
= \sum_{\mu\in\calP} \mathrm{Ad}_{B_\mu} \circ \mathrm{Ad}_{\Gamma^{|\mu|}} (x),\quad
x\in{{{\mathcal B}}({{\mathcal H}})},\end{aligned}$$ and $$\begin{aligned}
\label{eq:rho_and_E_using_V_even}
E_{\mu_0,\nu_0} \rho(E_{\mu_1,\nu_1}) \cdots \rho^N (E_{\mu_N,\mu_N}) =
(-1)^{\sum\limits_{k=1}^N (|\mu_k|+|\nu_k|) \sum\limits_{j=0}^{k-1} |\nu_j|}
B_{\mu_0} \cdots B_{\mu_N} B_{\nu_N}^* \cdots B_{\nu_0}^* \end{aligned}$$ for all $N\in{{\mathbb N}}\cup\{0\}$ and $\mu_0,\ldots,\mu_N,\nu_0,\ldots,\nu_N\in{{\mathcal P}}$. The operators $B_\mu$ have homogeneous parity such that ${\mathop{\mathrm{Ad}}\nolimits}_\Gamma{\left (}B_\mu{\right )}= (-1)^{|\mu|+\sigma_0}B_\mu$, with the same $\sigma_0$ as above. If there are isometries $\{T_\mu\}_{\mu \in \calP}$ such that $$\begin{aligned}
\label{buni}
T_\nu^* T_\mu = \delta_{\mu,\nu} {\mathbb I}, \quad T_\mu T_\nu^*= E_{\mu,\nu},
\quad\rho(x)
= \sum_{\mu\in\calP} \mathrm{Ad}_{T_\mu} \circ \mathrm{Ad}_{\Gamma^{|\mu|}} (x),\quad
x\in{{{\mathcal B}}({{\mathcal H}})},\end{aligned}$$ then there is some $c\in{{\mathbb T}}$ such that $T_\mu=cB_\mu$, for all $\mu\in {{\mathcal P}}$.
To study the situation, we note the following general property.
\[lem35\] Let ${{\mathcal H}}$ be a Hilbert space with a self-adjoint unitary $\Gamma$ that gives a grading for $\calB(\calH)$. Let ${{\mathcal M}}_1$, ${{\mathcal M}}_2$ be ${\mathop{\mathrm{Ad}}\nolimits}_{\Gamma}$-invariant von Neumann subalgebras of ${{{\mathcal B}}({{\mathcal H}})}$ with ${{\mathcal M}}_1 \vee {{\mathcal M}}_2 = {{{\mathcal B}}({{\mathcal H}})}$. Suppose that ${{\mathcal M}}_1$ is a type I factor with a self-adjoint unitary $\Gamma_1\in {{\mathcal M}}_1$ such that ${\mathop{\mathrm{Ad}}\nolimits}_{\Gamma_1}(x)={\mathop{\mathrm{Ad}}\nolimits}_{\Gamma}(x)$ for all $x\in{{\mathcal M}}_1$. Suppose further that $$\begin{aligned}
\label{anti}
ab- (-1)^{\partial a \partial b} b a = 0,\quad \text{for homogeneous}\quad a\in{{\mathcal M}}_1, b \in{{\mathcal M}}_2.\end{aligned}$$ Then there are Hilbert spaces ${{\mathcal H}}_1,{{\mathcal H}}_2$ and a unitary $V:{{\mathcal H}}\to{{\mathcal H}}_1\otimes{{\mathcal H}}_2$ such that $$\begin{aligned}
{\mathop{\mathrm{Ad}}\nolimits}_{V}({{\mathcal M}}_1)={{\mathcal B}}({{\mathcal H}}_1)\otimes {{\mathbb C}}{\mathbb I}_{{{\mathcal H}}_2}.\quad \label{mi}\end{aligned}$$ Furthermore, there are self-adjoint unitaries $\tilde\Gamma_i$ on ${{\mathcal H}}_i$ with $i=1,2$ such that $$\begin{aligned}
\label{ggg}
{\mathop{\mathrm{Ad}}\nolimits}_V(\Gamma)=\tilde\Gamma_1\otimes\tilde\Gamma_2,\quad
{\mathop{\mathrm{Ad}}\nolimits}_V(\Gamma_1)=\tilde\Gamma_1\otimes{\mathbb I}_{{{\mathcal H}}_2}.\end{aligned}$$ The commutant of ${{\mathcal M}}_2$ is given by $$\begin{aligned}
\label{mdes}
{{\mathcal M}}_2'
={{\mathcal M}}_1^{(0)}+{{\mathcal M}}_1^{(1)}\Gamma_1\Gamma.\end{aligned}$$ If $p$ is an even minimal projection in ${{\mathcal M}}_1$ then ${{\mathcal M}}_2\cdot p={{\mathcal B}}(p{{\mathcal H}})$.
We note that if ${{\mathcal M}}_1$ is a type I factor, Wigner’s Theorem guarantees the existence of a self-adjoint unitary $\Gamma_1\in {{\mathcal M}}_1$ such that ${\mathop{\mathrm{Ad}}\nolimits}_{\Gamma_1}(x)={\mathop{\mathrm{Ad}}\nolimits}_{\Gamma}(x)$ for all $x\in{{\mathcal M}}_1$.
Because ${{\mathcal M}}_1$ is a type I factor, by [@Takesaki1 Chapter V, Theorem 1.31] there are Hilbert spaces ${{\mathcal H}}_1,\,{{\mathcal H}}_2$ and a unitary $V:{{\mathcal H}}\to{{\mathcal H}}_1\otimes{{\mathcal H}}_2$ satisfying . Because $\Gamma_1\in{{\mathcal M}}_1$ and $\Gamma\Gamma_1\in {{\mathcal M}}_1'$, there are self-adjoint unitaries $\tilde\Gamma_i$ on ${{\mathcal H}}_i$ with $i=1,2$ satisfying . Clearly ${\mathop{\mathrm{Ad}}\nolimits}_{\Gamma_1}(\Gamma_1)=\Gamma_1$ and so $\Gamma_1$ is an even element of ${{\mathcal M}}_1$.
Note that ${{\mathcal N}}:={{\mathcal M}}_2^{(0)}+{{\mathcal M}}_2^{(1)}\Gamma_1$ is a von Neumann subalgebra of ${{\mathcal M}}_1'$ by . Therefore, ${\mathop{\mathrm{Ad}}\nolimits}_V({{\mathcal N}})$ is a von Neumann subalgebra of ${\mathbb I}_{{{\mathcal H}}_1}\otimes {{\mathcal B}}({{\mathcal H}}_2)$. Because $$\begin{aligned}
&{{\mathcal M}}_2={{\mathcal M}}_2^{(0)}+{{\mathcal M}}_2^{(1)}\Gamma_1\Gamma_1\subset {{\mathcal M}}_1\vee {{\mathcal N}},
&&{{\mathcal M}}_1\subset{{\mathcal M}}_1\vee {{\mathcal N}},
&&{{\mathcal M}}_1\vee{{\mathcal M}}_2={{\mathcal B}}({{\mathcal H}}),\end{aligned}$$ we have ${{\mathcal M}}_1\vee {{\mathcal N}}={{\mathcal B}}({{\mathcal H}})$. Combining with , this means $$\begin{aligned}
\label{mini}
{\mathop{\mathrm{Ad}}\nolimits}_{V}({{\mathcal M}}_2^{(0)}+{{\mathcal M}}_2^{(1)}\Gamma_1)={\mathop{\mathrm{Ad}}\nolimits}_V({{\mathcal N}})={{\mathbb C}}{\mathbb I}_{{{\mathcal H}}_1}\otimes{{\mathcal B}}({{\mathcal H}}_2).\end{aligned}$$
Now we associate the grading given by $\tilde\Gamma_i$ to ${{\mathcal B}}({{\mathcal H}}_i)$ for $i=1,2$, and regard ${{\mathcal B}}({{\mathcal H}}_1)\otimes{{\mathcal B}}({{\mathcal H}}_2)$ as ${{\mathcal B}}({{\mathcal H}}_1){{\,\hat{\otimes}\,}} {{\mathcal B}}({{\mathcal H}}_2)$, the graded tensor product of $({{\mathcal B}}({{\mathcal H}}_1),{{\mathcal H}}_1,\tilde{\Gamma}_1)$ and $({{\mathcal B}}({{\mathcal H}}_2),{{\mathcal H}}_2,{\tilde\Gamma_2})$. Because ${\mathop{\mathrm{Ad}}\nolimits}_V(\Gamma)=\tilde\Gamma_1\otimes\tilde\Gamma_2$, ${\mathop{\mathrm{Ad}}\nolimits}_V: {{\mathcal B}}({{\mathcal H}})\to {{\mathcal B}}({{\mathcal H}}_1){{\,\hat{\otimes}\,}} {{\mathcal B}}({{\mathcal H}}_2)$ is a graded $*$-isomorphism. Considering the even and odd subspaces of , we obtain $$\begin{aligned}
{\mathop{\mathrm{Ad}}\nolimits}_{V}({{\mathcal M}}_2^{(0)})={{\mathbb C}}{\mathbb I}_{{{\mathcal H}}_1}\otimes{{\mathcal B}}({{\mathcal H}}_2)^{(0)},\quad
{\mathop{\mathrm{Ad}}\nolimits}_{V}({{\mathcal M}}_2^{(1)}){\mathop{\mathrm{Ad}}\nolimits}_{V}(\Gamma_1)=
{{\mathbb C}}{\mathbb I}_{{{\mathcal H}}_1}\otimes{{\mathcal B}}({{\mathcal H}}_2)^{(1)}.\end{aligned}$$ and so $$\begin{aligned}
\label{m2}
{\mathop{\mathrm{Ad}}\nolimits}_{V}({{\mathcal M}}_2)={\mathop{\mathrm{Ad}}\nolimits}_{V}({{\mathcal M}}_2^{(0)}+{{\mathcal M}}_2^{(1)})
={{\mathbb C}}{\mathbb I}_{{{\mathcal H}}_1}\otimes{{\mathcal B}}({{\mathcal H}}_2)^{(0)}
+{{\mathbb C}}\tilde\Gamma_1\otimes{{\mathcal B}}({{\mathcal H}}_2)^{(1)}
= {{\mathbb C}}{\mathbb I}_{{{\mathcal H}}_1}{{\,\hat{\otimes}\,}} {{\mathcal B}}({{\mathcal H}}_2)\end{aligned}$$ where ${{\mathbb C}}{\mathbb I}_{{{\mathcal H}}_1}{{\,\hat{\otimes}\,}} {{\mathcal B}}({{\mathcal H}}_2)$ is a graded tensor product of $({{\mathbb C}}{\mathbb I}_{{{\mathcal H}}_1},{{\mathcal H}}_1, \tilde\Gamma_1)$ and $({{\mathcal B}}({{\mathcal H}}_2), {{\mathcal H}}_2,\tilde\Gamma_2)$.
We now consider the commutant of ${{\mathcal M}}_2$. Applying Lemma \[comgra\], we see that $$\begin{aligned}
{\mathop{\mathrm{Ad}}\nolimits}_{V}({{\mathcal M}}_2')
={{\mathcal B}}({{\mathcal H}}_1)^{(0)}\otimes {{\mathbb C}}{\mathbb I}_{{{\mathcal H}}_2}
+{{\mathcal B}}({{\mathcal H}}_1)^{(1)}\otimes {{\mathbb C}}\tilde\Gamma_2
={\mathop{\mathrm{Ad}}\nolimits}_{V}\big( {{\mathcal M}}_1^{(0)}+{{\mathcal M}}_1^{(1)}\Gamma_1\Gamma \big).\end{aligned}$$ Hence we obtain .
Let $p$ be a minimal projection in ${{\mathcal M}}_1$ and suppose that it is even. Then ${\mathop{\mathrm{Ad}}\nolimits}_V(p)$ is a minimal projection in ${{\mathcal B}}({{\mathcal H}}_1)\otimes{{\mathbb C}}{\mathbb I}_{{{\mathcal H}}_2}$. Therefore, there is a rank-one projection $r$ on ${{\mathcal H}}_1$ such that ${\mathop{\mathrm{Ad}}\nolimits}_V(p)=r\otimes{\mathbb I}_{{{\mathcal H}}_2}$. Because $p$ is even and ${\mathop{\mathrm{Ad}}\nolimits}_V$ is a graded $*$-isomorphism, we have ${\mathop{\mathrm{Ad}}\nolimits}_{\tilde \Gamma_1}(r)=r$. As $r$ is rank-one, this means that $\tilde \Gamma_1 r=\pm r$. Therefore, using , we have $$\begin{aligned}
{\mathop{\mathrm{Ad}}\nolimits}_V({{\mathcal M}}_2 p)
&=
{{\mathbb C}}r\otimes{{\mathcal B}}({{\mathcal H}}_2)^{(0)}
+{{\mathbb C}}\tilde\Gamma_1r\otimes{{\mathcal B}}({{\mathcal H}}_2)^{(1)} \nonumber \\
&={{\mathbb C}}r\otimes \big( {{\mathcal B}}({{\mathcal H}}_2)^{(0)}\pm {{\mathcal B}}({{\mathcal H}}_2)^{(1)}\big)
={{\mathbb C}}r\otimes {{\mathcal B}}({{\mathcal H}}_2)
={\mathop{\mathrm{Ad}}\nolimits}_V{\left (}p {{{\mathcal B}}({{\mathcal H}})}p{\right )}. \end{aligned}$$ Hence we obtain ${{\mathcal M}}_2 p=p{{{\mathcal B}}({{\mathcal H}})}p={{\mathcal B}}(p{{\mathcal H}})$.
\[lem:endo\_and\_commutant\_even\] Consider the setting of Proposition \[cuntzpropi\]. Then the following hold.
1. $\rho({{{\mathcal B}}({{\mathcal H}})})'={{\mathcal M}}^{(0)}+{{\mathcal M}}^{(1)}\Gamma_0\Gamma$.
2. Let $\hat{E}_{\mu,\nu} = E_{\mu,\nu} (\Gamma_0 \Gamma)^{|\mu|+|\nu|}$. Then $\{\hat{E}_{\mu,\nu}\}_{\mu,\nu\in\calP}$ are matrix units in $\rho(\calB(\calH))'$ spanning $\rho(\calB(\calH))'$,
3. For all $\mu \in \calP$, the map $$\begin{aligned}
\label{rmu1}
\rho_\mu: \calB(\calH) \ni x \mapsto \rho(x) E_{\mu,\mu} \in \calB( E_{\mu,\mu} \calH)
\end{aligned}$$ is a $\ast$-isomorphism.
Note that ${\mathop{\mathrm{Ad}}\nolimits}_{\Gamma}(x)={\mathop{\mathrm{Ad}}\nolimits}_{\Gamma_0}(x)$ for $x\in {{\mathcal M}}$. Applying Lemma \[lem35\] with ${{\mathcal M}}_1={{\mathcal M}}$, ${{\mathcal M}}_2=\rho({{{\mathcal B}}({{\mathcal H}})})$ and $\Gamma_1=\Gamma_0$, we immediately obtain (i). Because $\{E_{\mu,\nu}\}_{\mu,\nu\in\calP}$ are matrix units spanning ${{\mathcal M}}$ and satisfying , we see from (i) that $$\begin{aligned}
\rho({{\mathcal B}}({{\mathcal H}}))'={{\mathcal M}}^{(0)}+{{\mathcal M}}^{(1)}\Gamma_0\Gamma
= \mathrm{span}_{\mu,\nu\in {{\mathcal P}}}\left\{
E_{\mu,\nu}{\left (}\Gamma_0\Gamma{\right )}^{|\mu|+|\nu|}\right\}
= \mathrm{span}_{\mu,\nu\in {{\mathcal P}}}
\left\{\hat E_{\mu,\nu}\right\}.\end{aligned}$$ Because $\Gamma_0\Gamma$ commutes with $E_{\mu,\nu}$, it is straight forward to check that $\{\hat E_{\mu,\nu}\}_{\mu,\nu\in{{\mathcal P}}}$ are matrix units. Hence we obtain (ii).
For part (iii), we first note that because $E_{\mu,\mu}$ is even, $[\rho(x), E_{\mu,\mu}] = 0$ for all $x \in \calB(\calH)$. Therefore there is a well-defined $\ast$-homomorphism $$\rho_\mu : \calB(\calH) \to \calB(E_{\mu,\mu} \calH), \qquad
\rho_\mu(x) = \rho(x) E_{\mu,\mu}, \,\,\, x\in \calB(\calH).$$ Because $\calB(\calH)$ is a factor, $\rho_\mu$ is injective. To see that $\rho_\mu$ is surjective, we note that $E_{\mu\mu}$ is a minimal projection of ${{\mathcal M}}$ and it is even. Then applying Lemma \[lem35\] with ${{\mathcal M}}_1={{\mathcal M}}$ and ${{\mathcal M}}_2=\rho({{\mathcal B}}({{\mathcal H}}))$, we obtain $
\rho({{{\mathcal B}}({{\mathcal H}})})\cdot E_{\mu\mu}={{\mathcal B}}(E_{\mu\mu}{{\mathcal H}})
$ and so $\rho_\mu$ is surjective.
We now prove Proposition \[cuntzpropi\], which we split into two lemmas. We recall the matrix units $\{E_{\mu,\nu}\}_{\mu,\nu\in\calP}\subset {{\mathcal M}}$ and $\hat{E}_{\mu,\nu}= E_{\mu,\nu} (\Gamma_0 \Gamma)^{|\mu|+|\nu|}$ from Lemma \[lem:endo\_and\_commutant\_even\].
\[First part of Proposition \[cuntzpropi\]\]\[lemma:Even\_cuntz\_property\] Consider the setting of Proposition \[cuntzpropi\]. Then there exist isometries $\{S_\mu\}_{\mu \in \calP}$ on ${{\mathcal H}}$ with the property that for all $\mu,\nu \in\calP$ and $x \in \calB(\calH)$, $$\begin{aligned}
\label{sprop}
S_\nu^* S_\mu = \delta_{\mu,\nu} {\mathbb I}, \qquad S_\mu S_\nu^* = \hat{E}_{\mu,\nu}, \qquad
\rho(x) E_{\mu,\mu}= S_\mu x S_\mu^*, \qquad \rho(x) = \sum_{\mu\in\calP} S_\mu x S_\mu^*.
\end{aligned}$$ The operators $S_\mu$ have homogeneous parity and are such that ${\mathop{\mathrm{Ad}}\nolimits}_\Gamma{\left (}S_\mu{\right )}= (-1)^{|\mu|+\sigma_0}S_\mu$, with some uniform $\sigma_0\in\{0,1\}$. They also satisfy $\Gamma_0 S_\mu = (-1)^{|\mu|}S_\mu$.
By part (iii) of Lemma \[lem:endo\_and\_commutant\_even\], $\rho_\mu$ in is a $*$-isomorphism $\calB(\calH) \xrightarrow{\rho_\mu} \calB( {E}_{\mu,\mu} \calH)$. Therefore we can apply Wigner’s Theorem to obtain a unitary $w_\mu: \calH \to {E}_{\mu,\mu} \calH$ such that $ \rho_\mu = \mathrm{Ad}_{w_\mu}$. Note that $$w_\mu^* w_\nu = w_\mu^* E_{\mu,\mu} E_{\nu,\nu} w_\nu = \delta_{\mu,\nu} \, {\mathbb I}_{\calH}, \quad
\mu,\,\nu \in \calP.$$ We also see that, because $\sum_\mu E_{\mu,\mu} = {\mathbb I}$, $$\begin{aligned}
\rho(x) = \sum_\mu \rho(x) E_{\mu,\mu} = \sum_\mu \rho_\mu(x) = \sum_\mu w_\mu x w_\mu^*,\quad
x\in{{{\mathcal B}}({{\mathcal H}})}.\end{aligned}$$ We use the above property to compute that for any $x \in\calB(\calH)$, $$\begin{aligned}
w_\mu w_\nu^* \rho(x) &= w_\mu w_\nu^* \big( \sum_\lambda w_\lambda x w_\lambda^* \big)
= w_\mu x w_\nu^*
= \big( \sum_{\lambda} w_\lambda x w_\lambda^*\big) w_\mu w_\nu^*
= \rho(x) w_\mu w_\nu^*.\end{aligned}$$ Therefore $w_\mu w_\nu^* \in \rho(\calB(\calH))'$ for any $\mu, \nu \in \calP$.
Summarizing our results so far, we have obtained a collection of operators $\{w_\mu w_\nu^*\}_{\mu,\nu \in \calP}$ in $\rho(\calB(\calH))'$ such that $$\label{eq:Smu_prop}
\hat{E}_{\mu,\mu} w_\mu w_\nu^* \hat{E}_{\nu,\nu} = w_\mu w_\nu^* .$$ From , and (ii) of Lemma \[lem:endo\_and\_commutant\_even\], that there is some $c_{\mu\nu} \in {\ensuremath{\mathbb{C}}}$ such that $$w_\mu w_\nu^* = c_{\mu \nu} \hat{E}_{\mu,\nu}.$$ Note that $c_{\mu\nu}=\overline{c_{\nu\mu}}$. Because of the definition, we have $w_\mu w_\mu^*=\hat E_{\mu\mu}$ and we see that $c_{\mu\mu} = 1$. On the other hand, because of $w_\nu^* w_\nu={\mathbb I}_{{{\mathcal H}}}$, we have $$c_{\mu \lambda} \hat{E}_{\mu,\lambda} = w_\mu w_\lambda^*
= w_\mu w_\nu^* w_\nu w_\lambda^* = c_{\mu \nu} c_{\nu \lambda} \hat{E}_{\mu,\lambda}$$ and so $c_{\mu \lambda} = c_{\mu \nu} c_{\nu \lambda}$. In particular, $1=c_{\mu \mu} = c_{\mu \nu} c_{\nu \mu}=|c_{\mu\nu}|^2$ and so $c_{\mu\nu}\in{{\mathbb T}}$. Now setting $\mu_0 :=\emptyset\in \calP$ and define $S_\mu = c_{\mu_0 \mu} w_\mu$ for every $\mu \in \calP$. Then because of the above properties of $c_{\mu\nu}$, the collection $\{S_\mu\}_{\mu\in \calP}$ has the same algebraic properties as $\{w_\mu\}$ as well as that $S_\mu S_\nu^* = \hat{E}_{\mu,\nu}$ as required. Hence we obtain .
Next, we recall the grading operator $\Gamma_0 = \sum_\mu (-1)^{|\mu|} E_{\mu,\mu}$ of $\calM$. Because $S_\mu$ is an isometry onto $E_{\mu,\mu} \calH$ $$\Gamma_0 S_\mu = \Gamma_0 E_{\mu,\mu} S_\mu = (-1)^{|\mu|} E_{\mu,\mu} S_\mu = (-1)^{|\mu|} S_\mu.$$ We now consider the grading of $S_\mu$, $\mathrm{Ad}_{\Gamma}(S_\mu)$. We compute that for any $x \in\calB(\calH)$, $$\begin{aligned}
\Gamma S_\mu x S_\mu^* \Gamma &= \Gamma \rho(x) E_{\mu,\mu} \Gamma
= \Gamma \rho(x) \Gamma E_{\mu,\mu} = \rho( \Gamma x \Gamma) E_{\mu,\mu} =
S_\mu \Gamma x \Gamma S_\mu^*\end{aligned}$$ as $E_{\mu,\mu}$ is even and $\rho$ commutes with the grading. Multiplying $\Gamma S_\mu^*$ from the left and $\Gamma S_\mu$ from the right of the equation, we see that $\Gamma S_\mu^* \Gamma S_\mu \in \calB(\calH)' = {\ensuremath{\mathbb{C}}}{\mathbb I}_\calH$. Note that $\Gamma S_\mu^* \Gamma S_\mu$ is unitary because ${\mathop{\mathrm{Ad}}\nolimits}_\Gamma(E_{\mu\mu})=E_{\mu\mu}$. So $S_\mu^* \Gamma S_\mu= e^{i\varphi} \Gamma$ with some $e^{i\varphi}\in {{\mathbb T}}$. Multiplying $S_\mu$ from the left, and by $\Gamma$ from the right, we obtain $\Gamma S_\mu\Gamma=E_{\mu\mu}\Gamma S_\mu\Gamma= S_\mu S_\mu^* \Gamma S_\mu\Gamma=e^{i\varphi} S_\mu$. But because $({\mathop{\mathrm{Ad}}\nolimits}_{\Gamma})^2={\mathop{\mathrm{id}}\nolimits}$, $(e^{i\varphi})^2=1$ and ${\mathop{\mathrm{Ad}}\nolimits}_{\Gamma}(S_\mu) = (-1)^{b_\mu} S_\mu$ with some $b_\mu=0,1$.
Let us further examine the grading of the operator $S_\mu$. We compute that $$\begin{aligned}
\Gamma \hat{E}_{\mu,\nu} \Gamma
&= \Gamma E_{\mu,\nu} (\Gamma_0 \Gamma)^{|\mu|+|\nu|} \Gamma
= \Gamma E_{\mu,\nu} \Gamma (\Gamma_0 \Gamma)^{|\mu|+|\nu|} \\
&= (-1)^{|\mu|+|\nu|} E_{\mu,\nu} (\Gamma_0 \Gamma)^{|\mu|+|\nu|}
= (-1)^{|\mu|+|\nu|} \hat{E}_{\mu,\nu}\end{aligned}$$ while we also find $$\begin{aligned}
\Gamma \hat{E}_{\mu,\nu} \Gamma &= \Gamma S_\mu \Gamma \Gamma S_\nu^* \Gamma
= (-1)^{b_\mu} (-1)^{b_\nu} S_\mu S_\nu^*
= (-1)^{b_\mu+ b_\nu} \hat{E}_{\mu,\nu}.\end{aligned}$$ Therefore $|\mu|+|\nu| = b_\mu+b_\nu \in {\ensuremath{\mathbb{Z}}}_2$. By setting $\mu_0:=\emptyset\in \calP$ and $\sigma_0:=b_{\mu_0}$, we have that $\Gamma S_\mu \Gamma$ = $(-1)^{|\mu|+\sigma_0}S_\mu$ for all $\mu \in \calP$.
\[lem:Even\_Cuntz2\] Consider the setting of Proposition \[cuntzpropi\]. For $S_\mu$ of Lemma \[lemma:Even\_cuntz\_property\], set $B_\mu:=(\Gamma_0\Gamma)^{|\mu|} S_\mu$, for $\mu\in{{\mathcal P}}$. Then $B_\nu^* B_\mu = \delta_{\mu,\nu} {\mathbb I}$, $B_\mu B_\nu^*= E_{\mu,\nu}$, $$\begin{aligned}
\rho(x)
= \sum_{\mu\in\calP} \mathrm{Ad}_{B_\mu} \circ \mathrm{Ad}_{\Gamma^{|\mu|}} (x)
&=\sum_{\mu\in\calP} \mathrm{Ad}_{\Gamma^{|\mu|}}\circ\mathrm{Ad}_{B_\mu} ,\quad
x\in{{{\mathcal B}}({{\mathcal H}})}, \nonumber \\
\label{eq:rho_and_E_using_V_even_again}
E_{\mu_0,\nu_0} \rho(E_{\mu_1,\nu_1}) \cdots \rho^N (E_{\mu_N,\mu_N}) &=
(-1)^{\sum\limits_{k=1}^N (|\mu_k|+|\nu_k|) \sum\limits_{j=0}^{k-1} |\nu_j|}
B_{\mu_0} \cdots B_{\mu_N} B_{\nu_N}^* \cdots B_{\nu_0}^*,\end{aligned}$$ for all $N\in{{\mathbb N}}\cup\{0\}$ and $\mu_0,\ldots\mu_N,\nu_0,\ldots,\nu_N\in{{\mathcal P}}$. The operators $B_\mu$ have homogeneous parity and are such that ${\mathop{\mathrm{Ad}}\nolimits}_\Gamma{\left (}B_\mu{\right )}= (-1)^{|\mu|+\sigma_0}B_\mu$, with the same $\sigma_0\in\{0,1\}$ as in Lemma \[lemma:Even\_cuntz\_property\]. If there are isometries $\{T_\mu\}_{\mu \in \calP}$ satisfying , then there is some $c\in{{\mathbb T}}$ such that $T_\mu=cB_\mu$, for all $\mu\in {{\mathcal P}}$.
From Lemma \[lemma:Even\_cuntz\_property\], we check that $$\begin{aligned}
\label{bbn}
B_\mu^* B_\nu = S_\mu^* (\Gamma_0 \Gamma)^{|\mu|+|\nu|} S_\nu
= S_\mu^* S_\nu \Gamma^{|\mu|+|\nu|}(-1)^{(|\nu|+\sigma_0)(|\mu|+|\nu|)} (-1)^{|\nu|(|\mu|+|\nu|)}
= \delta_{\mu,\nu} {{{\mathbb I}}}.\end{aligned}$$ We also have from Lemma \[lemma:Even\_cuntz\_property\] that $$\begin{aligned}
\label{bbe}
B_\mu B_\nu^* &= (\Gamma_0 \Gamma)^{|\mu|} S_\mu S_\nu^* (\Gamma_0 \Gamma)^{|\nu|}
= (\Gamma_0 \Gamma)^{|\mu|} E_{\mu,\nu} (\Gamma_0\Gamma)^{|\mu|+|\nu|} (\Gamma_0 \Gamma)^{|\nu|}
= E_{\mu,\nu}\end{aligned}$$ because $\Gamma_0 \Gamma$ commutes with ${{\mathcal M}}$. Because $S_\mu$ has homogenous parity and $\Gamma_0 \Gamma$ is even, $B_\mu$ has the same homogeneous parity as $S_\mu$. In particular, ${\mathop{\mathrm{Ad}}\nolimits}_\Gamma{\left (}B_\mu{\right )}= (-1)^{|\mu|+\sigma_0}B_\mu$, with the same $\sigma_0\in\{0,1\}$ as in Lemma \[lemma:Even\_cuntz\_property\]. This implies that the endomorphism $\mathrm{Ad}_{B_\mu}$ respects the grading on $\calB({{\mathcal H}})$, i.e., ${\mathop{\mathrm{Ad}}\nolimits}_\Gamma\circ{{\mathop{\mathrm{Ad}}\nolimits}}_{B_\mu}={{\mathop{\mathrm{Ad}}\nolimits}}_{B_\mu}\circ{\mathop{\mathrm{Ad}}\nolimits}_\Gamma$. Furthermore, using that $\Gamma_0 S_\mu = (-1)^{|\mu|}S_\mu$, $\mathrm{Ad}_{S_\mu} = \mathrm{Ad}_{\Gamma^{|\mu|}B_\mu}= \mathrm{Ad}_{B_\mu \Gamma^{|\mu|}} $. We therefore see that for $x\in\calB({{\mathcal H}})$ $$\begin{aligned}
\rho(x) &= \sum_{\mu \in\calP} S_\mu x S_\mu^*
= \sum_{\mu\in\calP} \mathrm{Ad}_{B_\mu} \circ \mathrm{Ad}_{\Gamma^{|\mu|}} (x).\end{aligned}$$ A simple induction argument using that $\mathrm{Ad}_{B_\mu}$ commutes with $\mathrm{Ad}_\Gamma$ gives that $$\label{eq:iterated_rho_even_V}
\rho^N(x) = \sum_{\lambda_0,\ldots,\lambda_{N-1} \in\calP} \!\!
\mathrm{Ad}_{B_{\lambda_0}\cdots B_{\lambda_{N-1}}} \circ
\mathrm{Ad}_{\Gamma^{|\lambda_0|+\cdots |\lambda_{N-1}|}} (x).$$ We now consider $\rho(E_{\mu,\nu})$. Recalling and that $\mathrm{Ad}_\Gamma (E_{\mu,\nu}) = (-1)^{|\mu|+|\nu|} E_{\mu,\nu}$, we see that $$\rho(E_{\mu,\nu}) = \sum_\lambda B_\lambda \Gamma^{|\lambda|} E_{\mu,\nu} \Gamma^{|\lambda|} B_\lambda^*
= \sum_\lambda (-1)^{|\lambda|(|\mu|+|\nu|)} B_\lambda B_\mu B_\nu^* B_\lambda^*.$$ From this, and , we have $$E_{\mu_0,\nu_0} \rho( E_{\mu_1,\nu_1} )
= B_{\mu_0} B_{\nu_0}^* \sum_\lambda (-1)^{|\lambda|(|\mu_1|+|\nu_1|)}
B_\lambda B_{\mu_1} B_{\nu_1}^* B_\lambda^*
= (-1)^{|\nu_0|(|\mu_1|+|\nu_1|)} B_{\mu_0} B_{\mu_1} B_{\nu_1}^* B_{\nu_0}^*.$$ This proves Equation in the case of $N=1$. We now assume the equality is true for $N$ and consider $N+1$. Using Equation , , , we compute that $$\begin{aligned}
&E_{\mu_0,\nu_0} \rho(E_{\mu_1,\nu_1}) \cdots \rho^N(E_{\mu_N,\nu_N}) \rho^{N+1}(E_{\mu_{N+1},\nu_{N+1}}) \\
&=
(-1)^{\sum\limits_{k=1}^N (|\mu_k|+|\nu_k|) \sum\limits_{j=0}^{k-1} |\nu_j|}
B_{\mu_0} \cdots B_{\mu_N} B_{\nu_N}^* \cdots B_{\nu_0}^* \, \rho^{N+1}(E_{\mu_{N+1},\nu_{N+1}}) \\
&= (-1)^{\sum\limits_{k=1}^N (|\mu_k|+|\nu_k|) \sum\limits_{j=0}^{k-1} |\nu_j|}
B_{\mu_0} \cdots B_{\mu_N} B_{\nu_N}^* \cdots B_{\nu_0}^*
\Big( \sum_{\lambda_0,\ldots,\lambda_N} \mathrm{Ad}_{B_{\lambda_0}\cdots B_{\lambda_N}}
\circ \mathrm{Ad}_{\Gamma^{\sum\limits_{j=0}^N |\lambda_j|}} ( E_{\mu_{N+1},\nu_{N+1}}) \Big) \\
&= (-1)^{\sum\limits_{k=1}^N (|\mu_k|+|\nu_k|) \sum\limits_{j=0}^{k-1} |\nu_j|}
B_{\mu_0} \cdots B_{\mu_N}
\big( (-1)^{(|\mu_{N+1}|+|\nu_{N+1}|)\sum\limits_{j=0}^N |\nu_j|} B_{\mu_{N+1}}B_{\nu_{N+1}}^*\big)
B_{\nu_N}^* \cdots B_{\nu_0}^* \\
&= (-1)^{\sum\limits_{k=1}^{N+1} (|\mu_k|+|\nu_k|) \sum\limits_{j=0}^{k-1} |\nu_j|}
B_{\mu_0} \cdots B_{\mu_N}B_{\mu_{N+1}}B_{\nu_{N+1}}^* B_{\nu_N}^* \cdots B_{\nu_0}^* \end{aligned}$$ as required.
To show the last statement, suppose $\{{T}_\mu\}_{\mu\in\calP} \subset \calB({{\mathcal H}})$ satisfy . Because $$\begin{aligned}
\sum_{\lambda\in\calP} \mathrm{Ad}_{T_\lambda} \circ \mathrm{Ad}_{\Gamma^{|\lambda|}} (x)
=\rho(x)
= \sum_{\lambda\in\calP} \mathrm{Ad}_{B_\lambda} \circ \mathrm{Ad}_{\Gamma^{|\lambda|}} (x),\quad
x\in{{{\mathcal B}}({{\mathcal H}})},\end{aligned}$$ multiplying $T_\nu^*$ from the left and by $B_\nu$ from the right, we obtain $$\begin{aligned}
\mathrm{Ad}_{\Gamma^{|\nu|}} (x)\cdot T_\nu^*B_\nu
= T_\nu^* B_\nu\cdot \mathrm{Ad}_{\Gamma^{|\nu|}} (x),\quad
x\in{{{\mathcal B}}({{\mathcal H}})}.\end{aligned}$$ Hence we obtain $T_\nu^* B_\nu\in {{\mathbb C}}{\mathbb I}_{{{\mathcal H}}}$, i.e., we have $T_\nu^* B_\nu=c_\mu{\mathbb I}_{{{\mathcal H}}}$ for some $c_\mu\in{{\mathbb C}}$. We then have $$\begin{aligned}
B_\nu=
E_{\nu\nu}B_{\nu}=T_\nu T_\nu^*B_\nu
=c_\nu T_\nu.\end{aligned}$$ By $B_\nu^* B_\nu=T_\nu^* T_\nu={\mathbb I}_{{{\mathcal H}}}$, we see that $c_\nu\in{{\mathbb T}}$. Furthermore, from $B_\mu B_\nu^*=T_\mu T_\nu^*=E_{\mu\nu}$, we see that $c_\mu=c_\nu=:c\in{{\mathbb T}}$.
Lemma \[lemma:Even\_cuntz\_property\] and \[lem:Even\_Cuntz2\] complete the proof of Proposition \[cuntzpropi\]. We are ready to show Theorem \[cuntzthm0\].
We fix a $W^*$-$(G,{\mathfrak{p}})$-dynamical system $({{\mathcal R}}_{0,{{\mathcal K}}},{\mathop{\mathrm{Ad}}\nolimits}_{\Gamma_{{\mathcal K}}},{\mathop{\mathrm{Ad}}\nolimits}_{V_g})\in{{\mathcal S}}_{0}$ that is equivalent to $(\pi_\omega({{\mathcal A}}_R)'', {\mathop{\mathrm{Ad}}\nolimits}_{\Gamma_\omega}, \hat\alpha_\omega)$ and the endomorphism $\rho$ of Lemma \[iorho\]. Then the Hilbert space ${{\mathcal K}}\otimes{{{\mathbb C}}^2}$, self-adjoint unitary $\Gamma_{{\mathcal K}}$, finite type I factor $\iota\circ\pi_\omega({{\mathcal A}}_{\{0\}})$ with matrix units $\{\iota\circ\pi_\omega\circ \big( E_{\mu,\nu}^{(0)}\big) \}_{\mu,\nu\in\calP}
\subset \calB({{\mathcal K}}\otimes {{\mathbb C}}^2)$ and $\rho$ satisfy the hypothesis of Proposition \[cuntzpropi\]. Applying Proposition \[cuntzpropi\], we obtain the isometries $\{B_\mu\}$ such that $B_\mu^* B_\nu = \delta_{\mu,\nu}{\mathbb I}$ and which satisfy and from the statement of the Theorem. To show , set $\Gamma_0 := \iota\circ\pi_\omega \big( \mathfrak\Gamma(-{\mathbb I}) \big)=
\sum_\mu (-1)^{|\mu|} \iota\circ\pi_\omega \big( E_{\mu,\mu}^{(0)} \big)$. We claim for a homogeneous $x\in {{\mathcal R}}_{0,{{\mathcal K}}}$, and $N\in{{\mathbb N}}$ that $$\begin{aligned}
\label{pn}
T^N_{\bf B}(x) = {\Gamma}_0^{{\partial x}} \rho({\Gamma}_0^{{\partial x}} ) \cdots \rho^{N-1}({\Gamma}_0^{{\partial x}}) \rho^N(x).\end{aligned}$$ First set $\Gamma_1 := \sum_\mu \Gamma_{{{\mathcal K}}}^{|\mu|} \iota\circ\pi_\omega \big( E_{\mu,\mu}^{(0)} \big)$, which is a self-adjoint unitary. Because of with $N=0$, we have $\iota\circ\pi_\omega \big(E_{\mu\mu}^{(0)}\big)=B_\mu B_\mu^*$. Therefore, we have $$\begin{aligned}
\rho\circ\iota\circ\pi_\omega(A)
= \sum_{\mu\in\calP} \mathrm{Ad}_{\Gamma_{{{\mathcal K}}}^{|\mu|}B_\mu}\circ\iota\circ\pi_\omega (A)
={\mathop{\mathrm{Ad}}\nolimits}_{\Gamma_1}\circ T_{\bf B}\circ\iota\circ\pi_\omega (A),\quad
A\in{{\mathcal A}}_R.\end{aligned}$$ Hence we obtain for any homogeneous $x\in{{\mathcal R}}_{0,{{\mathcal K}}}$, $$\begin{aligned}
\label{tbx}
T_{\bf B}(x)={\mathop{\mathrm{Ad}}\nolimits}_{\Gamma_1}\circ \rho(x)
&=\sum_{\mu,\nu}
\Gamma_{{{\mathcal K}}}^{|\mu|}
{\left (}\iota\circ\pi_\omega \big( E_{\mu,\mu}^{(0)} \big) {\right )}\rho(x)
\Gamma_{{{\mathcal K}}}^{|\nu|}{\left (}\iota\circ\pi_\omega \big( E_{\nu,\nu}^{(0)} \big) {\right )}\\
&=\sum_{\mu}\iota\circ\pi_\omega\big( E_{\mu,\mu}^{(0)}\big)
\Gamma_{{{\mathcal K}}}^{|\mu|}
\rho(x)
\Gamma_{{{\mathcal K}}}^{|\mu|}. \nonumber \\
&=\sum_{\mu}\iota\circ\pi_\omega \big( E_{\mu,\mu}^{(0)}\big)
\rho\circ{\mathop{\mathrm{Ad}}\nolimits}_{\Gamma_{{{\mathcal K}}}^{|\mu|}}(x). \nonumber \\
&=\sum_{\mu}\iota\circ\pi_\omega \big( E_{\mu,\mu}^{(0)} \big)
(-1)^{|\mu|\partial x}
\rho(x)
=\Gamma_0^{\partial x}\rho(x), \nonumber\end{aligned}$$ where in the third equality we used that $ \iota\circ\pi_\omega \big( E_{\nu,\nu}^{(0)} \big)$ commutes with $\Gamma_{{\mathcal K}}$ and elements from $\rho({{\mathcal R}}_{0,{{\mathcal K}}})$. This proves for the case $N=1$. Now we proceed by induction and suppose that holds for $N$. Then using and the induction assumption, for any homogeneous $x\in{{\mathcal R}}_{0,{{\mathcal K}}}$, $$\begin{aligned}
T^{N+1}_{\bf B}(x)
&= T_{\bf B}{\left (}{\Gamma}_0^{{\partial x}} \rho({\Gamma}_0^{{\partial x}} ) \cdots \rho^{N-1}({\Gamma}_0^{{\partial x}}) \rho^N(x){\right )}\nonumber\\
&=\Gamma_0^{\partial {\left (}{\Gamma}_0^{{\partial x}} \rho({\Gamma}_0^{{\partial x}} ) \cdots \rho^{N-1}({\Gamma}_0^{{\partial x}}) \rho^N(x){\right )}}
\rho( {\Gamma}_0^{{\partial x}} )
\rho^2({\Gamma}_0^{{\partial x}} ) \cdots \rho^{N}({\Gamma}_0^{{\partial x}}) \rho^{N+1}(x) \nonumber \\
&=
\Gamma_0^{\partial x }\rho({\Gamma}_0^{{\partial x}} )
\rho^2({\Gamma}_0^{{\partial x}} ) \cdots \rho^{N}({\Gamma}_0^{{\partial x}}) \rho^{N+1}(x).\end{aligned}$$ Hence holds for $N+1$ and proves the claim.
Now we show . Because $\kappa_\omega = 0$, $\pi_\omega({{\mathcal A}}_R)''$ is a factor. Therefore, for any homogeneous $x\in{{\pi_{\omega}}}({{\mathcal A}}_R)''$, the sequence $$\begin{aligned}
T^N_{\bf B}\circ\iota(x)
&= {\Gamma}_0^{{\partial x}} \rho({\Gamma}_0^{{\partial x}} ) \cdots \rho^{N-1}({\Gamma}_0^{{\partial x}}) \rho^N\circ\iota(x) \nonumber\\
&=\iota\circ{\left (}\pi_\omega{\left (}{\mathfrak\Gamma}(-{\mathbb I})^{\partial x}
\beta_{S_1}\big({\mathfrak\Gamma}(-{\mathbb I})^{\partial x}\big) \cdots
\beta_{S_{N-1}}\big({\mathfrak\Gamma}(-{\mathbb I})^{\partial x}\big) {\right )}\hat \beta_{S_N}(x){\right )}\end{aligned}$$ converges to ${\langle{{\Omega_\omega}},x{{\Omega_\omega}}\rangle}{\mathbb I}_{{{\mathcal K}}\otimes{{{\mathbb C}}^2}}$ in the $\sigma$-weak topology by Lemma \[lem:trans\_weak\_conv\_to\_state\]. This proves .
To prove set $$\begin{aligned}
T_\nu := \sum_{\lambda\in\calP} {\overline}{ \langle \psi_\lambda , \mathfrak{\Gamma}(U_g)^* \psi_\nu \rangle }^{{{\mathfrak{p}}(g)}}
V_g B_\lambda V_g^* ,\quad \nu\in{{\mathcal P}}.
\end{aligned}$$ Recall that for $c\in{{\mathbb C}}$, $\overline{c}^{{\mathfrak{p}}(g)} = c$ for ${\mathfrak{p}}(g)=0$ and $\bar c$ if ${{\mathfrak{p}}}(g)=1$. We claim that $\{T_\mu\}_{\mu\in{{\mathcal P}}}$ satisfies with $E_{\mu\nu}$ and $\Gamma$ replaced by $\iota\circ\pi_\omega(E_{\mu\nu}^{(0)})$ and $\Gamma_{{\mathcal K}}$ respectively. We compute $$\begin{aligned}
T_\mu^* T_\nu &= \sum_{\lambda,\zeta} {\overline}{ \langle \psi_\lambda , \mathfrak{\Gamma}(U_g)^* \psi_\mu \rangle }^{{{\mathfrak{p}}(g)}+1}
{\overline}{ \langle \psi_\zeta , \mathfrak{\Gamma}(U_g)^* \psi_\nu \rangle }^{{{\mathfrak{p}}(g)}} V_g B_\lambda^* B_\zeta V_g^* \\
&= \sum_{\lambda} {\overline}{ \langle \mathfrak{\Gamma}(U_g)^* \psi_\mu , \psi_\lambda \rangle \,
\langle \psi_\lambda , \mathfrak{\Gamma}(U_g)^* \psi_\nu \rangle}^{{{\mathfrak{p}}(g)}}{\mathbb I}= \delta_{\mu,\nu} \, {{{\mathbb I}}}.\end{aligned}$$ To see the second property of , note that $$\begin{aligned}
{\mathfrak{\Gamma}(U_g)}^* E_{\mu,\nu}^{(0)} {\mathfrak{\Gamma}(U_g)} &=
\sum_{\lambda,\zeta\in\calP} {\overline}{ \langle \psi_\nu, {\mathfrak{\Gamma}(U_g)} \psi_\zeta\rangle }^{{\mathfrak{p}}(g)} \,
\langle \psi_\lambda, {\mathfrak{\Gamma}(U_g)}^* \psi_\mu \rangle E_{\lambda,\zeta}^{(0)}.
\end{aligned}$$ Using this, we obtain $$\begin{aligned}
T_\mu T_\nu^* &= \sum_{\lambda,\zeta} {\overline}{ \langle \psi_\lambda , \mathfrak{\Gamma}(U_g)^* \psi_\mu \rangle
\langle \mathfrak{\Gamma}(U_g)^* \psi_\nu, \psi_\zeta \rangle }^{{{\mathfrak{p}}(g)}}
V_g B_\lambda B_\zeta^* V_g^* \\
&= \sum_{\lambda,\zeta} {\overline}{ \langle \psi_\lambda , \mathfrak{\Gamma}(U_g)^* \psi_\mu \rangle
\langle \mathfrak{\Gamma}(U_g)^* \psi_\nu, \psi_\zeta \rangle }^{{{\mathfrak{p}}(g)}}
\iota\circ \pi_\omega\big( \mathfrak{\Gamma}(U_g) E_{\lambda,\zeta}^{(0)}\mathfrak{\Gamma}(U_g)^* \big) \\
&= \iota \circ \pi_\omega \circ \mathrm{Ad}_{\mathfrak{\Gamma}(U_g)} \Big(
\sum_{\lambda,\zeta} { \langle \psi_\lambda , \mathfrak{\Gamma}(U_g)^* \psi_\mu \rangle
\langle \mathfrak{\Gamma}(U_g)^* \psi_\nu, \psi_\zeta \rangle } E_{\lambda,\zeta}^{(0)} \Big) \\
&= \iota \circ \pi_\omega \circ \mathrm{Ad}_{\mathfrak{\Gamma}(U_g)} \big(
\mathfrak{\Gamma}(U_g)^* E_{\mu,\nu}^{(0)} \mathfrak{\Gamma}(U_g) \big)
= \iota\circ \pi_\omega( E_{\mu,\nu}^{(0)}).\end{aligned}$$
To check the third property of , note that $\langle \psi_\mu, {\mathfrak{\Gamma}(U_g)}^* \psi_\nu\rangle = 0$ if $|\mu|\neq |\nu|$, because ${\mathfrak{\Gamma}(U_g)}$ commutes with $\mathfrak{\Gamma}(-{\mathbb I}_{{{\mathbb C}}^d})$. Using this, we check that $$\begin{aligned}
&\sum_{\mu\in\calP} \mathrm{Ad}_{T_\mu} \circ \mathrm{Ad}_{\Gamma_{{{\mathcal K}}}^{|\mu|}} (\iota \circ \pi_\omega(A) ) \\
&= \sum_{\mu} \sum_{\lambda,\zeta} \langle {\overline}{\psi_\lambda, \mathfrak{\Gamma}(U_g)^* \psi_\mu\rangle }^{{{\mathfrak{p}}(g)}}
{\overline}{ \langle \psi_\zeta, \mathfrak{\Gamma}(U_g)^* \psi_\mu \rangle }^{{{\mathfrak{p}}(g)}+1}
\delta_{|\mu|,|\lambda|}
V_g B_\lambda V_g^* \Gamma_{{{\mathcal K}}}^{|\mu|} (\iota\circ \pi_\omega)(A) \Gamma_{{{\mathcal K}}}^{|\mu|}
V_g B_\zeta^* V_g^* \\
&= \sum_{\lambda,\zeta} \sum_\mu {\overline}{ \langle \psi_\lambda, \mathfrak{\Gamma}(U_g)^* \psi_\mu\rangle
\langle \mathfrak{\Gamma}(U_g)^* \psi_\mu, \psi_\zeta \rangle }^{{{\mathfrak{p}}(g)}}
V_g B_\lambda V_g^* \Gamma_{{{\mathcal K}}}^{|\lambda|} (\iota\circ \pi_\omega)(A) \Gamma_{{{\mathcal K}}}^{|\lambda|}
V_g B_\zeta^* V_g^* \\
&= \sum_\lambda V_g B_\lambda V_g^* \Gamma_{{{\mathcal K}}}^{|\lambda|} (\iota\circ \pi_\omega)(A) \Gamma_{{{\mathcal K}}}^{|\lambda|}
V_g B_\lambda^* V_g^* \\
&= \sum_{\lambda} V_g \big(\mathrm{Ad}_{B_\lambda} \circ \mathrm{Ad}_{\Gamma_{{{\mathcal K}}}^{|\lambda|}}\big)
\big( \iota\circ \pi_\omega \circ \alpha_g^{-1}(A) \big) V_g^*\end{aligned}$$ and recalling , $$\begin{aligned}
\sum_{\mu\in\calP} \mathrm{Ad}_{T_\mu} \circ \mathrm{Ad}_{\Gamma_{{{\mathcal K}}}^{|\mu|}} (\iota \circ \pi_\omega(A) )
&= \mathrm{Ad}_{V_g} \circ \rho\big( \iota\circ \pi_\omega \circ \alpha_g^{-1}(A) \big) \\
&= \mathrm{Ad}_{V_g} \circ \iota\circ \pi_\omega \big(\beta_{S_1} \circ \alpha_g^{-1}(A) \big) \\
&= \iota\circ \pi_\omega \circ \alpha_g \circ \beta_{S_1} \circ \alpha_g^{-1}(A)
= \rho\circ \iota\circ\pi_\omega(A),\end{aligned}$$ for all $A\in{{\mathcal A}}_R$. Hence we have proven that $\{T_\mu\}_{\mu\in{{\mathcal P}}}$ satisfies . Applying Proposition \[cuntzpropi\], there is some $c_g \in \mathbb{T}$ such that $B_\mu = c_g T_\mu$ for all $\mu\in{{\mathcal P}}$. Therefore $$\begin{aligned}
\sum_{\mu}{\overline}{c_g} \langle \psi_\mu, \mathfrak{\Gamma}(U_g)\psi_\nu \rangle
B_\mu
&= \sum_{\mu,\lambda} \langle \psi_\mu, \mathfrak{\Gamma}(U_g)\psi_\nu \rangle
{\overline}{ \langle \psi_\lambda, \mathfrak{\Gamma}(U_g)^* \psi_\mu \rangle }^{{{\mathfrak{p}}(g)}} V_g B_\lambda V_g^* \\
&= \sum_{\lambda,\mu} {\overline}{ \langle \mathfrak{\Gamma}(U_g)^* \psi_\mu, \psi_\nu \rangle
\langle \psi_\lambda, \mathfrak{\Gamma}(U_g)^* \psi_\mu \rangle }^{{{\mathfrak{p}}(g)}} V_g B_\lambda V_g^* \\
&= V_g B_\nu V_g^*. \end{aligned}$$ Hence $\sum_\mu \langle \psi_\mu, \, \mathfrak{\Gamma}(U_g) \psi_\nu \rangle B_\mu = c_g V_g B_\nu V_g^*$, which completes the proof.
Case: $\kappa_\omega=1$
-----------------------
We now consider endomorphisms on $W^*$-$(G,{\mathfrak{p}})$-dynamical systems that are equivalent to $({{\mathcal R}}_{1,{{\mathcal K}}},{\mathop{\mathrm{Ad}}\nolimits}_{\Gamma_{{\mathcal K}}},{\mathop{\mathrm{Ad}}\nolimits}_{V_g})\in{{\mathcal S}}_{1}$ from Example \[ni\]. Recall that $\mathfrak{\Gamma}(U_g)$ denotes the second quantization of $U_g$ on $\calF({\ensuremath{\mathbb{C}}}^d)$. Our aim is to prove the following.
\[cuntzthm1\] Let $\omega$ be a pure $\alpha$-invariant and translation invariant split state on ${{\mathcal A}}$. Suppose that the graded $W^*$-$(G,{\mathfrak{p}})$-dynamical system $(\pi_\omega({{\mathcal A}}_R)'', {\mathop{\mathrm{Ad}}\nolimits}_{\Gamma_\omega}, \hat\alpha_\omega)$ associated to $\omega$ is equivalent to $({{\mathcal R}}_{1,{{\mathcal K}}},{\mathop{\mathrm{Ad}}\nolimits}_{\Gamma_{{\mathcal K}}},{\mathop{\mathrm{Ad}}\nolimits}_{V_g})\in{{\mathcal S}}_{1}$ via a $*$-isomorphism $\iota: \pi({{\mathcal A}}_R)''\to {{\mathcal R}}_{1,{{\mathcal K}}}$. Let $\rho$ be the $*$-endomorphism on ${{\mathcal R}}_{1,{{\mathcal K}}}$ given in Lemma \[iorho\]. Then there is some $\sigma_0\in\{0,1\}$ such that $\rho{\left (}{\mathbb I}_{{\mathcal K}}\otimes \sigma_x{\right )}=(-1)^{\sigma_0}\iota\circ\pi_\omega{\left (}\mathfrak\Gamma(-{\mathbb I}){\right )}{\left (}{\mathbb I}_{{\mathcal K}}\otimes \sigma_x{\right )}$ and a set of isometries $\{S_\mu\}_{\mu \in \calP}$ on ${{\mathcal K}}$ such that $S_\nu^* S_\mu = \delta_{\mu,\nu} {\mathbb I}_{{\mathcal K}}$, $$\begin{aligned}
\label{bimpn}
\rho\circ\iota\circ\pi_\omega(A)
= \sum_{\mu\in\calP} \mathrm{Ad}_{\hat S_\mu}\circ\iota\circ\pi_\omega (A)
,\quad
A\in{{\mathcal A}}_R,
\quad \end{aligned}$$ with $\hat S_\mu:=S_\mu \otimes \sigma_z^{\sigma_0+|\mu|}$ and $$\begin{aligned}
\label{bbbon}
\iota\circ\pi_\omega \! {\left (}E_{\mu_0,\nu_0}^{(0)} E_{\mu_1,\nu_1}^{(1)}
\cdots E_{\mu_N,\mu_N}^{(N)}{\right )}=
(-1)^{\sum\limits_{k=1}^N (|\mu_k|+|\nu_k|) \sum\limits_{j=0}^{k-1}{\left (}\sigma_0+|\nu_j|{\right )}}
S_{\mu_0} \cdots S_{\mu_N} S_{\nu_N}^* \cdots S_{\nu_0}^*\otimes \sigma_x^{\sum\limits_{i=0}^{N} |\mu_i|+|\nu_i|} \end{aligned}$$ for all $N\in{{\mathbb N}}\cup\{0\}$ and $\mu_0,\ldots\mu_N,\nu_0,\ldots,\nu_N\in{{\mathcal P}}$. Furthermore, we have $$\begin{aligned}
\label{satn}
\sigma\mathrm{\hbox{-}weak}\lim_{N\to\infty}T_{\bf \hat S}^N\circ\iota(x)
={\langle{{\Omega_\omega}}, x{{\Omega_\omega}}\rangle} \, {\mathbb I}_{{{\mathcal K}}\otimes{{{\mathbb C}}^2}},\quad x\in{{\pi_{\omega}}}({{\mathcal A}}_R)''.\end{aligned}$$ For each $g\in G$, there is some $c_g \in \mathbb{T}$ such that $$\begin{aligned}
\label{gvbn}
(-1)^{{\mathfrak{q}}(g)|\nu|} \sum_{\mu\in\calP} \langle \psi_\mu, \, \mathfrak{\Gamma}(U_g) \psi_\nu \rangle S_\mu =
c_g V_g^{(0)} S_\nu (V_g^{(0)})^*,
\end{aligned}$$ where $V_g^{(0)}$ is given in Lemma \[zv\].
We again will prove this theorem in several steps. Parts of the proof follow the same argument as the case $\kappa_\omega =0$, so some details will be omitted.
\[cuntzpropii\] Let ${{\mathcal K}}$ be a Hilbert space and set $\Gamma_{{\mathcal K}}:={\mathbb I}_{{\mathcal K}}\otimes\sigma_z$ on ${{\mathcal K}}\otimes {{{\mathbb C}}^2}$. We give a grading to ${{\mathcal R}}_{1,{{\mathcal K}}}={{{\mathcal B}}({{\mathcal K}})}\otimes {\mathop{\mathfrak{C}}\nolimits}$ by ${\mathop{\mathrm{Ad}}\nolimits}_{\Gamma_{{\mathcal K}}}$. Suppose that ${{\mathcal N}}$ is a type I subfactor of ${{\mathcal R}}_{1,{{\mathcal K}}}$ with matrix units $\{E_{\mu,\nu}\}_{\mu,\nu\in\calP} \subset {{\mathcal N}}$ spanning ${{\mathcal N}}$. Assume that $$\begin{aligned}
\label{graen}
{\mathop{\mathrm{Ad}}\nolimits}_{\Gamma}(E_{\mu,\nu})=(-1)^{|\mu|+|\nu|} E_{\mu,\nu},\quad\text{for}\quad
\mu,\nu\in {{\mathcal P}}.\end{aligned}$$ Set $\Gamma_0:=\sum_{\mu\in{{\mathcal P}}} (-1)^{|\mu|}E_{\mu\mu}$. Let $\rho: {{\mathcal R}}_{1,{{\mathcal K}}} \to {{\mathcal R}}_{1,{{\mathcal K}}}$ be an injective graded, unital $\ast$-endomorphism such that $\rho(a)b- (-1)^{\partial a\partial b} b \rho(a) = 0$ for $b\in{{\mathcal N}}, a \in {{\mathcal R}}_{1,{{\mathcal K}}}$ with homogeneous grading. Suppose further that ${{\mathcal R}}_{1,{{\mathcal K}}}= \rho({{\mathcal R}}_{1,{{\mathcal K}}}) \vee {{\mathcal N}}$.
Then there is some $\sigma_0\in\{0,1\}$ such that $\rho{\left (}{\mathbb I}_{{\mathcal K}}\otimes \sigma_x{\right )}=(-1)^{\sigma_0}\Gamma_0{\left (}{\mathbb I}_{{\mathcal K}}\otimes \sigma_x{\right )}$ and there exist isometries $\{S_\mu\}_{\mu \in \calP}$ on ${{\mathcal K}}$ with the property that $$\begin{aligned}
\label{sss}
S_\nu^* S_\mu = \delta_{\mu,\nu} \, {\mathbb I}_{{{\mathcal K}}},
\qquad
\rho(b) = \sum_\mu {\mathop{\mathrm{Ad}}\nolimits}_{( S_\mu \otimes \sigma_z^{\sigma_0+|\mu|})}(b) \end{aligned}$$ for all $\mu,\nu \in\calP$ and $b \in {{\mathcal R}}_{1,{{\mathcal K}}}$. Furthermore, for $N\in{{\mathbb N}}$, $\mu_0,\ldots,\mu_{N-1},\nu_0,\ldots,\nu_{N-1}\in\calP$, the identity $$\begin{aligned}
\label{kansha}
& E_{\mu_0,\nu_0}\rho(E_{\mu_1,\nu_1}) \rho^2(E_{\mu_2,\nu_2}) \cdots \rho^{N-1}(E_{\mu_{N-1},\nu_{N-1}}) \nonumber\\
&\hspace{1.5cm} = (-1)^{\sum\limits_{j=1}^{N-1} (\sum\limits_{k=0}^{j-1}(\sigma_0+|\nu_k|) ) (|\mu_j|+|\nu_j|)}
S_{\mu_0}\cdots S_{\mu_{N-1}}S^*_{\nu_{N-1}} \cdots S_{\nu_0}^* \otimes \sigma_x^{\sum\limits_{i=0}^{N-1} |\mu_i|+|\nu_i|}
\end{aligned}$$ holds.
If there are isometries $\{T_\mu\}_{\mu \in \calP}$ on ${{\mathcal K}}$ such that $$\begin{aligned}
\label{buni2}
T_\nu^* T_\mu = \delta_{\mu,\nu} {\mathbb I}_{{\mathcal K}}, \quad T_\mu T_\nu^*\otimes \sigma_x^{|\mu|+|\nu|}= E_{\mu,\nu},
\quad\rho(b)
= \sum_{\mu\in\calP} \mathrm{Ad}_{T_\mu\otimes\sigma_z^{\sigma_0+|\mu|}}
(b),\quad
b\in {{\mathcal R}}_{1,{{\mathcal K}}},\end{aligned}$$ then there is some $c\in{{\mathbb T}}$ such that $T_\mu=c S_\mu$, for all $\mu\in {{\mathcal P}}$.
To study the situation, we note the following general property.
\[rmrm\] Let ${{\mathcal K}}$ be a Hilbert space and set ${\Gamma_{{\mathcal K}}}:={\mathbb I}_{{\mathcal K}}\otimes\sigma_z$ on ${{\mathcal K}}\otimes {{{\mathbb C}}^2}$. We give a grading to ${{\mathcal R}}_{1,{{\mathcal K}}}={{{\mathcal B}}({{\mathcal K}})}\otimes {\mathop{\mathfrak{C}}\nolimits}$ by ${\mathop{\mathrm{Ad}}\nolimits}_{{\Gamma_{{\mathcal K}}}}$. Let ${{\mathcal N}}$ and ${{\mathcal M}}$ be ${\mathop{\mathrm{Ad}}\nolimits}_{{\Gamma_{{\mathcal K}}}}$-invariant von Neumann subalgebras of ${{\mathcal R}}_{1,{{\mathcal K}}}={{{\mathcal B}}({{\mathcal K}})}\otimes {\mathop{\mathfrak{C}}\nolimits}$ satisfying $$\begin{aligned}
ab-(-1)^{\partial a\partial b} ba=0,\quad\text{for homogeneous }\quad a\in {{\mathcal N}}, \quad b\in {{\mathcal M}}.\end{aligned}$$ Suppose that ${{\mathcal N}}$ is a type I factor with a self-adjoint unitary ${\Gamma_1}\in{{\mathcal N}}$ satisfying ${\mathop{\mathrm{Ad}}\nolimits}_{{\Gamma_1}}(a)={\mathop{\mathrm{Ad}}\nolimits}_{{\Gamma_{{\mathcal K}}}}(a)$, for all $a\in{{\mathcal N}}$. Suppose $Z({{\mathcal M}})^{(1)}\neq\{0\}$ and ${{\mathcal N}}\vee {{\mathcal M}}={{{\mathcal B}}({{\mathcal K}})}\otimes {\mathop{\mathfrak{C}}\nolimits}$. Then the following holds.
1. There are Hilbert spaces ${{\mathcal H}}_1,{{\mathcal H}}_2$, a unitary $U:{{\mathcal K}}\otimes{{{\mathbb C}}^2}\to {{\mathcal H}}_1\otimes{{\mathcal H}}_2\otimes{{{\mathbb C}}^2}$ and a self-adjoint unitary ${\tilde \Gamma_1}$ on ${{\mathcal H}}_1$ such that $$\begin{aligned}
\label{ffs}
\mathrm{Ad}_U (\calN) &= \calB(\calH_1) \otimes {{\mathbb C}}{{{\mathbb I}}}_{\calH_2} \otimes {{\mathbb C}}{{{\mathbb I}}}_{{\ensuremath{\mathbb{C}}}^2},
\qquad \mathrm{Ad}_U ({{{\mathcal B}}({{\mathcal K}})}\otimes {\mathop{\mathfrak{C}}\nolimits}) = \calB(\calH_1 \otimes \calH_2) \otimes {\mathop{\mathfrak{C}}\nolimits},\nonumber\\
\mathrm{Ad}_{U}( {\Gamma_{{\mathcal K}}}) &=
{\tilde \Gamma_1} \otimes {{{\mathbb I}}}_{\calH_2} \otimes \sigma_z,
\quad \mathrm{Ad}_{U}( {\Gamma_1}) =
{\tilde \Gamma_1} \otimes {{{\mathbb I}}}_{\calH_2} \otimes {\mathbb I}_{{{{\mathbb C}}^2}},
\quad \mathrm{Ad}_{U}( {{{\mathbb I}}}_\calK \otimes \sigma_x)
= {\mathbb I}_{{{\mathcal H}}_1}\otimes {{{\mathbb I}}}_{\calH_2} \otimes \sigma_x,\end{aligned}$$ and $$\begin{aligned}
&\mathrm{Ad}_{U}( {{\mathcal M}})
= {{\mathbb C}}{{{\mathbb I}}}_{\calH_1} \otimes \calB(\calH_2) \otimes {\ensuremath{\mathbb{C}}}{\mathbb I}_{{{{\mathbb C}}^2}}
+ {{\mathbb C}}{\tilde \Gamma_1}\otimes \calB(\calH_2) \otimes {\ensuremath{\mathbb{C}}}\sigma_x.
\label{muu}\end{aligned}$$
2. ${{\mathcal M}}'={{\mathcal N}}^{(0)}{\left (}{{\mathbb C}}{\mathbb I}_{{{\mathcal K}}}\otimes {\mathop{\mathfrak{C}}\nolimits}{\right )}+{{\mathcal N}}^{(1)}{\left (}{{\mathbb C}}{\mathbb I}_{{{\mathcal K}}}\otimes {\mathop{\mathfrak{C}}\nolimits}{\right )}{\Gamma_1}{\Gamma_{{\mathcal K}}}$.
3. For any minimal projection $p$ of ${{\mathcal N}}$ which is even, we have ${{\mathcal M}}\cdot p={{\mathcal B}}(q {{\mathcal K}})\otimes{\mathop{\mathfrak{C}}\nolimits}$ with $q$ a projection on ${{\mathcal K}}$ satisfying $p=q\otimes {\mathbb I}_{{{\mathbb C}}^2}$. (Note that even $p$ is always of this form.)
4. $Z({{\mathcal M}})={{\mathbb C}}{{{\mathbb I}}}_{{{\mathcal K}}} \otimes {\mathbb I}_{{{{\mathbb C}}^2}}
+{{\mathbb C}}{\Gamma_1}{\left (}{\mathbb I}_{{{\mathcal K}}}\otimes \sigma_x{\right )}$.
\(i) As ${{\mathcal N}}$ is a type I factor, there are Hilbert spaces ${{\mathcal H}}_1,\tilde {{\mathcal H}}_2$, and a unitary $\tilde U:{{\mathcal K}}\otimes{{{\mathbb C}}^2}\to {{\mathcal H}}_1\otimes\tilde {{\mathcal H}}_2$ such that ${\mathop{\mathrm{Ad}}\nolimits}_{\tilde U}({{\mathcal N}})={{\mathcal B}}({{\mathcal H}}_1)\otimes{{\mathbb C}}{\mathbb I}_{\tilde {{\mathcal H}}_2}$. Because ${\Gamma_1}\in{{\mathcal N}}$, there is a self-adjoint unitary ${\tilde \Gamma_1}$ on ${{\mathcal H}}_1$ such that ${\mathop{\mathrm{Ad}}\nolimits}_{\tilde U}({\Gamma_1})={\tilde \Gamma_1}\otimes{\mathbb I}_{\tilde{{\mathcal H}}_2}$. Let ${{\mathcal D}}:= {\mathop{\mathrm{span}}\nolimits}_{{\mathbb C}}\{{\mathbb I},{\Gamma_1}{\Gamma_{{\mathcal K}}},({\mathbb I}_{{\mathcal K}}\otimes \sigma_x),
{\Gamma_1}{\Gamma_{{\mathcal K}}}({\mathbb I}_{{\mathcal K}}\otimes \sigma_x)\}$, a $*$-subalgebra of ${{\mathcal N}}'$. Let ${\Gamma_1}{\Gamma_{{\mathcal K}}}=e_{00}-e_{11}$ be a spectral decomposition of the self-adjoint unitary ${\Gamma_1}{\Gamma_{{\mathcal K}}}$. Set $e_{i,1-i}:=e_{ii}({\mathbb I}_{{\mathcal K}}\otimes \sigma_x) e_{1-i,1-i}$, $i=0,1$. Then because ${\Gamma_1}{\Gamma_{{\mathcal K}}}$ and ${\mathbb I}_{{\mathcal K}}\otimes \sigma_x$ anti-commute, we can check that $\{e_{ij}\}_{i,j=0,1}$ are matrix units in ${{\mathcal D}}$ spanning ${{\mathcal D}}$. Hence ${{\mathcal D}}$ is a type I${}_2$ factor in ${{\mathcal N}}'$ generated by the matrix units $\{e_{ij}\}_{i,j=0,1}$. Therefore, there is a type I${}_2$ factor ${{\mathcal D}}_1$ on $\tilde{{{\mathcal H}}}_2$ such that ${\mathop{\mathrm{Ad}}\nolimits}_{\tilde U}({{\mathcal D}})={{\mathbb C}}{\mathbb I}_{{{\mathcal H}}_1}\otimes {{\mathcal D}}_1$ and the generating matrix units $\{f_{ij}\}_{i,j=0,1}$ such that ${\mathop{\mathrm{Ad}}\nolimits}_{\tilde U}(e_{ij})={\mathbb I}_{{{\mathcal H}}_1}\otimes f_{ij}$. Then there is a Hilbert space ${{\mathcal H}}_2$ and a unitary $W:\tilde {{\mathcal H}}_2\to {{\mathcal H}}_2\otimes {{\mathbb C}}^2$ such that $$\begin{aligned}
{\mathop{\mathrm{Ad}}\nolimits}_{W}(f_{ij})={\mathbb I}_{{{\mathcal H}}_2}\otimes \hat e_{ij},\quad
{\mathop{\mathrm{Ad}}\nolimits}_{W}({{\mathcal D}}_1)={{\mathbb C}}{\mathbb I}_{{{\mathcal H}}_2}\otimes {\mathop{\mathrm{M}}\nolimits}_2.\end{aligned}$$ Here $\hat e_{ij}$ denotes the matrix unit of $2\times 2$ matrices ${\mathop{\mathrm{M}}\nolimits}_2$ with respect to the standard basis of ${{{\mathbb C}}^2}$. Setting $U:={\left (}{\mathbb I}_{{{\mathcal H}}_1}\otimes W{\right )}\tilde U: {{\mathcal K}}\otimes{{{\mathbb C}}^2}\to
{{\mathcal H}}_1\otimes {{\mathcal H}}_2\otimes{{{\mathbb C}}^2}$, we may check directly that $U$, ${{\mathcal H}}_1$, ${{\mathcal H}}_2$, ${\tilde \Gamma_1}$ satisfy the .
We now prove . Because ${{\mathcal M}}^{(0)}$ is a von Neumann subalgebra of ${{\mathcal N}}'\cap{\left (}{{{\mathcal B}}({{\mathcal K}})}\otimes{\mathop{\mathfrak{C}}\nolimits}{\right )}$, ${\mathop{\mathrm{Ad}}\nolimits}_U{\left (}{{\mathcal M}}^{(0)}{\right )}$ is a von Neumann subalgebra of $$\begin{aligned}
{\mathop{\mathrm{Ad}}\nolimits}_U({{\mathcal N}}')\cap {\mathop{\mathrm{Ad}}\nolimits}_U{\left (}{{{\mathcal B}}({{\mathcal K}})}\otimes{\mathop{\mathfrak{C}}\nolimits}{\right )}={{\mathbb C}}{\mathbb I}_{{{\mathcal H}}_1}\otimes {{\mathcal B}}({{\mathcal H}}_2)\otimes {\mathop{\mathfrak{C}}\nolimits}.\end{aligned}$$ Furthermore, because elements in ${{\mathcal M}}^{(0)}$ are even with respect to ${\mathop{\mathrm{Ad}}\nolimits}_{{\Gamma_{{\mathcal K}}}}$, elements in ${\mathop{\mathrm{Ad}}\nolimits}_U{\left (}{{\mathcal M}}^{(0)}{\right )}$ are even with respect to ${\mathop{\mathrm{Ad}}\nolimits}_{{\mathop{\mathrm{Ad}}\nolimits}_U({\Gamma_{{\mathcal K}}})}={\mathop{\mathrm{Ad}}\nolimits}_{{\tilde \Gamma_1}\otimes{\mathbb I}_{{{\mathcal H}}_2}\otimes \sigma_z}$. Therefore we have $
{\mathop{\mathrm{Ad}}\nolimits}_U{\left (}{{\mathcal M}}^{(0)}{\right )}\subset
{{\mathbb C}}{\mathbb I}_{{{\mathcal H}}_1}\otimes {{\mathcal B}}({{\mathcal H}}_2)\otimes {{\mathbb C}}{\mathbb I}_{{{{\mathbb C}}^2}}
$. Hence there is a von Neumann subalgebra $\tilde {{\mathcal M}}$ of ${{\mathcal B}}({{\mathcal H}}_2)$ such that $$\begin{aligned}
\label{mtm}
{\mathop{\mathrm{Ad}}\nolimits}_U \big({{\mathcal M}}^{(0)} \big) ={{\mathbb C}}{\mathbb I}_{{{\mathcal H}}_1}\otimes \tilde {{\mathcal M}}\otimes {{\mathbb C}}{\mathbb I}_{{{{\mathbb C}}^2}}.
\end{aligned}$$
Next we consider ${\mathop{\mathrm{Ad}}\nolimits}_U{\left (}{{\mathcal M}}^{(1)}{\right )}$. We claim $({\mathbb I}_{{\mathcal K}}\otimes \sigma_x) {\Gamma_1}\in Z({{\mathcal M}})^{(1)}$. To see this, let $b\in Z({{\mathcal M}})^{(1)}$ be a non-zero element, which exists because of the assumption, and set $\tilde b=({\mathbb I}_{{\mathcal K}}\otimes \sigma_x) {\Gamma_1}b$. Because $b\in Z({{\mathcal M}})^{(1)}$, ${\mathbb I}_{{\mathcal K}}\otimes \sigma_x\in Z({{{\mathcal B}}({{\mathcal K}})}\otimes{\mathop{\mathfrak{C}}\nolimits})$ and ${\Gamma_1}$ is an even element in ${{\mathcal N}}$ implementing the grading on ${{\mathcal N}}$, we see that $$\begin{aligned}
\tilde b\in {{\mathcal N}}'\cap{{\mathcal M}}'\cap\{{\Gamma_{{\mathcal K}}}\}'
={\left (}{{{\mathcal B}}({{\mathcal K}})}\otimes{\mathop{\mathfrak{C}}\nolimits}{\right )}'\cap \{{\Gamma_{{\mathcal K}}}\}'
={{\mathbb C}}{\mathbb I}_{{{\mathcal K}}\otimes {{{\mathbb C}}^2}}.\end{aligned}$$ Hence $({\mathbb I}_{{\mathcal K}}\otimes \sigma_x) {\Gamma_1}$ is proportional to $b\in Z({{\mathcal M}})^{(1)}$, i.e. it belongs to $Z({{\mathcal M}})^{(1)}$, proving the claim. From this and we have $$\begin{aligned}
\label{mtmi}
{\mathop{\mathrm{Ad}}\nolimits}_U({{\mathcal M}}^{(1)})
={\mathop{\mathrm{Ad}}\nolimits}_U \big( {{\mathcal M}}^{(0)}({\mathbb I}_{{\mathcal K}}\otimes \sigma_x) {\Gamma_1} \big)
={{\mathbb C}}{\tilde \Gamma_1}\otimes\tilde{{\mathcal M}}\otimes {{\mathbb C}}\sigma_x \end{aligned}$$ for $\tilde{{\mathcal M}}$ in . From and , to show , it suffices to show that $\tilde {{\mathcal M}}={{\mathcal B}}({{\mathcal H}}_2)$. For any $a\in\tilde{{\mathcal M}}'$ $$\begin{aligned}
{\mathop{\mathrm{Ad}}\nolimits}_{U^*}{\left (}{\mathbb I}_{{{\mathcal H}}_1}\otimes a\otimes{\mathbb I}_{{{{\mathbb C}}^2}}{\right )}\in \big( {{\mathcal M}}^{(0)}\big)' \cap \big( {{\mathcal M}}^{(1)}\big)' \cap {{\mathcal N}}'\cap \{{\Gamma_{{\mathcal K}}}\}'
={\left (}{{{\mathcal B}}({{\mathcal K}})}\otimes{\mathop{\mathfrak{C}}\nolimits}{\right )}'\cap\{{\Gamma_{{\mathcal K}}}\}'
={{\mathbb C}}{\mathbb I}_{{{\mathcal K}}\otimes{{{\mathbb C}}^2}}.\end{aligned}$$ Hence we obtain $a\in {{\mathbb C}}{\mathbb I}_{{{\mathcal H}}_2}$. This proves that $\tilde {{\mathcal M}}={{\mathcal B}}({{\mathcal H}}_2)$.
\(ii) We associate a spatial grading to ${{\mathbb C}}{\mathbb I}_{{{\mathcal H}}_1}$ and ${{\mathcal B}}({{\mathcal H}}_2)\otimes {\mathop{\mathfrak{C}}\nolimits}$ by ${\tilde \Gamma_1}$ and ${\mathbb I}_{{{\mathcal H}}_2}\otimes \sigma_z$ respectively. From , we see that ${\mathop{\mathrm{Ad}}\nolimits}_U({{\mathcal M}})$ is equal to the graded tensor product ${{\mathbb C}}{\mathbb I}_{{{\mathcal H}}_1}{{\,\hat{\otimes}\,}} {\left (}{{\mathcal B}}({{\mathcal H}}_2)\otimes {\mathop{\mathfrak{C}}\nolimits}{\right )}$ of $({{\mathbb C}}{\mathbb I}_{{{\mathcal H}}_1},{{\mathcal H}}_1, \tilde\Gamma_1)$ and $({{\mathcal B}}({{\mathcal H}}_2)\otimes{\mathop{\mathfrak{C}}\nolimits},{{\mathcal H}}_2\otimes {{{\mathbb C}}^2},{\mathbb I}_{{{\mathcal H}}_2}\otimes \sigma_z)$. By Lemma \[comgra\], its commutant ${\mathop{\mathrm{Ad}}\nolimits}_U({{\mathcal M}}')$ is equal to $$\begin{aligned}
\label{inh}
{\mathop{\mathrm{Ad}}\nolimits}_U({{\mathcal M}}')
&={{\mathcal B}}({{\mathcal H}}_1)^{(0)}\otimes {{\mathbb C}}{\mathbb I}_{{{\mathcal H}}_2}\otimes{\mathop{\mathfrak{C}}\nolimits}+{{\mathcal B}}({{\mathcal H}}_1)^{(1)}\otimes {{\mathbb C}}{\mathbb I}_{{{\mathcal H}}_2}\otimes{\mathop{\mathfrak{C}}\nolimits}\sigma_z \nonumber \\
&={\mathop{\mathrm{Ad}}\nolimits}_U \big( {{\mathcal N}}^{(0)}{\left (}{{\mathbb C}}{\mathbb I}_{{{\mathcal K}}}\otimes {\mathop{\mathfrak{C}}\nolimits}{\right )}+{{\mathcal N}}^{(1)}{\left (}{{\mathbb C}}{\mathbb I}_{{{\mathcal K}}}\otimes {\mathop{\mathfrak{C}}\nolimits}{\right )}{\Gamma_1}{\Gamma_{{\mathcal K}}} \big),\end{aligned}$$ where ${{\mathcal B}}({{\mathcal H}}_1)$ is given a grading by ${\tilde \Gamma_1}$. This proves the claim.
\(iii) Let $p$ be a minimal projection ${{\mathcal N}}$ which is even and hence of the form $p=q\otimes {\mathbb I}_{{{\mathbb C}}^2}$ with $q$ a projection on ${{\mathcal K}}$. Then because $p\in {{\mathcal N}}$ is minimal, we have ${\mathop{\mathrm{Ad}}\nolimits}_U(p)=r\otimes {\mathbb I}_{{{\mathcal H}}_2}\otimes {\mathbb I}_{{{{\mathbb C}}^2}}$ with a rank-one projection $r$ on ${{\mathcal H}}_1$. Because $p$ is even, $r$ is even with respect to ${\mathop{\mathrm{Ad}}\nolimits}_{{\tilde \Gamma_1}}$. Therefore, there is a $\sigma\in\{0,1\}$ such that ${\tilde \Gamma_1}r=(-1)^\sigma r$. Substituting , we then obtain $$\begin{aligned}
{\mathop{\mathrm{Ad}}\nolimits}_U({{\mathcal M}}p)
&= {{\mathbb C}}r \otimes \calB(\calH_2) \otimes {\ensuremath{\mathbb{C}}}{\mathbb I}_{{{{\mathbb C}}^2}}
+ {{\mathbb C}}{\tilde \Gamma_1} r\otimes \calB(\calH_2) \otimes {\ensuremath{\mathbb{C}}}\sigma_x\nonumber\\
&={{\mathbb C}}r \otimes \calB(\calH_2) \otimes {\ensuremath{\mathbb{C}}}{\mathbb I}_{{{{\mathbb C}}^2}}
+ {{\mathbb C}}(-1)^\sigma r\otimes \calB(\calH_2) \otimes {\ensuremath{\mathbb{C}}}\sigma_x \nonumber \\
&={\mathop{\mathrm{Ad}}\nolimits}_U{\left (}p {\left (}{{\mathcal B}}({{\mathcal K}})\otimes {\mathop{\mathfrak{C}}\nolimits}{\right )}p{\right )}={\mathop{\mathrm{Ad}}\nolimits}_U{\left (}{{\mathcal B}}(q{{\mathcal K}})\otimes {\mathop{\mathfrak{C}}\nolimits}{\right )}\end{aligned}$$ as required.
\(iv) From and , we have $$\begin{aligned}
&{\mathop{\mathrm{Ad}}\nolimits}_U \big( Z({{\mathcal M}})^{(0)} \big) \\
&\quad
={\left (}{{\mathbb C}}{{{\mathbb I}}}_{\calH_1} \otimes \calB(\calH_2) \otimes {\ensuremath{\mathbb{C}}}{\mathbb I}_{{{{\mathbb C}}^2}}{\right )}\cap \big(
{{\mathcal B}}({{\mathcal H}}_1)^{(0)}\otimes {{\mathbb C}}{\mathbb I}_{{{\mathcal H}}_2}\otimes{{\mathbb C}}{\mathbb I}_{{{{\mathbb C}}^2}}
+{{\mathcal B}}({{\mathcal H}}_1)^{(1)}\otimes {{\mathbb C}}{\mathbb I}_{{{\mathcal H}}_2}\otimes{{\mathbb C}}\sigma_x\sigma_z
\big)
={{\mathbb C}}{\mathbb I},\end{aligned}$$ and $$\begin{aligned}
{\mathop{\mathrm{Ad}}\nolimits}_U \big( Z({{\mathcal M}})^{(1)} \big) &= \big( {{\mathbb C}}{\tilde \Gamma_1}\otimes \calB(\calH_2) \otimes {\ensuremath{\mathbb{C}}}\sigma_x \big)
\cap
\big(
{{\mathcal B}}({{\mathcal H}}_1)^{(0)}\otimes {{\mathbb C}}{\mathbb I}_{{{\mathcal H}}_2}\otimes{{\mathbb C}}\sigma_x
+{{\mathcal B}}({{\mathcal H}}_1)^{(1)}\otimes {{\mathbb C}}{\mathbb I}_{{{\mathcal H}}_2}\otimes{{\mathbb C}}\sigma_z
\big) \\
&={{\mathbb C}}{\tilde \Gamma_1}\otimes{{\mathbb C}}{\mathbb I}_{{{\mathcal H}}_2}\otimes{{\mathbb C}}\sigma_x
={\mathop{\mathrm{Ad}}\nolimits}_U{\left (}{{\mathbb C}}{\Gamma_1}({\mathbb I}_{{{\mathcal K}}}\otimes\sigma_x){\right )}.\end{aligned}$$ This proves the claim.
We introduce some notation. Given a self-adjoint unitary $T$ on some Hilbert space, we write the $\pm 1$ eigenspace projections as $$\label{eq:pm_espace_proj}
{{\mathbb P}}_\varepsilon(T) = \frac{ {{{\mathbb I}}}+ (-1)^{\varepsilon} T}{2}, \quad \varepsilon \in \{0,1\}.$$ Note that because we use the presentation of ${\ensuremath{\mathbb{Z}}}_2$ as an additive group, ${{\mathbb P}}_1(T)$ is the projection onto the *negative* eigenspace. We also have that $T {{\mathbb P}}_\varepsilon(T) = (-1)^\varepsilon {{\mathbb P}}_\varepsilon(T) = {{\mathbb P}}_\varepsilon(T) T$.
Because ${\mathbb I}_{{\mathcal K}}\otimes \sigma_x$ belongs to $Z({{\mathcal R}}_{1,{{\mathcal K}}})^{(1)}$ and $\rho$ is graded, $\rho({\mathbb I}_{{\mathcal K}}\otimes \sigma_x)$ belongs to $ Z(\rho({{\mathcal R}}_{1,{{\mathcal K}}}))^{(1)}$. In particular, because $\rho$ is injective, $ Z(\rho({{\mathcal R}}_{1,{{\mathcal K}}}))^{(1)}$ is not zero. Therefore, we satisfy the hypothesis of Lemma \[rmrm\] with ${{\mathcal M}}$ and $\Gamma_1$ replaced by $\rho({{\mathcal R}}_{1,{{\mathcal K}}})$ and $\Gamma_0$ respectively. Applying the lemma, we have that
1. $Z{\left (}\rho({{\mathcal R}}_{1,{{\mathcal K}}}){\right )}={{\mathbb C}}{\mathbb I}+{{\mathbb C}}\Gamma_0{\left (}{\mathbb I}_{{\mathcal K}}\otimes \sigma_x{\right )}$.
2. For any $\mu\in{{\mathcal P}}$, $E_{\mu\mu}=e_{\mu\mu}\otimes{\mathbb I}_{{{{\mathbb C}}^2}}$ with $e_{\mu\mu}$ a projection on ${{\mathcal K}}$, $\rho{\left (}{{\mathcal R}}_{1,{{\mathcal K}}}{\right )}E_{\mu\mu}={{\mathcal B}}(e_{\mu\mu}{{\mathcal K}})\otimes {\mathop{\mathfrak{C}}\nolimits}$,
3. $\rho{\left (}{{\mathcal R}}_{1,{{\mathcal K}}}{\right )}'={{\mathcal N}}^{(0)} {\left (}{{\mathbb C}}{\mathbb I}_{{\mathcal K}}\otimes{\mathop{\mathfrak{C}}\nolimits}{\right )}+{{\mathcal N}}^{(1)}{\left (}{{\mathbb C}}{\mathbb I}_{{\mathcal K}}\otimes{\mathop{\mathfrak{C}}\nolimits}{\right )}\Gamma_0\Gamma_{{\mathcal K}}$.
Because of (i), $\rho({\mathbb I}_{{\mathcal K}}\otimes \sigma_x)$, an odd self-adjoint unitary in $Z{\left (}\rho({{\mathcal R}}_{1,{{\mathcal K}}}){\right )}$, should be either $\Gamma_0{\left (}{\mathbb I}_{{\mathcal K}}\otimes \sigma_x{\right )}$ or $-\Gamma_0{\left (}{\mathbb I}_{{\mathcal K}}\otimes \sigma_x{\right )}$. Therefore, there is $\sigma_0\in\{0,1\}$ such that $$\begin{aligned}
\label{rcc}
\rho{\left (}{\mathbb I}_{{\mathcal K}}\otimes \sigma_x{\right )}=(-1)^{\sigma_0}\Gamma_0{\left (}{\mathbb I}_{{\mathcal K}}\otimes \sigma_x{\right )}.\end{aligned}$$ By (ii), , and the fact that $E_{\mu\mu}\in {{\mathcal N}}^{(0)}$ commutes with $\rho({{\mathcal R}}_{1,{{\mathcal K}}})$, for each $\mu\in{{\mathcal P}}$, we have $$\begin{aligned}
\rho{\left (}{\left (}{{{\mathcal B}}({{\mathcal K}})}\otimes {{\mathbb C}}{\mathbb I}_{{{{\mathbb C}}^2}}{\right )}\cdot {{\mathbb P}}_0{\left (}{\mathbb I}_{{\mathcal K}}\otimes \sigma_x{\right )}{\right )}E_{\mu\mu}
=
\rho{\left (}{{\mathcal R}}_{1,{{\mathcal K}}} {{\mathbb P}}_0{\left (}{\mathbb I}_{{\mathcal K}}\otimes \sigma_x{\right )}{\right )}E_{\mu\mu}
={{\mathcal B}}(e_{\mu\mu}{{\mathcal K}})\otimes{{\mathbb C}}{{\mathbb P}}_{\sigma_0+|\mu|}{\left (}\sigma_x{\right )}.\end{aligned}$$ Therefore, there is a $*$-isomorphism $\rho_\mu:{{{\mathcal B}}({{\mathcal K}})}\to {{\mathcal B}}(e_{\mu\mu}{{\mathcal K}})$ such that $$\begin{aligned}
\label{rmu}
\rho{\left (}{\left (}a\otimes {\mathbb I}{\right )}\cdot {{\mathbb P}}_0{\left (}{\mathbb I}_{{\mathcal K}}\otimes \sigma_x{\right )}{\right )}E_{\mu\mu}
=\rho_\mu(a)\otimes {{\mathbb P}}_{\sigma_0+|\mu|}(\sigma_x),\quad a\in {{\mathcal B}}({{\mathcal K}}).\end{aligned}$$ Applying ${\mathop{\mathrm{Ad}}\nolimits}_{\Gamma_{{\mathcal K}}}$, we also get that $$\begin{aligned}
\label{rmu2}
\rho{\left (}{\left (}a\otimes {\mathbb I}{\right )}\cdot {{\mathbb P}}_1{\left (}{\mathbb I}_{{\mathcal K}}\otimes \sigma_x{\right )}{\right )}E_{\mu\mu}
=\rho_\mu(a)\otimes {{\mathbb P}}_{\sigma_0+|\mu|+1}(\sigma_x),\quad a\in {{\mathcal B}}({{\mathcal K}}).\end{aligned}$$ From and , we obtain $$\begin{aligned}
\label{sonoi}
\rho{\left (}a\otimes {\mathbb I}{\right )}E_{\mu\mu}
=\rho_\mu(a)\otimes {\mathbb I}_{{{{\mathbb C}}^2}},\quad a\in {{\mathcal B}}({{\mathcal K}}).\end{aligned}$$ Furthermore, by , we have $$\begin{aligned}
\label{sonon}
\rho{\left (}{\mathbb I}_{{\mathcal K}}\otimes \sigma_x{\right )}E_{\mu\mu}
=(-1)^{\sigma_0+|\mu|}{\left (}e_{\mu\mu}\otimes \sigma_x{\right )}.\end{aligned}$$
By Wigner’s Theorem, for each $\mu\in{{\mathcal P}}$, there is a unitary $T_\mu:{{\mathcal K}}\to e_{\mu\mu}{{\mathcal K}}$ such that $$\begin{aligned}
\label{ttt}
T_\mu^*T_\nu=\delta_{\mu,\nu} {\mathbb I}_{{{\mathcal K}}},\quad T_{\mu}T_\mu^*=e_{\mu\mu},\quad \mu,\nu\in {{\mathcal P}},\quad
{\mathop{\mathrm{Ad}}\nolimits}_{T_\mu}{\left (}a{\right )}=\rho_\mu(a),\quad a\in {{{\mathcal B}}({{\mathcal K}})}.\end{aligned}$$ From this, and , we obtain $$\begin{aligned}
\rho(b) E_{\mu\mu}={\mathop{\mathrm{Ad}}\nolimits}_{T_\mu\otimes \sigma_z^{\sigma_0+|\mu|}}{\left (}b{\right )},\quad
b\in {{\mathcal R}}_{1,{{\mathcal K}}}.\end{aligned}$$ Summing this over $\mu$, we obtain $$\begin{aligned}
\label{rrtran}
\rho(b) =\sum_{\mu\in {{\mathcal P}}}{\mathop{\mathrm{Ad}}\nolimits}_{T_\mu\otimes \sigma_z^{\sigma_0+|\mu|}}{\left (}b{\right )},\quad
b\in {{\mathcal R}}_{1,{{\mathcal K}}}.\end{aligned}$$ Multiplying $T_\nu T_\mu^*\otimes\sigma_z^{|\mu|+|\nu|}$ from left or right of , we obtain the same value for any $b\in {{\mathcal R}}_{1,{{\mathcal K}}}$. Therefore, $T_\nu T_\mu^*\otimes\sigma_z^{|\mu|+|\nu|}$ belongs to $\rho({{\mathcal R}}_{1,{{\mathcal K}}})'$. By (iii), we then have $$\begin{aligned}
T_\nu T_\mu^*\otimes\sigma_z^{|\mu|+|\nu|}
\in \rho({{\mathcal R}}_{1,{{\mathcal K}}})'
={{\mathcal N}}^{(0)} {\left (}{{\mathbb C}}{\mathbb I}_{{\mathcal K}}\otimes{\mathop{\mathfrak{C}}\nolimits}{\right )}+{{\mathcal N}}^{(1)}{\left (}{{\mathbb C}}{\mathbb I}_{{\mathcal K}}\otimes{\mathop{\mathfrak{C}}\nolimits}{\right )}\Gamma_0\Gamma_{{\mathcal K}}.\end{aligned}$$ Hence if $|\mu|=|\nu|$, $
T_\nu T_\mu^*\otimes {\mathbb I}_{{{{\mathbb C}}^2}} \in {{\mathcal N}}^{(0)},
$ while if $|\mu|\neq |\nu|$, this means $
T_\nu T_\mu^*\otimes {\mathbb I}_{{{{\mathbb C}}^2}} \in {{\mathcal N}}^{(1)}{\left (}{\mathbb I}_{{\mathcal K}}\otimes \sigma_x{\right )}$. From , $\{T_{\mu}T_\nu^*\otimes {{\mathbb P}}_0(\sigma_x)\}_{\mu,\nu\in{{\mathcal P}}}$ are matrix units in ${{\mathcal N}}{\left (}{\mathbb I}_{{\mathcal K}}\otimes {{\mathbb P}}_0(\sigma_x){\right )}$ with $ e_{\mu\mu}T_{\mu}T_\nu^* e_{\nu\nu}\otimes{{\mathbb P}}_0(\sigma_x)=
T_{\mu}T_\nu^*\otimes {{\mathbb P}}_0(\sigma_x)$. Then as in the proof of Proposition \[cuntzpropi\], there are $c_\mu\in {{\mathbb T}}$ such that $S_\mu S_\nu^*\otimes {{\mathbb P}}_0(\sigma_x)=E_{\mu\nu} {{\mathbb P}}_0({\mathbb I}_{{\mathcal K}}\otimes\sigma_x)$ for $S_\mu=c_\mu T_\mu$. Applying ${\mathop{\mathrm{Ad}}\nolimits}_{\Gamma_{{\mathcal K}}}$, we also obtain $S_\mu S_\nu^*\otimes {{\mathbb P}}_1(\sigma_x)=
(-1)^{|\mu|+|\nu|}E_{\mu\nu} {{\mathbb P}}_1({\mathbb I}_{{\mathcal K}}\otimes\sigma_x)$, which then implies that $$\begin{aligned}
\label{unitdesu}
\big( S_\mu\otimes \sigma_x^{|\mu|} \big)
\big( S_\nu\otimes \sigma_x^{|\nu|} \big)^*
=S_\mu S_\nu^*\otimes \sigma_x^{|\mu|+|\nu|}
=S_\mu S_\nu^*\otimes \big( {{\mathbb P}}_0(\sigma_x)+(-1)^{|\mu|+|\nu|}{{\mathbb P}}_1(\sigma_x) \big)
=E_{\mu\nu}.\end{aligned}$$ It is clear that $\{S_\mu\}_{\mu\in{{\mathcal P}}}$ are isometries satisfying . The proof of comes from an induction argument using and . As the argument is the same as in the proof of Proposition \[cuntzpropi\], we omit the details. Similarly, the proof that the isometries $\{S_\mu\}_{\mu\in{{\mathcal P}}}$ are unique up to scalar multiplication in ${{\mathbb T}}$ is the same as in Proposition \[cuntzpropi\].
The Hilbert space ${{\mathcal K}}$, finite type I factor $\iota\circ\pi_\omega({{\mathcal A}}_{\{0\}})$ with matrix units $\{\iota\circ\pi_\omega\circ \big( E_{\mu,\nu}^{(0)} \big) \}_{\mu,\nu\in\calP}
\subset \calB({{\mathcal K}}) \otimes \mathfrak{C}$ and $\rho$ satisfy the conditions of Proposition \[cuntzpropii\]. Applying the proposition, we obtain $\sigma_0\in\{0,1\}$ and $\{S_\mu\}$ satisfying and from the statement of the theorem. The property follows from and parts (i) and (iii) of Lemma \[lem:trans\_weak\_conv\_to\_state\]. For the proof of , we set $$\begin{aligned}
\label{gvbn1}
T_\nu:= (-1)^{{\mathfrak{q}}(g)|\nu|}\sum_{\mu\in\calP} \overline{\langle \psi_\mu, \, \mathfrak{\Gamma}(U_g) \psi_\nu \rangle}^{{\mathfrak{p}}(g)}
\big( V_g^{(0)}\big)^* S_\mu V_g^{(0)}.
\end{aligned}$$ As in the proof of Theorem \[cuntzthm0\], we then can check that $T_\mu$ satisfies for $E_{\mu\nu}$ replaced by $\iota\circ\pi_\omega(E_{\mu\nu}^{(0)})$. Applying the last statement of Proposition \[cuntzpropii\], there is some $c_g \in \mathbb{T}$ such that $S_\mu = c_g T_\mu$ for all $\mu\in{{\mathcal P}}$. The proof of is given by the same argument as in the proof of Theorem \[cuntzthm0\].
Fermionic matrix product states {#fmpssec}
===============================
Using our results from Section \[transsec\], in this section we consider a translation invariant split state $\omega$ of ${{\mathcal A}}$ whose density matrices have uniformly bounded rank on finite intervals. Our main result is that such states can be written as the thermodynamic limit of an even or odd fermionic matrix product state (MPS) depending on the value $\kappa_\omega \in {{\mathbb Z}}_2$. See [@BWHV; @KapustinfMPS] for the basic properties of fermionic MPS in the finite setting. The idea of the proof is the same as quantum spin case, cf. [@bjp; @Matsui3; @classA2], although anti-commutativity results in richer structures. We start with some preliminary results.
The following Lemma is immediate because each ${{\mathcal A}}_{[0,N-1]}$ is isomorphic to a matrix algebra.
Let $\omega$ be a $\Theta$-invariant state of ${{{\mathcal A}}}$. For each $N\in{{\mathbb N}}$, let $Q_N$ be the support projection of the density matrix of $\omega\vert_{{{\mathcal A}}_{[0,N-1]}}$, the restriction of $\omega$ to ${{\mathcal A}}_{[0,N-1]}$. Then $Q_N$ is even.
We consider the situation where the matrices $Q_N$ have uniformly bounded rank.
\[lem:intersection\_of\_kernel\_fin\_dim\] Let $\{Q_N\}$ be a sequence of orthogonal projections with $Q_N\in {{\mathcal A}}_{[0,N-1]}^{(0)}$. We suppose that the rank of $Q_N$ is uniformly bounded, i.e., $\sup_{N\in{{\mathbb N}}} {\mathop{\mathrm{rank}}\nolimits}(Q_N)<\infty$. Let $\pi$ be an irreducible representation of ${{{\mathcal A}}}_R$ or ${{{\mathcal A}}}_R^{(0)}$ on a Hilbert space $\calH$. Set $\calH_0 = \bigcap\limits_{N=1}^\infty\big( \pi(Q_N) {{\mathcal H}}\big)$. Then $\mathrm{dim}\,{{\mathcal H}}_0 < \infty$.
As the statement is trivial if ${{\mathcal H}}_0=\{0\}$, assume that ${{\mathcal H}}_0\neq\{0\}$. We fix a unit vector $\eta\in{{\mathcal H}}_0$ and let $\{\xi_j\}_{j=1}^l \subset {{\mathcal H}}_0$ be an orthonormal system. We let $\mathfrak{A}$ denote either ${{{\mathcal A}}}_R$ or ${{{\mathcal A}}}_R^{(0)}$ with $\pi: \mathfrak{A} \to \calB({{\mathcal H}})$ irreducible and let ${\mathfrak A}_{\rm loc}$ denote local elements in ${\mathfrak A}$. We similarly write $\mathfrak{A}_{[0,N-1]}$ to denote either ${{\mathcal A}}_{[0,N-1]}$ or its even subalgebra. Note that the $l\times l$ matrix $({\langle\xi_i,\xi_j\rangle})_{i,j=1,\ldots,l}$ is an identity. Because $\pi$ is irreducible, approximating $\xi_i$ with elements in $\pi({\mathfrak A}_{\rm loc})\eta$, there exists an $N\in \mathbb{N}$ and elements $a_{j,N} \in Q_N\mathfrak{A}_{[0,N-1]}Q_N$ such that for the $l\times l$-matrix $X_N = ( \langle \pi(a_{i,N})\eta, \pi(a_{j,N})\eta \rangle )_{i,j=1,\ldots,l}$, $$\begin{aligned}
\big\| X_N - {\mathbb I}_{{\mathop{\mathrm{M}}\nolimits}_l}\big\| < \frac{1}{2}
\end{aligned}$$ holds.
We now claim that $\{a_{j,N}\}_{j=1}^l$ are linearly independent within $Q_N\mathfrak{A}_{[0,N-1]} Q_N$. So we suppose that $\sum_j d_j a_{j,N} = 0$ for $\{d_j\}_{j=1}^l \subset {\ensuremath{\mathbb{C}}}$. Then taking the vector $d=(d_1,\ldots,d_l)$, $$\begin{aligned}
\langle d, X_N d \rangle &= \sum_{i,j=1}^l \langle \pi(a_{i,N})\eta, \pi(a_{j,N}) \eta\rangle \, {\overline}{d_i} d_j
= \big\| \pi\big( \sum_{j=1}^l d_j a_{j,N} \big) \eta \big\|^2 = 0.\end{aligned}$$ Therefore $$0 = \langle d, X_N d \rangle = \|d\|^2 + \langle d, (X_N - {{{\mathbb I}}}) d \rangle
\geq \|d\|^2 - \frac{1}{2} \|d\|^2 = \frac{1}{2} \|d\|^2$$ and so $d=0$ and $\{a_{j,N}\}_{j=1}^l$ are linearly independent.
By the assumption we have $\mathrm{dim}\big( Q_N \mathfrak{A}_{[0,N-1]} Q_N \big) \leq C^2$, for $C:=\sup_{N\in{{\mathbb N}}} {\mathop{\mathrm{rank}}\nolimits}(Q_N)<\infty$. This tells us that $l\leq C^2$ and so $\mathrm{dim}\, {{\mathcal H}}_0 \leq C^2$.
We now consider the case of even and odd fermionic MPS separately.
Case: $\kappa_\omega=0$ (even fermionic MPS)
--------------------------------------------
\[thmevenmps\] Let $\omega$ be a pure, split, translation invariant and $\alpha$-invariant state on ${{\mathcal A}}$ with index ${\mathop{\mathrm{Ind}}\nolimits}(\omega)=(0,{\mathfrak{q}},[\upsilon])$. For each $N\in{{\mathbb N}}$, let $Q_N$ be the support projection of the density matrix of $\omega\vert_{{{\mathcal A}}_{[0,N-1]}}$ and assume $\sup_{N\in{{\mathbb N}}} {\mathop{\mathrm{rank}}\nolimits}(Q_N)<\infty$. Then there is some $m\in{{\mathbb N}}$, a faithful density matrix $D\in {\mathop{\mathrm{M}}\nolimits}_m$, a self-adjoint unitary $\mathfrak\Theta\in{\mathop{\mathrm{M}}\nolimits}_m$ and a set of matrices $\{v_\mu\}_{\mu\in{{\mathcal P}}}$ in ${\mathop{\mathrm{M}}\nolimits}_m$ satisfying the following.
1. For all $x\in{\mathop{\mathrm{M}}\nolimits}_m$, $\lim_{N\to\infty}T_{\bf v}^N(x)={\mathop{\mathrm{Tr}}\nolimits}{\left (}D x{\right )}{\mathbb I}_{{\mathop{\mathrm{M}}\nolimits}_m}$ in the norm topology.
2. There is some $\sigma_0=0,1$ such that ${\mathop{\mathrm{Ad}}\nolimits}_\mathfrak \Theta{\left (}v_\mu {\right )}=(-1)^{|\mu|+\sigma_0}v_\mu$ for all $\mu\in{{\mathcal P}}$.
3. ${\mathop{\mathrm{Ad}}\nolimits}_{\mathfrak \Theta}(D)=D$.
4. For any $l\in{{\mathbb N}}\cup\{0\}$, and $\mu_0,\ldots\mu_l, \nu_0,\ldots\nu_l\in {{\mathcal P}}$, $$\begin{aligned}
\omega\big( E_{\mu_0,\nu_0}^{(0)} E_{\mu_1,\nu_1}^{(1)}\cdots E_{\mu_l,\nu_l}^{(l)} \big)
=(-1)^{\sum\limits_{k=1}^l (|\mu_k|+|\nu_k|) \sum\limits_{j=0}^{k-1} |\nu_j| }
{\mathop{\mathrm{Tr}}\nolimits}\big( D v_{\mu_0}\cdots v_{\mu_l} v_{\nu_l}^* \cdots v_{\nu_0}^* \big).
\end{aligned}$$
5. There is a projective unitary/anti-unitary representation $W$ on ${{\mathbb C}}^m$ relative to ${\mathfrak{p}}$ and $c_g \in \mathbb{T}$ such that $$\begin{aligned}
\sum_{\mu\in\calP} \langle \psi_\mu, \, \mathfrak{\Gamma}(U_g) \psi_\nu \rangle v_\mu =
c_g W_g v_\nu W_g^*.
\end{aligned}$$ The second cohomology class associated to $W$ is $[\upsilon]$ and $$\begin{aligned}
{\mathop{\mathrm{Ad}}\nolimits}_{W_g^*}(D)=D,\quad {\mathop{\mathrm{Ad}}\nolimits}_{W_g}{\left (}{\mathfrak \Theta}{\right )}=(-1)^{{\mathfrak{q}}(g)} \mathfrak \Theta,
\quad g\in G.
\end{aligned}$$
We first prove a preparatory lemma.
\[lem:even\_fmps\_from\_split\_main\] Consider the setting of Theorem \[thmevenmps\]. Suppose that the graded $W^*$-$(G,{\mathfrak{p}})$-dynamical system $(\pi_\omega({{\mathcal A}}_R)'', {\mathop{\mathrm{Ad}}\nolimits}_{\Gamma_\omega}, \hat\alpha_\omega)$ associated to $\omega$ is equivalent to $({{\mathcal R}}_{0,{{\mathcal K}}},{\mathop{\mathrm{Ad}}\nolimits}_{\Gamma_{{\mathcal K}}},{\mathop{\mathrm{Ad}}\nolimits}_{V_g})\in{{\mathcal S}}_{0}$, via a $*$-isomorphism $\iota: \pi_\omega({{\mathcal A}}_R)''\to {{\mathcal B}}({{\mathcal K}}\otimes{{{\mathbb C}}^2})$. Then the following holds.
1. There is a finite rank density operator $D$ on $\calK \otimes{{{\mathbb C}}^2}$ such that $$\begin{aligned}
\mathrm{Ad}_{\Gamma_{{{\mathcal K}}}}(D) = D,\quad\text{and}\quad
{\mathop{\mathrm{Tr}}\nolimits}_{{{\mathcal K}}\otimes{{{\mathbb C}}^2}}\big( D (\iota\circ \pi_\omega(A)) \big) = \omega(A)\end{aligned}$$ for all $A \in {{{\mathcal A}}_R}$. For $P_{\mathrm{Supp}(D)}$, the support projection of $D$, $ \mathrm{Ad}_{\Gamma_{{{\mathcal K}}}}{\left (}P_{\mathrm{Supp}(D)}{\right )}=P_{\mathrm{Supp}(D)}$.
2. Let $\{B_\mu\}_{\mu \in \calP}$ be the set of isometries given in Theorem \[cuntzthm0\]. Then we have $$\begin{aligned}
\label{vbv}
v_\mu:=P_{\mathrm{Supp}(D)} B_\mu=P_{\mathrm{Supp}(D)} B_\mu P_{\mathrm{Supp}(D)},\quad
\mu\in{{\mathcal P}}.
\end{aligned}$$
3. $P_{\mathrm{Supp}(D)}V_g = V_g P_{\mathrm{Supp}(D)}$ and $D V_g=V_g D$ for any $g\in G$.
\(i) Given the cyclic vector $\Omega_\omega$, $\langle \Omega_\omega, \iota^{-1}(x)\Omega_\omega\rangle$ defines a normal state on $\calB({{{\mathcal K}}\otimes{{{\mathbb C}}^2}})$. Let $D$ be a density operator on ${{\mathcal K}}\otimes{{{\mathbb C}}^2}$ such that ${\mathop{\mathrm{Tr}}\nolimits}_{{{{\mathcal K}}\otimes{{{\mathbb C}}^2}}}(D x) = \langle \Omega_\omega, \iota^{-1}(x)\Omega_\omega \rangle$. We then see that $${\mathop{\mathrm{Tr}}\nolimits}_{{{{\mathcal K}}\otimes{{{\mathbb C}}^2}}} \big( D (\iota\circ \pi_\omega)(A) \big)
= \langle \Omega_\omega, \pi_\omega(A) \Omega_\omega \rangle = \omega(A), \quad
A\in {{\mathcal A}}_R.$$ Because $\omega\circ \Theta = \omega$ and $\iota\circ \pi_\omega \circ \Theta\vert_{{{\mathcal A}}_R}=
\mathrm{Ad}_{\Gamma_{{\mathcal K}}} \circ \iota \circ \pi_\omega\vert_{{{\mathcal A}}_R}$, it follows that ${\mathop{\mathrm{Tr}}\nolimits}_{{{{\mathcal K}}\otimes{{{\mathbb C}}^2}}}( \mathrm{Ad}_{\Gamma_{{\mathcal K}}} (D) (\iota\circ \pi_\omega)(A))
= {\mathop{\mathrm{Tr}}\nolimits}_{{{{\mathcal K}}\otimes{{{\mathbb C}}^2}}} ( D (\iota\circ \pi_\omega)(A))$ for all $A\in {{\mathcal A}}_R$. As such, $\mathrm{Ad}_{\Gamma_{{\mathcal K}}} (D) = D$. From this, we have $ \mathrm{Ad}_{\Gamma_{{{\mathcal K}}}}{\left (}P_{\mathrm{Supp}(D)}{\right )}=P_{\mathrm{Supp}(D)}$.
Let $\calH_0 = \bigcap\limits_{N=1}^\infty {\left (}\iota \circ \pi_\omega(Q_N){\right )}{\left (}{{{\mathcal K}}\otimes{{{\mathbb C}}^2}}{\right )}$. Because $\iota\circ\pi_\omega$ is an irreducible representation of ${{\mathcal A}}_R$, from Lemma \[lem:intersection\_of\_kernel\_fin\_dim\], ${{\mathcal H}}_0$ is finite-dimensional. Because $\omega({\mathbb I}-Q_N)=0$, we have $ {\mathop{\mathrm{Tr}}\nolimits}_{{{{\mathcal K}}\otimes{{{\mathbb C}}^2}}} \big( D (\iota\circ \pi_\omega)({\mathbb I}-Q_N) \big)
= \omega({\mathbb I}-Q_N)=0$. This means $P_{\mathrm{Supp}(D)}$, the support projection of $D$, satisfies $P_{\mathrm{Supp}(D)}\le \iota\circ \pi_\omega(Q_N)$ for all $N\in{\mathbb{N}}$. Hence we have $P_{\mathrm{Supp}(D)}{\left (}{{\mathcal K}}\otimes{{{\mathbb C}}^2}{\right )}\subset{{\mathcal H}}_0$. Therefore $D$ is finite rank.
\(ii) Recall the endomorphism $\rho$ satisfying from Lemma \[iorho\]. Because $\omega(A) = \omega(\beta_{S_1}(A))$ for all $A\in{{\mathcal A}}_R$, the set of isometries $\{B_\mu\}_{\mu \in \calP}$ given in Theorem \[cuntzthm0\] are such that $$\begin{aligned}
{\mathop{\mathrm{Tr}}\nolimits}_{{{\mathcal K}}\otimes{{\mathbb C}}^2}\big( D(\iota\circ\pi_\omega)(A)\big)
&= {\mathop{\mathrm{Tr}}\nolimits}_{{{\mathcal K}}\otimes{{\mathbb C}}^2}\big( D( \rho \circ\iota\circ\pi_\omega)(A)\big) \\
&= \sum_{\mu} {\mathop{\mathrm{Tr}}\nolimits}_{{{\mathcal K}}\otimes{{\mathbb C}}^2}\big( {\mathop{\mathrm{Ad}}\nolimits}_{B_\mu^* }
\circ {\mathop{\mathrm{Ad}}\nolimits}_{ \Gamma_{{\mathcal K}}^{|\mu|} }(D) (\iota\circ \pi_\omega)(A) \big) \end{aligned}$$ for all $A\in{{\mathcal A}}_R$. This implies that $D=\sum_{\mu}{\mathop{\mathrm{Ad}}\nolimits}_{B_\mu^* }
\circ {\mathop{\mathrm{Ad}}\nolimits}_{ \Gamma_{{\mathcal K}}^{|\mu|} }(D)=\sum_{\mu}{\mathop{\mathrm{Ad}}\nolimits}_{B_\mu^*}(D)$ and so $$\sum_\mu {\left (}{\mathbb I}-P_{\mathrm{Supp}(D)}{\right )}B_\mu^* D B_\mu
{\left (}{\mathbb I}-P_{\mathrm{Supp}(D)}{\right )}=
{\left (}{\mathbb I}- P_{\mathrm{Supp}(D)} {\right )}D{\left (}{\mathbb I}- P_{\mathrm{Supp}(D)}{\right )}=0.$$ Hence we obtain $ P_{\mathrm{Supp}(D)}B_\mu {\left (}{\mathbb I}-P_{\mathrm{Supp}(D)}{\right )}=0$.
\(iii) For an element $A\in{{\mathcal A}}_R$ and ${\mathfrak{p}}(g) \in{{\mathbb Z}}_2$, we set $A^{{\mathfrak{p}}(g)\ast}$ as $A$ if ${\mathfrak{p}}(g)=0$ and $A^*$ if ${\mathfrak{p}}(g)=1$. Because $\omega( \alpha_g(A^{{\mathfrak{p}}(g)\ast})) = \omega(A) = {\mathop{\mathrm{Tr}}\nolimits}(D (\iota\circ\pi_\omega)(A))$, $A\in{{\mathcal A}}_R$, we have that $$\begin{aligned}
{\mathop{\mathrm{Tr}}\nolimits}_{{{\mathcal K}}\otimes{{{\mathbb C}}^2}}\big( D(\iota\circ\pi_\omega)(A) \big)
&= {\mathop{\mathrm{Tr}}\nolimits}_{{{\mathcal K}}\otimes{{{\mathbb C}}^2}}\big( D( \iota\circ \pi_\omega) {\left (}\alpha_g(A^{{\mathfrak{p}}(g)\ast}){\right )}\big)
= {\mathop{\mathrm{Tr}}\nolimits}_{{{\mathcal K}}\otimes{{{\mathbb C}}^2}}\big( D V_g \big( (\iota\circ \pi_\omega)(A^{{\mathfrak{p}}(g)\ast} ) \big) V_g^* \big).\end{aligned}$$ Given an orthonomal basis $\{\xi_j\}_j$ of ${{\mathcal K}}\otimes{{{\mathbb C}}^2}$, we see that for any $A\in {{\mathcal A}}_R$, $$\begin{aligned}
{\mathop{\mathrm{Tr}}\nolimits}_{{{\mathcal K}}\otimes{{{\mathbb C}}^2}}\big( D(\iota\circ\pi_\omega)(A) \big)
&={\mathop{\mathrm{Tr}}\nolimits}_{{{\mathcal K}}\otimes{{{\mathbb C}}^2}} \! {\left (}DV_g \big( (\iota\circ \pi_\omega)(A^{{\mathfrak{p}}(g)\ast}) \big) V_g^*
{\right )}= \sum_j \langle V_g \xi_j, D V_g (\iota\circ \pi_\omega)(A^{{\mathfrak{p}}(g)\ast}) \xi_j \rangle \\
&= \sum_j {\overline}{ \langle \xi_j, V_g^* D V_g(\iota\circ \pi_\omega)(A^{{\mathfrak{p}}(g)\ast}) \xi_j \rangle}^{{\mathfrak{p}}(g)}
= {\mathop{\mathrm{Tr}}\nolimits}_{\calK\otimes {{{\mathbb C}}^2}}\big( V_g^* D V_g (\iota\circ \pi_\omega)(A) \big),\end{aligned}$$ where for the second equality we used that $\{V_g \xi_j\}_j$ is an orthonomal basis of ${{\mathcal K}}\otimes{{{\mathbb C}}^2}$. Therefore, $V_g^* D V_g = D$ and so $P_{\mathrm{Supp}(D)}V_g = V_g P_{\mathrm{Supp}(D)}$.
We use the notation of Theorem \[cuntzthm0\] and Lemma \[lem:even\_fmps\_from\_split\_main\]. Let $m\in{{\mathbb N}}$ be the rank of $D$ from Lemma \[lem:even\_fmps\_from\_split\_main\]. We naturally identify $P_{\mathrm{Supp}(D)} {{\mathcal B}}({{\mathcal K}}\otimes{{{\mathbb C}}^2})P_{\mathrm{Supp}(D)} $ and ${\mathop{\mathrm{M}}\nolimits}_m$. Then we may regard $D$ as a faithful density matrix in ${\mathop{\mathrm{M}}\nolimits}_m$, and $\{v_\mu\}_{\mu\in{{\mathcal P}}}$ matrices in ${\mathop{\mathrm{M}}\nolimits}_m$. Because $\Gamma_{{{\mathcal K}}}$ commutes with $P_{\mathrm{Supp}(D)} $, $\mathfrak\Theta:=\Gamma_{{{\mathcal K}}}P_{\mathrm{Supp}(D)} $ defines a self-adjoint unitary in ${\mathop{\mathrm{M}}\nolimits}_m$. Similarly, because of (iii) of Lemma \[lem:even\_fmps\_from\_split\_main\], $W_g:=V_gP_{\mathrm{Supp}(D)}$ defines a projective unitary/anti-unitary representation of $G$ on $P_{\mathrm{Supp}(D)} $ relative to ${\mathfrak{p}}$. Clearly, the second cohomology class associated to $W$ is the same of that of $V$, i.e., $[\upsilon]$. From ${\mathop{\mathrm{Ad}}\nolimits}_{V_g}(\Gamma_{{{\mathcal K}}})=(-1)^{{\mathfrak{q}}(g) }\Gamma_{{{\mathcal K}}}$, we have that ${\mathop{\mathrm{Ad}}\nolimits}_{W_g}{\left (}{\mathfrak \Theta}{\right )}=(-1)^{{\mathfrak{q}}(g)} \mathfrak \Theta$.
Now we check the properties (i)-(v).\
Parts (ii) and (v) are immediate from the definition of $v_\mu$, $\mathfrak \Theta$, $W_g$, and the corresponding properties of $B_\mu$, $\Gamma_{{\mathcal K}}$, $V_g$. Part (iii) follows from Lemma \[lem:even\_fmps\_from\_split\_main\] (i), (iii). For part (i), using , and that $P_{\mathrm{Supp}(D)}$ is of finite rank, we have $$\begin{aligned}
T_{\bf v}^N(x)=
P_{\mathrm{Supp}(D)} \, T_{\bf B}^N(x) \, P_{\mathrm{Supp}(D)}
\xrightarrow[N\to \infty]{} \langle {{\Omega_\omega}},\iota^{-1}(x){{\Omega_\omega}}\rangle \, P_{\mathrm{Supp}(D)}
={\mathop{\mathrm{Tr}}\nolimits}_{{{\mathcal K}}\otimes{{{\mathbb C}}^2}}\big( D x \big) P_{\mathrm{Supp}(D)}\end{aligned}$$ for $x\in P_{\mathrm{Supp}(D)}{{\mathcal R}}_{0,{{\mathcal K}}} P_{\mathrm{Supp}(D)}={\mathop{\mathrm{M}}\nolimits}_m$ and convergence in the norm topology. For part (iv), and imply that $$\begin{aligned}
\omega{\left (}E_{\mu_0,\nu_0}^{(0)} E_{\mu_1,\nu_1}^{(1)}
\cdots E_{\mu_N,\mu_N}^{(N)}{\right )}&={\mathop{\mathrm{Tr}}\nolimits}_{{{\mathcal K}}\otimes{{{\mathbb C}}^2}} {\left (}D
{\left (}\iota\circ\pi_\omega{\left (}E_{\mu_0,\nu_0}^{(0)} E_{\mu_1,\nu_1}^{(1)}
\cdots E_{\mu_N,\mu_N}^{(N)}{\right )}{\right )}{\right )}\nonumber\\
& =
(-1)^{\sum\limits_{k=1}^N (|\mu_k|+|\nu_k|) \sum\limits_{j=0}^{k-1} |\nu_j|}
{\mathop{\mathrm{Tr}}\nolimits}_{{{\mathcal K}}\otimes{{{\mathbb C}}^2}} {\left (}D
B_{\mu_0} \cdots B_{\mu_N} B_{\nu_N}^* \cdots B_{\nu_0}^*{\right )}\nonumber\\
& =
(-1)^{\sum\limits_{k=1}^N (|\mu_k|+|\nu_k|) \sum\limits_{j=0}^{k-1} |\nu_j|}
{\mathop{\mathrm{Tr}}\nolimits}_{{\mathop{\mathrm{M}}\nolimits}_m} {\left (}D
v_{\mu_0} \cdots v_{\mu_N} v_{\nu_N}^* \cdots v_{\nu_0}^*{\right )}\end{aligned}$$ for all $N\in{{\mathbb N}}\cup\{0\}$ and $\mu_0,\ldots\mu_N,\nu_0,\ldots,\nu_N\in{{\mathcal P}}$. This proves (iv).
Case: $\kappa_\omega=1$ (odd fermionic MPS)
-------------------------------------------
\[thmoddmps\] Let $\omega$ be a pure, split, translation invariant and $\alpha$-invariant state on ${{\mathcal A}}$ with index ${\mathop{\mathrm{Ind}}\nolimits}(\omega)=(1,{\mathfrak{q}},[\upsilon])$. For each $N\in{{\mathbb N}}$, let $Q_N$ be the support projection of the density matrix of $\omega\vert_{{{\mathcal A}}_{[0,N-1]}}$ and assume $\sup_{N\in{{\mathbb N}}} {\mathop{\mathrm{rank}}\nolimits}(Q_N)<\infty$. Then there is some $m\in{{\mathbb N}}$, a faithful density matrix $D\in {\mathop{\mathrm{M}}\nolimits}_m$, a set of matrices $\{v_\mu\}_{\mu\in{{\mathcal P}}}$ in ${\mathop{\mathrm{M}}\nolimits}_m$ and $\sigma_0\in\{0,1\}$ satisfying the following.
1. Set $\hat v_\mu:=v_\mu\otimes \sigma_z^{\sigma_0+|\mu|}$ on ${{\mathbb C}}^m\otimes {{{\mathbb C}}^2}$. Then $\lim_{N\to\infty}T_{\bf \hat v}^N(b)={\mathop{\mathrm{Tr}}\nolimits}{\left (}{\left (}D\otimes \frac 12{\mathbb I}_{{{{\mathbb C}}^2}} {\right )}b{\right )}{\mathbb I}_{{\mathop{\mathrm{M}}\nolimits}_m}\otimes
{\mathbb I}_{{{{\mathbb C}}^2}}$ in norm for all $b\in{\mathop{\mathrm{M}}\nolimits}_m\otimes {\mathop{\mathfrak{C}}\nolimits}$.
2. For any $l\in{{\mathbb N}}\cup\{0\}$, and $\mu_0,\ldots\mu_l, \nu_0,\ldots\nu_l\in {{\mathcal P}}$, $$\begin{aligned}
& \omega\big( E_{\mu_0,\nu_0}^{(0)} E_{\mu_1,\nu_1}^{(1)}\cdots E_{\mu_l,\nu_l}^{(l)} \big) \nonumber\\
&\hspace{1.5cm} =(-1)^{\sum\limits_{k=1}^l (|\mu_k|+|\nu_k|)
\sum\limits_{j=0}^{k-1} {\left (}\sigma_0+|\nu_j|{\right )}} \delta_{\sum_{i=0}^l (|\mu_i|+|\nu_i|),\,0}
{\mathop{\mathrm{Tr}}\nolimits}{\left (}D
{\left (}v_{\mu_0}\cdots v_{\mu_l} v_{\nu_l}^* \cdots v_{\nu_0}^* {\right )}{\right )}.
\end{aligned}$$
3. There a projective unitary/anti-unitary representation $W$ of $G$ on ${{\mathbb C}}^m$ relative to ${\mathfrak{p}}$ and $c_g \in \mathbb{T}$ such that for all $g\in G$ and $\nu\in{{\mathcal P}}$ $$\begin{aligned}
(-1)^{{\mathfrak{q}}(g)|\nu|} \sum_{\mu\in\calP} \langle \psi_\mu, \, \mathfrak{\Gamma}(U_g) \psi_\nu \rangle v_\mu =
c_g W_g v_\nu W_g^*,
\qquad {\mathop{\mathrm{Ad}}\nolimits}_{W_g}(D)=D.
\end{aligned}$$ The second cohomology class associated to $W$ is $[\upsilon]$.
\[lem:odd\_fmps\_from\_split\_main\] Consider the setting of Theorem \[thmoddmps\]. Suppose that the graded $W^*$-$(G,{\mathfrak{p}})$-dynamical system $(\pi_\omega({{\mathcal A}}_R)'', {\mathop{\mathrm{Ad}}\nolimits}_{\Gamma_\omega}, \hat\alpha_\omega)$ associated to $\omega$ is equivalent to $({{\mathcal R}}_{1,{{\mathcal K}}},{\mathop{\mathrm{Ad}}\nolimits}_{\Gamma_{{\mathcal K}}},{\mathop{\mathrm{Ad}}\nolimits}_{V_g})\in{{\mathcal S}}_{1}$, via a $*$-isomorphism $\iota: \pi_\omega({{\mathcal A}}_R)''\to {{\mathcal R}}_{1,{{\mathcal K}}}$. Then the following holds.
1. There is a finite rank density operator $D$ on $\calK$ such that for all $A \in {{{\mathcal A}}_R}$, $$\begin{aligned}
\label{irg}
{\mathop{\mathrm{Tr}}\nolimits}_{{{\mathcal K}}\otimes{{{\mathbb C}}^2}}{\left (}\big( D\otimes \tfrac{1}{2} {\mathbb I}_{{{{\mathbb C}}^2}} \big) \big(\iota\circ \pi_\omega(A) \big) {\right )}= \omega(A).\end{aligned}$$
2. Let $\{S_\mu\}_{\mu \in \calP}$ be the set of isometries given in Theorem \[cuntzthm1\]. Then we have $$\begin{aligned}
\label{vbvn}
v_\mu:=P_{\mathrm{Supp}(D)} S_\mu=P_{\mathrm{Supp}(D)} S_\mu P_{\mathrm{Supp}(D)},\quad
\mu\in{{\mathcal P}}.
\end{aligned}$$
3. $P_{\mathrm{Supp}(D)}V_g^{(0)} = V_g^{(0)} P_{\mathrm{Supp}(D)}$ and ${\mathop{\mathrm{Ad}}\nolimits}_{V_g^{(0)}}(D)=D$ for any $g\in G$.
\(i) Given the cyclic vector $\Omega_\omega$, $\langle \Omega_\omega, \iota^{-1}(x)\Omega_\omega\rangle$, $x\in {{\mathcal R}}_{1,{{\mathcal K}}}$, defines a normal state on ${{\mathcal R}}_{1,{{\mathcal K}}}$. Let $\tilde D$ be a density operator on ${{\mathcal K}}\otimes{{{\mathbb C}}^2}$ such that ${\mathop{\mathrm{Tr}}\nolimits}_{{{{\mathcal K}}\otimes{{{\mathbb C}}^2}}}(\tilde D x) = \langle \Omega_\omega, \iota^{-1}(x)\Omega_\omega \rangle$ for $x\in {{\mathcal R}}_{1,{{\mathcal K}}}$. Because ${{\mathcal R}}_{1,{{\mathcal K}}}={{{\mathcal B}}({{\mathcal K}})}\otimes {\mathop{\mathfrak{C}}\nolimits}$ and recalling the notation ${{\mathbb P}}_\varepsilon$ from , we may assume that $\tilde D$ is of the form $\tilde D=D_0\otimes {{\mathbb P}}_0(\sigma_x)+D_1\otimes {{\mathbb P}}_1(\sigma_x)$. Because $\omega\circ \Theta = \omega$ and $\iota\circ \pi_\omega \circ \Theta\vert_{{{\mathcal A}}_R} = \mathrm{Ad}_{\Gamma_{{\mathcal K}}} \circ \iota \circ \pi_\omega\vert_{{{\mathcal A}}_R}$, it follows that ${\mathop{\mathrm{Tr}}\nolimits}_{{{{\mathcal K}}\otimes{{{\mathbb C}}^2}}}( \mathrm{Ad}_{\Gamma_{{\mathcal K}}} (\tilde D) (\iota\circ \pi_\omega)(A))
= {\mathop{\mathrm{Tr}}\nolimits}_{{{{\mathcal K}}\otimes{{{\mathbb C}}^2}}} ( \tilde D(\iota\circ \pi_\omega)(A))$ for all $A\in {{\mathcal A}}_R$. Therefore, we have $\mathrm{Ad}_{\Gamma_{{\mathcal K}}} (\tilde D)= \tilde D$, which implies $D_0=D_1$. We set $D:=2 D_0$, and see that $D$ is a density operator on ${{\mathcal K}}$ satisfying .
Let $\pi_0$ be the irreducible representation of ${{\mathcal A}}_R^{(0)}$ on ${{\mathcal K}}$ given by $$\begin{aligned}
\iota\circ\pi_\omega(a)=\pi_0(a)\otimes{\mathbb I}_{{{{\mathbb C}}^2}},\quad a\in {{\mathcal A}}_R^{(0)}.\end{aligned}$$ Let $\calH_0 = \bigcap\limits_{N=1}^\infty {\left (}\pi_0(Q_N)
{{\mathcal K}}{\right )}$. Because $\pi_0$ is an irreducible representation of ${{\mathcal A}}_R^{(0)}$, ${{\mathcal H}}_0$ is finite-dimensional by Lemma \[lem:intersection\_of\_kernel\_fin\_dim\]. Because $\omega({\mathbb I}-Q_N)=0$, we have $${\mathop{\mathrm{Tr}}\nolimits}_{{{{\mathcal K}}\otimes{{{\mathbb C}}^2}}}\big(
(D\otimes \tfrac{1}{2} {\mathbb I}_{{{{\mathbb C}}^2}} ) ( \pi_0({\mathbb I}-Q_N) \otimes {\mathbb I}_{{{{\mathbb C}}^2}}) \big)
= \omega({\mathbb I}-Q_N)=0.$$ This means $P_{\mathrm{Supp}(D)}$ satisfies $P_{\mathrm{Supp}(D)}\le \pi_0(Q_N)$ for all $N\in{\mathbb{N}}$. Hence we have $P_{\mathrm{Supp}(D)}{{\mathcal K}}\subset{{\mathcal H}}_0$ and $D$ is finite rank.
\(ii) Recall the endomorphism $\rho$ satisfying from Lemma \[iorho\]. Because $\omega(A) = \omega(\beta_{S_1}(A))$ for all $A\in{{\mathcal A}}_R$, the set of isometries $\{S_\mu\}_{\mu \in \calP}$ given in Theorem \[cuntzthm1\] and $\sigma_0$, gives that $$\begin{aligned}
{\mathop{\mathrm{Tr}}\nolimits}_{{{\mathcal K}}\otimes{{\mathbb C}}^2}{\left (}{\left (}D\otimes \tfrac 12{\mathbb I}_{{{{\mathbb C}}^2}}{\right )}(\iota\circ\pi_\omega)(A){\right )}&={\mathop{\mathrm{Tr}}\nolimits}_{{{\mathcal K}}\otimes{{\mathbb C}}^2}{\left (}{\left (}D\otimes\tfrac 12{\mathbb I}_{{{{\mathbb C}}^2}}{\right )}(\rho\circ \iota\circ\pi_\omega)(A){\right )}\nonumber\\
& = \sum_{\mu} {\mathop{\mathrm{Tr}}\nolimits}_{{{\mathcal K}}\otimes{{\mathbb C}}^2}{\left (}{\mathop{\mathrm{Ad}}\nolimits}_{( S_\mu^* \otimes \sigma_z^{\sigma_0+|\mu|} )}
{\left (}D\otimes\tfrac 12{\mathbb I}_{{{{\mathbb C}}^2}}{\right )}(\iota\circ \pi_\omega)(A) {\right )},\end{aligned}$$ which implies that $D=\sum_{\mu}{\mathop{\mathrm{Ad}}\nolimits}_{S_\mu^*}(D)$. We then obtain by the same proof as in Lemma \[lem:even\_fmps\_from\_split\_main\].
\(iii) By the same argument as in the proof of Lemma \[lem:even\_fmps\_from\_split\_main\], we obtain $(V_g^{(0)})^* D V_g^{(0)} = D$ and so $P_{\mathrm{Supp}(D)}V_g^{(0)} = V_g^{(0)} P_{\mathrm{Supp}(D)}$.
We use the notation of Theorem \[cuntzthm1\], and Lemma \[lem:odd\_fmps\_from\_split\_main\]. Let $m\in{{\mathbb N}}$ be the rank of $D$ from Lemma \[lem:odd\_fmps\_from\_split\_main\]. We naturally identify $P_{\mathrm{Supp}(D)} {{\mathcal B}}({{\mathcal K}})P_{\mathrm{Supp}(D)} $ and ${\mathop{\mathrm{M}}\nolimits}_m$. Then we may regard $D$ as a faithful density matrix in ${\mathop{\mathrm{M}}\nolimits}_m$, and $\{v_\mu\}_{\mu\in{{\mathcal P}}}$ matrices in ${\mathop{\mathrm{M}}\nolimits}_m$. Because of part (iii) of Lemma \[lem:odd\_fmps\_from\_split\_main\], $W_g:=V_g^{(0)}P_{\mathrm{Supp}(D)}$ defines a projective unitary/anti-unitary representation of $G$ on $P_{\mathrm{Supp}(D)} {{\mathcal K}}$ relative to ${\mathfrak{p}}$ whose cohomology class is the same as $V^{(0)}$, i.e. $[\upsilon]$. Now we check the properties (i)-(iii) of Theorem \[thmoddmps\].
Part (iii) is immediate from the definition of $v_\mu$, $W_g$, and the corresponding properties of $S_\mu$ and $V_g^{(0)}$.
For part (i), using , and that $P_{\mathrm{Supp}(D)}$ is finite rank, we have $$\begin{aligned}
T_{\bf \hat v}^N(x)&=
P_{\mathrm{Supp}(D)}T_{\bf \hat S}^N(x) P_{\mathrm{Supp}(D)}
\xrightarrow[N\to\infty]{} {\langle{{\Omega_\omega}}, \iota^{-1}(x){{\Omega_\omega}}\rangle} P_{\mathrm{Supp}(D)}=
{\mathop{\mathrm{Tr}}\nolimits}{\left (}{\left (}D\otimes \tfrac 12{\mathbb I}_{{{{\mathbb C}}^2}} {\right )}x{\right )}P_{\mathrm{Supp}(D)}\end{aligned}$$ for $x\in {\left (}P_{\mathrm{Supp}(D)}\otimes{\mathbb I}{\right )}{{\mathcal R}}_{1, {{\mathcal K}}}{\left (}P_{\mathrm{Supp}(D)}\otimes{\mathbb I}{\right )}={\mathop{\mathrm{M}}\nolimits}_m\otimes{\mathop{\mathfrak{C}}\nolimits}$ and convergence in the norm topology.
Part (ii) follows from and , as in the proof of Theorem \[thmevenmps\].
Acknowledgements {#acknowledgements .unnumbered}
================
Y.O. would like to thank Y. Kubota and T. Matsui for discussion. The present work was supported by JSPS Grants-in-Aid for Scientific Research no. 16K05171 and 19K03534 (Y.O.) and 19K14548 (C.B.). It was also supported by JST CREST Grant Number JPMJCR19T2 (Y.O.).
Graded von Neumann algebras {#sec:GradedvN}
===========================
For convenience, we collect some facts about graded von Neumann algebras and linear/anti-linear group actions. See Section \[Subsec:graded\_prelims\] and \[subsec:graded\_product\_def\] for basic definitions.
\[gradei\] Let $({{\mathcal M}}, \theta)$ be a balanced graded von Neumann algebra. Assume that ${{\mathcal M}}$ is of type $\mu$ and ${{\mathcal M}}^{(0)}$ is of type $\lambda$, with some $\mu,\lambda=\mathrm{I,II,III}$, and that both of ${{\mathcal M}}$ and ${{\mathcal M}}^{(0)}$ have finite-dimensional centers. Then $\lambda=\mu$.
Let $U\in {{\mathcal M}}^{(1)}$ be a self-adjoint unitary. Let ${{\mathbb E}}:{{\mathcal M}}\to{{\mathcal M}}^{(0)}$ be the conditional expectation $$\begin{aligned}
\label{defe}
{{\mathbb E}}(x):=\frac 12(x+\theta(x)),\quad x\in {{\mathcal M}}.\end{aligned}$$ If ${{\mathcal M}}^{(0)}$ has a faithful normal semifinite trace $\tau_0$ (i.e., ${{\mathcal M}}^{(0)}$ is semifinite), then $\tau:=(\tau_0+\tau_0\circ{\mathop{\mathrm{Ad}}\nolimits}_{U})\circ{{\mathbb E}}$ defines a faithful normal semifinite trace on ${{\mathcal M}}$. Hence if ${{\mathcal M}}^{(0)}$ is semifinite, then ${{\mathcal M}}$ is semifinite.
Let us denote by ${{\mathcal P}}({{\mathcal M}}), {{\mathcal P}}({{\mathcal M}}^{(0)})$, the set of all orthogonal projections in ${{\mathcal M}}, {{\mathcal M}}^{(0)}$. As $\tau\vert_{{{\mathcal M}}^{(0)}}$ is a faithful normal semifinite trace on ${{\mathcal M}}^{(0)}$, if $\lambda=\mathrm{II}$, then we have $\tau {\left (}{{\mathcal P}}({{\mathcal M}}^{(0)}){\right )}=[0,\tau(1)]$. As $\tau({{\mathcal P}}({{\mathcal M}}))$ contains $\tau{\left (}{{\mathcal P}}({{\mathcal M}}^{(0)}){\right )}$ and ${{\mathcal M}}$ is a finite direct sum of type $\mu$-factors, this means that $\mu=\mathrm{II}$.
If $\lambda=\mathrm{I}$, then there is a non-zero abelian projection $p$ of ${{\mathcal M}}^{(0)}$. We claim there is a non-zero abelian projection $r$ in ${{\mathcal M}}$ such that $r\le p$. If $p{{\mathcal M}}^{(1)}p=\{0\}$, then $p{{\mathcal M}}p={{\mathbb C}}p$ and $p$ itself is abelian in ${{\mathcal M}}$. If $p{{\mathcal M}}^{(1)}p\neq\{0\}$, then there is a self-adjoint odd element $b\in {{\mathcal M}}^{(1)}$ such that $pbp\neq 0$. Because $(pbp)^2=pbpbp\in p {{\mathcal M}}^{(0)}p={{\mathbb C}}p$, we may assume that $pbp$ is a non-zero self-adjoint unitary in $p{{\mathcal M}}p$. For any $x\in {{\mathcal M}}^{(1)}$, we also have $pxppbp\in p {{\mathcal M}}^{(0)}p={{\mathbb C}}p$. By the unitarity of $pbp$, we have $pxp\in {{\mathbb C}}pbp$, and $p{{\mathcal M}}^{(1)}p={{\mathbb C}}pbp$. As $pbp$ is self-adjoint unitary, we have a spectral decomposition $pbp=r_+-r_-$, with mutually orthogonal projections $r_\pm$ in ${{\mathcal M}}$ and at least one of $r_\pm$ is non-zero. From $p{{\mathcal M}}^{(1)}p={{\mathbb C}}pbp={{\mathbb C}}(r_+-r_-)$ and $p{{\mathcal M}}^{(0)}p={{\mathbb C}}p$, $r_\pm$ are abelian in ${{\mathcal M}}$ and $r_\pm\le p$, proving the claim. Hence ${{\mathcal M}}$ is type I as well, $\mu=\mathrm{I}$.
Conversely, if ${{\mathcal M}}$ has a faithful normal semifinite trace $\tau$ (i.e., if ${{\mathcal M}}$ is semifinite), then $\tau\vert_{{{\mathcal M}}^{(0)}}$ is a faithful normal semfinite trace on ${{\mathcal M}}^{(0)}$. Therefore, $\mu=\mathrm{III}$ if and only if $\lambda=\mathrm{III}$.
If $\mu=\mathrm{I}$, then $\lambda$ cannot be II or III and so is type I. If $\mu=\mathrm{II}$, then $\lambda$ cannot be I or III and so is type II.
\[jh\] Let $({{\mathcal M}},\theta)$ be a central graded von Neumann algebra. Then either $Z({{\mathcal M}})={{\mathbb C}}{\mathbb I}$ or $Z({{\mathcal M}})$ has a self-adjoint unitary $b\in Z({{\mathcal M}})\cap {{\mathcal M}}^{(1)}$ such that $$\begin{aligned}
\label{bba}
Z({{\mathcal M}})\cap {{\mathcal M}}^{(1)}={{\mathbb C}}b.\end{aligned}$$
Let us assume that ${{\mathcal M}}$ is not a factor. By the condition of centrality, $Z({{\mathcal M}})\cap {{\mathcal M}}^{(0)}={{\mathbb C}}{\mathbb I}$, there is a non-zero self-adjoint element $b\in Z({{\mathcal M}})\cap {{\mathcal M}}^{(1)}$. Because $b^2\in Z({{\mathcal M}})\cap {{\mathcal M}}^{(0)}={{\mathbb C}}{\mathbb I}$, we may assume that $b$ is unitary. For any $x\in Z({{\mathcal M}})\cap {{\mathcal M}}^{(1)}$, $xb$ also belongs to $Z({{\mathcal M}})\cap {{\mathcal M}}^{(0)}={{\mathbb C}}{\mathbb I}$, and by the unitarity of $b$, we obtain .
When $({{\mathcal M}},\theta)$ is spatially graded, an analogous result holds for ${{\mathcal M}}\cap {{\mathcal M}}'\Gamma$.
\[jh2\] Let $({{\mathcal M}},{\mathop{\mathrm{Ad}}\nolimits}_{\Gamma})$ be a central graded von Neumann algebra on ${{\mathcal H}}$, spatially graded by a self-adjoint untiary $\Gamma$. Then the following holds.
1. If ${{\mathcal M}}$ is not a factor, ${{\mathcal M}}\cap {{\mathcal M}}'\Gamma=\{0\}$.
2. If ${{\mathcal M}}\cap {{\mathcal M}}'\Gamma\neq \{0\}$, then there is a self-adjoint unitary $b\in {{\mathcal M}}\cap {{\mathcal M}}'\Gamma$ such that ${{\mathcal M}}\cap {{\mathcal M}}'\Gamma={{\mathbb C}}b$. In particular, if $\Gamma\in {{\mathcal M}}$, then ${{\mathcal M}}\cap {{\mathcal M}}'\Gamma={{\mathbb C}}\Gamma$.
\(i) If ${{\mathcal M}}$ is not a factor, from Lemma \[jh\], $Z({{\mathcal M}})$ has a self-adjoint unitary $b\in Z({{\mathcal M}})\cap {{\mathcal M}}^{(1)}$ such that $
Z({{\mathcal M}})\cap {{\mathcal M}}^{(1)}={{\mathbb C}}b
$. For any $a\in {{\mathcal M}}\cap {{\mathcal M}}'\Gamma$, we have $$\begin{aligned}
ba=ab=a\Gamma\Gamma b\Gamma\Gamma
=a\Gamma{\left (}-b{\right )}\Gamma
=-(a\Gamma) b\Gamma
=-b(a\Gamma) \Gamma
=-ba.\end{aligned}$$ The first equality is because $b\in Z({{\mathcal M}})$, and the fifth equality is because $a\Gamma\in{{\mathcal M}}'$. As $b$ is unitary, this means $a=0$.
\(ii) Note that for any $a,b\in {{\mathcal M}}\cap {{\mathcal M}}'\Gamma$, $ab\in Z({{\mathcal M}})$. From this observation and (i), the same proof as Lemma \[jh\] gives the claim. If $\Gamma\in {{\mathcal M}}$, as $\Gamma={\mathbb I}\Gamma$, we have $\Gamma\in {{\mathcal M}}\cap {{\mathcal M}}'\Gamma$.
Recall the graded tensor product product defined in Section \[subsec:graded\_product\_def\].
\[comgra\] For $i=1,2$, let $({{\mathcal M}}_i,{\mathop{\mathrm{Ad}}\nolimits}_{\Gamma_i})$ be a graded von Neumann algebra on ${{\mathcal H}}_i$ spatially graded by a self-adjoint unitary $\Gamma_i$ on ${{\mathcal H}}_i$. Let ${{\mathcal M}}_1{{\,\hat{\otimes}\,}}{{\mathcal M}}_2$ be the graded tensor product of $({{\mathcal M}}_1,{{\mathcal H}}_1,\Gamma_1)$ and $({{\mathcal M}}_2,{{\mathcal H}}_2,\Gamma_2)$. Then commutant of the graded tensor product $({{\mathcal M}}_1{{\,\hat{\otimes}\,}} {{\mathcal M}}_2)'$ is generated by $$\begin{aligned}
\label{mmfc}
({{\mathcal M}}_1')^{(0)}\odot {{\mathcal M}}_2',\qquad
({{\mathcal M}}_1')^{(1)}\odot {{\mathcal M}}_2'\Gamma_2.\end{aligned}$$
The proof is given by a modification of corresponding result for ungraded tensor products. Let ${{\mathcal M}}:={{\mathcal M}}_1{{\,\hat{\otimes}\,}}{{\mathcal M}}_2$ and ${{\mathcal N}}$ be a von Neumann algebra generated by . We would like to show ${{\mathcal N}}={{\mathcal M}}'$. A brief computation gives the inclusion ${{\mathcal M}}\subset {{\mathcal N}}'$.
We let $\sigma \in \{0,1\}$ and denote by ${{\mathcal R}}^{h,(\sigma)}$ the set of all self-adjoint elements with grading $\sigma$ in a graded von Neumann algebra ${{\mathcal R}}$. For a complex Hilbert space ${{\mathcal K}}$ and its real subspace ${{\mathcal V}}$, ${{\mathcal V}}_{{\mathbb R}}^\perp$ is the orthogonal complement of ${{\mathcal V}}$ in ${{\mathcal K}}$ regarding ${{\mathcal K}}$ as a real Hilbert space, with respect to the inner product $\langle \cdot ,\cdot \rangle _{{{\mathbb R}}}:=\Re \langle \cdot ,\cdot \rangle$.
First we assume that ${{\mathcal M}}_j$, $j=1,2$, has a cyclic vector $\Omega_j$ which is homogeneous in the sense that $\Gamma_j\Omega_j=(-1)^{\epsilon_j}\Omega_j$ for some $\epsilon_j\in \{0,1\}$.
As $\Omega:=\Omega_1\otimes \Omega_2$ is cyclic for ${{\mathcal M}}$ in ${{\mathcal H}}_1\otimes {{\mathcal H}}_2$, to show ${{\mathcal M}}'={{\mathcal N}}$, it suffices to show that ${{\mathcal M}}^h\Omega+i{{\mathcal N}}^h\Omega$ is dense in ${{\mathcal H}}_1\otimes{{\mathcal H}}_2$ by [@Takesaki1 Chapter IV, Lemma 5.7]. For $\sigma_j=0,1$, $j=1,2$, set ${{\mathcal L}}_{\sigma_j}^{(j)}:= ({\mathbb I}+(-1)^{\sigma_j}\Gamma_j){{\mathcal H}}_j$, $j=1,2$. Then by the cyclicity of $\Omega_j$, and $\Gamma_j\Omega_j=(-1)^{\epsilon_j}\Omega_j$, ${{\mathcal M}}_j^{(\sigma_j)}\Omega_j$ is a dense subspace of ${{\mathcal L}}_{\sigma_j+\epsilon_j}^{(j)}$. We also note $({{\mathcal M}}_j')^{(\sigma_j)}\Omega_j\subset {{\mathcal L}}_{\sigma_j+\epsilon_j}^{(j)}$. By [@Takesaki1 Chapter IV, Lemma 5.7], $i ({{\mathcal M}}_j')^h\Omega_j$ is dense in $({{\mathcal M}}_j^h\Omega_j)_{{\mathbb R}}^{\perp}$. Therefore, $i({{\mathcal M}}_j')^{h, (\sigma_j+\epsilon_j)}\Omega_j$ is dense in $(({{\mathcal M}}_j)^{h, (\sigma_j+\epsilon_j)}\Omega_j)_{{\mathbb R}}^{\perp}\cap {{\mathcal L}}_{\sigma_j}^{(j)}$. Set $Y_{\sigma_1}:=({{\mathcal M}}_1)^{h,(\sigma_1+\epsilon_1)}\Omega_1$ and $Z_{\sigma_2}:=({{\mathcal M}}_2)^{h, (\sigma_2+\epsilon_2)}\Omega_2$. By the above observation, $i({{\mathcal M}}_1')^{h, (\sigma_1+\epsilon_1)}\Omega_1$ is dense in $(Y_{\sigma_1})_{{\mathbb R}}^\perp\cap {{\mathcal L}}_{\sigma_1}^{(1)}$ and $i({{\mathcal M}}_2')^{h, (\sigma_2+\epsilon_2)}\Omega_2$ is dense in $(Z_{\sigma_2})_{{\mathbb R}}^\perp\cap {{\mathcal L}}_{\sigma_2}^{(2)}$. Because $Y_{\sigma_1}+iY_{\sigma_1}$ and $Z_{\sigma_2}+iZ_{\sigma_2}$ are dense in ${{\mathcal L}}_{\sigma_1}^{(1)}$ and ${{\mathcal L}}_{\sigma_2}^{(2)}$ respectively by [@Takesaki1 Chapter IV, Lemma 5.8], $Y_{\sigma_1}\odot Z_{\sigma_2}+i((Y_{\sigma_1})_{{\mathbb R}}^{\perp}\cap {{\mathcal L}}_{\sigma_1}^{(1)})\odot ((Z_{\sigma_2})_{{\mathbb R}}^{\perp}\cap {{\mathcal L}}_{\sigma_2}^{(2)})$ is dense in ${{\mathcal L}}_{\sigma_1}^{(1)}\otimes {{\mathcal L}}_{\sigma_2}^{(2)}$. Hence we conclude that $$\begin{aligned}
\label{vden}
({{\mathcal M}}_1)^{h, (\sigma_1+\epsilon_1)}\Omega_1\odot
({{\mathcal M}}_2)^{h, (\sigma_2+\epsilon_2)}\Omega_2
+i ({{\mathcal M}}_1')^{h, (\sigma_1+\epsilon_1)}\Omega_1\odot
({{\mathcal M}}_2')^{h, (\sigma_2+\epsilon_2)}\Omega_2=:{{\mathcal V}}_{\sigma_1,\sigma_2}\end{aligned}$$ is dense in ${{\mathcal L}}_{\sigma_1}^{(1)}\otimes {{\mathcal L}}_{\sigma_2}^{(2)}$. Using the homogeneity of $\Omega_j$, $\Gamma_j\Omega_j=(-1)^{\epsilon_j}\Omega_j$ we can prove that $
{{\mathcal M}}^h\Omega+i{{\mathcal N}}^h\Omega
$ includes $$\begin{aligned}
\sum_{\sigma_1,\sigma_2=0,1} i^{(\sigma_1+\epsilon_1)(\sigma_2+\epsilon_2)}
{{\mathcal V}}_{\sigma_1,\sigma_2}.\end{aligned}$$ By the density of ${{\mathcal V}}_{\sigma_1,\sigma_2}$ in ${{\mathcal L}}_{\sigma_1}^{(1)}\otimes {{\mathcal L}}_{\sigma_2}^{(2)}$, ${{\mathcal M}}^h\Omega+i{{\mathcal N}}^h\Omega$ is dense in ${{\mathcal H}}_1\otimes{{\mathcal H}}_2$ and this completes the proof for the case with cyclic vectors.
Now we drop the assumption of the existence of the cyclic vectors. Let $\{E_a'\}_a$ be a family of mutually orthogonal projections in ${{\mathcal M}}_1'$ such that each $E_a'$ is an orthogonal projection onto $\overline{{{\mathcal M}}_1\xi_a}$, with a homogeneous $\xi_a\in{{\mathcal H}}_1$, and $\sum_a E_a'={\mathbb I}_{{{\mathcal H}}_1}$. Let $\{F_b'\}_b$ be a family of mutually orthogonal projections in ${{\mathcal M}}_2'$ such that each $F_b'$ is an orthogonal projection onto $\overline{{{\mathcal M}}_2\eta_b}$, with a homogeneous $\eta_b\in{{\mathcal H}}_2$, and $\sum_b F_b'={\mathbb I}_{{{\mathcal H}}_2}$. Note that because $\xi_a$, $\eta_b$ are homogeneous, $E_a'$ and $F_b'$ are even with respect to ${\mathop{\mathrm{Ad}}\nolimits}_{\Gamma_1}$, ${\mathop{\mathrm{Ad}}\nolimits}_{\Gamma_2}$ respectively. Because $E_a'$ and $F_b'$ are even, the argument in [@KR Lemma 11.2.14] shows that the central support of $E_a'\otimes F_b'\in{{\mathcal N}}\subset {{\mathcal M}}'$ with respect to ${{\mathcal N}}$ and the central support of $E_a'\otimes F_b'\in{{\mathcal N}}\subset {{\mathcal M}}'$ with respect to ${{\mathcal M}}'$ coincides. We denote the common central support by $P_{a,b}$. By the first part of the proof, we know that ${\left (}E_a'\otimes F_b'{\right )}{{\mathcal N}}{\left (}E_a'\otimes F_b'{\right )}={\left (}E_a'\otimes F_b'{\right )}{{\mathcal M}}'{\left (}E_a'\otimes F_b'{\right )}$. We also have $\sum_{a,b} E_a'\otimes F_b'={\mathbb I}_{{{\mathcal H}}_1\otimes{{\mathcal H}}_2}$. Therefore, applying [@KR Lemma 11.2.15], we get ${{\mathcal N}}={{\mathcal M}}'$.
\[oisoiso\] Let $({{\mathcal M}}_i,{\mathop{\mathrm{Ad}}\nolimits}_{\Gamma_i})$ , $({{\mathcal N}}_i,{\mathop{\mathrm{Ad}}\nolimits}_{W_i})$, $i=1,2$, be spatially graded von Neumann algebras on ${{\mathcal H}}_i$ and ${{\mathcal K}}_i$ respectively, with grading operators $\Gamma_i$ and $W_i$. Let $\alpha_i: {{\mathcal M}}_i\to{{\mathcal N}}_i$, $i=1,2$ be graded $*$-isomorphisms. Suppose that ${{\mathcal M}}_2$ (hence ${{\mathcal N}}_2$ as well) is either balanced or trivially graded. Let ${{\mathcal M}}_1{{\,\hat{\otimes}\,}}{{\mathcal M}}_2$ be the graded tensor product of $({{\mathcal M}}_1,{{\mathcal H}}_1,\Gamma_1)$ and $({{\mathcal M}}_2,{{\mathcal H}}_2,\Gamma_2)$. Let ${{\mathcal N}}_1{{\,\hat{\otimes}\,}}{{\mathcal N}}_2$ be the graded tensor product of $({{\mathcal N}}_1,{{\mathcal K}}_1,W_1)$ and $({{\mathcal N}}_2,{{\mathcal K}}_2,W_2)$. Then there exists a unique $*$-isomorphism $\alpha_1{{\,\hat{\otimes}\,}}\alpha_2:{{\mathcal M}}_1{{\,\hat{\otimes}\,}}{{\mathcal M}}_2\to
{{\mathcal N}}_1{{\,\hat{\otimes}\,}}{{\mathcal N}}_2$ such that $$\begin{aligned}
\label{abab}
{\left (}\alpha_1{{\,\hat{\otimes}\,}}\alpha_2
{\right )}(a{{\,\hat{\otimes}\,}} b)
=\alpha_1(a) {\,\hat{\otimes}\,}\alpha_2(b),\end{aligned}$$ for all $a\in {{\mathcal M}}_1$ and homogeneous $b\in{{\mathcal M}}_2$.
As $\alpha_2^{(0)}:=\alpha_2\vert_{{{\mathcal M}}_2^{(0)}}$ is a normal $*$-isomorphism from ${{\mathcal M}}_2^{(0)}$ onto ${{\mathcal N}}_2^{(0)}$, by [@Takesaki1 Chapter IV, Corollary 5.3] there is a unique $*$-isomorphism $\alpha^{(0)}$ from ${{\mathcal M}}_1\otimes{{\mathcal M}}_2^{(0)}$ onto ${{\mathcal N}}_1\otimes{{\mathcal N}}_2^{(0)}$ such that $$\begin{aligned}
\alpha^{(0)}(a\otimes b)
=\alpha_1(a)\otimes\alpha_2(b),\quad a\in {{\mathcal M}}_1,\quad b\in{{\mathcal M}}_2^{(0)}.\end{aligned}$$ If ${{\mathcal M}}_2$ is trivially graded then we set $\alpha_1{{\,\hat{\otimes}\,}}\alpha_2:=\alpha^{(0)}$. If ${{\mathcal M}}_2$ is balanced, let $U$ be a self-adjoint unitary element in ${{\mathcal M}}_2^{(1)}$. As we have ${{\mathcal M}}=\big( {{\mathcal M}}_1\otimes {{\mathcal M}}_2^{(0)}\big) \oplus \big({{\mathcal M}}_1\otimes {{\mathcal M}}_2^{(0)} \big)
{\left (}\Gamma_1\otimes U
{\right )}$, we may define a linear map $\alpha_1{{\,\hat{\otimes}\,}}\alpha_2:{{\mathcal M}}\to{{\mathcal N}}$ by $$\begin{aligned}
(\alpha_1{{\,\hat{\otimes}\,}}\alpha_2) (x+y(\Gamma_1\otimes U))
=\alpha^{(0)}(x)+\alpha^{(0)}(y){\left (}W_1\otimes \alpha_2(U){\right )},\quad
x, y\in {{\mathcal M}}_1\otimes {{\mathcal M}}_2^{(0)}.\end{aligned}$$ It is straightforward to check that $\alpha_1{{\,\hat{\otimes}\,}}\alpha_2$ is a normal $*$-homomorphism. Similarly, we may define a normal $*$-homomorphism ${\left (}\alpha_1{\right )}^{-1}{{\,\hat{\otimes}\,}}{\left (}\alpha_2{\right )}^{-1}:{{\mathcal N}}\to{{\mathcal M}}$, which turns out to be the inverse of $\alpha_1{{\,\hat{\otimes}\,}}\alpha_2$. Hence $\alpha_1{{\,\hat{\otimes}\,}}\alpha_2$ is a $*$-isomorphism satisfying . The uniqueness is trivial from .
\[isoiso\] Let $({{\mathcal M}}_i,{\mathop{\mathrm{Ad}}\nolimits}_{\Gamma_i})$, $i=1,2$, be balanced and spatially graded von Neumann algebras on ${{\mathcal H}}_i$ with a grading operator $\Gamma_i$. Let ${{\mathcal M}}_1{{\,\hat{\otimes}\,}}{{\mathcal M}}_2$ be the graded tensor product of $({{\mathcal M}}_1,{{\mathcal H}}_1,\Gamma_1)$ and $({{\mathcal M}}_2,{{\mathcal H}}_2,\Gamma_2)$. For any graded $*$-automorphism $\beta_i$ on ${{\mathcal M}}_i$ implemented by a unitary $V_i$ on ${{\mathcal H}}_i$ satisfying $V_i\Gamma_i=(-1)^{\nu_i}\Gamma_i V_i$, $\nu_i\in\{0,1\}$ for each $i=1,2$, the automorphism $\beta_1{{\,\hat{\otimes}\,}}\beta_2$ on ${{\mathcal M}}_1{{\,\hat{\otimes}\,}}{{\mathcal M}}_2$ defined in Lemma \[oisoiso\] satisfies $$\begin{aligned}
\label{aiai}
{\left (}\beta_1{{\,\hat{\otimes}\,}}\beta_2
{\right )}(a{{\,\hat{\otimes}\,}} b)
={\mathop{\mathrm{Ad}}\nolimits}_{(V_1\otimes V_2\Gamma_2^{\nu_1})}{\left (}a{{\,\hat{\otimes}\,}} b{\right )},\end{aligned}$$ for all $a\in {{\mathcal M}}_1$ and homogeneous $b\in{{\mathcal M}}_2$.
We compute that $$\begin{aligned}
{\left (}\beta_1{{\,\hat{\otimes}\,}}\beta_2
{\right )}(a{{\,\hat{\otimes}\,}} b)
&=\beta_1(a)\Gamma_1^{\partial b}\otimes \beta_2(b)
={\mathop{\mathrm{Ad}}\nolimits}_{(V_1\otimes V_2)} {\left (}a\Gamma_1^{\partial b}(-1)^{\partial b\cdot \nu_1} \otimes b
{\right )}\\
&={\mathop{\mathrm{Ad}}\nolimits}_{(V_1\otimes V_2)} {\mathop{\mathrm{Ad}}\nolimits}_{({\mathbb I}\otimes \Gamma_2^{\nu_1})}
\big( a\Gamma_1^{\partial b}\otimes b \big)\end{aligned}$$ from which follows.
We also consider anti-linear $\ast$-automorphisms.
\[aisoaiso\] Let $({{\mathcal M}}_i,{\mathop{\mathrm{Ad}}\nolimits}_{\Gamma_i})$, $i=1,2$, be balanced and spatially graded von Neumann algebras on ${{\mathcal H}}_i$ with a grading operator $\Gamma_i$. Let ${{\mathcal M}}_1{{\,\hat{\otimes}\,}}{{\mathcal M}}_2$ be the graded tensor product of $({{\mathcal M}}_1,{{\mathcal H}}_1,\Gamma_1)$ and $({{\mathcal M}}_2,{{\mathcal H}}_2,\Gamma_2)$ Suppose that ${{\mathcal M}}_i$ has a faithful normal representation $({{\mathcal K}}_i,\pi_i)$ with a self-adjoint unitary $W_i$ on ${{\mathcal K}}_i$ satisfying ${\mathop{\mathrm{Ad}}\nolimits}_{W_i}\circ\pi_i(x)=\pi_i\circ {\mathop{\mathrm{Ad}}\nolimits}_{\Gamma_i}(x)$, $x\in{{\mathcal M}}_i$ and a complex conjugation ${{\mathcal C}}_i$ on ${{\mathcal K}}_i$ satisfying ${\mathop{\mathrm{Ad}}\nolimits}_{{{\mathcal C}}_i}{\left (}\pi_i({{\mathcal M}}_i){\right )}=\pi_i({{\mathcal M}}_i)$ and ${{\mathcal C}}_iW_i=W_i{{\mathcal C}}_i$, for $i=1,2$. Then for any graded anti-linear $*$-automorphism $\beta_i$ on ${{\mathcal M}}_i$, $i=1,2$, there exists a unique anti-linear $*$-automorphism $\beta_1{{\,\hat{\otimes}\,}}\beta_2$ on ${{\mathcal M}}_1{{\,\hat{\otimes}\,}} {{\mathcal M}}_2$ such that $$\begin{aligned}
\label{abab3}
{\left (}\beta_1{{\,\hat{\otimes}\,}}\beta_2
{\right )}{\left (}a{{\,\hat{\otimes}\,}} b{\right )}=\beta_1(a){{\,\hat{\otimes}\,}} \beta_2(b),\end{aligned}$$ for all $a\in {{\mathcal M}}_1$ and homogeneous $b\in{{\mathcal M}}_2$.
If $\beta_i$ is implemented by an anti-unitary $V_i$ on ${{\mathcal H}}_i$ satisfying $V_i\Gamma_i=(-1)^{\nu_i}\Gamma_i V_i$, $\nu_i\in\{0,1\}$ for each $i=1,2$, then $$\begin{aligned}
{\left (}\beta_1{{\,\hat{\otimes}\,}}\beta_2
{\right )}{\left (}a {\,\hat{\otimes}\,}b{\right )}={\mathop{\mathrm{Ad}}\nolimits}_{(V_1\otimes V_2\Gamma_2^{\nu_1})}{\left (}a{{\,\hat{\otimes}\,}} b{\right )}.\end{aligned}$$
Let $\pi_1({{\mathcal M}}_1){{\,\hat{\otimes}\,}}\pi_2({{\mathcal M}}_2)$ be the graded tensor product of the $(\pi_1({{\mathcal M}}_1),{{\mathcal K}}_1,{W_1})$ and $(\pi_2({{\mathcal M}}_2),{{\mathcal K}}_2,{W_2})$. By Lemma \[oisoiso\], there is a $*$-isomorphism $\pi:=\pi_1{{\,\hat{\otimes}\,}} \pi_2$ from ${{\mathcal M}}_1{{\,\hat{\otimes}\,}}{{\mathcal M}}_2$ onto $\pi_1({{\mathcal M}}_1){{\,\hat{\otimes}\,}}\pi_2({{\mathcal M}}_2)$ satisfying ${\left (}\pi_1{{\,\hat{\otimes}\,}}\pi_2{\right )}(a{{\,\hat{\otimes}\,}} b)=\pi_1(a){{\,\hat{\otimes}\,}} \pi_2(b)$ for $a\in{{\mathcal M}}_1$ and homogeneous $b\in {{\mathcal M}}_2$. Because $\beta_i$, ${\mathop{\mathrm{Ad}}\nolimits}_{{{\mathcal C}}_i}$ and $\pi_i$ preserve the grading, $\alpha_i:={\mathop{\mathrm{Ad}}\nolimits}_{{{\mathcal C}}_i}\circ\pi_i\circ\beta_i\circ\pi_i^{-1}$ is a graded (linear) $*$-automorphism on $\pi_i({{\mathcal M}}_i)$. By Lemma \[oisoiso\], there is a $*$-automorphism $\alpha:=\alpha_1{{\,\hat{\otimes}\,}}\alpha_2$ on $\pi_1({{\mathcal M}}_1){{\,\hat{\otimes}\,}}\pi_2({{\mathcal M}}_2)$ such that ${\left (}\alpha_1{{\,\hat{\otimes}\,}}\alpha_2{\right )}(a{{\,\hat{\otimes}\,}} b)=\alpha_1(a){{\,\hat{\otimes}\,}} \alpha_2(b)$ for $a\in\pi_1({{\mathcal M}}_1)$ and homogeneous $b\in \pi_2({{\mathcal M}}_2)$. Furthermore, for ${{\mathcal C}}:={{\mathcal C}}_1\otimes{{\mathcal C}}_2$, ${\mathop{\mathrm{Ad}}\nolimits}_{{{\mathcal C}}}$ preserves $\pi_1({{\mathcal M}}_1){{\,\hat{\otimes}\,}}\pi_2({{\mathcal M}}_2)$. Therefore, $\beta_1{{\,\hat{\otimes}\,}}\beta_1:=\pi^{-1}\circ{\mathop{\mathrm{Ad}}\nolimits}_{{{\mathcal C}}}\circ\alpha\circ\pi$ defines an anti-linear $*$-automorphism on ${{\mathcal M}}_1{{\,\hat{\otimes}\,}} {{\mathcal M}}_2$ and it satisfies .
The proof for the second half of the lemma is the same as in Lemma \[isoiso\].
\[ga8\] Let $G$ be a finite group and ${\mathfrak{p}}: G\to{{\mathbb Z}}_2$ be a group homomorphism. Let $({{\mathcal M}}_1,{\mathop{\mathrm{Ad}}\nolimits}_{\Gamma_1},\alpha_1)$, $({{\mathcal M}}_2,{\mathop{\mathrm{Ad}}\nolimits}_{\Gamma_2},\alpha_2)$ be graded $W^*$-$(G,{\mathfrak{p}})$-dynamical systems such that, for $i=1,2$, $\calM_i$ is a balanced, central, spatially graded and type I von Neumann algebra with grading operator $\Gamma_i$. Let ${{\mathcal M}}_1{{\,\hat{\otimes}\,}}{{\mathcal M}}_2$ be the graded tensor product of $({{\mathcal M}}_1,{{\mathcal H}}_1,\Gamma_1)$ and $({{\mathcal M}}_2,{{\mathcal H}}_2,\Gamma_2)$. Then for every $g\in G$, there exists a linear $*$-automorphism (${{\mathfrak{p}}(g)}=0$) or anti-linear automorphism (${{\mathfrak{p}}(g)}=1$), $(\alpha_{1}{{\,\hat{\otimes}\,}}\alpha_{2})_g$ on ${{\mathcal M}}_1{{\,\hat{\otimes}\,}}{{\mathcal M}}_2$ such that $$\begin{aligned}
{\left (}\alpha_{1}{{\,\hat{\otimes}\,}}\alpha_{2}{\right )}_g
(a{{\,\hat{\otimes}\,}} b)
=\alpha_{1,g}(a){{\,\hat{\otimes}\,}} \alpha_{2,g}(b),\end{aligned}$$ for all homogeneous $a\in {{\mathcal M}}_1$ and $b\in{{\mathcal M}}_2$.
By Lemma \[casebycase\], there are graded $*$-isomorphisms $\iota_i: {{\mathcal M}}_i\to {{\mathcal R}}_{\kappa_i, {{\mathcal K}}_i}$ with some\
$({{\mathcal R}}_{\kappa_i, {{\mathcal K}}_i},{\mathop{\mathrm{Ad}}\nolimits}_{\Gamma_{{{\mathcal K}}_i}},{\mathop{\mathrm{Ad}}\nolimits}_{V_{i,g}})\in{{\mathcal S}}_{\kappa_i}$ for each $i=1,2$. Hence, $({{\mathcal K}}_i\otimes {{{\mathbb C}}^2}, \iota_i)$ is a faithful normal representation with a self-adjoint unitary $\Gamma_{{{\mathcal K}}_i}$ implementing ${\mathop{\mathrm{Ad}}\nolimits}_{\Gamma_i}$ on ${{\mathcal K}}_i\otimes{{{\mathbb C}}^2}$. Let $C$ be a complex conjugation with respect to the standard basis of ${{{\mathbb C}}^2}$ and ${{\mathcal C}}_i$ be any complex conjugation on ${{\mathcal K}}_i$. Then ${{\mathcal C}}_i\otimes C$ is a complex conjugation on ${{\mathcal K}}_i\otimes {{{\mathbb C}}^2}$ commuting with $\Gamma_{{{\mathcal K}}_i}={\mathbb I}_{{{\mathcal K}}_i}\otimes\sigma_z$, preserving ${{\mathcal R}}_{\kappa_i, {{\mathcal K}}_i} = \iota_i{\left (}{{\mathcal M}}_i{\right )}$. Hence we may apply Lemma \[oisoiso\] and Lemma \[aisoaiso\], which gives the result.
Lieb-Robinson bound for lattice fermion systems {#flr}
===============================================
In this section, prove the Lieb-Robinson bound for one-dimensional lattice fermion systems. While this result is not new, see [@BSP; @NSY17], our method of using an odd self-adjoint unitary to derive the Lieb-Robinson bound for odd elements from even elements is new. The result holds for more general metric graphs, but to avoid the introduction of further notation, we restrict ourselves to one-dimensional case. Let us recall the basic setting for Lieb-Robinson bound, see [@BMNS; @NSY17; @NSY18] for details.
\[deff\] An $F$-function $F$ on ${{\mathbb Z}}$ is a non-increasing function $F:[0,\infty)\to (0,\infty)$ such that
1. ${\left \Vert}F{\right \Vert}:=\sup_{x\in{{\mathbb Z}}}{\left (}\sum_{y\in{{\mathbb Z}}}F{\left (}d(x,y){\right )}{\right )}<\infty$, and
2. $C_{F}:=\sup_{x,y\in{{\mathbb Z}}}{\left (}\sum_{z\in{{\mathbb Z}}}
\frac{F{\left (}d(x,z){\right )}F{\left (}d(z,y){\right )}}{F{\left (}d(x,y){\right )}}{\right )}<\infty$.
Let $F$ be an $F$-function on ${{\mathbb Z}}$, and $I$ an interval in ${{\mathbb R}}$. We denote by ${{\mathcal B}}_{F}^e(I)$ the set of all norm continuous paths of even interactions on ${{\mathcal A}}$ defined on an interval $I$ such that the function ${\left \Vert}\Phi
{\right \Vert}_F: I\to {{\mathbb R}}$ defined by $$\begin{aligned}
\label{pnb}
{\left \Vert}\Phi
{\right \Vert}_F(t):=
\sup_{x,y\in{{\mathbb Z}}}\frac{1}{F{\left (}d(x,y){\right )}}\sum_{Z\in{{\mathfrak S}_{{{\mathbb Z}}}}, Z\ni x,y}
{\left \Vert}\Phi(Z;t){\right \Vert},\quad t\in I,\end{aligned}$$ is uniformly bounded, i.e., $\sup_{t\in I}{\left \Vert}\Phi {\right \Vert}(t)<\infty$.
For the rest of this Appendix, we fix some $\Phi\in {{\mathcal B}}_{F}^e(I)$. For each $s\in I$, we define a local Hamiltonian by . We denote by $U_{\Lambda,\Phi}(t;s)$ the solution of $$\begin{aligned}
\frac{d}{dt} U_{\Lambda,\Phi}(t;s)=-iH_{\Lambda,\Phi}(t) U_{\Lambda,\Phi}(t;s),\quad t,s\in I,\quad
U_{\Lambda,\Phi}(s;s)={\mathbb I}.\end{aligned}$$ We define the corresponding automorphisms $\tau_{t,s}^{(\Lambda),\Phi}$ on ${{\mathcal A}}_{{{\mathbb Z}}}$ by $$\begin{aligned}
\tau_{t,s}^{(\Lambda), \Phi}(A):=U_{\Lambda,\Phi}(t;s)^{*}AU_{\Lambda,\Phi}(t;s)\end{aligned}$$ with $A \in {{\mathcal A}}_{{\mathbb Z}}$. Note that ${\tau}_{s,t}^{(\Lambda), \Phi}$ is the inverse of $\tau_{t,s}^{(\Lambda),\Phi}$. Because $\Phi(s)$ is even, the proof of [@NSY18 Theorem 3.1] gives the following.
\[lre\] Let $X,Y\in{\mathfrak S}_{{\mathbb Z}}$ with $X\cap Y=\emptyset$. If either $A\in{{\mathcal A}}_{X}$ or $B\in{{\mathcal A}}_Y$ is even, then $$\begin{aligned}
{\left \Vert}\left[
\tau_{t,s}^{(\Lambda), \Phi}(A), B
\right]
{\right \Vert}\le
\frac{2{\left \Vert}A{\right \Vert}{\left \Vert}B{\right \Vert}}{C_F} {\left (}e^{v|t-s|}-1
{\right )}D_0(X,Y),\end{aligned}$$ where $v>0$ is some constant and $$\begin{aligned}
D_0(X,Y):=\sum_{x\in X}\sum_{y\in Y} F(|x-y|).\end{aligned}$$
Using this lemma and because $\Phi$ is even, the proof of [@NSY18 Theorem 3.4] guarantees the existence of the limit $$\begin{aligned}
\tau_{t,s}^{\Phi}(A):=\lim_{\Lambda \nearrow{{\mathbb Z}}}\tau_{t,s}^{(\Lambda), \Phi}(A),\quad
A\in{{\mathcal A}}, \quad t,s\in[0,1].\end{aligned}$$ Clearly the limit dynamics $\tau_{t,s}^{\Phi}$ satisfy the same Lieb-Robinson bound as in Lemma \[lre\]. We would like to have an analogous bound as Lemma \[lre\] for odd $A,B$. To do this, fix an odd self-adjoint unitary $U_0\in{{\mathcal A}}_{\{0\}}$. For each $m\in{{\mathbb Z}}$, $\beta_{S_{m}}(U_0)$ is a self-adjoint unitary in ${{\mathcal A}}_{\{m\}}$. Define an interaction $\tilde\Phi_m(s)$ by $$\begin{aligned}
\tilde\Phi_m(Z; s):={\mathop{\mathrm{Ad}}\nolimits}_{\beta_{S_{m}}(U_0)}{\left (}\Phi(Z; s)
{\right )},\quad Z\in{\mathfrak S}_{{{\mathbb Z}}},\quad s\in I,\quad m\in{{\mathbb N}}.\end{aligned}$$ Note that $\tilde\Phi_m(Z; s)=\Phi(Z; s)$ if $Z$ does not include $m$. Because $\tilde\Phi_m$ and $\Phi$ are even, Lemma \[lre\] and the proof of [@NSY18 Theorem 3.4] implies the bound $$\begin{aligned}
{\left \Vert}\tau_{t,s}^{\Phi}(A)-\tau_{t,s}^{\tilde\Phi_m}(A)
{\right \Vert}&\le
\frac{4{\left \Vert}A{\right \Vert}}{C_F} \sum_{Z\ni m}
\int_{[s,t]} dr {\left \Vert}\Phi(Z; r){\right \Vert}D_0(X,Z){\left (}e^{v|t-r|}-1{\right )}\nonumber\\
&\le
{4{\left \Vert}A{\right \Vert}}
\int_{[s,t]} {\left (}e^{v|t-r|}-1{\right )}{\left \Vert}\Phi{\right \Vert}_F(r)\sum_{x\in X} F(|x-m|)=:g(m), \end{aligned}$$ for any $A\in {{\mathcal A}}_X^{(1)}$, where the last inequality uses (i) and (ii) of Definition \[deff\] as well as Equation . Note that $\lim_{m\to\infty} g(m)=0$. Therefore, we have $$\begin{aligned}
\label{ulrb}
{\left \Vert}\left\{
\tau_{t,s}^{\Phi}(A), \beta_{S_{m}}(U_0)
\right\}
{\right \Vert}={\left \Vert}\tau_{t,s}^{\Phi}(A)-\tau_{t,s}^{\tilde\Phi_m}(A)
{\right \Vert}\le g(m),\end{aligned}$$ for any $A\in {{\mathcal A}}_X^{(1)}$ and $X\in{\mathfrak S}_{{{\mathbb Z}}}$ with $m \notin X$. Let $X,Y\in{\mathfrak S}_{{\mathbb Z}}$ with $X\cap Y=\emptyset$, $A\in{{\mathcal A}}_X^{(1)}$, $B\in {{\mathcal A}}_Y^{(1)}$ and $m\notin X$. Because $B\beta_{S_m}(U_0)\in {{\mathcal A}}_{Y\cup \{m\}}^{(0)}$, Lemma \[lre\] and implies $$\begin{aligned}
{\left \Vert}\left\{
\tau_{t,s}^{\Phi}(A), B
\right\}
{\right \Vert}&={\left \Vert}\left[
\tau_{t,s}^{\Phi}(A), B\beta_{S_m}(U_0)
\right] \beta_{S_m}(U_0)
+B\beta_{S_m}(U_0)
\left\{
\tau_{t,s}^{\Phi}(A), \beta_{S_{m}}(U_0)
\right\}
{\right \Vert}\nonumber\\
&\le
\frac{2{\left \Vert}A{\right \Vert}{\left \Vert}B{\right \Vert}}{C_F} {\left (}e^{v|t-s|}-1
{\right )}D_0(X,Y\cup\{m\})
+g(m){\left \Vert}B{\right \Vert}.\end{aligned}$$ Taking the limit $m\to\infty$ and using Lemma \[lre\], we obtain the following.
\[lruni\] Let $X,Y\in{\mathfrak S}_{{\mathbb Z}}$ with $X\cap Y=\emptyset$. For homogeneous $A\in{{\mathcal A}}_{X}$ and $B\in{{\mathcal A}}_Y$ we have $$\begin{aligned}
{\left \Vert}\tau_{t,s}^{\Phi}(A) B-(-1)^{\partial A\partial B} B\tau_{t,s}^{\Phi}(A)
{\right \Vert}\le
\frac{2{\left \Vert}A{\right \Vert}{\left \Vert}B{\right \Vert}}{C_F} {\left (}e^{v|t-s|}-1
{\right )}D_0(X,Y).\end{aligned}$$
As in quantum spin systems, we can estimate the locality of the time evolved observables from Lieb-Robinson bounds. To do this, let $\{ \mathbb{E}_N : \mathcal{A} \to \mathcal{A}_{ \Lambda_N} ~|~ N\in \mathbb{N} \}$ be the family of conditional expectations with respect to the trace on ${{\mathcal A}}$, see [@am]. By the same argument as [@NSY18 Corollary 4.4], if $A \in \mathcal{A}^{(0)}$ is such that $$\begin{split}
\big\| [A, B] \big\| \leq C \lVert B \rVert.
\end{split},$$ for all $B \in \bigcup_{\substack{X \in \mathfrak{S}_{\mathbb{Z}^\nu}\\ X \cap [-N,N]= \emptyset}} \mathcal{A}_X$, then $\lVert A - \mathbb{E}_N (A) \rVert \leq C$. We extend this bound to odd elements.
Suppose that $A \in \mathcal{A}^{(1)}$ is such that $$\begin{split}
\big\| AB-(-1)^{\partial B} BA \big\| \leq C \lVert B \rVert.
\end{split}$$ for all homogeneous $B \in \bigcup_{\substack{X \in \mathfrak{S}_{\mathbb{Z}^\nu}\\ X \cap [-N,N]= \emptyset}} \mathcal{A}_X$. Let $U_0\in {{\mathcal A}}_{\{0\}}^{(1)}$ be a self-adjoint unitary. Then we have $AU_0\in {{\mathcal A}}^{(0)}$ and $$\begin{aligned}
\big\| [AU_0, B] \big\|
=
\big\| \big( AB-(-1)^{\partial B} B A \big) U_0 \big\|
\le C {\left \Vert}B{\right \Vert}\end{aligned}$$ for all homogeneous $B \in \bigcup_{\substack{X \in \mathfrak{S}_{\mathbb{Z}^\nu}\\ X \cap [-N,N]= \emptyset}} \mathcal{A}_X$. Hence we have that ${\left \Vert}[AU_0, B]
{\right \Vert}\le 2C{\left \Vert}B{\right \Vert}$ for any $B \in \bigcup_{\substack{X \in \mathfrak{S}_{\mathbb{Z}^\nu}\\ X \cap [-N,N]= \emptyset}} \mathcal{A}_X$. Therefore, by the even case, we obtain that $$\begin{aligned}
{\left \Vert}A- \mathbb{E}_N (A)
{\right \Vert}={\left \Vert}{\left (}A- \mathbb{E}_N (A)
{\right )}U_0
{\right \Vert}=
\lVert AU_0 - \mathbb{E}_N (AU_0) \rVert \leq 2C, \end{aligned}$$ where we used the fact that $U_0\in {{\mathcal A}}_{\Lambda_N}$. From this and Lemma \[lruni\], we have shown the following.
\[lruni2\] For any $N\in{{\mathbb N}}$, $X\in{\mathfrak S}_{{\mathbb Z}}$ with $X\subset [-N,N]$ and $A\in{{\mathcal A}}_{X}$, we have $$\begin{aligned}
\big\|
{{\mathbb E}}_N{\left (}\tau_{t,s}^{\Phi}(A)
{\right )}-\tau_{t,s}^{\Phi}(A)
\big\|
\le
\frac{8{\left \Vert}A{\right \Vert}}{C_F} {\left (}e^{v|t-s|}-1
{\right )}D_0(X,[-N,N]^c).\end{aligned}$$
Having Lemma \[lruni\] and Lemma \[lruni2\] as input, we can carry out all the arguments in [@mo Theorem 1.3] and [@OgataTRI Proposition 3.5].
[99]{}
I. Affleck, T. Kennedy, E. H. Lieb, and H. Tasaki. [*Valence bond ground states in isotropic quantum antiferromagnets*]{}. Comm. Math. Phys., **115**, 477–528, (1988).
H. Araki, H. Moriya. [*Equilibrium statistical mechanics of fermion lattice systems*]{}. [Rev. Math. Phys.]{}, [**15**]{}, no. 02, 93–198 (2003).
S. Bachmann, S. Michalakis, B. Nachtergaele, R. Sims. [*Automorphic equivalence of gapped phases of quantum lattice systems*]{}. Comm. Math. Phys., **309**, no. 3, 835–871 (2012).
C. Bourne, H. Schulz-Baldes. [*On ${\ensuremath{\mathbb{Z}}}_2$-indices for ground states of fermionic chains*]{}. Rev. Math. Phys., **32**, 2050028 (2020).
O. Bratteli, P. E. T. Jorgensen. [*Endomorphisms of $B(H)$ II. Finitely Correlated States on $O_n$.*]{} J. Funct. Anal., [**145**]{}, 323–373 1997.
O. Bratteli P. Jorgensen, G. Price. [*Endomorphisms of $B({{\mathcal H}})$.*]{} In Quantization, nonlinear partial differential equations, and operator algebra (Cambridge, MA, 1994), 93–138, Proc. Sympos. Pure Math., **59**, Amer. Math. Soc., Providence (1996).
O. Bratteli, P. E. T. Jorgensen, A. Kishimoto, R.F Werner. [*Pure states on ${{\mathcal O}}_d$*]{}. [ J. Operator Theory]{}, [**43**]{}, 97–143 (2000).
O. Bratelli, D. R. Robinson. [*Operators Algebras and Quantum Statistical Mechanics 1*]{}. 2nd edition, Springer, Berlin (1997).
O. Bratelli, D. R. Robinson. [*Operators Algebras and Quantum Statistical Mechanics 2*]{}, 2nd edition, Springer, Berlin (1997).
J.-B. Bru, W. de Siqueira Pedra. [*Lieb-Robinson Bounds for Multi-Commutators and Applications to Response Theory*]{}. Volume 13 of Springer Briefs in Mathematical Physics, Springer, Berlin (2017).
N. Bultinck, D. J Williamson, J. Haegeman, F. Verstraete. [*Fermionic matrix product states and one-dimensional topological phases*]{}, Phys. Rev. B, [**95**]{}, 075108 (2017).
X. Chen, Z.-C. Gu, and X.-G. Wen. [*Classification of gapped symmetric phases in one-dimensional spin systems*]{}. Phys. Rev. B [**83**]{}, 035107 (2011).
M. Fannes, B. Nachtergaele, R.F. Werner. [*Finitely correlated states on quantum spin chains*]{}. [Comm. Math. Phys.]{}, [**144**]{}, 443–490 (1992).
L. Fidkowski, A. Kitaev. [*The effects of interactions on the topological classification of free fermion systems*]{}. [Phys. Rev. B]{} [**81**]{}, 134509 (2009).
L. Fidkowski, A. Kitaev. [*Topological phases of fermions in one dimension*]{}. Phys. Rev. B, **83**, 075103 (2011).
Z.-C. Gu, X.-G. Wen. [*Tensor-entanglement-filtering renormalization approach and symmetry-protected topological order*]{}. Phys. Rev. B, [**80**]{}, 155131 (2009).
M. Hastings. [*An area law for one-dimensional quantum systems.*]{} J. Stat. Mech. Theory Exp., no. 8, P08024, 14 pp. (2007).
R. V. Kadison, J. R. Ringrose. [*Fundamentals of the theory of operator algebras II.*]{} Volume 16 of Graduate Studies in Mathematics, Amer. Math. Soc., Providence (1997).
A. Kapustin, R. Thorngren. [*Fermionic SPT phases in higher dimensions and bosonization.* ]{} J. High Energy Phys., no. 10, 080, 48pp. (2017).
A. Kapustin, A. Turzillo, M. You. [*Spin topological field theory and fermionic matrix product states*]{}. Phys. Rev. B, **98**, 125101 (2018).
T. Matsui. [*A characterization of pure finitely correlated states*]{}. Infin. Dimens. Anal. Quantum Probab. Relat. Top., **1**, no. 4, 647–661 (1998).
T. Matsui. [*The split property and the symmetry breaking of the quantum spin chain.*]{} Comm. Math. Phys., [**218**]{}, 393–416 (2001).
T. Matsui. [*Boundedness of entanglement entropy and split property of quantum spin chains*]{}. Rev. Math. Phys., **26**, no. 9, 1350017 (2013).
T. Matsui. [*Split property and fermionic string order*]{}. arXiv:2003.13778 (2020).
A. Moon, Y. Ogata. [*Automorphic equivalence within gapped phases in the bulk.*]{} J. Funct. Anal., [**278**]{}, no. 8, 108422 (2020).
M. E. Moutuou. [*Graded Brauer groups of a groupoid with involution*]{}. J. Funct. Anal., **266**, no. 5, 2689–2739 (2014).
B. Nachtergaele, R. Sims, A. Young. [*Lieb–Robinson bounds, the spectral flow, and stability of the spectral gap for lattice fermion systems*]{}. In Mathematical Problems in Quantum Physics, Vol. 717 of Contemp. Math., Amer. Math. Soc., Providence (2018).
B. Nachtergaele, R. Sims, A. Young. [*Quasi-locality bounds for quantum lattice systems. Part 1. Lieb–Robinson bounds, quasi-local maps, and spectral flow automorphisms*]{}. J. Math. Phys., **60**, 061101 (2019).
Y. Ogata. [*A $\mathbb{Z}_2$-index of symmetry protected topological phases with time reversal symmetry for quantum spin chains*]{}. [Comm. Math. Phys.]{}, **374**, no. 2, 705–734 (2020).
Y. Ogata. [*A ${{\mathbb Z}}_2$-index of symmetry protected topological phases with reflection symmetry for quantum spin chains.*]{} arXiv:1904.01669 (2019)
Y. Ogata. [*A classification of pure states on quantum spin chains satisfying the split property with on-site finite group symmetries*]{}. arXiv:1908.08621 (2019).
Y. Ogata, A class of asymmetric gapped Hamiltonians on quantum spin chains and its characterization II, Comm. Math. Phys. [**348**]{} (2016), 897–957.
F. Pollmann, A. Turner, E. Berg, M. Oshikawa. [*Entanglement spectrum of a topological phase in one dimension*]{}. Phys. Rev. B [**81**]{}, 064439 (2010).
F. Pollmann, A. Turner, E. Berg, M. Oshikawa. [*Symmetry protection of topological phases in one-dimensional quantum spin systems*]{}. Phys. Rev. B [**81**]{}, 075125 (2012).
D. Perez-Garcia, M. M. Wolf, M. Sanz, F. Verstraete, J. I. Cirac. [*String order and symmetries in quantum spin lattices*]{}. Phys. Rev. Lett. [**100**]{}, 167202 (2008).
M. Takesaki. [*Theory of operator algebras. I.*]{} Encyclopaedia of Mathematical Sciences. Springer-Verlag, Berlin (2002).
A. Turzillo, M. You. [*Fermionic matrix product states and one-dimensional short-range entangled phases with antiunitary symmetries*]{}. Phys. Rev. B, **99**, 035103 (2019).
C. T. C. Wall. [*Graded Brauer groups*]{}, J. Reine Angew. Math., **213**, 187–199 (1963/64).
[^1]: Throughout this paper, we use the presentation of ${{\mathbb Z}}_2$ as the additive group $\{0,1\}$.
[^2]: In this article we use the following notation of Pauli matrices $$\begin{aligned}
\sigma_x:=\begin{pmatrix}
0&1\\1&0
\end{pmatrix}, \qquad
\sigma_y:=
\begin{pmatrix}
0&-i\\i&0
\end{pmatrix}, \qquad
\sigma_z:=
\begin{pmatrix}
1&0\\0&-1
\end{pmatrix}.\end{aligned}$$
[^3]: We may regard ${\mathop{\mathfrak{C}}\nolimits}$ as Clifford algebra ${{\mathbb C}}l_1$ generated by $e_1:=\sigma_x$.
| {
"pile_set_name": "ArXiv"
} | ArXiv |
---
abstract: 'We propose a renormalizable multi-Higgs model with $A_{4}\otimes Z_{2}\otimes Z^{\prime}_{2}$ symmetry, accounting for the experimental deviation from the tribimaximal mixing pattern of the neutrino mixing matrix. In this framework we study the charged lepton and neutrino masses and mixings. The light neutrino masses are generated via a radiative seesaw mechanism, which involves a single heavy Majorana neutrino and neutral scalars running in the loops. The obtained neutrino mixings and mass squared splittings are in good agreement with the neutrino oscillation experimental data for both normal and inverted hierarchy. The model predicts an effective Majorana neutrino mass $m_{\beta\beta}=$ 4 meV and 50 meV for the normal and the inverted neutrino spectrum, respectively. The model also features a suppression of CP violation in neutrino oscillations, a low scale for the heavy Majorana neutrino (few TeV) and, due to the unbroken $Z_2$ symmetry, a natural dark matter candidate.'
author:
- 'A. E. Cárcamo Hernández'
- 'I. de Medeiros Varzielas'
- 'S. Kovalenko'
- 'H. Päs'
- Iván Schmidt
title: 'Lepton masses and mixings in an $A_{4}$ multi-Higgs model with radiative seesaw mechanism'
---
Introduction
============
The existence of three generations of fermions, as well as their particular pattern of masses and mixing cannot be understood within the Standard Model (SM), and makes it appealing to consider a more fundamental theory addressing these issues. This problem is especially challenging in the neutrino sector, where the striking smallness of neutrino masses and large mixing between generations suggest a different kind of underlying physics than what should be responsible for the masses and mixings of the quarks. Unlike in the quark sector, where the mixing angles are very small, two of the three neutrino mixing angles, the atmospheric $\theta_{23}$ and the solar $\theta_{12}$ are large, while the reactor angle $\theta_{13}$ is comparatively small [@PDG; @Abe:2011sj; @Adamson:2011qu; @Abe:2011fz; @An:2012eh; @Ahn:2012nd; @Tortola:2012te; @Fogli:2012ua; @GonzalezGarcia:2012sz].
In the literature there has been a formidable amount of effort to understand the origin of the leptonic flavor structure, with various proposed scenarios and models of neutrino mass generation. Among those approaches to understand the pattern of neutrino mixing, models with discrete flavor symmetries are particularly popular (for recent reviews see Refs. [@King:2013eh; @Altarelli:2010gt; @Ishimori:2010au]). There is a great variety of such models, some with Multi-Higgs sectors [@textures; @Ma:2006km; @Ma:2006fn; @Hernandez:2013mcf; @Machado:2010uc; @Ma:2001dn; @Babu:2002dz; @Altarelli:2005yx; @Altarelli:2009gn; @Bazzocchi:2009pv; @He:2006dk; @Fukuyama:2010mz; @Fukuyama:2010ff; @Holthausen:2012wz; @Ahn:2012tv; @Ahn:2013mva; @Chen:2012st; @Toorop:2010ex; @Memenga:2013vc; @Ferreira:2013oga; @Felipe:2013vwa; @Machado:2007ng; @Machado:2011gn; @Felipe:2013ie; @Varzielas:2012ai; @Ishimori:2012fg; @Bhattacharya:2013mpa; @Teshima:2005bk; @Mohapatra:2006pu; @Ma:2013zca; @Canales:2013cga; @Canales:2012dr; @Dong:2011vb; @Kajiyama:2013sza; @Hernandez:2013hea; @Mohapatra:2012tb; @Varzielas:2012pa; @Ding:2013hpa; @Cooper:2012bd; @King:2013hj; @Morisi:2013qna; @Morisi:2013eca; @Varzielas:2012nn; @Bhattacharyya:2012pi; @Ma:2013xqa; @Nishi:2013jqa; @Varzielas:2013sla; @Aranda:2013gga], Extra Dimensions [@Rius:2001dd; @Dobrescu:1998dg; @Altarelli:2005yp; @Ishimori:2010fs; @Kadosh:2010rm; @Kadosh:2013nra; @Ding:2013eca; @CarcamoHernandez:2012xy], Grand Unification [@GUT] or Superstrings [@String]. Another approach attempts to describe certain phenomenological features of the fermion mass hierarchy by postulating particular zero-texture Yukawa matrices [@textures].
In this context, the groups explored recently in the literature include $A_4$ [@He:2006dk; @Fukuyama:2010mz; @Fukuyama:2010ff; @Ahn:2012tv; @Ahn:2013mva; @Chen:2012st; @Toorop:2010ex; @Memenga:2013vc; @Holthausen:2012wz; @Ferreira:2013oga; @Felipe:2013vwa; @Machado:2007ng; @Machado:2011gn; @Felipe:2013ie; @Varzielas:2012ai; @Ishimori:2012fg; @Bhattacharya:2013mpa; @King:2013hj; @Morisi:2013qna; @Morisi:2013eca; @Altarelli:2005yp; @Kadosh:2010rm; @Kadosh:2013nra], $\Delta(27)$ [@Varzielas:2012nn; @Bhattacharyya:2012pi; @Ma:2013xqa; @Nishi:2013jqa; @Varzielas:2013sla; @Aranda:2013gga] , $S_3$ [@Teshima:2005bk; @Mohapatra:2006pu; @Ma:2013zca; @Canales:2013cga; @Canales:2012dr; @Dong:2011vb; @Kajiyama:2013sza; @Hernandez:2013hea], $S_4$ [@Altarelli:2009gn; @Bazzocchi:2009pv; @Mohapatra:2012tb; @Varzielas:2012pa; @Ding:2013hpa; @Ding:2013eca; @Ishimori:2010fs] and $A_5$ [@Cooper:2012bd]. These models can be implemented in a supersymmetric framework [@Babu:2002dz; @Altarelli:2005yx; @Altarelli:2009gn; @Bazzocchi:2009pv; @Morisi:2013eca], or in extra dimensional scenarios with $S_4$ [@Ishimori:2010fs; @Ding:2013eca] or $A_4$ [@Altarelli:2005yp; @Kadosh:2010rm; @Kadosh:2013nra].
The popular tribimaximal (TBM) ansatz for the leptonic mixing matrix $$\label{TBM-ansatz}
U_{TBM}=\left(
\begin{array}{ccc}
\sqrt{\frac{2}{3}} & \frac{1}{\sqrt{3}} & 0 \\
-\frac{1}{\sqrt{6}} & \frac{1}{\sqrt{3}} & -\frac{1}{\sqrt{2}} \\
-\frac{1}{\sqrt{6}} & \frac{1}{\sqrt{3}} & \frac{1}{\sqrt{2}}\end{array}\right) ,$$ can originate, in particular, from $A_{4}$. TBM corresponds to mixing angles with $\left(\sin ^{2}\theta_{12}\right) _{TBM}=\frac1{3}$, $\left(\sin^{2}\theta _{23}\right)_{TBM}=\frac1{2}$, and $\left(\sin ^{2}\theta_{13}\right)_{TBM}=0$. On the other hand the T2K [@Abe:2011sj], MINOS [@Adamson:2011qu], Double Chooz [@Abe:2011fz], Daya Bay [@An:2012eh] and RENO [@Ahn:2012nd] experiments have recently measured a non-vanishing mixing angle $\theta_{13}$, ruling out the exact TBM pattern. The global fits of the available data from neutrino oscillation experiments [@Tortola:2012te; @Fogli:2012ua; @GonzalezGarcia:2012sz] give experimental constraints on the neutrino mass squared splittings and mixing parameters. We use the values from [@Tortola:2012te], shown in Tables \[NH\] and \[IH\], for the cases of normal and inverted hierarchy, respectively. It can be seen that the data deviate significantly from the TBM pattern.
--------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
Parameter $\Delta m_{21}^{2}$($10^{-5}$eV$^2$) $\Delta m_{31}^{2}$($10^{-3}$eV$^2$) $\left( \sin ^{2}\theta _{12}\right) _{\exp }$ $\left( $\left( \sin ^{2}\theta
\sin ^{2}\theta _{23}\right) _{\exp }$ _{13}\right) _{\exp }$
------------------ -------------------------------------- -------------------------------------- ------------------------------------------------ ---------------------------------------- -------------------------
Best fit $7.62$ $2.55$ $0.320$ $0.613$ $0.0246$
$1\sigma $ range $7.43-7.81$ $2.46-2.61$ $0.303-0.336$ $0.573-0.635$ $0.0218-0.0275$
$2\sigma $ range $7.27-8.01$ $2.38-2.68$ $0.29-0.35$ $0.38-0.66$ $0.019-0.030$
$3\sigma $ range $7.12-8.20$ $2.31-2.74$ $0.27-0.37$ $0.36-0.68$
--------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
: Range for experimental values of neutrino mass squared splittings and leptonic mixing parameters taken from Ref. [@Tortola:2012te] for the case of normal hierarchy.[]{data-label="NH"}
------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
Parameter $\Delta m_{21}^{2}$($10^{-5}$eV$^2$) $\Delta m_{13}^{2}$($10^{-3}$eV$^2$) $\left( \sin ^{2}\theta _{12}\right) _{\exp }$ $\left( $\left( \sin ^{2}\theta_{13}\right) _{\exp }$
\sin ^{2}\theta _{23}\right) _{\exp }$
------------------ -------------------------------------- -------------------------------------- ------------------------------------------------ ---------------------------------------- -----------------------------------------------
Best fit $7.62$ $2.43$ $0.320$ $0.600$ $0.0250$
$1\sigma $ range $7.43-7.81$ $2.37-2.50$ $0.303-0.336$ $0.569-0.626$ $0.0223-0.0276$
$2\sigma $ range $7.27-8.01$ $2.29-2.58$ $0.29-0.35$ $0.39-0.65 $ $0.020-0.030$
$3\sigma $ range $7.12-8.20$ $2.21-2.64$ $0.27-0.37$ $0.37-0.67$ $0.017-0.033$
------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
: Range for experimental values of neutrino mass squared splittings and leptonic mixing parameters taken from Ref. [@Tortola:2012te] for the case of inverted hierarchy.[]{data-label="IH"}
Here we present a renormalizable model with $A_{4}\otimes Z_{2}\otimes Z^{\prime}_{2}$ discrete flavor symmetry, which is consistent with the current neutrino data for the neutrino masses and mixings shown in Tables \[NH\],\[IH\] and which has less effective model parameters than other similar models, as discussed in section \[Numerical Analysis\]. We choose $A_4$ since it is the smallest symmetry with one three-dimensional and three distinct one-dimensional irreducible representations, where the three families of fermions can be accommodated rather naturally. Thereby we unify the left-handed leptons in the $A_4$ triplet representation and assign the right-handed leptons to $A_4$ singlets. This type of setup was proposed for the first time in Ref. [@Ma:2001dn]. In our model there is only one right-handed SM singlet Majorana neutrino $N_{R}$, and the scalar sector includes three $A_4$ triplets, one of which is a SM singlet while the other two are $SU(2)_L$ doublets. We further impose on the model a $Z_2$ discrete symmetry, in order to separate the two $A_{4}$ triplets transforming as $SU(2)_{L}$ doublets, so that one of them participates only in those Yukawa interactions which involve right-handed $SU(2)_{L}$ singlets $e_{R}, \mu_{R}, \tau_{R}$, while the other one participates only in those with the right-handed SM sterile neutrino $N_{R}$. Finally, a (spontaneously broken) $Z^{\prime}_2$ symmetry is introduced to forbid terms in the scalar potential with odd powers of the SM singlet scalar field $\chi$, the only one transforming non-trivially under $Z^{\prime}_2$. We assume that the $Z_2$ symmetry is not affected by the Electroweak Symmetry Breaking. Therefore the scalar fields coupled to the neutrinos have vanishing vacuum expectation values, which implies that the light neutrino masses are not generated at tree level via the usual seesaw mechanism, but instead are generated through loop corrections in a variant of the so-called radiative seesaw mechanism. The loops involve a heavy Majorana neutrino and neutral scalars, which in turn couple through quartic interactions with other neutral scalars in the external lines. The smallness of neutrino masses generated via a radiative seesaw mechanism is attributed to the smallness of the loop factor and to the quadratic dependence on the small neutrino Yukawa coupling. The scale of new physics can therefore be kept low, with the heavy Majorana neutrino mass of a few TeV. The radiative seesaw mechanism has been discussed in Refs. [@Ahn:2012tv; @Ahn:2013mva] in the context of a similar $A_4$ model, but with a field content quite different from ours: we introduce only one SM singlet Majorana neutrino instead of an $A_4$ triplet, with the lepton doublets as $A_{4}$ triplets, as in Ref. [@He:2006dk] and many other models, but not as in Ref. [@Ahn:2012tv], where they are assigned to $A_4$ singlets. Our scalar content is also distinct, with one additional $A_4$ triplet (and no $A_4$ singlets), which acquires a VEV in a different direction of the group space.
The paper is organized as follows. In section \[model\] we outline the proposed model. The results, in terms of neutrino masses and mixing, are presented in section \[massmix\]. This is followed by a numerical analysis in section \[Numerical Analysis\]. We conclude with discussions and a summary in \[Summary\]. Several technical details are presented in appendices: appendix \[A\] collects some necessary facts about the $A_4$ group, appendix \[B\] contains a discussion of the full $A_4$ invariant scalar potential, and appendix \[C\] deals with the mass spectrum for the physical scalars that enter in the radiative seesaw loops.
The Model {#model}
=========
Our model is a multi-Higgs doublet extension of the Standard Model (SM), with the full symmetry ${\cal G}$ experiencing a two-step spontaneous breaking $$\begin{aligned}
\label{Group}
&&{\cal G} = SU\left( 3\right) _{C}\otimes
SU\left( 2\right) _{L}\otimes U\left( 1\right) _{Y}\otimes A_{4}\otimes
Z_{2}\otimes Z^{\prime}_{2}\\
\nonumber\\[-3mm]
\nonumber
&& \hspace{35mm} \Downarrow \Lambda_{int}\\[3mm]
\nonumber
&& \hspace{15mm}SU\left( 3\right) _{C}\otimes
SU\left( 2\right) _{L}\otimes U\left(1\right) _{Y}\otimes Z_{2}\\[3mm]
\nonumber
&& \hspace{35mm} \Downarrow \Lambda_{EW}\\[3mm]
\nonumber
&& \hspace{23mm} SU\left( 3\right) _{C}\otimes U\left(1\right) _{em}\otimes Z_{2}\end{aligned}$$ We extend the fermion sector of the SM by introducing only one additional field, a SM singlet Majorana neutrino, $N_{R}$. The scalar sector is significantly enlarged and contains the six $SU(2)_{L}$ doublets $\Phi_{1,2,3}^{(1,2)}$ and three singlets $\chi_{1,2,3}$. We group them in triplets of $A_{4}$. The complete field content and its ${\cal G}$ assignments is given below: $$\begin{aligned}
\label{FieldContent-S}
&& \Phi^{(k=1,2)}: \left({\bf 1}, {\bf 2}, 1, {\bf 3}, (-1)^{k}, 1\right), \ \ \
\chi: \left({\bf 1}, {\bf 1}, 1, {\bf 3}, 1, -1\right),\\
\label{FieldContent-lL}
&&l_{L}:\hspace{8.5mm} \left({\bf 1}, {\bf 2}, -1, {\bf 3}, 1, 1\right),\\
\label{FieldContent-ER}
&& e_{R}: \hspace{8.5mm} \left({\bf 1}, \, {\bf 1}, -2, {\bf 1}, 1, 1\right), \hspace{6mm}
\mu_{R}: \left({\bf 1}, \, {\bf 1}, -2, {\bf 1^{\prime}}, 1, 1\right), \ \ \
\tau_{R}:\left({\bf 1}, {\bf 1}, -2, {\bf 1^{\prime\prime}}, 1, 1\right), \\
\label{FieldContent-NR}
&& N_{R}: \hspace{7.5mm} \left({\bf 1}, {\bf 1}, 0, {\bf 1}, -1, 1\right).
$$ Here the numbers in bold face are dimensions of representations of the corresponding group factor in Eq. (\[Group\]), the third number from the left is the weak hypercharge and the last two numbers are $Z_{2}$ and $Z'_{2}$ parities, respectively. The three families of the left-handed SM doublet leptons $l_{L}^{1,2,3}$ are unified in a single $A_{4}$ triplet $l_{L}$ while the right-handed SM singlet charged leptons $e_{R}, \mu_{R}, \tau_{R}$ are accommodated in the three distinct $A_{4}$ singlets ${\bf 1, 1' , 1''}$. The only right-handed SM singlet neutrino $N_{R}$ introduced in our model is assigned to ${\bf 1}$ of $A_{4}$ in order for its Majorana mass term be invariant under this symmetry. The presence of this term is crucial for our construction as explained below. Note that neither the ${\bf 1'}$ nor ${\bf 1''}$ singlet representations of $A_4$ satisfy this condition as can be seen from the multiplication rules in Eq. (\[A4-singlet-multiplication\]).
The two SM doublet $A_4$ triplet scalars $\Phi^{(k=1,2)}$ are distinguished by their $Z_2$ parities $(-1)^{k}$. We require that this $Z_{2}$ symmetry remains unbroken after the electroweak symmetry breaking. Therefore, $\Phi ^{\left(1\right) }$, which transforms non-trivially under $Z_2$, does not acquire a vacuum expectation value. The preserved $Z_2$ discrete symmetry also allows for stable dark matter candidates, as in [@Ma:2006km; @Ma:2006fn]. In our model they are either the lightest neutral component of $\Phi^{(1)}$ or the Majorana neutrino $N_{R}$. We do not address this question in the present paper. We introduce two SM doublet $A_4$ triplets, in order to ensure that one $A_4$ scalar triplet $\Phi^{(2)}$ gives masses to the charged leptons, while the other one $\Phi^{(1)}$, with vanishing VEV, couples to the SM singlet neutrino $N_{R}$. Thus neutrinos do not receive masses at tree level. The SM singlet $A_4$ triplet $\chi$ is introduced in order to generate a neutrino mass matrix texture compatible with the experimentally observed deviation from the TBM pattern. As we will explain in the following, the neutrino mass matrix texture generated via the one loop seesaw mechanism is mainly due to the VEV of the SM singlet $A_4$ triplet scalar $\langle \chi\rangle = \Lambda_{int}$, which is assumed to be much larger than the scale of the electroweak symmetry breaking $\Lambda_{int}\gg \Lambda_{EW}=246$ GeV. In this way, the contribution associated with the $(1,1,1)$ direction in $A_{4}$-space that shapes the charged lepton mass matrix is suppressed and effectively absent in the neutrino mass matrix, leading to a mixing matrix that is TBM to a good approximation. The $Z^{\prime}_{2}$ discrete symmetry is also an important ingredient of our approach, as will be shown below. Once it is imposed it forbids the terms in the scalar potential involving odd powers of the SM singlet $A_4$ triplet scalar $\chi$. This results in a reduction of the number of free model parameters and selects a particular direction of symmetry breaking in the group space. The $Z_{2}^{\prime}$ symmetry is broken after the $\chi$ field acquires a non vanishing vacuum expectation value.
With the field content of Eqs. (\[FieldContent-S\])-(\[FieldContent-NR\]), the Yukawa part of the model Lagrangian for the lepton sector takes the form $$\mathcal{L}_{Y}=y_{\nu }\left( \overline{l}_{L}\widetilde{\Phi }^{\left(
1\right) }\right) _{\mathbf{1}}N_{R}+M_{N}\overline{N}_{R}N_{R}^{c}+y_{e}
\left( \overline{l}_{L}\Phi ^{\left( 2\right) }\right) _{\mathbf{\bf 1}
}e_{R}+y_{\mu }\left( \overline{l}_{L}\Phi ^{\left( 2\right) }\right) _{
\mathbf{1}^{\prime \prime }}\mu _{R}+y_{\tau }\left( \overline{l}_{L}\Phi
^{\left( 2\right) }\right) _{\bf 1^{\prime }}\tau _{R}+h.c,
\label{LYlepton}$$with $\widetilde{\Phi }^{\left(k \right) }= i\sigma _{2}\left( \Phi ^{\left( k\right)
}\right) ^{\ast }$ ($k=1,2$). The subscripts ${\bf 1, 1', 1''}$ denote projecting out the corresponding $A_{4}$ singlet in the product of the two triplets.
Note that the assignment of the charged right-handed leptons (\[FieldContent-ER\]) to different $A_{4}$ singlets leads, as can be seen in Eq. (\[LYlepton\]), to different Yukawa couplings $y_{e,\mu,\tau}$ of the electrically neutral components of the $\Phi^{(2)0}$ to the different charged leptons $e, \mu, \tau$. The lightest of the $\Phi^{(2)0}$ should be interpreted as the SM-like 125 GeV Higgs observed at the LHC [@LHC-H-discovery], and the mentioned non-universality of its couplings to the charged leptons is in agreement with the recent ATLAS result [@ATLAS-CONF-2013-010], strongly disfavoring the case of coupling universality.
As can be seen from the Appendix \[C\], the masses of all the neutral scalar states from the $A_4$ triplets $\Phi^{(1)}$ and $\Phi^{(2)}$, except for the SM-like Higgs $\Phi^{(2)0}$, are proportional to $\langle \chi\rangle = \Lambda_{int} \gg \Lambda_{EW} = 246$ GeV and consequently are very heavy. Our model is not predictive in the scalar sector, having numerous free uncorrelated parameters in the scalar potential. We simply choose the scale $\Lambda_{int}$ such that the heavy scalars are pushed outside the LHC reach. The loop effects of the heavy scalars contributing to certain observables can be suppressed by the appropriate choice of the other free parameters. All these adjustments, as will be shown in Sec. \[Numerical Analysis\], do not affect the neutrino sector, which is totally controlled by [*three*]{} effective parameters, depending in turn on the scalar potential parameters and the lepton-Higgs Yukawa couplings.
The scalar fields $\Phi _{m}^{\left(k \right) }$ can be decomposed as: $$\label{decomp}
\Phi _{m}^{\left( k \right) }=\left(
\begin{array}{c}
\frac{1}{\sqrt{2}}\left( \omega _{m}^{\left( k \right) }+i\xi _{m}^{\left(k \right) }\right) \\
\frac{1}{\sqrt{2}}\left( v_{m}^{\left(k \right) }+\rho _{m}^{\left(k \right)
}+i\eta _{m}^{\left(k \right) }\right)
\end{array}\right) ,\hspace{1cm} k = 1,2,\hspace{1cm}m=1,2,3.$$ with $$\left\langle \rho _{m}^{\left(k \right) }\right\rangle =\left\langle \eta
_{m}^{\left(k \right) }\right\rangle =\left\langle \omega _{m}^{\left(
k\right) }\right\rangle =\left\langle \xi _{m}^{\left(k \right)
}\right\rangle =0,\hspace{2cm}\hspace{35mm} k = 1,2,\hspace{1cm}m=1,2,3.$$The Higgs doublets and the singlet fields can acquire vacuum expectation values: $$\left\langle \Phi _{m}^{\left( k\right) }\right\rangle =\left(
\begin{array}{c}
0 \\
\frac{v_{m}^{\left( k\right) }}{\sqrt{2}}\end{array}\right) ,\hspace{1cm}\left\langle \chi \right\rangle =\left( v_{\chi
_{1}},v_{\chi _{2}},v_{\chi _{3}}\right) ,\hspace{1cm}k=1,2,\hspace{1cm}m=1,2,3.$$Our requirement (see (\[Group\])) that $Z_{2}$ is preserved implies, according to the field assignment of (\[FieldContent-S\]), that $$\label{VEV1}
v_{m}^{\left( 1\right) }=0,\hspace{1cm}m=1,2,3.$$This can be achieved by having a positive squared mass term of $\Phi^{(1)}$ in the scalar potential. As a consequence of (\[LYlepton\]) and (\[VEV1\]) neutrinos do not acquire masses at tree level. As will be discussed in more detail in section \[massmix\], their masses are radiatively generated through loop diagrams involving virtual neutral scalars and the heavy Majorana neutrino in the internal lines. The aforementioned virtual scalars couple to real scalars due to the scalar quartic interactions, leading to the radiative seesaw mechanism of neutrino mass generation [@Ma:2006km; @Ma:2006fn].
We assume the following VEV pattern for the neutral components of the SM Higgs doublets $\Phi _{m}^{\left( 2\right) }$ ($m=1,2,3$) and for the components of the $A_{4}$ triplet SM singlet scalar $\chi$ : $$v_{1}^{\left( 2\right) }=v_{2}^{\left( 2\right) }=v_{3}^{\left( 2\right) }=
\frac{v}{\sqrt{3}},\hspace{2cm}\left\langle \chi \right\rangle =\frac{
v_{\chi }}{\sqrt{2}}\left( 1,0,-1\right) . \label{VEV}$$ Here $v=\Lambda_{EW}$ and $v_{\chi} = \Lambda_{int}$. This choice of directions in the $A_4$ space is justified by the observation that they describe a natural solution of the scalar potential minimization equations. Indeed, in the single-field case, $A_{4}$ invariance readily favors the $(1,1,1)$ direction over e.g. the $(1,0,0)$ solution for large regions of parameter space. The vacuum $\left\langle \Phi ^{\left( 2\right) }\right\rangle $ is a configuration that preserves a $Z_{3}$ subgroup of $A_{4}$, which has been extensively studied by many authors (see for example Refs. [@Altarelli:2005yp; @Altarelli:2005yx; @He:2006dk; @Toorop:2010ex; @Ahn:2012tv; @Mohapatra:2012tb; @Chen:2012st]). In our model we have more fields, but there are also classes of the $A_{4}$ invariants favoring respective VEVs of two fields in orthogonal directions, as desired for our analysis. Therefore our assumption is essentially that the quartic couplings in the potential involving $\chi $ and $\Phi ^{(2)}$ are within the range of the parameter space where these directions are the global minimum. More details are presented in Appendix \[B\], where the minimization conditions of the full scalar potential of our model are considered, showing that the $\left\langle \chi \right\rangle $ vacuum (\[VEV\]), together with the $\left\langle \Phi ^{(2)}\right\rangle
$ vacuum (\[VEV\]), are consistent.
As follows from Eqs. (\[LYlepton\]) and (\[decomp\]), the neutrino Yukawa interactions are described by the following Lagrangian: $$\mathcal{L}_{\nu \overline{\nu }S}=\frac{y_{\nu }}{\sqrt{2}}\left[ \overline{\nu }_{1L}\left( \rho _{1}^{\left( 1\right) }+i\eta _{1}^{\left( 1\right)
}\right) N_{R}+\overline{\nu }_{2L}\left( \rho _{2}^{\left( 1\right) }+i\eta
_{2}^{\left( 1\right) }\right) N_{R}+\overline{\nu }_{3L}\left( \rho
_{3}^{\left( 1\right) }+i\eta _{3}^{\left( 1\right) }\right) N_{R}\right]
+h.c.$$We consider the scenario where $v_{\chi }\gg v$. A moderate hierarchy in the VEVs is quite natural, given that $\chi$ is a SM gauge singlet and its VEV does not have to be related to the electroweak scale. The scale of $v_{\chi }$ is ultimately controlled by the $\chi $ squared mass term in the potential. From Eq. (\[LYlepton\]) and Eq. (\[Physicalscalars\]) it follows (for details see Appendix \[C\]) that the neutrino Yukawa interactions, in terms of the physical scalar fields, can be approximately written as: $$\begin{aligned}
\mathcal{L}_{\nu \overline{\nu }S} &=&\frac{y_{\nu }e^{i\psi}}{2}\overline{\nu }
_{1L}\left[\left( H_{1}^{0}-A_{3}^{0}\right)+i\left( H_{3}^{0}+A_{1}^{0}\right)\right] N_{R}+\frac{y_{\nu }}{\sqrt{2}}\overline{\nu }_{2L}\left(
H_{2}^{0}+iA_{2}^{0}\right) N_{R}\notag\\
&&+\frac{y_{\nu }e^{-i\psi}}{2}\overline{\nu }_{3L}\left[\left( H_{1}^{0}-A_{3}^{0}\right)+i\left( H_{3}^{0}+A_{1}^{0}\right)\right] N_{R}
+h.c. \label{Yukawaneutrinos}\end{aligned}$$ When the subleading effects are considered, there is some mixing between the scalar states, so that $H_{2}^0$, $A_{2}^{0}$ will appear in the Yukawa couplings of $\overline{\nu }_{1L}$, $\overline{\nu }_{3L}$, and the other scalars will also appear in the Yukawa couplings to $\overline{\nu }_{2L}$. As described in more detail in Appendix \[C\], the parameter $\psi $ is given by: $$\tan 2\psi \simeq \frac{1}{\sqrt{\frac{9}{4}\left( \frac{M_{A_{3}^{0}}^{2}-M_{A_{1}^{0}}^{2}}{M_{A_{3}^{0}}^{2}+M_{A_{1}^{0}}^{2}-2M_{A_{2}^{0}}^{2}}\right) ^{2}-1}}\,,
\label{tan2psi}$$in terms of the masses $M_{A_{m}^{0}}$ ($m=1,2,3$) of the neutral CP-odd scalar fields.
Lepton masses and mixing \[massmix\]
====================================
From Eq. (\[LYlepton\]), and by using the product rules for the $A_{4}$ group given in Appendix \[A\], it follows that the charged lepton mass matrix is given by $$M_{l}=V_{lL}^{\dag }diag\left( m_{e},m_{\mu },m_{\tau }\right) ,\hspace{2cm}
V_{lL}=\frac{1}{\sqrt{3}}\left(
\begin{array}{ccc}
1 & 1 & 1 \\
1 & \omega & \omega ^{2} \\
1 & \omega ^{2} & \omega
\end{array}
\right) ,\hspace{2cm}\omega =e^{\frac{2\pi i}{3}}.
\label{Ml}$$The neutrino mass term does not appear at tree-level due to vanishing v.e.v. of $\Phi^{(1)}$ field (\[VEV1\]). It arises in the form of a Majorana mass term $$\begin{aligned}
\label{Nu-Mass-Term}
&&-\frac{1}{2} \bar{\nu} M_{\nu} \nu^{C} + \mbox{h.c.}\end{aligned}$$ from radiative corrections at 1-loop level. The leading 1-loop contributions to the complex symmetric Majorana neutrino mass matrix $M_{\nu}$ are derived from Eqs. (\[Yukawaneutrinos\]) and (\[L2\]). The corresponding diagrams are shown in Fig. \[figMu\].
![One loop Feynman diagrams contributing to the entries of the neutrino mass matrix.[]{data-label="figMu"}](Diag-2.pdf)
In the approximation discussed in Appendix C we obtain $$M_{\nu }\simeq \left(
\begin{array}{ccc}
Ae^{2i\psi } & 0 & A \\
0 & B & 0 \\
A & 0 & Ae^{-2i\psi }\end{array}\right) .
\label{Mnu}$$ where: $$\begin{aligned}
\label{A-param}
A &\simeq &\frac{y_{\nu }^{2}}{16\pi ^{2}M_{N}}\left\{ \left(
M_{A_{1}^{0}}^{2}-M_{A_{2}^{0}}^{2}+\frac{\varepsilon v_{\chi }^{2}}{2}\right) \left[ D_{0}\left( \frac{M_{H_{1}^{0}}}{M_{N}}\right) -D_{0}\left(
\frac{M_{A_{1}^{0}}}{M_{N}}\right) \right] \right. \notag \\
&&\hspace{14mm}+\left. \left( M_{A_{3}^{0}}^{2}-M_{A_{2}^{0}}^{2}+\frac{\varepsilon
v_{\chi }^{2}}{2}\right) \left[ D_{0}\left( \frac{M_{A_{3}^{0}}}{M_{N}}\right) -D_{0}\left( \frac{M_{H_{3}^{0}}}{M_{N}}\right) \right] \right\} \\[3mm],
\label{B-param}
B&\simeq& \frac{\varepsilon y_{\nu }^{2}v_{\chi }^{2}}{16\pi ^{2}M_{N}}\left[
D_{0}\left( \frac{M_{H_{2}^{0}}}{M_{N}}\right) -D_{0}\left( \frac{M_{A_{2}^{0}}}{M_{N}}\right) \right].\end{aligned}$$ Here $\varepsilon $ is a dimensionless parameter, which takes into account the difference between a pair of quartic couplings of the scalar potential (see Appendix \[C\] for details). We introduced the function [@Hernandez:2013mcf]: $$D_{0}(x) =\frac{-1+x^{2}-\ln x^{2}}{\left( 1-x^{2}\right) ^{2}}.$$ Since $M_{\nu }$ depends only on the square of the VEVs, even a moderate hierarchy in the VEVs significantly suppresses contributions related to $\Phi ^{\left( 2\right) }$. Furthermore, because $\langle \chi _{2}\rangle =0$, $\left\vert \left( M_{\nu }\right) _{12}\right\vert $ and $\left\vert \left(
M_{\nu }\right) _{23}\right\vert $ are only generated through $\Phi ^{\left(
2\right) }$ and are then much smaller than $\left\vert \left( M_{\nu
}\right) _{13}\right\vert $ and $\left\vert \left( M_{\nu }\right)
_{mm}\right\vert $ ($m=1,2,3$). Consequently the zero entries in Eq. (\[Mnu\]) become $$\begin{aligned}
\label{Zero-entries}
\left( M_{\nu }\right) _{12}&\sim& \left( M_{\nu }\right) _{23}\sim \frac{v^{2}}{v_{\chi }^{2}}\left( M_{\nu }\right) _{13}\end{aligned}$$ and are strongly suppressed in comparison to the other entries if $v_{\chi}\gg v$, as assumed in our model. Note that a similar neutrino mass matrix texture was obtained in Ref.[@Memenga:2013vc] from higher dimensional operators.
The neutrino mass matrix given in Eq. (\[Mnu\]) depends effectively only on three parameters: $A$, $B$ and $\psi$. As seen from Eqs. (\[A-param\]), (\[B-param\]), the parameters $A$ and $B$ contain the dependence on various model parameters. It is relevant that $A$ and $B$ are loop suppressed and are approximately inverse proportional to $M_N$. As seen from Eqs. (\[A-param\]) and (\[B-param\]), a non-vanishing mass splitting between the CP even $H^{0}_{i}$ and CP odd $A^{0}_{i}$ neutral scalars is crucial. Its absence would lead to massless neutrinos at one loop level. Note also that universality in the quartic couplings of the scalar potential, which would correspond to $\varepsilon=0$, would imply $B\sim 0$ and lead to only one massive neutrino. For simplicity, we parametrize the non-universality of the relevant couplings through the parameter $\varepsilon$, defined in Eq. (\[univ-eps\]). As will be shown below, the Pontecorvo-Maki-Nakagawa-Sakata (PMNS) mixing matrix depends only on the parameter $\psi$, while the neutrino mass squared splittings are controlled by parameters $A$ and $B$.
A complex symmetric Majorana mass matrix $M_{\nu}$, as in Eq. (\[Nu-Mass-Term\]), can be diagonalized by a unitary rotation of the neutrino fields so that $$\begin{aligned}
\label{diagonalisation}
&& \nu' = V_{\nu}\cdot \nu \hspace{3mm} \longrightarrow \hspace{3mm} V^{\dagger}_{\nu} M_{\nu}(V^{\dagger}_{\nu})^T = \mbox{Diag}\{m_{\nu_{1}}, m_{\nu_{2}}, m_{\nu_{3}} \}\ \ \ \ \mbox{with} \ \ \ \ V_{\nu} V^{\dagger}_{\nu} = {\bf 1},\end{aligned}$$ where $m_{1,2,3}$ are real and positive. The rotation matrix has the form $$V_{\nu } = \left(
\begin{array}{ccc}
\cos \theta & 0 & \sin \theta e^{-i\phi } \\
0 & 1 & 0 \\
-\sin \theta e^{i\phi } & 0 & \cos \theta
\end{array}
\right)P_{\nu},\hspace{0.5cm}\mbox{with}\hspace{0.5cm}P_{\nu}=diag\left(e^{i\alpha_1/2},e^{i\alpha_2/2},e^{i\alpha_3/2}\right),\hspace{0.5cm} \theta =\pm \frac{\pi }{4},\hspace{1cm}\phi =-2\psi .
\label{tanphi}$$ We identify the Majorana neutrino masses and Majorana phases for the two possible solutions with $\theta = \pi/4, -\pi/4$ with the normal (NH) and inverted (IH) mass hierarchies, respectively. They are $$\begin{aligned}
\label{mass-spectrum-Normal}
\mbox{NH}&:& \theta = + \frac{\pi}{4}: \hspace{10mm} m_{\nu_{1}} = 0, \hspace{10mm} m_{\nu_{2}} = B, \hspace{10mm} m_{\nu_{3}} = 2 A, \hspace{10mm}\alpha_1=\alpha_2=0,\hspace{10mm}\alpha_3=\phi, \\[3mm]
\label{mass-spectrum-Inverted}
\mbox{IH}&:& \theta = - \frac{\pi}{4}: \hspace{10mm} m_{\nu_{1}} = 2 A, \hspace{8mm} m_{\nu_{2}} = B, \hspace{10mm} m_{\nu_{3}} = 0, \hspace{12.5mm}\alpha_2=\alpha_3=0,\hspace{10mm}\alpha_1=-\phi.
$$
Note that the nonvanishing Majorana phases are $\phi$ and $-\phi$ for normal and inverted mass hierarchies, respectively.
With the rotation matrices in the charged lepton sector $V_{lL}$, given in Eq. (\[Ml\]), and in the neutrino sector $V_{\nu}$ , given in Eq. (\[tanphi\]), we find the PMNS mixing matrix: $$U=V_{lL}^{\dag }V_{\nu }\simeq \left(
\begin{array}{ccc}
\frac{\cos \theta }{\sqrt{3}}-\frac{e^{i\phi }\sin \theta }{\sqrt{3}} &
\frac{1}{\sqrt{3}} & \frac{\cos \theta }{\sqrt{3}}+\frac{e^{-i\phi }\sin
\theta }{\sqrt{3}} \\
& & \\
\frac{\cos \theta }{\sqrt{3}}-\frac{e^{i\phi +\frac{2i\pi }{3}}\sin \theta }{\sqrt{3}} & \frac{e^{-\frac{2i\pi }{3}}}{\sqrt{3}} & \frac{e^{\frac{2i\pi }{3}}\cos \theta }{\sqrt{3}}+\frac{e^{-i\phi }\sin \theta }{\sqrt{3}} \\
& & \\
\frac{\cos \theta }{\sqrt{3}}-\frac{e^{i\phi -\frac{2i\pi }{3}}\sin \theta }{\sqrt{3}} & \frac{e^{\frac{2i\pi }{3}}}{\sqrt{3}} & \frac{e^{-\frac{2i\pi }{3}}\cos \theta }{\sqrt{3}}+\frac{e^{-i\phi }\sin \theta }{\sqrt{3}}\end{array}\right)P_{\nu}. \label{PMNS}$$From the standard parametrization of the leptonic mixing matrix, it follows that the lepton mixing angles are [@PDG]: $$\begin{aligned}
\label{theta-ij}
&&\sin ^{2}\theta _{12}=\frac{\left\vert U_{e2}\right\vert ^{2}}{1-\left\vert
U_{e3}\right\vert ^{2}} = \frac{1}{2\mp \cos\phi}, \hspace{20mm}
\sin ^{2}\theta _{13}=\left\vert
U_{e3}\right\vert ^{2} = \frac{1}{3}(1\pm \cos\phi), \\[3mm]
\nonumber
&&\sin ^{2}\theta _{23}=\frac{\left\vert
U_{\mu 3}\right\vert ^{2}}{1-\left\vert U_{e3}\right\vert ^{2}} = \frac{2 \mp (\cos\phi + \sqrt{3} \sin\phi)}{4 \mp 2\cos\phi},\end{aligned}$$ where the upper sign corresponds to normal ($\theta = +\pi/4$) and the lower one to inverted ($\theta = -\pi/4$) hierarchy, respectively. The PMNS matrix (\[PMNS\]) of our model reproduces the magnitudes of the corresponding matrix elements of the TBM ansatz (\[TBM-ansatz\]) in the limit $\phi =0$ and $\phi=\pi$ for the inverted and the normal hierarchy, respectively. In both cases the special value for $\phi$ implies that the physical neutral scalars originating from $\Phi^{(1)}$ are degenerate in mass. Notice that the lepton mixing angles are controlled by the Majorana phases $\pm\phi$, where the plus and minus signs correspond to normal and inverted mass hierarchy, respectively.
The Jarlskog invariant and the CP violating phase are given by [@PDG]: $$J=\func{Im}\left( U_{e1}U_{\mu 2}U_{e2}^{\ast }U_{\mu 1}^{\ast }\right)\simeq-\frac{1}{6\sqrt{3}}\cos 2\theta ,\hspace{2cm}\sin \delta =\frac{8J}{\cos
\theta _{13}\sin 2\theta _{12}\sin 2\theta _{23}\sin 2\theta _{13}}.$$Since $\theta =\pm \frac{\pi }{4}$, we predict $J\simeq 0$ and $\delta\simeq 0$ for $v_{\chi}\gg v$, implying that in our model CP violation is suppressed in neutrino oscillations.
Phenomenological implications {#Numerical Analysis}
=============================
In the following we adjust the free parameters of our model to reproduce the experimental values given in and discuss some implications of this choice of the parameters.
As seen from Eqs. (\[mass-spectrum-Normal\]), (\[mass-spectrum-Inverted\]) and (\[PMNS\]), (\[theta-ij\]) we have only [*three*]{} effective free parameters to fit: $\phi$, $A$ and $B$. It is noteworthy that in our model a single parameter ($\phi$) determines all three neutrino mixing parameters $\sin ^{2}\theta _{13}$, $\sin ^{2}\theta _{12}$ and $\sin ^{2}\theta _{23}$ as well as the Majorana phases $\alpha_{i}$. The parameters $A$ and $B$ control the two mass squared splittings $\Delta m^{2}_{ij}$. Therefore we actually fit only $\phi$ to adjust the values of $\sin^{2}\theta _{ij}$, while $A$ and $B$ for the NH and the IH hierarchies are simply $$\begin{aligned}
\label{AB-Delta-NH}
&&\mbox{NH}:\ m_{\nu_{1}} = 0, \ \ \ m_{\nu_{2}} = B = \sqrt{\Delta m^{2}_{21}} \approx 9 \mbox{meV}, \ \ \ m_{\nu_{3}} = 2 A = \sqrt{\Delta m^{2}_{31}} \approx 51 \mbox{meV};\\[3mm]
\label{AB-Delta-IH}
&&\mbox{IH}\hspace{2mm} :\ m_{\nu_{2}} = B = \sqrt{\Delta m^{2}_{21} +\Delta m^{2}_{13}} \approx 50 \mbox{meV}, \ \ \ \ \ m_{\nu_{1}} = 2 A = \sqrt{\Delta m^{2}_{13}}
\approx 49 \mbox{meV}, \ \ \ m_{\nu_{3}} = 0,
$$ as follows from Eqs. (\[mass-spectrum-Normal\]), (\[mass-spectrum-Inverted\]) and the definition $\Delta m^{2}_{ij} = m^{2}_{i}-m^{2}_{j}$. In Eqs. (\[AB-Delta-NH\]), (\[AB-Delta-IH\]) we assumed the best fit values of $\Delta m^{2}_{ij}$ from the Tables \[NH\], \[IH\].
Varying the model parameter $\phi$ in Eq. (\[theta-ij\]) we have fitted the $\sin ^{2}\theta _{ij}$ to the experimental values in Tables \[NH\], \[IH\]. The best fit result is: $$\begin{aligned}
\label{parameter-fit-NH}
&&\mbox{NH}\ :\ \phi = - 0.877\, \pi, \ \ \ \sin^{2}\theta _{12} \approx 0.34 , \ \ \ \sin^{2}\theta _{23} \approx 0.61 , \ \ \
\sin^{2}\theta _{13} \approx 0.0246; \\[3mm]
\label{parameter-fit-IH}
&&\mbox{IH}\hspace{2.5mm} :\ \phi=\ \ 0.12\, \pi, \ \ \ \ \ \sin^{2}\theta _{12} \approx 0.34 , \ \ \ \sin^{2}\theta _{23} \approx 0.6, \ \ \ \ \,
\sin^{2}\theta _{13} \approx 0.025 .\end{aligned}$$
Comparing Eqs. (\[parameter-fit-NH\]), (\[parameter-fit-IH\]) with Tables \[NH\], \[IH\] we see that $\sin^{2}\theta _{13}$ and $\sin ^{2}\theta _{23}$ are in excellent agreement with the experimental data, for both NH and IH, with $\sin^{2}\theta _{12}$ within a $2\sigma $ deviation from its best fit values.
The effective parameters $A$, $B$ and $\tan\phi$ depend on various model parameters: the SM singlet neutrino Majorana mass $M_{N}$, the quartic and bilinear couplings of the model Lagrangian (\[LYlepton\]), (\[V\]), as well as on the scale of $A_{4}$ symmetry breaking $v_{\chi}$. It is worth checking that the solution in Eqs. (\[AB-Delta-NH\])-(\[parameter-fit-IH\]) does imply neither fine-tuning or very large values of dimensionful parameters. For this purpose consider a point in the model parameter space with all the relevant dimensionless quartic couplings in Eqs. (\[V\]), (\[univ-eps\]) given by $$\begin{aligned}
\label{quartic}
\lambda = \tau_{i} = \lambda_{b1} = \alpha_{1} = \lambda_{a1} - \varepsilon \sim 1,\end{aligned}$$ compatible with the perturbative regime ($\lambda/4\pi < 1$). Absence of fine-tuning in this sector favors $\varepsilon \sim 1$. Using Eqs. (\[MH10\])-(\[abLB\]) one may derive an order of magnitude estimate $$\begin{aligned}
\label{estimate-AB}
A\sim B \sim \left(\frac{y_{\nu}}{4\pi}\right)^{2} z(\lambda, \epsilon)\ \frac{v^2_{\chi}}{M_{N}},\end{aligned}$$ where the function $z(\lambda, \varepsilon) \sim 1$ for the values chosen in Eq. (\[quartic\]). In this estimation we assumed $\mu_{1}\leq v_{\chi}$ and $v_{\chi}\gg v$. Let us also assume that the neutrino and electron Yukawa couplings in Eq. (\[LYlepton\]) are comparable $y_{\nu} \sim y_{e}$. From Eqs. (\[LYlepton\]), (\[decomp\]), (\[VEV\]) and the value of the electron mass we estimate $$\begin{aligned}
\label{estimate-Yukawa}
y_{\nu} \sim y_{e} \sim 10^{-6}.
$$ Then from Eqs. (\[AB-Delta-NH\]), (\[AB-Delta-IH\]) and (\[estimate-AB\]) we roughly estimate $$\begin{aligned}
\label{Fin-estimate}
m_{\nu} \sim A\sim B \sim \frac{v_{\chi}}{M_{N}} \cdot \mbox{meV}.
$$ Therefore for any value $M_{N} \sim v_{\chi} \gg v \sim 250$ GeV and without special tuning of the model parameters we are in the ballpark of the neutrino mass squared splittings measured in neutrino oscillation experiments (Tables \[NH\], \[IH\]). Both the scale of new physics $v_{\chi}$, related to the $A_{4}$ symmetry, and the SM singlet Majorana neutrino mass $M_{N}$ could be comparatively low, around a few TeV.
With the values of the model parameters given in Eqs. (\[AB-Delta-NH\])-(\[parameter-fit-IH\]), derived from the oscillation experiments, we can predict the amplitude for neutrinoless double beta ($0\nu\beta\beta$) decay, which is proportional to the effective Majorana neutrino mass $$m_{\beta\beta}=\sum_jU^2_{ek}m_{\nu_k},
\label{mee}$$ where $U^2_{ej}$ and $m_{\nu_k}$ are the PMNS mixing matrix elements and the Majorana neutrino masses, respectively.
Then, from Eqs. (\[tanphi\])-(\[PMNS\]) and (\[AB-Delta-NH\])-(\[parameter-fit-IH\]), we predict the following effective neutrino mass for both hierarchies: $$\begin{aligned}
\label{eff-mass-pred}
m_{\beta\beta}=\frac1{3}\left(B+4A\cos^2\frac{\phi}{2}\right)=
\left\{ \begin{array}{l}
4 \ \mbox{meV}\ \ \ \ \ \ \ \mbox{for \ \ \ \ NH}\\
50 \ \mbox{meV}\ \ \ \ \ \ \ \mbox{for \ \ \ \ IH} \\
\end{array} \right.\end{aligned}$$ This is beyond the reach of the present and forthcoming neutrinoless double beta decay experiments. The presently best upper limit on this parameter $m_{\beta\beta}\leq 160$ meV comes from the recently quoted EXO-200 experiment [@Auger:2012ar] $T^{0\nu\beta\beta}_{1/2}(^{136}{\rm Xe}) \ge 1.6 \times 10^{25}$ yr at the 90 % CL. This limit will be improved within the not too distant future. The GERDA experiment [@Abt:2004yk; @Ackermann:2012xja] is currently moving to “phase-II”, at the end of which it is expected to reach , corresponding to $m_{\beta\beta}\leq 100$ MeV. A bolometric CUORE experiment, using ${}^{130}Te$ [@Alessandria:2011rc], is currently under construction. Its estimated sensitivity is around $T^{0\nu\beta\beta}_{1/2}(^{130}{\rm Te})\sim 10^{26}$ yr corresponding to There are also proposals for ton-scale next-to-next generation $0\nu\beta\beta$ experiments with $^{136}$Xe [@KamLANDZen:2012aa; @Auty:2013:zz] and $^{76}$Ge [@Abt:2004yk; @Guiseppe:2011me] claiming sensitivities over $T^{0\nu\beta\beta}_{1/2} \sim 10^{27}$ yr, corresponding to $m_{\beta\beta}\sim 12-30$ meV. For recent experimental reviews, see for example Ref. [@Barabash:1209.4241] and references therein. Thus, according to Eq. (\[eff-mass-pred\]) our model predicts $T^{0\nu\beta\beta}_{1/2}$ at the level of sensitivities of the next generation or next-to-next generation $0\nu\beta\beta$ experiments.
Conclusions {#Summary}
===========
We have presented a simple renormalizable model that successfully accounts for the charged lepton and neutrino masses and mixings. The neutrino masses arise from a radiative seesaw mechanism, which explains their smallness, while keeping the scale of new physics $\Lambda_{int}$ at the comparatively low values, which could be about a few TeV (for the single SM singlet neutrino $N_{R}$). The neutrino mixing is approximately tribimaximal due to the spontaneously broken $A_4$ symmetry of the model. The experimentally observed deviation from the TBM pattern is implemented by introducing the SM singlet $A_{4}$ triplet $\chi$. Its VEV $\langle\chi\rangle = \Lambda_{int} \gg \Lambda_{EW}$ breaks $A_{4}$ symmetry and properly shapes the neutrino mass matrix at 1-loop level. CP violation in neutrino oscillations is suppressed.
The model has only 3 effective free parameters in the neutrino sector, which, nevertheless, allowed us to reproduce with good accuracy the mass squared splittings and all mixing angles measured in neutrino oscillation experiments for both normal and inverted neutrino spectrum.
The model predicts the effective Majorana neutrino mass $m_{\beta\beta}$ for neutrinoless double beta decay to be 4 meV and 50 meV for the normal and the inverted neutrino spectrum, respectively. The lightest neutral scalar of our model, $\Phi^{(2)0}$, interpreted as the SM-like 125 GeV Higgs boson observed at the LHC, has non-universal Yukawa couplings to the charged leptons $e, \mu, \tau$. This is in agreement with the recent ATLAS result [@ATLAS-CONF-2013-010], strongly disfavoring the case of Yukawa coupling universality.
An unbroken $Z_2$ discrete symmetry of our model also allows for stable dark matter candidates, as in Refs. [@Ma:2006km; @Ma:2006fn]. The candidate could be either the lightest neutral component of $\Phi^{(1)}$ or the right-handed Majorana neutrino $N_{R}$. We do not address this possibility further in the present paper.
Acknowledgments {#acknowledgments .unnumbered}
===============
This work was partially supported by Fondecyt (Chile) under grants 1100582 and 1100287. IdMV was supported by DFG grant PA 803/6-1 and by the Swiss National Science Foundation. HP was supported by DFG grant PA 803/6-1. AECH thanks Dortmund University for hospitality where part of this work was done. The visit of AECH to Dortmund University was supported by Dortmund University and DFG-CONICYT grant PA-803/7-1.
The product rules for $A_4$ \[A\]
=================================
The following product rules for the $A_{4}$ group were used in the construction of our model Lagrangian: $$\begin{aligned}
\label{prod-rule-1}
&& \hspace{18mm }\mathbf{3}\otimes \mathbf{3}=\mathbf{3}_{s}\oplus \mathbf{3}_{a}\oplus
\mathbf{1}\oplus \mathbf{1}^{\prime }\oplus \mathbf{1}^{\prime \prime },\\[3mm]
\label{A4-singlet-multiplication}
&&\mathbf{1}\otimes \mathbf{1}=\mathbf{1},\hspace{5mm}\mathbf{1}^{\prime}\otimes \mathbf{1}^{\prime \prime }=\mathbf{1},\hspace{5mm}
\mathbf{1}^{\prime }\otimes \mathbf{1}^{\prime }=\mathbf{1}^{\prime \prime },
\hspace{5mm}\mathbf{1}^{\prime \prime }\otimes \mathbf{1}^{\prime \prime }=\mathbf{1}^{\prime },\end{aligned}$$ Denoting $\left( x_{1},y_{1},z_{1}\right) $ and $\left(x_{2},y_{2},z_{2}\right) $ as the basis vectors for two $A_{4}$-triplets $\mathbf{3}$, one finds:
$$\begin{aligned}
\label{triplet-vectors}
&&\left( \mathbf{3}\otimes \mathbf{3}\right) _{\mathbf{1}}=x_{1}y_{1}+x_{2}y_{2}+x_{3}y_{3},\\
&&\left( \mathbf{3}\otimes \mathbf{3}\right) _{\mathbf{3}_{s}}=\left(
x_{2}y_{3}+x_{3}y_{2},x_{3}y_{1}+x_{1}y_{3},x_{1}y_{2}+x_{2}y_{1}\right) ,
\ \ \ \
\left( \mathbf{3}\otimes \mathbf{3}\right) _{\mathbf{1}^{\prime}}=x_{1}y_{1}+\omega x_{2}y_{2}+\omega ^{2}x_{3}y_{3},\\
&&\left( \mathbf{3}\otimes \mathbf{3}\right) _{\mathbf{3}_{a}}=\left(x_{2}y_{3}-x_{3}y_{2},x_{3}y_{1}-x_{1}y_{3},x_{1}y_{2}-x_{2}y_{1}\right),
\ \ \ \left( \mathbf{3}\otimes \mathbf{3}\right) _{\mathbf{1}^{\prime\prime }}=x_{1}y_{1}+\omega ^{2}x_{2}y_{2}+\omega x_{3}y_{3},
$$
where $\omega =e^{i \frac{ 2 \pi }{3}}$. The representation $\mathbf{1}$ is trivial, while the non-trivial $\mathbf{1}^{\prime }$ and $\mathbf{1}^{\prime \prime }$ are complex conjugate to each other. Comprehensive reviews of discrete symmetries in particle physics can be found in Refs. [@King:2013eh; @Altarelli:2010gt; @Ishimori:2010au; @Discret-Group-Review].
Scalar Potential \[B\]
======================
The scalar potential of the model is constructed of the three $A_{4}$ triplet fields $\Phi^{(1,2)}$ and $\chi$ in the way invariant under the group ${\cal G}$ in Eq. (\[Group\]).
For convenience we separate its terms into the three different groups as
$$V=V\left( \Phi ^{\left( 1\right) },\Phi ^{\left( 2\right) }\right) +V\left(
\Phi ^{\left( 1\right) },\Phi ^{\left( 2\right) },\chi \right) +V\left( \chi
\right) ,
\label{V}$$
where $$\begin{aligned}
V\left( \Phi ^{\left( 1\right) },\Phi ^{\left( 2\right) }\right)
&=&\dsum\limits_{l=1}^{2}\left[ \mu _{l}^{2}\left( \left( \Phi ^{\left(
l\right) }\right) ^{\dagger }\Phi ^{\left( l\right) }\right) _{\mathbf{1}}+\kappa _{l}\left( \left( \Phi ^{\left( l\right) }\right) ^{\dagger }\Phi
^{\left( l\right) }\right) _{\mathbf{1}}\left( \left( \Phi ^{\left( l\right)
}\right) ^{\dagger }\Phi ^{\left( l\right) }\right) _{\mathbf{1}}+\sigma
_{l}\left( \left( \Phi ^{\left( l\right) }\right) ^{\dagger }\Phi ^{\left(
l\right) }\right) _{\mathbf{1}^{\prime }}\left( \left( \Phi ^{\left(
l\right) }\right) ^{\dagger }\Phi ^{\left( l\right) }\right) _{\mathbf{1}^{\prime \prime }}\right. \notag \\
&&+\left. \gamma _{l}\left( \left( \Phi ^{\left( l\right) }\right) ^{\dagger
}\Phi ^{\left( l\right) }\right) _{\mathbf{3s}}\left( \left( \Phi ^{\left(
l\right) }\right) ^{\dagger }\Phi ^{\left( l\right) }\right) _{\mathbf{3s}}+\delta _{l}\left( \left( \Phi ^{\left( l\right) }\right) ^{\dagger }\Phi
^{\left( l\right) }\right) _{\mathbf{3a}}\left( \left( \Phi ^{\left(
l\right) }\right) ^{\dagger }\Phi ^{\left( l\right) }\right) _{\mathbf{3a}}\right] \notag \\
&&+\zeta _{1}\left( \left( \Phi ^{\left( 1\right) }\right) ^{\dagger }\Phi
^{\left( 1\right) }\right) _{\mathbf{3s}}\left( \left( \Phi ^{\left(
2\right) }\right) ^{\dagger }\Phi ^{\left( 2\right) }\right) _{\mathbf{3s}}+\zeta _{2}\left( \left( \Phi ^{\left( 1\right) }\right) ^{\dagger }\Phi
^{\left( 1\right) }\right) _{\mathbf{3a}}\left( \left( \Phi ^{\left(
2\right) }\right) ^{\dagger }\Phi ^{\left( 2\right) }\right) _{\mathbf{3a}}
\notag \\
&&+\zeta _{3}\left[ \left( \left( \Phi ^{\left( 1\right) }\right) ^{\dagger
}\Phi ^{\left( 1\right) }\right) _{\mathbf{1}^{\prime }}\left( \left( \Phi
^{\left( 2\right) }\right) ^{\dagger }\Phi ^{\left( 2\right) }\right) _{\mathbf{1}^{\prime \prime }}+\left( \left( \Phi ^{\left( 1\right) }\right)
^{\dagger }\Phi ^{\left( 1\right) }\right) _{\mathbf{1}^{\prime \prime
}}\left( \left( \Phi ^{\left( 2\right) }\right) ^{\dagger }\Phi ^{\left(
2\right) }\right) _{\mathbf{1}^{\prime }}\right] \notag \\
&&+\zeta _{4}\left( \left( \Phi ^{\left( 1\right) }\right) ^{\dagger }\Phi
^{\left( 1\right) }\right) _{\mathbf{1}}\left( \left( \Phi ^{\left( 2\right)
}\right) ^{\dagger }\Phi ^{\left( 2\right) }\right) _{\mathbf{1}}+\left[
\tau _{1}\left( \left( \Phi ^{\left( 1\right) }\right) ^{\dagger }\Phi
^{\left( 2\right) }\right) _{\mathbf{3s}}\left( \left( \Phi ^{\left(
2\right) }\right) ^{\dagger }\Phi ^{\left( 1\right) }\right) _{\mathbf{3s}}+h.c\right] \notag \\
&&+\left[ \tau _{2}\left( \left( \Phi ^{\left( 1\right) }\right) ^{\dagger
}\Phi ^{\left( 2\right) }\right) _{\mathbf{3a}}\left( \left( \Phi ^{\left(
2\right) }\right) ^{\dagger }\Phi ^{\left( 1\right) }\right) _{\mathbf{3a}}+\tau _{3}\left( \left( \Phi ^{\left( 1\right) }\right) ^{\dagger
}\Phi ^{\left( 2\right) }\right) _{\mathbf{3a}}\left( \left( \Phi ^{\left(
2\right) }\right) ^{\dagger }\Phi ^{\left( 1\right) }\right) _{\mathbf{3s}}+h.c\right] \notag \\
&&+\tau _{4}\left( \left( \Phi ^{\left( 1\right) }\right)
^{\dagger }\Phi ^{\left( 2\right) }\right) _{\mathbf{1}}\left( \left( \Phi
^{\left( 2\right) }\right) ^{\dagger }\Phi ^{\left( 1\right) }\right) _{\mathbf{1}}+\tau _{5}\left( \left( \Phi ^{\left( 1\right) }\right) ^{\dagger }\Phi
^{\left( 2\right) }\right) _{\mathbf{1}^{\prime }}\left( \left( \Phi
^{\left( 2\right) }\right) ^{\dagger }\Phi ^{\left( 1\right) }\right) _{\mathbf{1}^{\prime \prime }}\notag\\
&&+\tau _{6}\left( \left( \Phi ^{\left( 1\right)
}\right) ^{\dagger }\Phi ^{\left( 2\right) }\right) _{\mathbf{1}^{\prime
\prime }}\left( \left( \Phi ^{\left( 2\right) }\right) ^{\dagger }\Phi
^{\left( 1\right) }\right) _{\mathbf{1}^{\prime }},
\label{V1}\end{aligned}$$$$\begin{aligned}
V\left( \Phi ^{\left( 1\right) },\Phi ^{\left( 2\right) },\chi \right)
&=&\dsum\limits_{l=1}^{2}\left\{ \lambda _{al}\left( \left( \Phi ^{\left(
l\right) }\right) ^{\dagger }\Phi ^{\left( l\right) }\right) _{\mathbf{1}}\left( \chi \chi \right) _{\mathbf{1}}+\lambda _{bl}\left[ \left( \left(
\Phi ^{\left( l\right) }\right) ^{\dagger }\Phi ^{\left( l\right) }\right) _{\mathbf{1}^{\prime }}\left( \chi \chi \right) _{\mathbf{1}^{\prime \prime
}}+\left( \left( \Phi ^{\left( l\right) }\right) ^{\dagger }\Phi ^{\left(
l\right) }\right) _{\mathbf{1}^{\prime \prime }}\left( \chi \chi \right) _{\mathbf{1}^{\prime }}\right] \right.\notag \\
&&+\left. \alpha _{l}\left( \left( \Phi ^{\left( l\right) }\right) ^{\dagger
}\Phi ^{\left( l\right) }\right) _{\mathbf{3s}}\left( \chi \chi \right) _{\mathbf{3s}}+\left[ \beta
_{l}e^{i\frac{\pi }{2}}\left( \left( \Phi ^{\left( l\right) }\right)
^{\dagger }\Phi ^{\left( l\right) }\right) _{\mathbf{3a}}\left( \chi \chi
\right) _{\mathbf{3s}}+h.c\right] \right\} ,
\label{V2}\end{aligned}$$$$V\left( \chi \right) =D^{2}\left( \chi \chi \right) _{\mathbf{1}}+d_{1}\left( \chi \chi \right) _{\mathbf{1}}\left( \chi \chi \right) _{\mathbf{1}}+d_{2}\left( \chi \chi \right) _{\mathbf{1}^{\prime }}\left( \chi
\chi \right) _{\mathbf{1}^{\prime \prime }}+d_{3}\left( \chi \chi \right) _{\mathbf{3s}}\left( \chi \chi \right) _{\mathbf{3s}}.
\label{V3}$$
Where all parameters of the scalar potential have to be real.
Now we are going to determine the conditions under which the VEV pattern for the components of the $A_{4}$ triplet $\chi $, given in Eq. (\[VEV\]), is a solution of the scalar potential, assuming that the $\left\langle \Phi
^{\left( 2\right) }\right\rangle $ vacuum preserves the appropriate $Z_{3}$ subgroup of $A_4$ as in Eq. (\[VEV\]). Then, from the previous expressions and from the minimization conditions of the scalar potential, the following relations are obtained:
$$\begin{aligned}
\frac{\partial V}{\partial \chi _{1}}\biggl|_{\substack{ \left\langle \chi
_{m}\right\rangle =v_{\chi _{m}} \\ m=1,2,3}} &=&2v_{\chi _{1}}\left[ \frac{1}{2}\lambda _{a2}v^{2}+D^{2}+\left( 2d_{1}-d_{2}+4d_{3}\right) \left(
v_{\chi _{2}}^{2}+v_{\chi _{3}}^{2}\right) +2\left( d_{1}+d_{2}\right)
v_{\chi _{1}}^{2}\right]+\frac{2}{3}\alpha _{2}\left( v_{\chi _{2}}+v_{\chi _{3}}\right) v^{2}\notag \\
&=&0\end{aligned}$$
$$\begin{aligned}
\frac{\partial V}{\partial \chi _{2}}\biggl|_{\substack{ \left\langle \chi
_{m}\right\rangle =v_{\chi _{m}} \\ m=1,2,3}} &=&2v_{\chi _{2}}\left[ \frac{1}{2}\lambda _{a2}v^{2}+D^{2}+\left( 2d_{1}-d_{2}+4d_{3}\right) \left(
v_{\chi _{1}}^{2}+v_{\chi _{3}}^{2}\right) +2\left( d_{1}+d_{2}\right)
v_{\chi _{2}}^{2}\right] +\frac{2}{3}\alpha _{2}\left( v_{\chi _{1}}+v_{\chi _{3}}\right) v^{2} \notag \\
&=&0\end{aligned}$$
$$\begin{aligned}
\frac{\partial V}{\partial \chi _{3}}\biggl|_{\substack{ \left\langle \chi
_{m}\right\rangle =v_{\chi _{m}} \\ m=1,2,3}} &=&2v_{\chi _{3}}\left[ \frac{1}{2}\lambda _{a2}v^{2}+D^{2}+\left( 2d_{1}-d_{2}+4d_{3}\right) \left(
v_{\chi _{1}}^{2}+v_{\chi _{2}}^{2}\right) +2\left( d_{1}+d_{2}\right)
v_{\chi _{3}}^{2}\right]+\frac{2}{3}\alpha _{2}\left( v_{\chi _{1}}+v_{\chi _{2}}\right) v^{2} \notag \\
&=&0\end{aligned}$$
From the expressions given above, and using the vacuum configuration for the components of the $A_{4}$ triplet $\chi $ given in Eq. (\[VEV\]), the following relation is obtained: $$D^{2}=-\left( \frac{1}{2}\lambda_{a2}-\frac{1}{3}\alpha _{2}\right) v^{2} -\left(4d_{1}+d_{2}+4d_{3}\right) \frac{v_\chi^2}{2} \,,
\label{conditionminVEVchi}$$ which clearly shows that the hierarchy between the VEVs depends on the $\chi \chi$ mass term ($D^2$), and that the $Z_{3}$ invariant $\left\langle \Phi^{\left( 2\right) }\right\rangle $ vacuum given in Eq. (\[VEV\]) satisfies the minimization conditions of the scalar potential, in a way that is consistent with the desired direction for $\left\langle \chi \right\rangle $ This demonstrates that the VEV directions in Eq. (\[VEV\]) are consistent with a global minimum of the scalar potential (\[V\]) of our model, for a not fine-tuned region of parameter space.
Mass spectrum of the neutral scalar fields contained in the $A_{4}$ triplet $\Phi ^{\left( 1\right) }$. \[C\]
=============================================================================================================
In this section we proceed to compute the squared mass matrix for the neutral scalars, coming from the $A_{4}$ triplet $\Phi ^{\left( 1\right) }$. We assume, to simplify the analysis, that the couplings are nearly universal, i.e. $$\begin{aligned}
\label{univ-eps}
\lambda =\tau_i=\lambda _{b1}=\alpha _{1}=\lambda
_{a1}-\varepsilon, \ \ \ \ \ \ \ \ \ i=(1-6).\end{aligned}$$ In practice the coefficients do not need to be equal and indeed non-universality is required, with non-zero $\varepsilon$, necessary to generate two neutrino mass squared differences. Using the simplified assumptions a semi-analytical treatment is possible.
As mentioned in the text, we restrict to the scenario to $v_{\chi }\gg v$. Then, the dominant contribution to the mass Lagrangian for the neutral scalars contained in $\Phi ^{\left( 1\right) }$ will come from $\mu
_{1}^{2}\left( \left( \Phi ^{\left( 1\right) }\right) ^{\dagger }\Phi
^{\left( 1\right) }\right) _{\mathbf{1}}+V\left( \Phi ^{\left( 1\right)
},\Phi ^{\left( 2\right) },\chi \right) $. Using the relations: $$\left( \left( \Phi _{m}^{\left( 1\right) }\right) ^{\dagger }\Phi
_{m}^{\left( 1\right) }\right) =\frac{1}{2}\left[ \left( \omega _{m}^{\left(
1\right) }\right) ^{2}+\left( \xi _{m}^{\left( 1\right) }\right) ^{2}+\left(
\rho _{m}^{\left( 1\right) }\right) ^{2}+\left( \eta _{m}^{\left( 1\right)
}\right) ^{2}\right] ,\hspace{1cm}\hspace{1cm}m=1,2,3,$$$$\left( \left( \Phi ^{\left( 1\right) }\right) ^{\dagger }\Phi ^{\left(
1\right) }\right) _{\mathbf{1}}\left\langle \left( \chi \chi \right) _{\mathbf{1}}\right\rangle =v_{\chi }^{2}\left[ \left( \Phi _{1}^{\left(
1\right) }\right) ^{\dagger }\Phi _{1}^{\left( 1\right) }+\left( \Phi
_{2}^{\left( 1\right) }\right) ^{\dagger }\Phi _{2}^{\left( 1\right)
}+\left( \Phi _{3}^{\left( 1\right) }\right) ^{\dagger }\Phi _{3}^{\left(
1\right) }\right] ,$$$$\left( \left( \Phi ^{\left( 1\right) }\right) ^{\dagger }\Phi ^{\left(
1\right) }\right) _{\mathbf{1}^{\prime }}\left\langle \left( \chi \chi
\right) \right\rangle _{\mathbf{1}^{\prime \prime }}=\frac{v_{\chi }^{2}}{2}\left[ \left( \Phi _{1}^{\left( 1\right) }\right) ^{\dagger }\Phi
_{1}^{\left( 1\right) }+\omega \left( \Phi _{2}^{\left( 1\right) }\right)
^{\dagger }\Phi _{2}^{\left( 1\right) }+\omega ^{2}\left( \Phi _{3}^{\left(
1\right) }\right) ^{\dagger }\Phi _{3}^{\left( 1\right) }\right] \left(
1+\omega \right) ,$$$$\left( \left( \Phi ^{\left( 1\right) }\right) ^{\dagger }\Phi ^{\left(
1\right) }\right) _{\mathbf{1}^{\prime \prime }}\left\langle \left( \chi
\chi \right) _{\mathbf{1}^{\prime }}\right\rangle =\frac{v_{\chi }^{2}}{2}\left[ \left( \Phi _{1}^{\left( 1\right) }\right) ^{\dagger }\Phi
_{1}^{\left( 1\right) }+\omega ^{2}\left( \Phi _{2}^{\left( 1\right)
}\right) ^{\dagger }\Phi _{2}^{\left( 1\right) }+\omega \left( \Phi
_{3}^{\left( 1\right) }\right) ^{\dagger }\Phi _{3}^{\left( 1\right) }\right]
\left( 1+\omega ^{2}\right) ,$$$$\left( \left( \Phi ^{\left( 1\right) }\right) ^{\dagger }\Phi ^{\left(
1\right) }\right) _{\mathbf{3s}}\left\langle \left( \chi \chi \right) _{\mathbf{3s}}\right\rangle =-v_{\chi }^{2}\left[ \omega _{1}^{\left( 1\right)
}\omega _{3}^{\left( 1\right) }+\xi _{1}^{\left( 1\right) }\xi _{3}^{\left(
1\right) }+\rho _{1}^{\left( 1\right) }\rho _{3}^{\left( 1\right) }+\eta
_{1}^{\left( 1\right) }\eta _{3}^{\left( 1\right) }\right] ,$$ $$\beta _{1}e^{i\frac{\pi }{2}}\left( \left( \Phi ^{\left( 1\right) }\right)
^{\dagger }\Phi ^{\left( 1\right) }\right) _{\mathbf{3a}}\left\langle \left(
\chi \chi \right) _{\mathbf{3s}}\right\rangle +h.c=4\beta _{1}v_{\chi
}^{2}\left( \rho _{3}^{\left( 1\right) }\eta _{1}^{\left( 1\right) }-\rho
_{1}^{\left( 1\right) }\eta _{3}^{\left( 1\right) }\right) ,$$
$$\left( \left( \Phi ^{\left( 1\right) }\right) ^{\dagger }\Phi ^{\left(
2\right) }\right) _{\mathbf{3s}}\left( \left( \Phi ^{\left( 2\right)
}\right) ^{\dagger }\Phi ^{\left( 1\right) }\right) _{\mathbf{3s}}+\left(
\left( \Phi ^{\left( 1\right) }\right) ^{\dagger }\Phi ^{\left( 2\right)
}\right) _{\mathbf{3a}}\left( \left( \Phi ^{\left( 2\right) }\right)
^{\dagger }\Phi ^{\left( 1\right) }\right) _{\mathbf{3a}}+h.c\supset \frac{v^{2}}{2\sqrt{3}}\left( \rho _{3}^{\left( 1\right) }\rho _{2}^{\left(
1\right) }+\rho _{1}^{\left( 1\right) }\rho _{2}^{\left( 1\right) }+\rho
_{1}^{\left( 1\right) }\rho _{3}^{\left( 1\right) }\right) ,$$
$$\left( \left( \Phi ^{\left( 1\right) }\right) ^{\dagger }\Phi ^{\left(
2\right) }\right) _{\mathbf{1}}\left( \left( \Phi ^{\left( 2\right) }\right)
^{\dagger }\Phi ^{\left( 1\right) }\right) _{\mathbf{1}}\supset \frac{v^{2}}{4\sqrt{3}}\left( \rho _{1}^{\left( 1\right) }\rho _{1}^{\left( 1\right)
}+\rho _{2}^{\left( 1\right) }\rho _{2}^{\left( 1\right) }+\rho _{3}^{\left(
1\right) }\rho _{3}^{\left( 1\right) }+2\rho _{3}^{\left( 1\right) }\rho
_{2}^{\left( 1\right) }+2\rho _{1}^{\left( 1\right) }\rho _{2}^{\left(
1\right) }+2\rho _{1}^{\left( 1\right) }\rho _{3}^{\left( 1\right) }\right) ,$$
$$\begin{aligned}
&&\left( \left( \Phi ^{\left( 1\right) }\right) ^{\dagger }\Phi ^{\left(
2\right) }\right) _{\mathbf{1}^{\prime }}\left( \left( \Phi ^{\left(
2\right) }\right) ^{\dagger }\Phi ^{\left( 1\right) }\right) _{\mathbf{1}^{\prime \prime }}+\left( \left( \Phi ^{\left( 1\right) }\right) ^{\dagger
}\Phi ^{\left( 2\right) }\right) _{\mathbf{1}^{\prime \prime }}\left( \left(
\Phi ^{\left( 2\right) }\right) ^{\dagger }\Phi ^{\left( 1\right) }\right) _{\mathbf{1}^{\prime }} \notag \\
&\supset &\frac{2v^{2}}{4\sqrt{3}}\left( \rho _{1}^{\left( 1\right) }\rho
_{1}^{\left( 1\right) }+\rho _{2}^{\left( 1\right) }\rho _{2}^{\left(
1\right) }+\rho _{3}^{\left( 1\right) }\rho _{3}^{\left( 1\right) }-\rho
_{3}^{\left( 1\right) }\rho _{2}^{\left( 1\right) }-\rho _{1}^{\left(
1\right) }\rho _{2}^{\left( 1\right) }-\rho _{1}^{\left( 1\right) }\rho
_{3}^{\left( 1\right) }\right) ,\end{aligned}$$
we obtain that the mass Lagrangian for the neutral scalars contained in $\Phi ^{\left( 1\right) }$ is given by: $$\begin{aligned}
-\mathcal{L}_{mass}^{\left( 1\right) neutral} &=&\frac{\mu
_{1}^{2}+\varepsilon v_{\chi }^{2}}{2}\left[ \left( \rho _{1}^{\left(
1\right) }\right) ^{2}+\left( \rho _{2}^{\left( 1\right) }\right)
^{2}+\left( \rho _{3}^{\left( 1\right) }\right) ^{2}+\left( \eta
_{1}^{\left( 1\right) }\right) ^{2}+\left( \eta _{2}^{\left( 1\right)
}\right) ^{2}+\left( \eta _{3}^{\left( 1\right) }\right) ^{2}\right] \notag
\\
&&+\frac{3\lambda v_{\chi }^{2}}{4}\left[ \left( \rho _{1}^{\left( 1\right)
}\right) ^{2}+\left( \eta _{1}^{\left( 1\right) }\right) ^{2}+\left( \rho
_{3}^{\left( 1\right) }\right) ^{2}+\left( \eta _{3}^{\left( 1\right)
}\right) ^{2}\right]\notag \\
&&+4\beta
_{1}v_{\chi }^{2}\left( \rho _{3}^{\left( 1\right) }\eta _{1}^{\left(
1\right) }-\rho _{1}^{\left( 1\right) }\eta _{3}^{\left( 1\right) }\right) -\lambda v_{\chi }^{2}\left[ \rho _{1}^{\left( 1\right)
}\rho _{3}^{\left( 1\right) }+\eta _{1}^{\left( 1\right) }\eta _{3}^{\left(
1\right) }\right]
\notag \\
&&+\frac{\lambda v^{2}}{4\sqrt{3}}\left( 3\rho _{1}^{\left( 1\right) }\rho
_{1}^{\left( 1\right) }+3\rho _{2}^{\left( 1\right) }\rho _{2}^{\left(
1\right) }+3\rho _{3}^{\left( 1\right) }\rho _{3}^{\left( 1\right) }+2\rho
_{3}^{\left( 1\right) }\rho _{2}^{\left( 1\right) }+2\rho _{1}^{\left(
1\right) }\rho _{2}^{\left( 1\right) }+2\rho _{1}^{\left( 1\right) }\rho
_{3}^{\left( 1\right) }\right) .\end{aligned}$$ The squared mass matrix for the neutral scalars $(\rho_1,\rho_2,\rho_3,\eta_1,\eta_2,\eta_3)$ is given by: $$M^{2}\simeq \left(
\begin{array}{cccccc}
\frac{\mu _{1}^{2}+\varepsilon v_{\chi }^{2}}{2}+\frac{3\lambda v_{\chi }^{2}}{4} & \frac{\lambda v^{2}}{4\sqrt{3}} & -\frac{\lambda v_{\chi }^{2}}{2} & 0
& 0 & -2\beta _{1}v_{\chi }^{2} \\
\frac{\lambda v^{2}}{4\sqrt{3}} & \frac{\mu _{1}^{2}+\varepsilon v_{\chi
}^{2}}{2} & \frac{\lambda v^{2}}{4\sqrt{3}} & 0 & 0 & 0 \\
-\frac{\lambda v_{\chi }^{2}}{2} & \frac{\lambda v^{2}}{4\sqrt{3}} & \frac{\mu _{1}^{2}+\varepsilon v_{\chi }^{2}}{2}+\frac{3\lambda v_{\chi }^{2}}{4}
& 2\beta _{1}v_{\chi }^{2} & 0 & 0 \\
0 & 0 & 2\beta _{1}v_{\chi }^{2} & \frac{\mu _{1}^{2}+\varepsilon v_{\chi
}^{2}}{2}+\frac{3\lambda v_{\chi }^{2}}{4} & 0 & -\frac{\lambda v_{\chi }^{2}}{2} \\
0 & 0 & 0 & 0 & \frac{\mu _{1}^{2}+\varepsilon v_{\chi }^{2}}{2} & 0 \\
-2\beta _{1}v_{\chi }^{2} & 0 & 0 & -\frac{\lambda v_{\chi }^{2}}{2} & 0 &
\frac{\mu _{1}^{2}+\varepsilon v_{\chi }^{2}}{2}+\frac{3\lambda v_{\chi }^{2}
}{4}
\end{array}
\right) .$$Our near-universality assumption (\[univ-eps\]) allows for an approximate analytical diagonalization of this squared mass matrix, by a rotation matrix $R$: $$R^{T}M^{2}R\simeq diag\left(
M_{H_{1}^{0}}^{2},M_{H_{2}^{0}}^{2},M_{H_{3}^{0}}^{2},M_{A_{1}^{0}}^{2},M_{A_{2}^{0}}^{2},M_{A_{3}^{0}}^{2}\right) .$$ Due to the structure of the dominant $\chi$ VEV (\[VEV\]), the rotation matrix mixes the 1st and 3rd components of the scalars. If we lift the universality condition on the quartic couplings, there will be subleading mixing of the 2nd component of the scalars as well. Within the near-universality approximation, the rotation matrix $R$ is: $$R\simeq \left(
\begin{array}{cccccc}
\frac{\cos \psi \cos \theta _{1}}{\sqrt{2}} & -\frac{\sin \theta _{1}}{\sqrt{2}} & -\frac{\cos \theta _{1}\sin \psi }{\sqrt{2}} & -\frac{\sin \psi }{\sqrt{2}} & 0 & -\frac{\cos \psi }{\sqrt{2}} \\
\cos \psi \cos \theta _{2}\sin \theta _{1}-\sin \psi \sin \theta _{2} & \cos
\theta _{1}\cos \theta _{2} & -\cos \theta _{2}\sin \psi \sin \theta
_{1}-\cos \psi \sin \theta _{2} & 0 & 0 & 0 \\
\frac{\cos \theta _{2}\sin \psi }{\sqrt{2}}+\frac{\cos \psi \sin \theta
_{1}\sin \theta _{2}}{\sqrt{2}} & \frac{\cos \theta _{1}\sin \theta _{2}}{\sqrt{2}} & \frac{\cos \psi \cos \theta _{2}}{\sqrt{2}}-\frac{\sin \psi \sin
\theta _{1}\sin \theta _{2}}{\sqrt{2}} & -\frac{\cos \psi }{\sqrt{2}} & 0 &
\frac{\sin \psi }{\sqrt{2}} \\
\frac{\cos \theta _{2}\sin \psi }{\sqrt{2}}+\frac{\cos \psi \sin \theta
_{1}\sin \theta _{2}}{\sqrt{2}} & \frac{\cos \theta _{1}\sin \theta _{2}}{\sqrt{2}} & \frac{\cos \psi \cos \theta _{2}}{\sqrt{2}}-\frac{\sin \psi \sin
\theta _{1}\sin \theta _{2}}{\sqrt{2}} & \frac{\cos \psi }{\sqrt{2}} & 0 & -\frac{\sin \psi }{\sqrt{2}} \\
0 & 0 & 0 & 0 & 1 & 0 \\
\frac{\cos \psi \cos \theta _{1}}{\sqrt{2}} & -\frac{\sin \theta _{1}}{\sqrt{2}} & -\frac{\cos \theta _{1}\sin \psi }{\sqrt{2}} & \frac{\sin \psi }{\sqrt{2}} & 0 & \frac{\cos \psi }{\sqrt{2}}\end{array}\right) ,$$with: $$\tan 2\psi \simeq \frac{\lambda }{4\beta _{1}},\hspace{1cm}\hspace{1cm}\tan
2\theta _{1}\simeq \frac{2\lambda v^{2}}{\sqrt{6}\left( 3\lambda -8\beta
_{1}\right) v_{\chi }^{2}},\hspace{1cm}\hspace{1cm}\tan 2\theta _{2}\simeq
\frac{4\lambda ^{2}v}{\sqrt{6}\left( 9\lambda ^{2}-64\beta _{1}^{2}\right)
v_{\chi }}.$$The masses of the physical neutral scalars are given by: $$\label{MH10}
M_{H_{1}^{0}}^{2}\simeq M_{A_{1}^{0}}^{2}+\left\{ \left[ \frac{a^{2}+b^{2}}{4}+\frac{\left( a^{2}-b^{2}\right) \beta _{1}}{\sqrt{\lambda ^{2}+16\beta
_{1}^{2}}}\right] \frac{\mu _{1}^{2}+\varepsilon v_{\chi }^{2}}{v_{\chi }^{2}}-\frac{1}{2}ab\lambda +\frac{2ab\lambda \beta _{1}}{\sqrt{\lambda
^{2}+16\beta _{1}^{2}}}\right\} v^{2},$$ $$M_{H_{2}^{0}}^{2}\simeq M_{A_{2}^{0}}^{2}+\frac{1}{4}\left[ a^{2}\left(
3\lambda -8\beta _{1}\right) +b^{2}\left( 3\lambda +8\beta _{1}\right)
+2\left( a^{2}+b^{2}\right) \frac{\mu _{1}^{2}+\varepsilon v_{\chi }^{2}}{v_{\chi }^{2}}+4ab\lambda \right] v^{2},$$ $$M_{H_{3}^{0}}^{2}\simeq M_{A_{3}^{0}}^{2}+\left\{ \left[ \frac{a^{2}+b^{2}}{4}-\frac{\left( a^{2}-b^{2}\right) \beta _{1}}{\sqrt{\lambda ^{2}+16\beta
_{1}^{2}}}\right] \frac{\mu _{1}^{2}+\varepsilon v_{\chi }^{2}}{v_{\chi }^{2}}-\frac{1}{2}ab\lambda -\frac{2ab\lambda \beta _{1}}{\sqrt{\lambda
^{2}+16\beta _{1}^{2}}}\right\} v^{2},$$ $$M_{A_{1}^{0}}^{2}\simeq \frac{1}{4}\left( 2\mu _{1}^{2}+2\varepsilon v_{\chi
}^{2}+3\lambda v_{\chi }^{2}-2\sqrt{\lambda ^{2}+16\beta _{1}^{2}}v_{\chi
}^{2}\right) ,$$$$M_{A_{2}^{0}}^{2}\simeq \frac{\mu _{1}^{2}+\varepsilon v_{\chi }^{2}}{2},$$$$M_{A_{3}^{0}}^{2}\simeq \frac{1}{4}\left( 2\mu _{1}^{2}+2\varepsilon v_{\chi
}^{2}+3\lambda v_{\chi }^{2}+2\sqrt{\lambda ^{2}+16\beta _{1}^{2}}v_{\chi
}^{2}\right) ,$$ where $$\label{abLB}
a\simeq \frac{\lambda }{\sqrt{6}\left( 3\lambda -8\beta _{1}\right) },\hspace{1cm}\hspace{1cm}b\simeq \frac{2\lambda ^{2}}{\sqrt{6}\left( 9\lambda
^{2}-64\beta _{1}^{2}\right) }.$$ It is worth mentioning that the last five terms of Eq. (\[V1\]), involving distinct $A_4$ invariant contractions, are responsible for the mass splitting between the CP even and CP odd neutral scalars.
From the previous expressions we obtain the relation connecting the parameter $\psi$ with the neutral scalar masses: $$\tan 2\psi \simeq \frac{1}{\sqrt{\frac{9}{4}\left( \frac{M_{A_{3}^{0}}^{2}-M_{A_{1}^{0}}^{2}}{ M_{A_{3}^{0}}^{2}+M_{A_{1}^{0}}^{2}-2M_{A_{2}^{0}}^{2}}\right) ^{2}-1}}.$$The physical scalars $H_{1}^{0}$, $H_{2}^{0}$, $H_{3}^{0}$, $A_{1}^{0}$, $A_{2}^{0}$ and $A_{3}^{0}$ are given by: $$\begin{aligned}
&&\left(
\begin{array}{c}
H_{1}^{0} \\
H_{2}^{0} \\
H_{3}^{0} \\
A_{1}^{0} \\
A_{2}^{0} \\
A_{3}^{0}\end{array}\right) \simeq \notag \\
&\simeq &\left(
\begin{array}{cccccc}
\frac{\cos \psi \cos \theta _{1}}{\sqrt{2}} & \cos \psi \cos \theta _{2}\sin
\theta _{1}-\sin \psi \sin \theta _{2} & \frac{\cos \theta _{2}\sin \psi }{\sqrt{2}}+\frac{\cos \psi \sin \theta _{1}\sin \theta _{2}}{\sqrt{2}} &
\frac{\cos \theta _{2}\sin \psi }{\sqrt{2}}+\frac{\cos \psi \sin \theta
_{1}\sin \theta _{2}}{\sqrt{2}} & 0 & \frac{\cos \psi \cos \theta _{1}}{\sqrt{2}} \\
-\frac{\sin \theta _{1}}{\sqrt{2}} & \cos \theta _{1}\cos \theta _{2} &
\frac{\cos \theta _{1}\sin \theta _{2}}{\sqrt{2}} & \frac{\cos \theta
_{1}\sin \theta _{2}}{\sqrt{2}} & 0 & -\frac{\sin \theta _{1}}{\sqrt{2}} \\
-\frac{\cos \theta _{1}\sin \psi }{\sqrt{2}} & -\cos \theta _{2}\sin \psi
\sin \theta _{1}-\cos \psi \sin \theta _{2} & \frac{\cos \psi \cos \theta
_{2}}{\sqrt{2}}-\frac{\sin \psi \sin \theta _{1}\sin \theta _{2}}{\sqrt{2}}
& \frac{\cos \psi \cos \theta _{2}}{\sqrt{2}}-\frac{\sin \psi \sin \theta
_{1}\sin \theta _{2}}{\sqrt{2}} & 0 & -\frac{\cos \theta _{1}\sin \psi }{\sqrt{2}} \\
-\frac{\sin \psi }{\sqrt{2}} & 0 & -\frac{\cos \psi }{\sqrt{2}} & \frac{\cos
\psi }{\sqrt{2}} & 0 & \frac{\sin \psi }{\sqrt{2}} \\
0 & 0 & 0 & 0 & 1 & 0 \\
-\frac{\cos \psi }{\sqrt{2}} & 0 & \frac{\sin \psi }{\sqrt{2}} & -\frac{\sin
\psi }{\sqrt{2}} & 0 & \frac{\cos \psi }{\sqrt{2}}\end{array}\right) \allowbreak \notag \\
&&\times \left(
\begin{array}{c}
\rho _{1}^{\left( 1\right) } \\
\rho _{2}^{\left( 1\right) } \\
\rho _{3}^{\left( 1\right) } \\
\eta _{1}^{\left( 1\right) } \\
\eta _{2}^{\left( 1\right) } \\
\eta _{3}^{\left( 1\right) }\end{array}\right) . \label{Physicalscalars}\end{aligned}$$After $\chi $ gets its VEV, $v_{\chi}$, and neglecting terms suppressed by powers of $v/v_{\chi }$, the part of the Lagrangian that is obtained from the quartic interactions with $\chi $ becomes $$\begin{aligned}
-\mathcal{L}_{mass}^{\left( 1\right) neutral} &\supset &\left(
M_{A_{1}^{0}}^{2}-M_{A_{2}^{0}}^{2}+\frac{\varepsilon v_{\chi }^{2}}{2}\right) \left[ \left( H_{1}^{0}\right) ^{2}+\left( A_{1}^{0}\right) ^{2}\right] +\frac{\varepsilon v_{\chi }^{2}}{2}\left[ \left( H_{2}^{0}\right)
^{2}+\left( A_{2}^{0}\right) ^{2}\right] \notag \\
&&+\left( M_{A_{3}^{0}}^{2}-M_{A_{2}^{0}}^{2}+\frac{\varepsilon v_{\chi }^{2}}{2}\right) \left[ \left( H_{3}^{0}\right) ^{2}+\left( A_{3}^{0}\right) ^{2}\right] . \label{L2}\end{aligned}$$
[99]{}
J. Beringer et al. (Particle Data Group), Phys. Rev. D **86** 010001 (2012).
K. Abe *et al.* \[T2K Collaboration\], Phys. Rev. Lett. **107** 041801 (2011) \[arXiv:1106.2822 \[hep-ex\]\]. P. Adamson *et al.* \[MINOS Collaboration\], Phys. Rev. Lett. **107** 181802 (2011) \[arXiv:1108.0015 \[hep-ex\]\]. Y. Abe *et al.* \[DOUBLE-CHOOZ Collaboration\], Phys. Rev. Lett. **108** 131801 (2012) \[arXiv:1112.6353 \[hep-ex\]\].
F. P. An *et al.* \[DAYA-BAY Collaboration\], Phys. Rev. Lett. **108** 171803 (2012) \[arXiv:1203.1669 \[hep-ex\]\]. J. K. Ahn *et al.* \[RENO Collaboration\], Phys. Rev. Lett. **108** 191802 (2012) \[arXiv:1204.0626 \[hep-ex\]\]. D. V. Forero, M. Tortola and J. W. F. Valle, Phys. Rev. D **86** 073012 (2012) \[arXiv:1205.4018 \[hep-ph\]\]. G. L. Fogli, E. Lisi, A. Marrone, D. Montanino, A. Palazzo and A. M. Rotunno, Phys. Rev. D [**86**]{} 013012 (2012) \[arXiv:1205.5254 \[hep-ph\]\]. M. C. Gonzalez-Garcia, M. Maltoni, J. Salvado and T. Schwetz, JHEP [**1212**]{}, 123 (2012) \[arXiv:1209.3023 \[hep-ph\]\]. S. F. King and C. Luhn, Rept. Prog. Phys. [**76**]{}, 056201 (2013) \[arXiv:1301.1340 \[hep-ph\]\]. G. Altarelli and F. Feruglio, Rev. Mod. Phys. **82** 2701 (2010) \[arXiv:1002.0211 \[hep-ph\]\]. H. Ishimori, T. Kobayashi, H. Ohki, Y. Shimizu, H. Okada and M. Tanimoto, Prog. Theor. Phys. Suppl. [**183**]{}, 1 (2010) \[arXiv:1003.3552 \[hep-th\]\]. H. Fritzsch, Phys. Lett. B70, 436 (1977) ; B73, 317 (1978); Nucl. Phys. B155, 189 (1979); T.P. Cheng and M. Sher, Phys. Rev. D35, 3484 (1987); D. DU and Z. Z. Xing, Phys. Rev. D48, 2349 (1993); H. Fritzsch and Z. Z. Xing, Phys. Lett. 555, 63 (2003); J.L. Diaz-Cruz, R. Noriega-Papaqui and A. Rosado, Phys. Rev. D71, 015014 (2005); K. Matsuda and H. Nishiura, Phys. Rev. D74, 033014 (2006); A. Carcamo, R. Martinez and J.-A. Rodriguez, Eur. Phys. J. C50, 935 (2007).
E. Ma, Phys. Rev. D [**73**]{}, 077301 (2006) \[hep-ph/0601225\]. E. Ma, Mod. Phys. Lett. A **21** 1777 (2006) \[hep-ph/0605180\]. A. E. Cárcamo Hernández, R. Martínez and F. Ochoa, Phys. Rev. D **87** 075009 (2013) \[arXiv:1302.1757 \[hep-ph\]\]. A. C. B. Machado, J. C. Montero and V. Pleitez, Phys. Lett. B **697** 318 (2011) \[arXiv:1011.5855 \[hep-ph\]\]. E. Ma and G. Rajasekaran, Phys. Rev. D **64** 113012 (2001) \[hep-ph/0106291\]. K. S. Babu, E. Ma and J. W. F. Valle, Phys. Lett. B **552** 207 (2003) \[hep-ph/0206292\]. G. Altarelli and F. Feruglio, Nucl. Phys. B **741** 215 (2006) \[hep-ph/0512103\]. G. Altarelli, F. Feruglio and L. Merlo, JHEP [**0905**]{}, 020 (2009) \[arXiv:0903.1940 \[hep-ph\]\]. F. Bazzocchi, L. Merlo and S. Morisi, Nucl. Phys. B [**816**]{}, 204 (2009) \[arXiv:0901.2086 \[hep-ph\]\]. X. -G. He, Y. -Y. Keum and R. R. Volkas, JHEP **0604** 039 (2006) \[hep-ph/0601001\]. T. Fukuyama, H. Sugiyama and K. Tsumura, Phys. Rev. D [**82**]{}, 036004 (2010) \[arXiv:1005.5338 \[hep-ph\]\]. T. Fukuyama, H. Sugiyama and K. Tsumura, Phys. Rev. D [**83**]{}, 056016 (2011) \[arXiv:1012.4886 \[hep-ph\]\]. Y. H. Ahn and S. K. Kang, Phys. Rev. D **86** 093003 (2012) \[arXiv:1203.4185 \[hep-ph\]\]. Y. H. Ahn, S. K. Kang and C. S. Kim, Phys. Rev. D **87** 113012 (2013) \[arXiv:1304.0921 \[hep-ph\]\]
M. -C. Chen, J. Huang, J. -M. O’Bryan, A. M. Wijangco and F. Yu, JHEP **1302** 021 (2013) \[arXiv:1210.6982 \[hep-ph\]\]. R. de Adelhart Toorop, F. Bazzocchi, L. Merlo and A. Paris, JHEP [**1103**]{}, 035 (2011) \[Erratum-ibid. [**1301**]{}, 098 (2013)\] \[arXiv:1012.1791 \[hep-ph\]\]. N. Memenga, W. Rodejohann and H. Zhang, Phys. Rev. D **87** 053021 (2013) \[arXiv:1301.2963 \[hep-ph\]\]. M. Holthausen, M. Lindner and M. A. Schmidt, Phys. Rev. D [**87**]{}, no. 3, 033006 (2013) \[arXiv:1211.5143 \[hep-ph\]\]. P. M. Ferreira, L. Lavoura and P. O. Ludl, arXiv:1306.1500 \[hep-ph\]. R. Gonzalez Felipe, H. Serodio and J. P. Silva, Phys. Rev. D [**88**]{}, 015015 (2013), \[arXiv:1304.3468 \[hep-ph\]\]. A. C. B. Machado and V. Pleitez, Phys. Lett. B **674** 223 (2009) \[arXiv:0712.0781 \[hep-ph\]\]. A. C. B. Machado, J. C. Montero and V. Pleitez, Int. J. Mod. Phys. A **27** 1250068 (2012) \[arXiv:1108.1767 \[hep-ph\]\]. R. G. Felipe, H. Serodio and J. P. Silva, Phys. Rev. D [**87**]{}, 055010 (2013) \[arXiv:1302.0861 \[hep-ph\]\]. I. de Medeiros Varzielas and D. Pidt, JHEP **1303** 065 (2013) \[arXiv:1211.5370 \[hep-ph\]\]. H. Ishimori and E. Ma, Phys. Rev. D **86** 045030 (2012) \[arXiv:1205.0075 \[hep-ph\]\]. S. Bhattacharya, E. Ma, A. Natale and A. Rashed, Phys. Rev. D [**87**]{}, 097301 (2013) \[arXiv:1302.6266 \[hep-ph\]\]. S. F. King, S. Morisi, E. Peinado and J. W. F. Valle, Phys. Lett. B [**724**]{}, 68 (2013) \[arXiv:1301.7065 \[hep-ph\]\]. S. Morisi, D. V. Forero, J. C. Romao and J. W. F. Valle, Phys. Rev. D [**88**]{}, 016003 (2013) \[arXiv:1305.6774 \[hep-ph\]\]. S. Morisi, M. Nebot, K. M. Patel, E. Peinado and J. W. F. Valle, Phys. Rev. D [**88**]{}, 036001 (2013) \[arXiv:1303.4394 \[hep-ph\]\]. T. Teshima, Phys. Rev. D [**73**]{}, 045019 (2006) \[arXiv:hep-ph/0509094\]; R. N. Mohapatra, S. Nasri and H. B. Yu, Phys. Lett. B [**639**]{}, 318 (2006) \[arXiv:hep-ph/0605020\];
E. Ma and B. Melic, Phys. Lett. B [**725**]{}, 402 (2013) \[arXiv:1303.6928 \[hep-ph\]\]. F. G. Canales, A. Mondragón, M. Mondragón, U. J. S. Salazar and L. Velasco-Sevilla, arXiv:1304.6644 F. Gonzalez Canales, A. Mondragon and M. Mondragon, Fortsch. Phys. [**61**]{}, 546 (2013) \[arXiv:1205.4755 \[hep-ph\]\]. P. V. Dong, H. N. Long, C. H. Nam and V. V. Vien, Phys. Rev. D [**85**]{}, 053001 (2012) \[arXiv:1111.6360 \[hep-ph\]\]. Y. Kajiyama, H. Okada and K. Yagyu, arXiv:1309.6234 \[hep-ph\]. A. E. Cárcamo Hernández, R. Martínez and F. Ochoa, arXiv:1309.6567 \[hep-ph\].
R. N. Mohapatra and C. C. Nishi, Phys. Rev. D **86** 073007 (2012) \[arXiv:1208.2875 \[hep-ph\]\]. I. de Medeiros Varzielas and Luís Lavoura, J. Phys. G [**40**]{}, 085002 (2013) \[arXiv:1212.3247 \[hep-ph\]\]. G. -J. Ding, S. F. King, C. Luhn and A. J. Stuart, JHEP [**1305**]{}, 084 (2013) \[arXiv:1303.6180 \[hep-ph\]\]. I. K. Cooper, S. F. King and A. J. Stuart, Nucl. Phys. B [**875**]{}, 650 (2013) \[arXiv:1212.1066 \[hep-ph\]\]. I. de Medeiros Varzielas, D. Emmanuel-Costa and P. Leser, Phys. Lett. B [**716**]{}, 193 (2012) \[arXiv:1204.3633 \[hep-ph\]\]. G. Bhattacharyya, I. de Medeiros Varzielas and P. Leser, Phys. Rev. Lett. [**109**]{}, 241603 (2012) \[arXiv:1210.0545 \[hep-ph\]\]. E. Ma, Phys. Lett. B [**723**]{}, 161 (2013) \[arXiv:1304.1603 \[hep-ph\]\]. C. C. Nishi, Phys. Rev. D [**88**]{}, 033010 (2013) \[arXiv:1306.0877 \[hep-ph\]\]. I. de Medeiros Varzielas and D. Pidt, arXiv:1307.0711 \[hep-ph\]. A. Aranda, C. Bonilla, S. Morisi, E. Peinado and J. W. F. Valle, Nucl. Phys. B [**876**]{}, 418 (2013) \[arXiv:1304.2645 \[hep-ph\]\].
N. Rius and V. Sanz, Phys. Rev. D **64** 075006 (2001) \[hep-ph/0103086\].
B. A. Dobrescu, Phys. Lett. B **461** 99 (1999) \[hep-ph/9812349\].
G. Altarelli and F. Feruglio, Nucl. Phys. B **720** 64 (2005) \[hep-ph/0504165\]. H. Ishimori, Y. Shimizu, M. Tanimoto and A. Watanabe, Phys. Rev. D **83** 033004 (2011) \[arXiv:1010.3805 \[hep-ph\]\]. A. Kadosh and E. Pallante, JHEP **1008** 115 (2010) \[arXiv:1004.0321 \[hep-ph\]\]. A. Kadosh, JHEP [**1306**]{}, 114 (2013) \[arXiv:1303.2645 \[hep-ph\]\]. G. -J. Ding and Y. -L. Zhou, Nucl. Phys. B [**876**]{}, 418 (2013) \[arXiv:1304.2645 \[hep-ph\]\]. A. E. Cárcamo Hernández, C. O. Dib, N. Neill H. and A. R. Zerwekh, JHEP **1202** 132 (2012) \[arXiv:1201.0878 \[hep-ph\]\]. R. Barbieri, G. R. Dvali, A. Strumia, Z. Berezhiani and L. J. Hall, Nucl. Phys. B **432**, 49 (1994) \[arXiv:hep-ph/9405428\]; Z. Berezhiani, Phys. Lett. B **355**, 481 (1995) \[arXiv:hep-ph/9503366\]; H. C. Cheng, Phys. Rev. D **60**, 075015 (1999) \[arXiv:hep-ph/9904252\]; A. E. Cárcamo Hernández and Rakibur Rahman \[arXiv:hep-ph/1007.0447\]. K. S. Babu and R. N. Mohapatra, Phys. Rev. Lett. **74**, 2418 (1995) \[arXiv:hep-ph/9410326\].
G. Aad et al. \[ATLAS Collaboration\], Phys. Lett. B [**716**]{}, 1 (2012); S. Chatrchyan et al. \[CMS Collaboration\], Phys. Lett. B [**716**]{}, 30 (2012).
ATLAS-CONF-2013-010, http://cds.cern.ch/record/1523695/files/ATLAS-CONF-2013-010.pdf
EXO Collaboration, M. Auger [*et al.*]{}, Phys.Rev.Lett. [**109**]{}, 032505 (2012), arXiv:1205.5608.
GERDA Collaboration, I. Abt [*et al.*]{}, (2004), arXiv:hep-ex/0404039. GERDA Collaboration, K.-H. Ackermann [*et al.*]{}, (2012), arXiv:1212.4067. F. Alessandria [*et al.*]{}, (2011), arXiv:1109.0494.
KamLAND-Zen Collaboration, A. Gando [*et al.*]{}, Phys.Rev. [**C85**]{}, 045504 (2012), arXiv:1201.4664. EXO-200 Collaboration, D. Auty, Recontres de Moriond, http://moriond.in2p3.fr/ (2013).
Majorana Collaboration, C. Aalseth [*et al.*]{}, Nucl.Phys.Proc.Suppl. [**217**]{}, 44 (2011), arXiv:1101.0119. A. Barabash, (2012), arXiv:1209.4241.
P. Ramond, [*Group Theory: A Physicist’s Survey*]{}, Cambridge University Press, UK (2010).
| {
"pile_set_name": "ArXiv"
} | ArXiv |
---
abstract: 'We calculate the number PI from the problem of two billiard balls colliding with a wall (Galperin billiard method). We provide a complete explicit solution for the balls’ positions and velocities as a function of the collision number and time. The relation between collision number and time can be further simplified in the limit of large base $b$ or mantissa length $N$. We find new invariants of motion. Also, we recover previously known ones for which we provide a simple physical explanation in terms of the square of the angular momentum and demonstrate that they coincide with the action invariant derived within the adiabatic approximation close to the return point. We show that for general values of the parameters the system is integrable and for some special values of the parameters it is maximally superintegrable. The portrait of the system is close to a circle in velocity-velocity and velocity-inverse position coordinates and to a hyperbola in position-time variables. A differential equation describing the heavy-ball trajectory close to the point of return is derived and results in a parabolic trajectory. A generalization of the system to finite-size balls (hard rods) is provided. We propose to treat the possible error in the last digit as a systematic error. Examples of integer and non-integer bases are considered. In the intriguing case of expressing $\pi$ in the base $b=\pi$, the generated expression is different from a finite number, $\pi = 1\times \pi^1$, and instead is given by an infinite representation, $\pi = 3 + 1/\pi^2 + 1/\pi^3+\cdots$. The difference between finite and infinite representation is similar to that of $1 = 0.999(9)$ in the decimal system. Finally we note that the finite representation is not unique in the base of the golden number.'
author:
- 'X. M. Aretxabaleta'
- Marina Gonchenko
- 'N.L. Harshman'
- Steven Glenn Jackson
- Maxim Olshanii
- 'G. E. Astrakharchik'
title: |
Two-ball billiard predicts digits of the number PI\
in non-integer numerical bases
---
=1
Introduction
============
The invention of numbers was probably one of the most influential discoveries in the history of science leading to the foundation and development of mathematics. In many ancient cultures, the symbols for the first digits correspond to a graphical representation of counting. In Babylonian, Roman and Japanese numerals, digit “1” contains one counting object, digit “2” two objects, digit “3” three objects, see Fig. \[Fig:numerals\]. After counting, the next important concept is that of positioning system (position of a digit defines its value) and the number base representation. Throughout history different bases were used, including the modern decimal system and the sexagesimal one introduced in Babylon around second millennium BC. Its legacy can still be found in modern units of time, with 60 seconds in one minute and 60 minutes in one hour.
![[]{data-label="Fig:numerals"}](FigRomanJapaneaseNumbers.pdf){width="0.4\columnwidth"}
It was realized that some numbers, referred to as [*irrational*]{}, cannot be written as a simple ratio of integer numbers and in this sense they are the most difficult to be calculated precisely. Probably, the most important and fascinating irrational numbers are $\sqrt{2}$, $\pi$, Euler’s constant $e$ and golden ratio $\varphi$. Already in the antiquity there was a practical interest in representing some of those numbers explicitly [@ArndtBook; @Goyanes2007; @BerggrenBook]. In a Babylonian clay tablet from second millennium BC, the first four digits of $\sqrt{2}$ are explicitly given in sexagesimal system as $1, 24, 51, 10$. In decimal system the error appears in the eighth digit as can be appreciated by comparing $1 + 24/60 +51/60^2+10/60^3 = 1.4142130$ with the proper value $1.4142135\dots$. Another irrational number, $\pi$, naturally appears when calculating the ratio between a circumference of a circle and its diameter [@BeckmannBook]. In the Old Testament \[1 Kings 7:23\], the ratio between a circumference of a round vessel and its diameter is said to be equal to $3$. While in many practical situations, it is sufficient to use an approximate value, it was of fundamental interest to find a method of finding the next digits. Some other ancient estimations come from an Egyptian papyrus which implies $\pi = 256/81 = 3.160\dots$ and a Babylonian clay tablet leading to the value $25/8 = 3.125\dots$ . Archimedes calculated the upper bound as $22/7 = 3.1428\dots$. The fascination with the number $\pi$ makes scientists compete for the largest number of digits calculated. Simon Newcomb (1835-1909) is quoted for having said “[*Ten decimal places of $\pi$ are sufficient to give the circumference of the earth to a fraction of an inch, and thirty decimal places would give the circumference of the visible universe to a quantity imperceptible to the most powerful microscope*]{}”. The current world record [@Trueb2016] consists in calculating first 22,459,157,718,361 ($\pi^e$ trillion) digits. Such a task manifestly goes beyond any practical purpose but can be justified by the great attractiveness of the number $\pi$ itself. Apparently the distribution of digits is flat in different bases [@Trueb2016] and it was tested that the sequence of $\pi$ digits makes a good random number generator which can be used for practical scientific and engineering computations [@SHU-JU2005]. An alternative popular idea is that, in contrast, special information might be coded in digits of $\pi$ [@ShumikhinBook], or even God’s name as in the plot of “Pi” film from 1990. Recently, analogies between anomalies in the cosmic microwave background and patterns in the digits of $\pi$ were pointed out in “Pi in the Sky” article [@Frolop2016], which appeared on the 1st of April.
While the number $\pi$ elegantly arises in a large variety of trigonometric relations, integrals, series, products, continued fractions, far fewer experimental methods are known of how to obtain its digits by performing measurements according to some procedure. A stochastic method, stemming from Comte de Buffon dates back to the eighteenth century and consists in dropping $N$ needles of length $l$ on parallel lines separated by length $L$ and experimentally determining the number of times $N_{cross}$ that those needles were crossing the lines. The number $\pi$ can be then approximated by $\pi \approx 2l\cdot N/(L N_{cross})$ with the error in its estimation proportional to $1/\sqrt{N}$. It means that in order to get the first $D$ digits right, one has to perform more than $100^{D}$ trials. This makes it extremely difficult to obtain the precise value in a real-world experiment although an equivalent computer experiment can be easily performed with modern computational power.
Still, a simple mechanical system which would provide explicitly the digits of number $\pi$ was lacking for a long time. A completely new perspective has emerged when G. A. Galperin formulated a deterministic method based on a two-ball billiard [@Galperin2001]. The scheme of the method is summarized in Fig. \[Fig:balls\]. Two balls, heavy and light, move along a groove which ends with a wall. The heavy ball collides with the stationary light ball and the number of collisions $\Pi$ is counted for different mass ratio of the heavy to light ball. It was shown by Galperin that the number of collisions is intimately related to the number $\pi$, providing the first digits of the irrational number. Thus, for equal masses, $M = m$, the number of collisions is $\Pi = 3$, which corresponds to the first digit of number $\pi$. For masses $M = 100 m$ the number of collisions is $\Pi = 31$, giving the first two digits. The case of $M = 10 000 m$ results in $\Pi = 314$ thus providing three digits and so on. To a certain extent, finding digits of number $\pi$ became conceptually as simple and elegant as enumerating the counting objects like Roman or Japanese digits shown in Fig. \[Fig:numerals\].
![Schematic picture of a billiard system, consisting of a heavy ball $M$, light ball $m$ and a wall[]{data-label="Fig:balls"}](FigBalls.pdf){width="0.5\columnwidth"}
The problem of the ergodic motion of two balls within two walls was posed in the book by Sinai[@SinaiBook]. He showed that the configuration space of the system is limited to a triangle and, thus, the problem is equivalent to a billiard with the same opening angle. Also Sinai used “billiard variables” such that the absolute value of the rescaled velocity is conserved and the product of the vector of the rescaled velocities with the vector $(\sqrt{M},\sqrt{m})$ is constant. The number of collisions in a “gas of two molecules” was given in the book by Galperin and Zemplyakov [@GalperinBook] in 1990, although no relation to the digits of $\pi$ was given at that moment. Similarly, Tabachnikov in his 1995 book[@TabachnikovBook] argued that the number of collisions is always finite and provided the same expression for it. In the 90s in Galperin’s seminars the way to extract digits of $\pi$ from the billiard was explained. In year 2001 Galperin published a short article on that procedure in Russian[@Galperin2001] and in year 2003 in English [@Galperin2003]. This fascinating problem was given as a motivating example in the introduction of another book by Tabachnikov, Ref. [@TabachnikovBook2005], illustrating trajectory unfolding. Gorelyshev in Ref. [@Gorelyshev2006] applied adiabatic approximation to the two-ball problem and found a conserved quantity, namely the action, close to the point of return. Weidman [@Weidman2013] found two invariants of motion, corresponding to ball-ball and ball-wall collisions. He noted that the terminal collision distinguishes between even and odd digits. Davis in Ref. [@Davis2015] solved the equations of motion as a system of two linear equations for ball-ball and ball-wall collisions, finding the rotation angle from determinantal equation. In addition to the energetic circle [@Galperin2001], defining the velocities, he expressed the balls positions as a function of the number of collisions. A number of related systems was recently studied, including two balls in one dimension with gravity [@Whelan1990], dynamics of polygonal billiards [@Gutkin1996] and a ping-pong ball between two cannonballs [@Redner2004].
In the present work we describe how Galperin’s billiard method for mass ratio $M / m = b^{2N}$ generates first $N$ digits of the fractional part (i.e. digits beyond the radix point) of number $\pi$ in base-$b$ numeral system and address the underlying assumptions. We consider the cases of integer and non-integer base systems, including a compelling case of representing number $\pi$ in a system base of $\pi$.
The article is organized as follows. In Sec. \[Sec:Galperin billiard method\] we review the Galperin billiard method. The basic model is introduced in Sec \[Sec:model\]. The relation between the number of collisions and the number $\pi$ in a system with base $b$ is explained in Sec. \[Sec:Number of collisions\]. Different properties of the trajectory of balls are derived and discussed in Sec. \[Sec:ball trajectory\]. An iterative discrete solution to the equations of motion is given in Sec. \[Sec:equations of motion\]. Minimal and maximal values of the velocities are obtained in Sec. \[Sec:max:V\]. A differential equation describing the motion close to the point of return is derived and solved in Sec. \[Sec:point of return\]. The analytical vs non-analytical shapes of the envelope of the trajectory are discussed in Sec. \[Sec:non-analytic\]. The adiabatic approximation and action-phase variables are introduced in Sec. \[Sec:adiabatic approximation\] in the vicinity of the point of return. In Sec. \[Sec:invariants\] it is shown that the adiabatic action invariant is preserved not only close to the point of return, but for all collisions. In Sec. \[Sec:Unfolding\] the unfolding of the trajectory is explained. The integrability of the model is addressed in Sec. \[Sec:integrability\], and superintegrable mass ratios are identified. Minimal and maximal distances from the wall are obtained in Sec. \[Sec:max:X\] and the hyperbolic shape for the position vs. time for the large ball in the large mass ratio limit is derived in Sec. \[Sec:xvs.t\]. In Sec. \[Sec:(V,1/x)circle\] we show that the phase portrait of the heavy particle in velocity - inverse position coordinates takes a circular form. In Sec. \[Sec:exact solution\] we provide an explicit solution for the positions and velocities as a function of the number of collision and introduce the exact relation between the moment of time $t$ and the collision number $n$. In Sec. \[Sec:approximate solution\] we derive an approximate relation between $t$ and $n$ which is expressed by a simple formula. The relation between the terminal collision and the parity of the digits is discussed in Sec. \[Sec:terminal collision\]. In Sec. \[Sec:physical realizations\] we discuss different systems to which the Galperin method can be applied. We generalize the point-size billiard to finite-size spheres (“hard rods”) in \[Sec:hard rods\]. Other equivalent systems include a single-ball billiard in a wedge (Sec. \[Sec:billiard\]) and a four-ball chain (\[Sec:four ball chain\]) with appropriate masses and initial conditions. In Sec. \[Sec:error\] we introduce the concept of systematic error and analyze the possible differences between digits generated by the Galperin billiard method and the usual methods of expressing the number $\pi$ in some base. In Sec. \[Sec:integer bases\] we address the case of integer bases. In Sec. \[Sec:integer representation\] we start by reviewing usual methods which might be employed for representing a number in an integer base. In Sec. \[Sec:degeneracy\] we consider the case of identical balls which should be treated separately from the other cases due to the presence of a degeneracy. As relevant examples of integer bases we consider the cases of decimal (Sec. \[Sec:Decimal base\]), binary and ternary (Sec. \[Sec:binary base\]) bases. We conclude this section by discussing in Sec. \[Sec:experiment\] which bases are the most appropriate to carrying out a real experiment. Section \[Sec:Non-integer bases\] is dedicated to non-integer base systems. In Sec. \[Sec:representation\] we briefly overview how a number can be represented in such number systems. Examples of non-integer base representation are reported in Sec. \[Sec:irrational base\] and include a compelling case of representing the number $\pi$ in a system base of $\pi$. Finally, we draw the conclusions in Sec. \[Sec:conclusions\].
Galperin billiard method\[Sec:Galperin billiard method\]
========================================================
In this Section we introduce the Galperin billiard model and review how it can be used to calculate the digits of $\pi$. As a general rule, we assign capital letters to the variables of the heavy ball and lower-case letters to those of the light ball.
Model \[Sec:model\]
-------------------
The system consists of two balls of the same size but of a different mass running along a straight line and a hard wall, located at the origin. We assume that the positions of the heavy and the light balls satisfy $X<x<0$ at any moment of time. The system is unrestricted on one side (semifinite billard). The heavy ball with mass $M$ is thrown against the stationary light ball of mass $m<M$ with some initial velocity $V_0>0$, its precise value being irrelevant for the total number of collisions. The initial position of the light ball $x_0$ defines the length scale for the rest of the processes. The mass of the heavy ball is chosen as $$\begin{aligned}
M = b^{2N}m\;,\end{aligned}$$ with parameter $b$ (integer or not) referred to as the [*base*]{} and an integer parameter $N$ which we call [*mantissa*]{} which eventually will define the number of obtained digits. Once the collision sequence is started, one counts the total number of collisions denoted as $\Pi(b,N)$. All collisions are considered to be elastic so that the total kinetic energy is conserved.
Number of collisions\[Sec:Number of collisions\]
------------------------------------------------
It will be demonstrated in this Section that the velocities of the balls after a certain number of collisions and the total number of collisions can be found explicitly by using conservation laws. Instead, the ball positions cannot be expressed by a simple expression although they can be obtained by integrating the equations of motion.
The kinetic energy $T$ is conserved during elastic ball-ball and ball-wall collisions, $$\begin{aligned}
\frac{1}{2}M V^2 + \frac{1}{2}m v^2 = T \;.
\label{Eq:concervation law:Ekin}\end{aligned}$$ For two balls, energy conservation law (\[Eq:concervation law:Ekin\]) can be given a geometrical interpretation as a mathematical equation defining the shape of an ellipse. Instead of the particle coordinates $X$ and $x$ it is convenient to introduce [*billiard variables*]{}[@SinaiBook] defined as $$\begin{split}
Y &= \sqrt{M} X,\\
y &= \sqrt{m} x
\end{split}
\label{Eq:billiard variables}$$ with the scope of reducing the ellipse to a circle. The billiard velocities (or configuration speed in Ref. [@GalperinBook]) are defined as time derivative of the position (\[Eq:billiard variables\]) and are also scaled with the square root of balls’ masses $$\begin{aligned}
\nonumber
W&=&\frac{dY}{dt}=\sqrt{M}\frac{dX}{dt}=\sqrt{M}\cdot V \\
w&=&\frac{dy}{dt}=\sqrt{m}\frac{dY}{dt}=\sqrt{m}\cdot v
\label{Eq:billiard velocities}\end{aligned}$$
Energy conservation law (\[Eq:concervation law:Ekin\]) expressed in billiard velocities (\[Eq:billiard velocities\]) reads as $$W^2(t) + w^2(t) =
2T \; ,
\label{Eq:concervation law:circle}$$ with the geometric position of allowed values of $W(t)$ and $w(t)$ forming a circle and the square of its radius defined by twice the kinetic energy of the system. A vector of velocities, defined as $\mathbf{w}=(W,w)$, rotates on the circle forming some angle $\phi$ with the horizontal axis. The goal is to find the values of the angle after the first collision, $\phi_1$, second collision, $\phi_2$, and so on which will completely define the velocities. In order to find out how angle $\phi$ changes during collisions we will analyze how the momentum is changed. There are two types of the collisions: (i) ball—ball (ii) ball—wall.
In the first type of collision the total momentum of the system is conserved $$MV + mv = const\;,
\label{Eq:concervation law:momentum}$$ while momentum of each individual ball is changed. It is convenient to recast the constraint (\[Eq:concervation law:momentum\]) in a vector form as $$MV + mv = \sqrt{M}\cdot W+\sqrt{m}\cdot w= \\
=\begin{pmatrix}
\sqrt{M}\\
\sqrt{m}
\end{pmatrix}\cdot
\begin{pmatrix}
W\\
w
\end{pmatrix} = |\mathbf{m}| \cdot |\mathbf{w}|\cos\varphi = const
\label{Eq:rotation}$$ where we have introduced so far an unknown angle of the rotation $\varphi$ and a vector $\mathbf{m} =(\sqrt{M},\sqrt{m})$ with its elements corresponding to the square root of each mass. Since the mass vector $\mathbf{m}$ does not change and according to Eq. (\[Eq:concervation law:circle\]) the absolute value of the velocity vector $\mathbf{w}$ remains invariant, one immediately realizes that $\cos\varphi$ has to remain constant. Therefore, during a collision the angle between the two vectors changes from $\varphi$ to $-\varphi$, keeping the value of $\cos\varphi$ constant. The angle $\phi$ between the vector $\mathbf{w}$ and the horizontal axis is incremented by $2\varphi$ during each ball-ball collision. The specific value of $\varphi$ depends on the mass ratio, as will be discussed later.
In the second type of collisions, when the light ball hits the wall, its velocity gets inverted $w \to -w$ and the momentum of the system changes. The vector of velocities gets flipped vertically, $\left(W,w\right)\to \left(W,-w\right)$ and the angle between the vector $\mathbf{w}$ and the horizontal axis is reflected, $\phi \to -\phi$.
Using two rules, describing ball-ball and ball-wall collisions, it is straightforward to obtain the angle $\phi_{n}$ describing the velocities after $n$ collisions. Typical changes of the vector $\mathbf{w}$ are depicted in Fig. \[Fig:energetic circle\] and can be summarized as follows:
- before any collision has happened, the light particle is at rest, $v = w = 0$, as shown with the horizontal vector with $\phi_0 = 0$
- after the fist ball-ball collision, vector $\mathbf{w}$ is rotated by $2\varphi$, resulting in $\phi_1 = 2\varphi$
- after the fist ball-wall collision, vector $\mathbf{w}$ is flipped vertically, resulting in $\phi_2 = -2\varphi$
- after the second ball-ball collision, vector $\mathbf{w}$ is rotated by $2\varphi$, resulting in $\phi_3 = 4\varphi$
- after the second ball-wall collision, vector $\mathbf{w}$ is flipped vertically, resulting in $\phi_2 = -4\varphi$
- after $(n+1)/2$ ball-ball collisions $\phi_n = (n+1)\varphi$
- after $n/2$ ball-wall collisions $\phi_n = -n \varphi$.
During each ball-ball collision, the velocity $V$ of the heavy ball is changed by a negative amount, eventually stopping and reversing the heavy ball. After the angle has crossed $\pi/2$ position, the velocity of the heavy mass becomes negative (the ball is moving away from the wall) and its absolute value is increased with each consecutive collision. Collisions continue until the closest angle to $\pi$ position is reached. After that the iterations end, as continuing further would result in a decrease of the velocity of the heavy mass, which is physically impossible. The total number of collisions, labeled by $\Pi$, is then given by $$\Pi = {\mathop{\rm int}\nolimits}\left[\frac{2\pi}{\phi}\right]\;,
\label{Eq:Pi(phi)}$$ where function ${\mathop{\rm int}\nolimits}[x]$ denotes the integer part of $x$, the greatest integer less than $x$ or equal to it.
[0.45]{} ![A characteristic example of the dependence of the vector of velocities on collision number $n$. Left panel, vector of velocities $\mathbf{v} = (V, v)$ forming an ellipse; right panel, vector of rescaled velocities $\mathbf{w} = (V, v)$ forming a circle. The shown data is obtained for $b=3$ and $N=1$. []{data-label="Fig:energetic circle"}](Fig3V.pdf "fig:"){width="\linewidth"}
[0.45]{} ![A characteristic example of the dependence of the vector of velocities on collision number $n$. Left panel, vector of velocities $\mathbf{v} = (V, v)$ forming an ellipse; right panel, vector of rescaled velocities $\mathbf{w} = (V, v)$ forming a circle. The shown data is obtained for $b=3$ and $N=1$. []{data-label="Fig:energetic circle"}](Fig4W.pdf "fig:"){width="\linewidth"}
Let us find out how the pivot angle $\varphi$ is related to the mass ratio. Suppose there is a ball-ball collision which corresponds to a rotation of $\mathbf{v}$ vector from $\alpha - \varphi$ to $\alpha+\varphi$. The change in the momentum of the heavy mass is $-\sqrt{M}(\cos(\alpha+\varphi)+\cos(\alpha-\varphi)) = -2\sqrt{M}\cos\alpha\cos\varphi$, while the change in the momentum of the light mass is $\sqrt{m}(\sin(\alpha+\varphi)-\sin(\alpha-\varphi))=2\sqrt{m}\sin\varphi\cos\alpha$. The sum of the two should be equal to zero in the ball-ball collision, resulting in a condition which is independent of the actual value of $\alpha$, $$\begin{aligned}
\tan \varphi = \sqrt{\frac{m}{M}}\;.\end{aligned}$$ That is the pivot angle is fully determined by the square root of the mass ratio.
Hence, the angle between the velocity vectors during a ball-ball collision will be: $$\phi=2\varphi=2\cdot \arctan\left(\frac{\sqrt{m}}{\sqrt{M}}\right)=2\cdot \arctan(b^{-N})
\label{Eq:phi}$$ Consequently, the number of collisions (\[Eq:Pi(phi)\]) can be explicitly evaluated as a function of parameters $b$ and $N$ as $$\Pi={\mathop{\rm int}\nolimits}\left[\frac{2\pi}{2\arctan(b^{-N})}\right]={\mathop{\rm int}\nolimits}\left[\frac{\pi}{\arctan(b^{-N})}\right] \;.
\label{Eq:pi:acrctg}$$ Moreover, for large base $b$ and large mantissa $N$ the argument of the inverse tangent function is small, $x = b^{-N}\ll 1$, and the inverse tangent function can expanded as $\arctan (x) \approx x$, resulting in an elegant expression $$\Pi(b,N)\approx {\mathop{\rm int}\nolimits}\left[\frac{\pi}{b^{-N}}\right] \;.
\label{Eq:hurbilketa}$$ This equation provides the basis for expressing the number $\pi$ in systems with integer and non-integer bases.
In Ref. [@Davis2015] the new velocities after a one round of ball-ball and ball-wall collisions are found in terms of an eigenproblem of a system of two linear equations. The eigenvalues $e^{i2\varphi}$ correspond to a rotation by angle $2\varphi$ as can be seen in Fig. \[Fig:energetic circle\].
Ball trajectory\[Sec:ball trajectory\]
======================================
Solving the equations of motion\[Sec:equations of motion\]
----------------------------------------------------------
In Sec. \[Sec:Number of collisions\], it was shown that the total number of collisions can be explicitly obtained from the conservation laws resulting in Eq. (\[Eq:hurbilketa\]), and it does not depend on the exact positions of the balls and the initial velocity of the incident ball. Here we outline how the trajectory of the balls can be obtained.
First, one has to define the initial conditions (positions and velocities) and then integrate the equations of motion. Let $X_0$ and $x_0$ ($X_0<x_0<0$) be the initial coordinates of the heavy and light balls, and $V_0>0$ and $v_0=0$ be their initial velocities. As the velocities change only at contact, the balls move with constant velocities between the collisions. The velocities change according to the rules (the energy and momentum conservation laws) provided in Sec. \[Sec:Number of collisions\]. All odd collisions, $n=2k+1$, correspond to balls hitting each other, while even collisions, $n=2k$, to the light ball hitting the wall. Let $X_n$, $x_n$, $V_n$, $v_n$ be the coordinates and velocities of the heavy and light balls at the time $t_n$ of the $n$-th collision, respectively. Then the time between the consecutive collisions is $$\begin{split}
\tau_{2k} &= \frac{x_{2k}-X_{2k}}{V_{2k}-v_{2k}} \\
\tau_{2k-1} &= - \frac{x_{2k-1}}{v_{2k-1}},
\end{split}
\label{Eq:tau(n):exact}$$ where $\tau_{2k} = t_{2k+1}- t_{2k}$ is the time interval passed between ball-ball $2k$ and the subsequent ball-wall $2k+1$ collision, and $\tau_{2k-1} = t_{2k} - t_{2k-1}$ is the time interval between ball-wall $2k-1$ and ball-ball $2k$ collisions. The time moment of the $n$-th collision can be calculated as the sum of the preceding time intervals as $$t_n = \sum\limits_{l=1}^{n-1} \tau_l.
\label{Eq:t(n):exact}$$
The solution to the equations of motion can be expressed as the following iterative formulas for a ball-ball collision ($n=2k+1, k=0,1,\dots$) $$\begin{split}
X_{2k+1} &= X_{2k} + (X_{2k}-x_{2k}) \frac{V_{2k}}{v_{2k}-V_{2k}} \\
x_{2k+1} &= x_{2k} + (X_{2k}-x_{2k}) \frac{v_{2k}}{v_{2k}-V_{2k}} \\
V_{2k+1} &= \frac{M-m}{m+M} V_{2k} + \frac{2m}{m+M} v_{2k} \\
v_{2k+1} &= \frac{2M}{m+M} V_{2k}+ \frac{m-M}{m+M} v_{2k},
\end{split}
\label{Eq:equation of motion:BB}$$ and for a ball-wall collision ($n=2k, k=1,2,\dots$) $$\begin{split}
X_{2k} &=X_{2k-1}-\frac{V_{2k-1}}{v_{2k-1}} x_{2k-1} \\
x_{2k} &=0 \\
V_{2k} &=V_{2k-1} \\
v_{2k} &= -v_{2k-1}. \\
\end{split}
\label{Eq:equation of motion:BW}$$ The iterative process stops when one of the following equivalent conditions holds: (i) $X_n>0$, (ii) $V_n$ is not monotone and starts decreasing after the point of return, see Sec. \[Sec:point of return\] for the discussion on the point of return, (iii) after a ball-ball collision $v_n<0$, which physically means that the light ball goes to $-\infty$.
In Fig. \[Fig:positions\] we display a typical example of heavy and light ball trajectories, $(t_n, X_n)$ and $(t_n, x_n)$, calculated by iteratively solving Eqs. (\[Eq:equation of motion:BB\]-\[Eq:equation of motion:BW\]) for $b=2$ and different $N$.
![Distance of the heavy and light balls from the wall as a function of time for base $b=2$ and different values of $N$ (in arbitrary units). Solid lines and solid symbols, heavy ball $X$; dashed lines and open symbols, light ball $x$. []{data-label="Fig:positions"}](FigTrajectory.pdf){width="0.6\columnwidth"}
Minimal and maximal velocities\[Sec:max:V\]
-------------------------------------------
In this Section we discuss which smallest or largest value of velocity and momentum the balls might have. The characteristic value of velocity is set by the initial velocity $V_0$ of the heavy ball. The energy conservation law (\[Eq:concervation law:Ekin\]) relates the velocities of the heavy and light balls, the maximal velocity of the light ball, $v_{max}$, which is reached close to the point of return, where the heavy ball stops, $V=0$. The maximal possible velocity of the light ball is $$v_{max} \approx \sqrt{\frac{M}{m}}V_0 = b^N V_0\;.
\label{Eq:v:max}$$ The light ball might move much faster than the heavy ball. The larger is the mass ratio, the more pronounced is the effect. Accordingly, the maximal momentum of the heavy $P_{max}$ and light $p_{max}$ balls are $$\begin{aligned}
\label{Eq:P:max}
P_{max} &=& M V_0 \\
p_{max} &\approx& m v_{max} = b^{-N} P_{max}.
\label{Eq:p:max}\end{aligned}$$
It should be noted that while $P_{max}$ is always reached as it corresponds to the initial momentum before the first collision happens, $p_{max}$ is obtained by assuming that the heavy ball completely stops. Due to finite discretization of the angle $\varphi$, the heavy does not necessarily comes to a complete stop, although the larger the mass ratio $M/m$, the better Eqs. (\[Eq:v:max\],\[Eq:p:max\]) are satisfied.
Differential equation for motion close to the point of return\[Sec:point of return\]
------------------------------------------------------------------------------------
In this section we derive a differential equation describing the motion of the slow (heavy) particle close to the point of return and find its solution.
The direction of the light ball is inverted at each collision, refer to Fig. \[Fig:positions\]. It is positive after the first or any odd collision, $n=2k+1$, and is negative after an even number of collisions, $n=2k$, $$\begin{aligned}
w^{(2k)}&=& -w_{max} \sin(k\phi)\\
w^{(2k+1)}&=&w_{max} \sin(k\phi),
\label{Eq:w(n)}\end{aligned}$$ where $w_{max}$ is the maximal value of the scaled velocity that the light ball might have. Instead, the velocity of the heavy ball is gradually reduced and is inverted at a certain moment. Also, the velocity of the heavy ball is not changed when the light ball hits the wall $$\begin{aligned}
W^{(2k)}&=& W_{max} \cos(k\phi)\\
W^{(2k+1)}&=&W_{max} \cos(k\phi)
\label{Eq:W(n)}\end{aligned}$$ with $W_{max} = w_{max}$. The point of return is signaled by the inversion of the velocity of the heavy ball and is closely defined by the condition that the cosine function in Eq. (\[Eq:W(n)\]) has a node, $W=0$, $$k_{inversion}\phi\approx\frac{\pi}{2}\;.
\label{Eq:kinversion}$$ We assume that the “discretization” $\varphi$ is rather small and the light ball makes a large number of collisions. In this situation, Eq. (\[Eq:kinversion\]) predicts well the collision index $k_{inversion}$ corresponding to the point of return. At that moment, the heavy ball stops moving and reaches the smallest distance $X_{min}$ from the wall.
Counting the number of collisions $k'$ from the return point (primed variables are counted from the point of return), $k' = k - k_{inversion}$, the trigonometric functions in Eqs. (\[Eq:w(n)\]-\[Eq:W(n)\]) can be expanded leading to $$\begin{aligned}
w^{(2k')}
&= -w_{max} \sin(\frac{\pi}{2}+k'\phi)
&= -w_{max} \cos(k'\phi)
\approx w_{max}\\
W^{(2k')}
&= W_{max} \cos(\frac{\pi}{2}+k'\varphi)
&= -W_{max} \sin(k' \varphi)
\approx -W_{max}k'\varphi
\label{Eq:w(n')}\end{aligned}$$ In terms of the velocities we obtain $$\begin{aligned}
v^{(2k')} &\approx& - v_{max}\\
V^{(2k')} &\approx& - V_{max}k'\varphi
\label{Eq:v(n')}\end{aligned}$$ As at this point the absolute value of the velocity of the light ball practically does not change and the heavy ball is almost not moving, the time $$\begin{aligned}
\tau_{inversion} = \frac{|X_{min}|}{v_{max}}
\label{Eq:tau:inversion}\end{aligned}$$ between consecutive collisions is constant. Velocity of the heavy ball is changed only at ball-ball collisions with time $\Delta t'$ between consecutive BB collisions equal to $$\begin{aligned}
\Delta t' = 2\tau_{inversion} = \frac{2|X_{min}|}{v_{max}} = const\;.
\label{Eq:Delta t}\end{aligned}$$ The time $t'$ passed after the return point is $t'=k'\Delta t'$. The position of the heavy ball changes as $$\begin{aligned}
\frac{\Delta X}{\Delta t'}
= -k' V_{max}\varphi
\approx
- \frac{V_{max} v_{max} \varphi}{|X_{min}|} t'
\;,
\label{Eq:return:discrete}\end{aligned}$$ where the negative sign shows that the heavy ball goes to the left direction, away from the wall.
The heavier is the ball, the smaller is its minimal distance from the wall, $X_{min}$, and the larger is the maximal velocity of the light ball, $v_{max}$. According to Eq. (\[Eq:Delta t\]) both tendencies make time $\Delta t'$ between collisions smaller. By treating it as an infinitesimal increment $dt'$ in Eq. (\[Eq:return:discrete\]), we find the differential equation describing the trajectory of the heavy ball $$\begin{aligned}
\frac{dX}{dt'}
= - \frac{V_{max} v_{max} \varphi}{|X_{min}|} t'
\;.
\label{Eq:return:differential}\end{aligned}$$ Equation (\[Eq:return:differential\]) can be explicitly integrated, giving the trajectory of the heavy particle close to the return point in the continuous approximation, $$\begin{aligned}
X(t')
= X_{min} - \frac{V_{max} v_{max} \varphi}{2X_{min}} (t')^2
\approx
X_{min} - \frac{V_{max}^2}{2|X_{min}|}(t')^2
\;.
$$ where we have used Eqs. (\[Eq:phi\],\[Eq:v:max\]) and made an expansion of $\phi$ assuming $M\gg m$.
The relative displacement of the heavy ball is expressed as $$\begin{aligned}
\frac{X(t')}{|X_{min}|}
= 1 + \frac{1}{2} \left(\frac{V_{max} t'}{X_{min}}\right)^2\;.
\label{Eq:return:parabola}\end{aligned}$$
![Distance of the heavy ball close to the return point, for $b=2$ and $N=3$. Solid symbols, heavy ball $X$; dashed line, limit of an infinitely massive ball, Eq. (\[Eq:return:heavyball\]); solid line, parabola defined by Eq. (\[Eq:return:parabola\]). []{data-label="Fig:parabola"}](FigParabola.pdf){width="0.6\columnwidth"}
Figure \[Fig:positions\] shows a characteristic example of the time evolution of the position of the heavy ball for $b=2$ and different values of $N$. For $N=0$ there are only three collisions, the first ball-ball collision stops the first ball; ball-wall collision inverts the velocity of the second ball, while the heavy ball remains immobile; the last ball-ball collision sends the first ball back. For $N=1$ no clear regular structure is visible. For $N=2$ a parabola, predicted by Eq. (\[Eq:return:parabola\]), starts being formed. For $N=3$, shown in Fig. \[Fig:parabola\] the parabolic dependence becomes clearly visible.
Analytic vs non-analytic trajectory envelope\[Sec:non-analytic\]
----------------------------------------------------------------
The trajectory is a piece-wise function, as the velocities are constant between the collisions. The resulting “edges” in the trajectory are clearly visible for the small number of collisions, similarly to $N=0$ case, schematically shown in Fig. \[Fig:non-analytic trajectory\]a. It is interesting to note that such trajectory is described by a [*non-analytic*]{} function, as it experiences kinks and the first derivative is undefined at any of the collision points. However, when the number of collisions is large, the envelope of the trajectory becomes smooth and is described by an [*analytic*]{} function, which close to the point of return has a parabolic shape (\[Eq:return:parabola\]). Such a parabolic dependence is schematically shown in Fig. \[Fig:non-analytic trajectory\]b. In the limit of an infinitely heavy ball ($N\to\infty$ and $M\to\infty$), the trajectory again becomes non-analytic with a kink in the point where the heavy ball hits the wall, $$\begin{aligned}
X(t') = V_{max}|t'|
\label{Eq:return:heavyball}\end{aligned}$$ shown in Fig. \[Fig:non-analytic trajectory\]c with two straight lines.
[0.2]{} ![ Schematic shape of the trajectory of the heavy particle close to the point of return, given by (a) a non-analytic function for $N=0$; (b) analytic function (\[Eq:return:parabola\]); (c) non-analytic function (\[Eq:return:heavyball\]) in $N\to\infty$ limit. []{data-label="Fig:non-analytic trajectory"}](FigNonAnalytic1.pdf "fig:"){width="\linewidth"}
[0.2]{} ![ Schematic shape of the trajectory of the heavy particle close to the point of return, given by (a) a non-analytic function for $N=0$; (b) analytic function (\[Eq:return:parabola\]); (c) non-analytic function (\[Eq:return:heavyball\]) in $N\to\infty$ limit. []{data-label="Fig:non-analytic trajectory"}](FigNonAnalytic2.pdf "fig:"){width="\linewidth"}
[0.2]{} ![ Schematic shape of the trajectory of the heavy particle close to the point of return, given by (a) a non-analytic function for $N=0$; (b) analytic function (\[Eq:return:parabola\]); (c) non-analytic function (\[Eq:return:heavyball\]) in $N\to\infty$ limit. []{data-label="Fig:non-analytic trajectory"}](FigNonAnalytic3.pdf "fig:"){width="\linewidth"}
Adiabatic approximation\[Sec:adiabatic approximation\]
------------------------------------------------------
From the point of view of Hamiltonian systems, the problem of two balls has two degrees of freedom, namely two positions $X$, $x$ while momenta $P = MV$ and $p=mv$ are conjugate variables. As it was discussed in Sec. \[Sec:point of return\], when the heavy ball approaches the point of return, it slows down while the light ball wildly oscillates between the heavy ball and the wall. This separates the scales into [*fast*]{} and [*slow*]{} variables so that during a single oscillation of a light ball, the position of the heavy ball is only slightly changed. It was argued by Kapitza [@Kapitza51UFN; @Kapitza51JETP] in his work on driven pendulum (Kapitza pendulum) that by averaging over the fast variables it might be possible to simplify the problem and provide a solution if the separation of scales is large enough. In our case the parameter which defines the separation of scales is the mass ratio $M/m = b^{2N}$, so for any base $b$ by increasing $N$ the needed condition $M/m \gg 1$ is well satisfied. The systems with different scales can be studied in the theory of adiabatic invariants[@ArnoldKozlovNeishtadtBook].
It is useful to analyze the $(p,x)$ portrait of the system, corresponding to the fast variables. A typical example is shown in Fig. \[Fig:xp\]. After the ball-ball collision, (for example, $n=1$), the light ball moves with a constant momentum $p$ until it hits the wall. This results in a horizontal line with some momentum $p$ and $0<x/x_0 < X_1/x_0$. During the ball-wall ($n=2$) collision the momentum of the light mass is inverted, resulting in a vertical line $x=0$, $p\to -p$. After that the light ball travels with constant momentum until it hits the heavy ball ($n=3$), corresponding to a horizontal line at $-p$ from $0<x/x_0 < X_3/x_0$. At the next ball-ball collision, the velocity of the light particle inverts the sign but its absolute value is slightly changed due to a small but finite momentum transfer from the heavy ball. During a single cycle (or “period”) consisting of four collisions the light ball draws an almost closed rectangle. The larger is the mass of the heavy ball and the smaller is its velocity, the more similar is the trajectory during a cycle to a closed rectangle.
![(x,p) portrait for $b=10$ and $N=1$; Red symbols, light particle during ball-wall ($x=0$) and ball-ball ($x\ne 0$) collisions. Green thick lines, constant action curve defined by Eq. (\[Eq:action\]). Blue thin lines, trajectory. Index $n=1,2,3,\cdots$ denotes the state after $n$ collisions while primed index $n'$ correspond to an intermediate state in which the velocity of the light ball is not yet reflected. Area, covered by the trajectory between two consecutive collusion of the same type (BB or BW) define the action (\[Eq:action\]). []{data-label="Fig:xp"}](Figxp.pdf){width="0.6\columnwidth"}
The area covered by the light particle during a cycle in $(p,x)$ space has units of \[energy $\times$ sec\] and is called [*action*]{} $I$, defined as $$\begin{aligned}
I = \frac{X p}{2\pi} \,,
\label{Eq:action}\end{aligned}$$ where $p$ is the maximal momentum of the light particle and $X$ is its maximal distance from the wall (defined by the position of the heavy particle) during a single cycle. Within the adiabatic approximation the action is conserved, implying the following relation between the momentum of the light particle $p$ and the position of the heavy particle $X$ $$\begin{aligned}
p = \frac{2\pi I}{X}.
\label{Eq:adiabatic:p(X)}\end{aligned}$$
It is shown in Ref. [@Gorelyshev2006] that for times of the order of $\varepsilon^2$, action (\[Eq:action\]) is conserved with accuracy $\varepsilon$ where $\varepsilon = \sqrt{m/M}$ is treated as a small parameter. In the same limit the Hamiltonian can be written as $$\begin{aligned}
H = \frac{P^2}{2M} + \frac{\pi^2 I^2}{2mX^2}\;.
\label{Eq:adiabatic:H}\end{aligned}$$ At the point of return the heavy particle has zero momentum and the energy can be expressed in terms of the minimal distance $X_{min}$ between the heavy particle and the wall, $$\begin{aligned}
H = \frac{\pi^2 I^2}{2mX_{min}^2}\;.
\label{Eq:adiabatic:Xmin}\end{aligned}$$
From Eqs. (\[Eq:adiabatic:H\]-\[Eq:adiabatic:Xmin\]) it follows that the dependence of the momentum $P$ of the heavy particle on its inverse coordinate $1/X$ has the semicircular form [@Gorelyshev2006], $$\begin{aligned}
P = \frac{\pi I}{X_{min}} \sqrt{1 - \frac{X_{min}^2}{X^2}}\;,
\label{Eq:adiabatic:P}\end{aligned}$$ which we will address in more detail in Sec. \[Sec:(V,1/x)circle\].
Variables $I$ and $\phi_{\rm phase}$ are conjugate, and time derivative of the phase can be obtained from Hamiltonian (\[Eq:adiabatic:H\]) as $\dot\phi_{\rm phase} = dH / dI$. The integration over the time gives the final phase after all collisions have happened as $\phi_{\rm phase}^{\rm final} = \pi^2 \sqrt{M}/\sqrt{m}$ [@Gorelyshev2006]. During each cycle there are two collisions (BB and BW) and the phase changes by $2\pi$, so the total number of collisions $\Pi$ can be infer as $\phi_{\rm phase}^{\rm final} = \Pi \pi$, resulting in $\Pi = \pi / \varphi + O(\varphi)$ where $\varphi \approx \sqrt{m} /\sqrt{M}$. This formula should be compared with Eq. (\[Eq:Pi(phi)\]) and, indeed, correctly relates total number of collisions $\Pi$ with $\pi$. At the same time it is not a priori obvious that the adiabatic approximation should be precise far from the return point, that is for times $t\gg \varepsilon^2$, especially at the time of the final collision. Indeed, it might be observed in Fig. \[Fig:xp\] that while action (\[Eq:action\]) is a good adiabatic invariant close to the return point (shown with thick green line), the first few collisions ($n=1; 3; \cdots$) are quite off. In Sec. \[Sec:invariants\] it will be shown that in the present problem it is possible to find two invariants (for BB and BW collisions), which coincide close to the point of return with adiabatic invariant (action) given by Eq. (\[Eq:action\]), and, in particular, this clarifies why the adiabatic approximation leads to the correct number of collisions even if the region of applicability of the approximation is violated.
Finally we note that the time dependence of the phase $\phi_{\rm phase}(t)$ is related to the time dependence of the collision number $n(t)$ according to $\phi_{\rm phase}(t_n) = \pi n(t_n)$. In the continuous limit of many collisions, the phase increases as an inverse tangent function, as shown in Fig. \[Fig:n\] of Sec. \[Sec:approximate solution\] below.
Action invariants\[Sec:invariants\]
-----------------------------------
We have seen in Sec. \[Sec:adiabatic approximation\], that action (\[Eq:action\]) is an adiabatic invariant and is not changed in the vicinity of the return point. Consequently, the iterative solution (\[Eq:equation of motion:BB\]-\[Eq:equation of motion:BW\]) to the equations of motion should also preserve the action $I$. The change of coordinates within one cycle of collisions is obtained by applying consequently ball-ball (\[Eq:equation of motion:BB\]) and ball-wall (\[Eq:equation of motion:BW\]) movements, and it can be straightforwardly verified that the action remains constant for a cycle which starts and ends with a ball-wall collision, $n=2k$, $$X_{2k} v_{2k} = \frac{\pi I}{m} = const\;.
\label{Eq:invariant:BW}$$ Importantly, action (\[Eq:invariant:BW\]) is [*always*]{} conserved on a ball-wall cycle and not only close to the point of return. The absolute value of the velocity $v_{2k}$ of the light particle is constant on the paths BB-BW and BW-BB between consecutive collisions, while the edge position of the light particle is displaced from $x_{2k-1}$ to $x_{2k+1}$, see Fig. \[Fig:xp\]. As a result, the trajectory during a single BB-BW-BB loop is not closed, and it is rather natural that the exact invariant (\[Eq:invariant:BW\]) should be satisfied for some value of $x$, lying between $x_{2k-1}$ and $x_{2k+1}$. The position of the heavy particle $X$ coincides with $X_{2k-1} = x_{2k-1}$ and $X_{2k+1}=x_{2k+1}$ and $x = X_{2k}$, indeed, lies between $x_{2k-1}$ and $x_{2k+1}$. One also might note that a simple average position $x = (x_{2k-1}+x_{2k+1})/2$ does not lead to an invariant.
It was discovered by Weidman [@Weidman2013] that there also exists a second invariant which remains unchanged during a ball-ball collision, $n=2k+1$, given by $$X_{2k+1} (V_{2k+1}-v_{2k+1}) = \frac{\pi I}{m} = const\;.
\label{Eq:invariant:BB}$$ Furthermore, it can be shown[@Weidman2013] from the equations of motion (\[Eq:equation of motion:BB\]-\[Eq:equation of motion:BW\]) that the action $I$, entering into Eqs. (\[Eq:invariant:BW\]-\[Eq:invariant:BB\]), can be expressed in terms of the initial conditions as $$I = \frac{|x_0| V_0 m}{\pi}
\label{Eq:action invariant}$$
Figure \[Fig:Xv\] shows the $(X,v)$ portrait of the system. By using Eq. (\[Eq:action invariant\]), the ball-wall invariant (\[Eq:invariant:BW\]) reduces to an elegant expression, $X_{2k} v_{2k} = - x_0 V_0$ shown with a solid line. It can be appreciated that the ball-wall invariant is, indeed, conserved for any BW collision. Instead, the mirrored line, $X_{2k} v_{2k} = + x_0 V_0$, describes correctly the ball-ball process only close to the point of return. Indeed, there it was derived within the adiabatic approximation, as given by Eq. (\[Eq:adiabatic:p(X)\]). Formally, the error arising within the adiabatic approach can be explicitly seen from the BB invariant (\[Eq:invariant:BB\]), which reduces to $X_{2k+1}v_{2k+1} = const$ only when $V_{2k+1} = 0$, that is exactly at the point of return.
![$(X,v)$ portrait for $b=10$ and $N=1$; Red symbols, heavy particle during ball-ball ($v>0$) and ball-ball ($v < 0$) collisions. Green thick lines, ball-wall invariant (\[Eq:invariant:BW\]), corresponding to action (\[Eq:action invariant\]). []{data-label="Fig:Xv"}](FigInvariant1.pdf){width="0.6\columnwidth"}
Finally, it is worth recalling that during ball-ball collision, $x$ and $X$ coordinates obviously coincide and BB invariant (\[Eq:invariant:BB\]) is applicable not only for the coordinate of the heavy particle, but also to the light one, $$x_{2k+1} (V_{2k+1}-v_{2k+1}) = \frac{2\pi I}{m} = const\;.
\label{Eq:invariant:BB:x}$$
Unfolding the trajectory\[Sec:Unfolding\]
-----------------------------------------
The trajectory of the balls in the phase space can be given a simple geometrical interpretation which also clarifies one more time how the number of collisions is related to the opening angle[@Galperin2003; @TabachnikovBook2005]. The original particle coordinates are restricted to the region $0\leq |x| \leq |X|$, where boundary $x=0$ corresponds to the light ball hitting the wall and $x=X$ the ball-ball collision, see Fig. \[Fig:unfolding\]a. The opening angle is $45\degree$ but the reflections do not obey the laws of optics, as the incident angle differs from the angle of reflection and neither is the velocity $V$ conserved. Instead, the opening angle in billiard coordinates $Y$ and $y$ is equal to $\varphi$, see Fig. \[Fig:unfolding\]b. Now the absolute value of the scaled velocity is conserved and reflections obey the optical laws. In this way the original two-body problem is mapped to a problem of a single ball moving in a wedge with opening angle $\varphi$ with specular reflections from the mirrors. It was demonstrated in Sec. \[Sec:Number of collisions\] that the vector $\mathbf{m} =(\sqrt{M},\sqrt{m})$ is preserved during any ball-ball collision and ball-wall collisions do not affect the trajectory of the heavy ball. The boundary line of the configuration space corresponds to the collision condition $X=x$, written in billiard coordinates as $Y / \sqrt{M} - y / \sqrt{m} = 0$. The scalar product of its normal vector $(1/\sqrt{M}, -1/\sqrt{m})$ and the vector $\mathbf{m}$ is equal to zero. It means that in the billiard coordinates, the vector $\mathbf{m}$ is tangent to the boundary line and if the the ricocheting trajectory in the wedge is unfolded, it results in a straight line. In Fig. \[Fig:unfolding\] we show a typical example. It provides a simple geometrical interpretation for the number of collisions as the number of times the opening angle can fit into the maximal angle of $180$ degrees or $\pi$ radian.
![ The trajectory in different phase spaces, (a) original coordinates $0\leq |x| \leq |X|$, (b) variables of billiard in a wedge, $0\leq |y| \leq |Y|\tan\varphi$, (c) unfolded trajectory. The parameters are $b=2$ and $N=1$ and correspond to time-dependent data shown in Fig. \[Fig:positions\]. []{data-label="Fig:unfolding"}](FigXx.pdf "fig:"){width="0.4\columnwidth"} ![ The trajectory in different phase spaces, (a) original coordinates $0\leq |x| \leq |X|$, (b) variables of billiard in a wedge, $0\leq |y| \leq |Y|\tan\varphi$, (c) unfolded trajectory. The parameters are $b=2$ and $N=1$ and correspond to time-dependent data shown in Fig. \[Fig:positions\]. []{data-label="Fig:unfolding"}](FigYy.pdf "fig:"){width="0.4\columnwidth"} ![ The trajectory in different phase spaces, (a) original coordinates $0\leq |x| \leq |X|$, (b) variables of billiard in a wedge, $0\leq |y| \leq |Y|\tan\varphi$, (c) unfolded trajectory. The parameters are $b=2$ and $N=1$ and correspond to time-dependent data shown in Fig. \[Fig:positions\]. []{data-label="Fig:unfolding"}](FigUnfolded.pdf "fig:"){width="0.8\columnwidth"}
Liouville integrability, superintegrability, and maximal superintegrability\[Sec:integrability\]
------------------------------------------------------------------------------------------------
Our system can be shown to be *Liouville integrable*, i.e. it possesses as many exact constants (first integrals) of motion in involution as it has degrees of freedom. For our two-degree-of-freedom system, this implies an existence of only one constant of motion in addition to the total energy [^1]. To identify this conserved quantity, let us consider the system in the billiard coordinates (\[Eq:billiard variables\]). It is represented by a two-dimensional particle of a unit mass moving a wedge of an opening $\varphi = \arctan(\sqrt{m/M})$ as shown in Fig. \[Fig:unfolding\]. In between the collisions, the angular momentum, $L=(Y w - y W)$, is conserved. Upon a ball-wall or a ball-ball collision, the angular momentum changes sign. However, its square, $$L^2 = (Y w - y W)^2 = m M (X v - x V)^2
\label{LSq}$$ remains invariant throughout the evolution.
At the instances of a ball-wall collision, where $x=0$, the invariant (\[LSq\]) is proportional to the square of the adiabatic invariant (\[Eq:invariant:BW\]): $$\begin{aligned}
L^2 \Big|_{2k} = m M (X_{2k} v_{2k})^2 = \pi^2 b^{2N} I^2\,\,.\end{aligned}$$ Likewise, on a ball-ball collision, the angular momentum square assumes a value proportional to the corresponding adiabatic invariant (\[Eq:invariant:BB\]): $$\begin{aligned}
L^2 \Big|_{2k+1} = m M \left(X_{2k+1} (v_{2k+1}-V_{2k+1})\right)^2 = \pi^2 b^{2N} I^2\,\,,\end{aligned}$$ with the same coefficient of proportionality.
For a discrete set of mass ratios with a commensurate opening angle ($\varphi = \pi/q$), $$\begin{aligned}
&
\frac{m}{M} = \tan^2(\pi/q)
\label{kaleidoscopes}
\\
&
n=3,\,4,\,5,\,\ldots
\nonumber
\,\,,\end{aligned}$$ a third constant of motion appears, promoting our system to *superintegrable*; i.e. it will have more functionally independent constant of motion than degrees of freedom. In fact, two-dimensional superintegrable systems are also always *maximally superintegrable*: they have a maximally allowed number of functionally independent conserved quantities that is one less the dimensionality of the phase space, coordinates and velocities combined.
For bounded orbits, the superintegrability manifests itself as a reduction of the dimensionality of the phase space available from a given initial condition. Maximal superintegrability results in closed one-dimensional orbits. For unbounded orbits, the manifestation of the maximal superintegrability is more subtle, but still—as we will see below—tangible.
For the mass ratios (\[kaleidoscopes\]), the wedge in the billiard coordinates $(y,Y)$ depicted in Fig. \[Fig:unfolding\]b acquires an opening of $\pi/q$, with $q \ge 3$ being an integer. In this case, sequences of reflections about the cavity walls form a finite group with order $2q$ known as the reflection group $I_{2}(q)$. As it has been shown in Ref. [@olshanii2015_105005], in such a situation a new constant of motion can be constructed: it is represented by the first nontrivial invariant (or Chevalley) polynomial of the group [@chevalley1955_778; @mehta1988_1083], evaluated on the momentum vector. The constant of motion $J$ produced by this construction in our case is as follows: $$\begin{aligned}
\begin{split}
J
&
= \frac{1}{2 M^{q/2}}\left((W+iw)^{q}+(W-iw)^{q}\right)
\\
&
= \frac{1}{2}
\left(
(V + i \tan(\pi/q) v)^{q} + (V - i \tan(\pi/q) v)^{q}
\right)
\end{split}
\label{J}
\,\,.\end{aligned}$$ Some notable examples include $$\begin{aligned}
&
q=3
&
\frac{m}{M} = 3
&&
J = V^3-9 V v^2
&
\\
&
q=4
&
\frac{m}{M} = 1
&&
J = V^4 - 6 V^2 v^2 +v^4
&
\\
&
q=5
&
\frac{m}{M} = 5 - 2 \sqrt{5}
&&
J = V^5 - 10(5-2\sqrt{5}) V^3 v^2 + 25(9-4\sqrt{5}) V v^4
&
\\
&
q=6
&
\frac{m}{M} = \frac{1}{3}
&&
J = V^6 - 5 V^4 v^2 + \frac{5}{3} V^2 v^4 - \frac{1}{27} v^6
&
\,\,.
$$ Observe that in these examples and in general, even(odd) $q$, produces an even(odd) constant of motion $J$, with respect to the $V\to -V,\,v\to -v$ inversion. This difference between the even and odd cases, will lead to a difference between the maximal superintegrability manifestations between these two cases.
To discuss the consequences of the maximal superintegrability, we will enlarge, temporarily, the set of the initial conditions considered, allowing for a nonzero initial velocity of the light particle. Generally, the allowed sets of incident (in) velocities, i.e. the states where no collisions occurred in the past, would require positive initial velocities ordered according to $$V_{\text{in}} > v_{\text{in}} > 0 \;.
\label{incident_state}$$ Likewise, an outgoing state (out), i.e. a state that does not lead to any collisions in the future, is characterized by negative final velocities ordered according to $$V_{\text{out}} < v_{\text{out}} < 0\;.
\label{outgoing_state}$$ It can be shown that the conservation of energy and the observable $J$, $$\begin{aligned}
T_{\text{out}}& =& T_{\text{in}}\\
J_{\text{out}}& =& J_{\text{in}}\;\,\end{aligned}$$ both being a function of the velocities *only*, restricts the set of the allowed outgoing velocity pairs produced by the given incident pair, to one value only [^2]. Notice that in this case, the outgoing velocities do not depend on the order of collisions: depending on the initial coordinates $X_0<x_0<0$, the first collision in the chain can be represented by either a ball-wall or a ball-ball collision. This independence can be regarded as a classical (as opposed to quantum) manifestation of the so-called Yang-Baxter property [@gaudin1983_book_english; @sutherland2004_book] for the three-body system where the wall is considered a third, infinitely massive body.
In contrast to the superintegrable mass ratios, a generic mass ratio produces two different outcomes, depending on the order of collisions. Notice that two qualitatively different trajectories may even originate from two infinitely close initial conditions. In the maximally superintegrable case of integer $q$, these two trajectories collapse to a single one-dimensional line. This phenomenon can be regarded as an unbounded orbit analogue of the closing the orbits in the bounded case.
The actual sets of the outgoing velocities are very different in the even and in the odd cases. In the even case, the initial velocities are simply inverted: $$q=\text{even} \to
\begin{array}{l}
\quad V_{\text{out}} = - V_{\text{in}}\\
\quad v_{\text{out}} = - v_{\text{in}}
\end{array}
\;.$$ Indeed, since the energy and, in this case, the observable $J$ are even functions of the velocities, the above connection protects the conservation laws. The odd case is much more involved. One can show that $$q=\text{odd} \to
\begin{array}{l}
\quad V_{\text{out}} = - \cos(\pi/q) V_{\text{in}} - \tan(\pi/q) \sin(\pi/q) v_{\text{in}}\\
\quad v_{\text{out}} = - \cos(\pi/q) ( V_{\text{in}} - v_{\text{in}})
\end{array}
\;.$$ Remark that the case $v_{\text{in}} = V_{\text{in}}$, where $v_{\text{out}}$ vanishes, may be regarded as a generalization of a notion of a *Galilean Cannon* [@olshanii2016_161001060]: a system of balls that arrives at the wall with the same speed and transfers all the energy to the far-most one in the end.
Minimal and maximal distances\[Sec:max:X\]
------------------------------------------
In this section we discuss what the extreme positions of the balls are. The characteristic unit of length is set by the initial position of the light ball, $x_0$. The largest distance to the wall corresponds to the asymptotically large separations $|X|\to\infty$ and $|x|\to\infty$ which are asymptotically approached after the terminal collision, except for the very special situation when the final velocity of the light ball is equal to zero (this situation happens for $N=0$). The minimal distance of the light ball is $x=0$ reached at any ball-wall collision.
The closest position $X_{min}$ of the heavy ball to the wall is reached at the collision in which the heavy ball inverts its velocity, while the light ball achieves its maximal velocity approximated by Eq. (\[Eq:v:max\]). As can be seen from Eq. (\[Eq:adiabatic:Xmin\]) derived within the adiabatic approximation, $X_{min}$ is inversely proportional to the action $I$. According to the discussion in Sec. \[Sec:invariants\], the action is conserved during the whole length of the processes and its value is given by Eqs. (\[Eq:v:max\]), (\[Eq:invariant:BW\]) and (\[Eq:action invariant\]). As a result, the minimal distance of the heavy ball can be expressed as $$X_{min}
= \frac{V_0}{v_{max}}x_0
= \sqrt{\frac{m}{M}}x_0
= \frac{x_0}{b^N}\;.
\label{Eq:X:min}$$ Expression (\[Eq:X:min\]) is approximate and it can be made exact by using the ball-wall invariant (\[Eq:invariant:BB\]). From a practical point of view, even in its simple form it works rather precisely. For example for $b=10$ one finds $X_{min} = 0.0998$ instead of $1/10$ already for $N=1$ and the accuracy is further improved as $N$ is increased. The curvature of parabola (\[Eq:return:parabola\]) close to the point of return can be expressed using Eq. (\[Eq:X:min\]) in a simple form $$\begin{aligned}
\frac{X(t')}{|X_{min}|}
= 1 + \frac{b^{2N}}{2}\left(\frac{t'}{t_0}\right)^2
= 1 + \frac{1}{2}\frac{M}{m}\left(\frac{t'}{t_0}\right)^2
\;,\end{aligned}$$ where $t_0 = |x_0| / V_0$ is the characteristic timescale for the period between the first collision and the reflection. The larger is the mass ratio, the “sharper” is the trajectory close to the point of return.
Position as a function of time: hyperbolic shape\[Sec:xvs.t\]
-------------------------------------------------------------
Here we will demonstrate that a hyperbolic curve describes the positions of the light ball at BB collisions and of the heavy ball both at BB and BW collisions.
In the description of a billiard in a wedge, the trajectory is bound to the phase space $0\leq |y| \leq |Y| \tan(\varphi)$, as shown in Fig. \[Fig:unfolding\]. The collisions happen when either $y=0$, i.e. when the light particle hits the wall (BW collision) or when $y=Y$ and the light particle hits the heavy one (BB collision). The unfolded trajectory is formed by reflecting the wedge, so that its angle $\varphi$ is preserved. The collisions in the unfolded trajectory occur when the straight line intersects one of the mirrors, corresponding to an angle $n\varphi$ with $n$ the number of the collision. For any intersection its distance from the origin is the same in unfolded picture and that of the billiard in a wedge. In particular, for a ball-ball collision, this distance is equal to $\sqrt{Y^2(t)+y^2(t)}$. Instead, in the moment of a ball-wall collision, the light ball has coordinate $y(t)=0$ and this distance is equal to the position of the heavy ball $Y(t)$. The minimal possible distance $Y_{min}$ of the heavy ball from the wall corresponds to the point of return, which is located on the vertical line directly above the origin. The projection to the horizontal axis is given by $Wt'$, where $t'$ is the time counted from the point of return and $W$ is the constant velocity, equal to the initial velocity of the heavy ball $W = W_{max}$. Catheti $Y_{min}$, $W_{max}t'$ and hypotenuse $Y(t)$ forming a right-angled triangle are related as $Y^2(t) = Y^2_{min} + (W_{max}t')^2$. The same expression written in terms of the original coordinate $X(t')$ and velocity $V$ leads to the hyperbolic relation $$\begin{aligned}
\left(\frac{X(t')}{X_{min}}\right)^2 - \left(\frac{t'}{X_{min}/V_{max}}\right)^2 = 1\;,
\label{Eq:hyperbola:BW}\end{aligned}$$ exactly satisfied for any ball-wall collision. Here $X_{min}$ is given by Eq. (\[Eq:X:min\]).
Instead, for a ball-ball collision both coordinates of the heavy and light particles are equal, $X=x$, and lie on a hyperbola of a slightly smaller semi-axis $$\begin{aligned}
\left(\frac{X(t')}{\sqrt{\frac{M}{M+m}}X_{min}}\right)^2 - \left(\frac{t'}{X_{min}/V_{max}}\right)^2 = 1\;,
\label{Eq:hyperbola:BB}\end{aligned}$$ In the limit of large mass, Eqs. (\[Eq:hyperbola:BW\]-\[Eq:hyperbola:BB\]) coincide.
![Distance of the heavy ball close to the return point, for $b=2$ and $N=3$. Solid symbols, heavy ball $X$; dashed line, limit of an infinitely massive ball, Eq. (\[Eq:return:heavyball\]); solid line, parabola defined by Eq. (\[Eq:hyperbola:BB\]). []{data-label="Fig:hyperbola"}](FigHyperbola.pdf){width="0.6\columnwidth"}
We compare predictions of Eq. (\[Eq:hyperbola:BB\]) with the exact results in Fig. \[Fig:hyperbola\]. Close to the return point, parabolic dependence (\[Eq:return:parabola\]) shown in Fig. \[Fig:parabola\] is recovered. Instead, far from the return point, the limit of an infinitely massive ball, Eq. (\[Eq:return:heavyball\]), is satisfied.
The minimal possible distance $X_{min}$ is actually reached only if there is a crossing of the unfolded trajectory at the vertical line above the origin (see Fig. \[Fig:unfolding\]), otherwise the actual minimal distance is larger.
Circle in ($V$,$1/X$) variables \[Sec:(V,1/x)circle\]
-----------------------------------------------------
Within the adiabatic approximation, introduced in Sec. \[Sec:adiabatic approximation\], the $(P, 1/X)$ portrait has a semicircular shape given by Eq. (\[Eq:adiabatic:P\]) with the coefficient of proportionality linear in the action $I$. In Sec. \[Sec:invariants\] it was verified that some of the predictions of the adiabatic theory actually remain exact even far from the point of return, effectively expanding the limits of its applicability. In particular, the action $I$ is conserved for any ball-wall collision throughout the whole process, and its value can be expressed in terms of the initial position of the light ball $x_0$ and the initial velocity of the heavy ball $V_0$ according to Eq. (\[Eq:action invariant\]). This suggests that the portraits in ($P$, $1/X$) and ($V$, $1/X$) coordinates are close to ellipses.
A straightforward way to see it is to use the Hamiltonian (\[Eq:adiabatic:H\]) obtained within the adiabatic approximation, $$\begin{aligned}
\frac{MV^2}{2} + \frac{\pi^2 I^2}{2mX^2}
= \frac{MV_0^2}{2}
\label{Eq:H:(V,X)}\end{aligned}$$ where we extended its validity to any BW collision, in particular to collisions happening far from the point of return, $X\to\infty$. Equation (\[Eq:H:(V,X)\]) can be recast in the form of an ellipse for $(V,1/X)$ coordinates as $$\begin{aligned}
\frac{1}{b^{2N}}
\left(\frac{x_0}{X}\right)^2
+
\left(\frac{V}{V_0}\right)^2
=
1\;.
\label{Eq:ellipse:(V,X)}\end{aligned}$$
Figure \[Fig:ellipse\] shows an example of the trajectory in $(V,1/X)$ coordinates. The first collision happens at $V/V_0 = 1$ and $x_0/X = 1$ corresponding to the initial velocity and the initial (large) distance from the wall. As the collisions go on, the heavy ball comes closer to the wall until it inverts its velocity at the point $V=0$, which corresponds to the point of return. At this moment the heavy ball is located at the closest distance to the wall. It might be appreciated that Eq. (\[Eq:X:min\]) describing this distance is quite precise from the practical point of view. The case illustrated in Fig. \[Fig:ellipse\] corresponds to the binary base, $b=2$, and for mantissa length $N=1;2;3;4$ the heavy ball is expected to come closer to the wall by a factor of $2;4;8;16$ compared to the initial position of the light ball. Once the point of return is passed, the heavy ball has a negative velocity which increases in absolute value up to $V/V_0 \approx -1$, while the ball moves far away from the ball $x_0/X\to 0$.
![ $(V,1/X)$ portrait for $b=2$ and $N=0,1,2,3,4$ (from bottom to top). Solid symbols, ball-wall collision; open symbols, ball-ball collision. Solid lines, ellipse (\[Eq:ellipse:(V,X)\]). Dashed horizontal lines, inverse of the minimal distance (\[Eq:X:min\]). []{data-label="Fig:ellipse"}](FigEllipse.pdf){width="0.5\columnwidth"}
Overall, the shapes obtained are quite similar to ellipses predicted by Eq. (\[Eq:ellipse:(V,X)\]). The “discretization” becomes smaller as $N$ is increased. The pairs with same velocity $V$ but different values of $X$ correspond to light ball-wall collisions (velocity of the heavy ball is not changed) and ball-ball collisions, shown in Fig. \[Fig:ellipse\] with closed and open symbols. The points which correspond to the ball-wall collisions lie exactly on the top of the ellipse due to presence of ball-wall invariant (\[Eq:invariant:BW\]). Instead, for ball-ball collision there is some shift, with a different sign for the heavy ball moving towards the wall or away from it. A similar effect was observed in Fig. \[Fig:Xv\] and it originates from an additional contribution containing the velocity of the heavy ball in the ball-ball invariant (\[Eq:invariant:BB\]). At the point of return this correction vanishes and the adiabatic theory becomes fully applicable.
Exact solution\[Sec:exact solution\]
------------------------------------
A peculiarity of this problem is that it allows an explicit solution, so that the velocities and positions can be explicitly expressed as a function of the collision number. According to Sinai[@SinaiBook], the billiard variables (\[Eq:billiard variables\]-\[Eq:billiard velocities\]) reduce the rescaled velocities to a circle, so that the velocities of the light and the heavy balls at collision $n$ is fully defined by the angle $\phi_n$. This angle is flipped at any BW collision and flipped and increased at each BB collision, leading to a simple dependence of $\phi_n$ on $n$. We note that action invariants relate the position of the balls with the momentum of the heavy ball, thus permitting to extract particle positions once their velocities are known. The time $\tau_n$ passed between collision $n$ and the next one can be inferred from the balls positions and their velocities according to Eq. (\[Eq:tau(n):exact\]).
The exact solution for the positions, velocities and period of time between the collisions can be explicitly written as a function of the collision number, The explicit solution can be summarized as follows $$\begin{aligned}
\phi_n &= &(-1)^{n+1}2\arctan\left(\frac{1}{b^N}\right) {\mathop{\rm int}\nolimits}\left[\frac{n+1}{2}\right], \\
\phi_{2k} &= & - 2k\arctan\left(\frac{1}{b^N}\right), \;\;\; k=0, 1, \ldots,\\
V_n &= &V_0 \cos \phi_n,\\
v_n &= &V_0b^N \sin \phi_n,\\
X_{2k} &= &-\frac{x_0}{b^N \sin \phi_{2k}}, \;\;\; k= 1, 2, \ldots,\\
X_{2k+1} &= &\frac{x_0}{b^N \sin \phi_{2k}-\cos \phi_{2k}}, \;\;\; k=0, 1, \ldots,\\
x_{2k} & = & 0, \;\;\; k= 1, 2, \ldots,\\
x_{2k+1} &= & X_{2k+1}, \;\;\; k=0, 1, \ldots,\\
\tau_{2k-1} &= & - \frac{x_0}{V_0b^N\sin\phi_{2k}(b^N\sin\phi_{2k} + \cos\phi_{2k})}, \;\;\; k= 1, 2, \ldots,\\
\tau_{2k} &=& - \frac{x_0}{V_0b^N\sin\phi_{2k}(b^N\sin\phi_{2k} - \cos\phi_{2k})}, \;\;\; k= 1, 2, \ldots.\end{aligned}$$ Here the collision number is $n = 0, 1, \ldots$ with $n=0$ corresponding to the initial conditions. It is also useful to express the quantities as a function of time.
The exact relation between collision number $n$ and the moment of time $t(n)$ when the collision happened can be obtained by summation of the time intervals $\tau_n$ passed between collisions, see Eq. (\[Eq:t(n):exact\]). Although the result of the summation can be explicitly written in terms of $q$-digamma function, its presentation is quite cumbersome, and we prefer to keep the exact function $t(n)$ as a sum, while in the next section we provide an elegant approximate expression.
Approximate solution\[Sec:approximate solution\]
------------------------------------------------
It is also of interest to obtain an explicit relation in the continuous limit of many collisions, $b^N\gg 1$. In this case one can neglect cosine function in $\tau$ obtaining the following simple expression for the inverse of time $\tau_n$ between consecutive collisions, $$\begin{aligned}
\frac{t_0}{\tau_n}
\approx b^{2N}\sin^2(n/b^N)
\approx b^{2N}\cos^2(n'/b^N)
\label{Eq:tau(n):approx}\end{aligned}$$ where the characteristic time $$\begin{aligned}
t_0 = \left|\frac{x_0}{V_0}\right|\end{aligned}$$ defines the appropriate units of time, i.e. how long the heavy ball would take to hit the wall in the absence of the light ball.
Figure \[Fig:tau\] shows an example of how the inverse time $t_0 / \tau_n$ depends on collision number. In the considered case with $b=10$ and $N=1$, the total number of collisions is $\Pi = 31$. The first and last collisions are “slow”, the time between them is large and the inverse time goes to zero. Close to the return point the oscillations are fast and the inverse time reaches its maximal value, where the time between collisions is by factor of $b^{2N}$ faster than the characteristic time $t_0$ of the whole process. The parabolic shape of the trajectory derived in Sec. \[Sec:point of return\] was obtained by ignoring dependence of $\tau$ on $n$, see Eq. (\[Eq:tau:inversion\]), and corresponds to the dashed horizontal line.
![ Inverse of the time between adjacent collisions for $b=10$ and $N=1$. Symbols, exact results; solid blue line, Eq, (\[Eq:tau(n):approx\]); dashed green line, $b^{2N}$. []{data-label="Fig:tau"}](FigTau.pdf){width="0.5\columnwidth"}
Integrating Eq. (\[Eq:tau(n):approx\]) close to the point of return and treating the number of collisions after the return point $n' = n - \Pi /2$ as a continuous variable we get $$\begin{aligned}
t_{n'} /t_0
= 1 + \int\limits_0^{n'} \frac{dn}{b^{2N}\cos^2(n/b^N)}
= 1 + \frac{1}{b^N}\tan(n'/b^N).
\label{Eq:t:approx}\end{aligned}$$
Figure \[Fig:n\] shows the dependence of the number of collisions $n$ on time $t$ for $b=10$ and $N=1$. The function has an inclination point at $t_0$ with the value of the point of return $n_{inversion}\approx \Pi/2$. The final collision takes place for a large $t$ and the function tends to $\Pi$ as $t\to +\infty$. This function is related to the phase $\phi_{\rm phase}$, since $\phi_{\rm phase}(t)=\pi n(t)$.
![ The collision number $n$ as a function of time $t$ for the same parameters as in Fig. \[Fig:tau\]. Symbols, exact results; solid blue line, Eq. (\[Eq:t:approx\]); dashed line, inversion collision approximated by $n_{inversion}\approx \Pi/2$. []{data-label="Fig:n"}](Fign.pdf){width="0.5\columnwidth"}
Terminal collision\[Sec:terminal collision\]
--------------------------------------------
The last collision defines if the number of collisions is an odd or an even number. Depending on its value, the corresponding digit of $\pi$ is either odd or even. Physically, its parity depends if the last collision was ball-wall with no more ball-ball impacts or if it was a ball-ball collision. In Ref. [@Davis2015] it is shown that an even number of collisions occurs, $\Pi=2k$, when $2k\varphi < \pi < (2k+1)\varphi$.
Physical realizations\[Sec:physical realizations\]
==================================================
Finite-size balls\[Sec:hard rods\]
----------------------------------
The pair $(X, x)$ of positions generates a configuration point, and the set of all configuration points form the configuration space[@GalperinBook]. For point-size balls, it is bounded by the position of the wall, $|X|> 0$ and $|x|>0$, and the condition of the impenetrability of the balls, preserving their order, $0<|x|<|X|$. More realistically, the real balls must have some finite size which we denote as $R$ and $r$ for the radii of the heavy and light balls, respectively. Still, we argue that if all collisions are elastic, the problem can be effectively reduced to the previous one of point-size balls. One might note that finite-size impenetrable balls have a smaller configuration space, schematically shown in Fig. \[Fig:excluded volume\], which contains an [*excluded volume*]{}[@Girardeau60]. The configuration space of finite-size balls is $|x|>r$ and $|X|>|x|+r+R$. In other words, mapping which removes the excluded volume, $$\begin{aligned}
x' &=& x + r\\
X' &=& X + R + r,
\label{Eq:excluded volume}\end{aligned}$$ reduces the problem of finite-size hard spheres to the problem of point-like objects, via a simple scaling which does not affect the balls’ velocities.
As sphere is a three-dimensional object, sometimes finite width one-dimensional balls are referred to as [*hard rods*]{}.
![Configuration space for (a) two point balls (b) balls of size $r$ and $R$. Mapping (\[Eq:excluded volume\]) translates configuration space (a) into (b). []{data-label="Fig:excluded volume"}](FigExcludedA.pdf "fig:"){width="0.45\columnwidth"} ![Configuration space for (a) two point balls (b) balls of size $r$ and $R$. Mapping (\[Eq:excluded volume\]) translates configuration space (a) into (b). []{data-label="Fig:excluded volume"}](FigExcludedB.pdf "fig:"){width="0.45\columnwidth"}
Billiard\[Sec:billiard\]
------------------------
The restricted domain of the available phase space (half of a quadrant) together with the specular reflection laws makes the system consisting of two identical balls and a wall mappable to a problem of a [*billiard*]{} with opening angle of $45^\circ$. In a billiard, the balls move in straight lines and collide with the boundaries (mirrors), where the incident and reflected angles are equal[@KozlovTreshchevBook]. It might be shown[@Galperin2003; @TabachnikovBook2005] that billiard variables (\[Eq:billiard variables\]) change the opening angle to $\phi$ and have a special property which is that the reflections result in a straight trajectory. This unfolding creates a straight-line trajectory which intersects a certain number of lines, each of them rotated by the angle $\phi$. Each intersection corresponds to a single collision and the total number of intersections defines the total number of collisions $\Pi$. Altogether, this picture provides an intuitive visualization of the relation between $\pi$, corresponding to the angle of $180^\circ$, and the number of collisions.
Four-ball chain\[Sec:four ball chain\]
--------------------------------------
Another physical system which conceptually is related to the present system consisting of two balls and a wall, is a problem of four balls on a line. The action of the wall consists in reflecting the mass $m$ ball with the same absolute value of the velocity, $v \to -v$. The same effect can be achieved by replacing the rigid wall by another ball of mass $m$, moving with velocity $-v$. During an elastic collision, both balls will exchange their velocities. In order to make the system completely symmetric, one has also to add an additional heavy ball, resulting in $M -m -m - M$ chain. The distance between $1-2$ and $3-4$ balls must be the same, while $2-3$ distance can be arbitrary chosen. Finally, the initial velocities should be chosen such $v_2 = v_3$ and $v_1 - v_2 = v_3 - v_4$.
Systematic error\[Sec:error\]
=============================
Any real experimental procedure should contain an error analysis. For example, the stochastic method of Buffon provides not only an approximate value of $\pi$, but also the statistical error associated with it. After $N$ trials of dropping the needle, $\pi$ is estimated as an average value while the statistical error is $\varepsilon_{stat} = \sigma / \sqrt{N-1}$, where $\sigma$ is the variance. Although in each experiment the realizations are different, the statistical error can be estimated and its value can be controllably reduced by increasing the number of trials.
In the present study we do not report results of a real experiment, in which the number of collisions will be limited by friction, non-perfect elasticity of collisions, etc. Nevertheless, the relation (\[Eq:hurbilketa\]) between the number of collisions and the Galperin billiard relies on the Taylor expansion of inverse tangent function in Eq. (\[Eq:pi:acrctg\]) and on taking its integer part, and these might induce a certain error to the final result. Accuracy of the approximations used is reported in Fig. \[Fig:error\] as a function of the base $b$ and mantissa $N$. For completeness, here we consider $N$ not limited to integer values but as a continuous variable $N\ge 0$ and the base $b>1$. The analyzed data gives error $\varepsilon$ limited to two values $\varepsilon = 1$ (light color) and $\varepsilon = 0$ (black). It becomes evident that for large $N$ the approximate formula always works correctly, while for small $N$ there appears a complicated structure as a function of $b$. For large system base (for example, decimal $b=10$ and hexadecimal $b=16$ cases) formula (\[Eq:hurbilketa\]) works correctly for any length of mantissa apart from $N=0$ case, which in any case should be treated separately due to degeneracy as will be discussed in Sec. \[Sec:degeneracy\].
![ Difference between the exact number of collisions (\[Eq:pi:acrctg\]) and the approximation (\[Eq:hurbilketa\]), which relates it to the digits of pi, as a function of base $b$ and mantissa $N$. Two possible values are denoted with the dark (0) and light (1) colors. []{data-label="Fig:error"}](FigAccuracy.pdf){width="0.6\columnwidth"}
The error $\varepsilon$ is a complicated non-analytic function of $N$ and $b$, as can be perceived from Fig. \[Fig:error\]. It turns out that for some integer bases expressions (\[Eq:pi:acrctg\]) and (\[Eq:hurbilketa\]) lead to different results. Namely, the error is $\varepsilon = 1$ for integer bases $b=6 ; 7; 14$ and $N=1$. It means that for the mentioned combinations, Galperin billiard method does not provide the digits of $\pi$ exactly, as there is an error of $\varepsilon = 1$ in the last digit. The cardinality of irrational numbers is greater than that of the integer numbers. For irrational numbers it is possible to find examples where the error is different from zero for different values of $N$ and the same value of the base $b$. Namely, $\varepsilon=1$ for $b=3.7823797$ and $N=1, 2, 3, 4$ and $6$. In general, it is clear that the closer is the base to $b=1$ the worse is the description, and for a larger number of values of $N$ Galperin billiard gives digits different from $\pi$.
We propose to treat a possible difference between (\[Eq:pi:acrctg\]) and (\[Eq:hurbilketa\]) as a [*systematic error*]{}, so that the final result of each “measurement” is $\varepsilon / b^N$ with $\varepsilon \le 1$. That is, the approximation of $\pi$ in a base $b$ can be expressed from the number of collisions $\Pi(b,N)$ as $$\pi_b = \frac{\Pi(b,N)}{b^N} \pm \frac{\varepsilon}{b^N}\;.
\label{Eq:error}$$ Such a classification is closer to a spirit of a real measurement, where different effects might contribute to the error. Another advantage of the proposed idea of introducing the concept of the systematic error, is that it solves the problem of the number of digits which are predicted correctly using Galperin billiard. It was noted by Galperin in Ref. [@Galperin2003] (see also Ref. [@TabachnikovBook2005]) that if there is a string of nines, that might lead to a situation when more than one digit is different. In a similar sense, the numbers $0.999$ and $1.000$ differ by all four digits. If instead, one allows an error of $0.001$, both numbers become compatible. Indeed, from a practical point of view (suppose we calculate perimeter of a circle knowing its radius), the use of the incorrect value would lead to a relative error of 0.001 and not to completely incorrect result as all the original digits are different.
In the next sections we consider the cases of integer and non-integer bases.
Integer bases\[Sec:integer bases\]
==================================
Equation (\[Eq:pi:acrctg\]) has a profound mathematical meaning, as the number of collisions $\Pi(b,N)$ provides the first $N$ digits of the fractional part (i.e. digits beyond the radix point) of the number $\pi$ in base $b$. It might be immediately realized that as the number of collisions is obviously an integer number, its integer base representation can be chosen to be finite.
In Sections \[Sec:integer bases\]-\[Sec:Non-integer bases\] we use number of collisions $\Pi(b, N)$, as given by Eq. (\[Eq:pi:acrctg\]), to approximate the digits of $\pi$ for different integer bases $b$, then $(\Pi/b^N)_b$ yields the base-$b$ representation of $\pi$ with $N$ digit beyond the radix point.
Representing a number in integer bases\[Sec:integer representation\]
--------------------------------------------------------------------
Let $b>1$ be an integer number. Any positive number $x$ has the integer expansion in base $b$, i.e. can be represented in powers of $b$ as $$\label{Eq:integer}
x =(a_n a_{n-1} \ldots a_0.a_{-1}a_{-2} \ldots)_b = \sum\limits_{i=-\infty}^n a_i b^i,$$ where $n={\mathop{\rm int}\nolimits}[\log_b x]$ and $a_i=\{0,1, \ldots, b-1\}$ are the digits in the corresponding numeral system and we use form $x_b$ to denote the representation of number $x$ in base $b$. For bases with $b>10$, the symbols $A, B, \ldots$ are commonly used to denote $10, 11, \ldots$. In order to obtain the digits $a_i$, one can use the following iterative process: $a_i={\mathop{\rm int}\nolimits}[ r_i/b^i ]$, $i\leq n$ with $r_n=x$ and $r_{j-1}=r_j-a_j\cdot b^j$, $j\leq n-1$. Multiplying the base-$b$ representation (\[Eq:integer\]) by $b^i$ shifts the radix point by $i$ digits. Thus, approximation (\[Eq:hurbilketa\]) gives the integer part and first $N$ digits of the fractional part of $\pi$ in base $b$.
The most frequently used integer systems are decimal ($b=10$) and binary $b=2$ systems. Occasionally, also ternary $b=3$, octal ($b=8$), hexadecimal ($b=16$) and others systems are used. Importantly, for integer bases, finite representations are unique, while infinite representations might be not unique. For example, the finite number $1_{10}$ in the decimal base can be written as $1.000(0)_{10}=0.999(9)_{10}$.
Degenerate case of equal masses and submultiple angles\[Sec:degeneracy\]
------------------------------------------------------------------------
Before considering in detail the representation in bases $b=10; 2; 3$ reported in Tables \[table:b=10\],\[table:b=2\],\[table:b=3\], we note that $N=0$ case is universal as the mass ratio $M/m = b^N = 1$ does not depend on the base $b$. In other words, the digit in front of the radix point always correspond to the same number. The Eq. (\[Eq:pi:acrctg\]) formally gives 4 collisions, which is different from the physically correct number of 3 collisions. The reason for such a difference comes from a degeneracy between the third and fourth collision. While for $N>0$, the direction of the light ball is always inverted in the last two collisions ($\phi\to -\phi$), for $N=0$ the light ball completely stops exactly at the third collision. In physical sense there is no difference between $v_3 = -0$ and $v_4 = +0$ velocities.
Thus, Eqs. (\[Eq:pi:acrctg\]-\[Eq:hurbilketa\]) are applicable only for $N\geq1$ while $N=0$ is a special case and it should be treated separately.
The analogous result takes place in the case of the angle $\phi$ being submultiple of $\pi$, i.e. when the ratio $\pi/\phi$ is an integer number. The number of collisions is not given correctly by (\[Eq:pi:acrctg\]) as the last collision is degenerate as well.
Decimal base \[Sec:Decimal base\]
---------------------------------
For the most natural case of the decimal base system, $b=10$, the number of collisions $\Pi(10, N)$ is given in Table \[table:b=10\]. It is easy to follow, how Galperin billiard generates digit of $\pi$. For $N=0$, Eq. (\[Eq:pi:acrctg\]) results in the first digit of $\pi$ approximated by 4, while due to degeneracy discussed in Sec. \[Sec:degeneracy\], physically there are 3 collisions. For $N=1$, there are $31$ collisions, resulting in expression with 1 digit after the radix point, $3.1$. From $N=2$, the number of collisions in 314 giving the number $\pi$ with 2 digits after the radix point. One can see that the billiard method correctly approximates the number $\pi$ as 3 plus $N$ more digits in the decimal base.
Conceptually, one might ask if there is a difference between the number of collisions Eq. (\[Eq:pi:acrctg\]) which depend on $\arctan(b^{-N})$ rather than $b^{-N}$, as in Eq. (\[Eq:hurbilketa\]). It turns out that the base $b=10$ is large enough (see Fig. \[Fig:error\]) so that there is no any difference in the integer part of the expansion.
N $\Pi(10,N)_{10}$ $(\Pi(10,N)/10^N)_{10}$ $(1/10^N)_{10}$
---- ------------------ ------------------------- -----------------
0 4 4 1
1 31 3.1 0.1
2 314 3.14 0.01
3 3141 3.141 0.001
4 31415 3.1415 0.0001
5 314159 3.14159 0.00001
6 3141592 3.141592 0.000001
7 31415926 3.1415926 0.0000001
8 314159265 3.14159265 0.00000001
9 3141592653 3.141592653 0.000000001
10 31415926535 3.1415926535 0.0000000001
: Number of collisions $\Pi(10,N)$ given by Eq. (\[Eq:pi:acrctg\]) for the decimal base, $b=10$. The first column reports the value of mantissa $N$. The second column is the resulting number of collisions in the decimal base. The third column is the number $\pi$ with $N$ digits in the fractional part in the decimal representation. The fourth column gives the systematic error according to Eq. (\[Eq:error\]). []{data-label="table:b=10"}
Binary and ternary bases \[Sec:binary base\]
--------------------------------------------
Other important examples of number systems include the binary ($b=2$) and ternary ($b=3$) base systems. The binary system lies in the core of modern computers which operate with [*bits*]{} $0, 1$. Interestingly, base-$3$ computer named Setun was built 1958 under leadership of mathematician Sergei Sobolev and operated with [*trits*]{}, $0,1,2$.
Table \[table:b=2\] reports the number of collisions $\Pi(2,N)$ obtained for $b=2$ base. By expressing the number of collisions in binary base using zeros and ones, one obtains the representation of the number $\pi$ in binary base. In the ternary base, the number of collisions are written using the three allowed digits, $0,1,2$, see Table \[table:b=3\].
N $\Pi(2,N)_{10}$ $\Pi(2,N)_{2}$ $(\Pi(2,N)/2^N)_{2}$ $(1/2^N)_{2}$
---- ----------------- ---------------- ---------------------- ---------------
0 4 100 100 1
1 6 110 11.0 0.1
2 12 1100 11.00 0.01
3 25 11001 11.001 0.001
4 50 110010 11.0010 0.0001
5 100 1100100 11.00100 0.00001
6 201 11001001 11.001001 0.000001
7 402 110010010 11.0010010 0.0000001
8 804 1100100100 11.00100100 0.00000001
9 1608 11001001000 11.001001000 0.000000001
10 3216 110010010000 11.0010010000 0.0000000001
: Number of collisions $\Pi$ given by Eq. (\[Eq:pi:acrctg\]) for the binary base, $b=2$. The first column reports the value of mantissa $N$. The second column is the resulting number of collisions in the decimal base. The third column is the number of collisions written in the binary representation. The fourth column is the binary representation of the number $\pi$ with $N$ digits in the fractional part. The fifth column gives the systematic error according to Eq. (\[Eq:error\]). []{data-label="table:b=2"}
N $\Pi(3,N)_{10}$ $\Pi(3,N)_{3}$ $(\Pi(3,N)/3^N)_{3}$ $(1/3^N)_{3}$
---- ----------------- ---------------- ---------------------- ---------------
0 4 11 11 1
1 9 100 10.0 0.1
2 28 1001 10.01 0.01
3 84 10010 10.010 0.001
4 254 100102 10.0102 0.0001
5 763 1001021 10.01021 0.00001
6 2290 10010211 10.010211 0.000001
7 6870 100102110 10.0102110 0.0000001
8 20611 1001021101 10.01021101 0.00000001
9 61835 10010211012 10.010211012 0.000000001
10 185507 100102110122 10.0102110122 0.0000000001
: Number of collisions $\Pi$ given by Eq. (\[Eq:pi:acrctg\]) for the ternary base, $b=3$. The first column reports the value of mantissa $N$. The second column is the resulting number of collisions in the decimal base. The third column is the number of collisions written in the binary representation. The fourth column is the ternary representation of the number $\pi$ with $N$ digits in the fractional part. The fifth column gives the systematic error according to Eq. (\[Eq:error\]). []{data-label="table:b=3"}
Best bases for a possible experiment\[Sec:experiment\]
------------------------------------------------------
As concerning the effects of the friction and other sources of energy dissipation, it is easier to perform experiments for small base $b$. While for $N=0$ (the mass ratio is $M/m = 1$ independently of $b$) there are 3 collisions which can be easily observed with identical balls, for larger $N$ the number of collisions grows exponentially fast. The decimal system has a rather “large” base $b=10$ which already for $N=1$ results in 31 collisions and $N=2$ even in 314 collisions. It might be notoriously hard to create a clean system in which such a large number of collisions can be reliably measured.
For the binary base $b=2$ and $N=1$ the number of collisions to be observed is much smaller, $3; 6; 12; 25; \ldots$, see Table \[table:b=2\], making such system more suitable for an experimental observation.
Non-integer bases \[Sec:Non-integer bases\]
===========================================
As anticipated above, the Galperin billiard method should provide digits of $\pi$ in an arbitrary base $b$, even for a non-integer one. In this Section we consider a number of examples.
Representing a number in a non-integer base\[Sec:representation\]
-----------------------------------------------------------------
For a non-integer base $b>1$, any positive number $x$ can be written in the base-$b$ representation according to $$x_b=d_n...d_2d_1d_0.d_{-1}d_{-2}\ldots\;,
\label{Eq:non-integer base:x_b}$$ where digits $d_i$ can take only non-negative integer values smaller than non-integer base, $d_i<\lceil b \rceil$ ($\lceil x \rceil$ stands for the least integer which is greater than or equal to $x$), and $$x = \sum\limits_{i=-\infty}^n d_i b^i\;.
\label{Eq:non-integer base:x}$$
Unlike the integer bases, for a non-integer base $b$, even finite fractions might have different $b$ representations. For example, in the golden mean $\varphi\approx 1.61803$ base, due to the equality $\varphi^2=1+\varphi$, one has $100_\varphi = 11_\varphi$. With Eqs. (\[Eq:non-integer base:x\_b\]-\[Eq:non-integer base:x\]) we can find at least one representation for $x$.
Number systems with irrational bases {#Sec:irrational base}
------------------------------------
Some notable examples of non-integer bases include the fundamental cases of the bases with the golden mean $b=\phi$, the natural logarithm $b=e$ and a curious situation when the number $\pi$ is used itself as a base, $b=\pi$. The number of collisions $\Pi(b,N)$ is obviously an integer number, and it always can be written with a finite representation in any integer base system. Contrarily, in a non-integer base $b$ it is a common situation that an integer number needs an infinite representation which corresponds to the number.
In Table \[table\_phi\] we give two different representations for $\pi_\varphi$, since its integer part $100_\varphi=11_\varphi$. The fourth and fifth columns in Table report the $\varphi$-representation with the integer part $100_\varphi$ and $11_\varphi$, respectively. The allowed digits for both representations are $0$ and $1$. Table \[table\_e\] reports the resulting number of collisions $\Pi(e, N)$ in base $e$ with the allowed digits $0, 1$ and $2$. We get the representation $\pi=(10.1010020200\ldots)_e=e+e^{-1}+e^{-3}+2 e^{-6} + 2 e^{-8}+\ldots$. One can see the influence of the error of the computation by the Galpelin billiard which in the case of base $b=e$ is $1/e^N=0.00\ldots01_e$. Due to this error, the last digit may be incorrectly predicted by the method. Especially, when the last digit is the maximum allowed ($2$ for $b=e$, see the cases $N=4$ and $N=9$ in Table \[table\_e\]), then the two last digit may be incorrect.
N $\Pi(\varphi, N)_{10}$ $\Pi(\varphi, N)_{\varphi}$ $(\Pi(\varphi, N)/\varphi^N)_{\varphi}$ (I) $(\Pi(\varphi, N)/\varphi^N)_{\varphi}$ (II) $(1/\varphi^N)_{\varphi}$
---- ------------------------ ----------------------------- --------------------------------------------- ---------------------------------------------- ---------------------------
0 4 101. 100. 11. 1
1 5 1000. 100.0 11.0 0.1
2 8 10001. 100.01 11.01 0.01
3 13 100010. 100.010 11.010 0.001
4 21 1000100. 100.0100 11.0100 0.0001
5 34 10001000. 100.01001 11.01001 0.00001
6 56 100010010. 100.010010 11.010010 0.000001
7 91 1000100101. 100.0100101 11.0100101 0.0000001
8 147 10001001010. 100.01001010 11.01001010 0.00000001
9 238 100010010100. 100.010010101 11.010010101 0.000000001
10 386 1000100101010. 100.0100101010 11.0100101010 0.0000000001
: Number of collisions $\Pi(\varphi,N)$ given by (\[Eq:pi:acrctg\]) for $b=\varphi$. The first column is $N$. The second column is $\Pi(\varphi,N)$ in the decimal base. The third column is the integer part of $\Pi(\varphi,N)$ written in the base $\pi$. The fourth column is the number $\pi$ with $N$ digits in the fractional part (Type I) in the base $\pi$. The fifth column is the number $\pi$ with $N$ digits in the fractional part (Type II) in the base $\pi$. The sixth column gives the systematic error according to Eq. (\[Eq:error\]). []{data-label="table_phi"}
N $\Pi(e,N)_{10}$ $\Pi(e,N)_{e}$ $(\Pi(e,N)/e^N)_{e}$ $(1/e^N)_{e}$
---- ----------------- ---------------- ---------------------- ---------------
0 4 11. 11. 1
1 8 100. 10.0 0.1
2 23 1010. 10.10 0.01
3 63 10101. 10.101 0.001
4 171 101002. 10.1002 0.0001
5 466 1010100. 10.10100 0.00001
6 1267 10101001. 10.101001 0.000001
7 3445 101010020. 10.1010020 0.0000001
8 9364 1010100201. 10.10100201 0.00000001
9 25456 10101002012. 10.101002012 0.000000001
10 69198 101010020200. 10.1010020200 0.0000000001
: Number of collisions $\Pi(e,N)$ given by (\[Eq:pi:acrctg\]) and approximation of $\pi$ for $b=e$. The first column is $N$. The second column is $\Pi(e,N)$ in the decimal base. The third column is the integer part of the number of collisions $\Pi(e,N)$ written in the base $e$. The fourth column is the number $\pi$ with $N$ digits in the fractional part in the base $e$. The fifth column gives the systematic error according to Eq. (\[Eq:error\]). []{data-label="table_e"}
In Table \[table\_pi\] we show approximations of $\pi$ in the base $b=\pi$ The allowed digits in this base are $0, 1, 2$ and $3$. The Galperin billiard does not provide an integer-number representation for the number $\pi$ even in this case, as instead of the “natural” possibility $\pi = 10_\pi$ one obtains an infinitely long representation $$\pi = (3.0110211100\ldots)_{\pi}=3+\pi^{-2}+\pi^{-3}+2\pi^{-5}+\pi^{-6} + \pi^{-7} + \pi^{-8} +\ldots.$$ This non-unique representation is similar to the infinite representation $0.999(9)\ldots$ of $1$ in the decimal system.
Another peculiarity of this base is that for $N=1$ (line marked with bold in Table \[table\_pi\]) the number of collisions $\Pi(\pi, 1)$ in the Galperin billiard, given by Eq. (\[Eq:pi:acrctg\]), does not coincide with expression (\[Eq:hurbilketa\]) which is used to transcribe $N$ digits of $\pi$ in the base $b$. The resulting difference in the last digit can be interpreted as a the systematic error $\varepsilon=1$ in the spirit of Section \[Sec:error\].
N $\Pi(\pi,N)_{10}$ $\Pi(\pi,N)_{\pi}$ $(\Pi(\pi, N)/\pi^N)_{\pi}$ $(1/\pi^N)_{\pi}$
------- ------------------- -------------------- ----------------------------- -------------------
0 4 10. 10. 1
**1** **10** **100.** **10.0** 0.1
2 31 301. 3.01 0.01
3 97 3010. 3.010 0.001
4 306 30110. 3.0110 0.0001
5 961 301102. 3.01102 0.00001
6 3020 3011021. 3.011021 0.000001
7 9488 30110210. 3.0110210 0.000001
8 29809 301102110. 3.01102110 0.0000001
9 93648 3011021110. 3.011021110 0.00000001
10 294204 30110211100. 3.0110211100 0.000000001
: Number of collisions $\Pi(\pi,N)$ given by (\[Eq:pi:acrctg\]) for $b=\pi$. The first column is $N$. The second column is the number of collisions in the decimal base. The third column is the integer part of the number of collisions $\Pi(\pi,N)$ written in the base $\pi$. The fourth column is the number $\pi$ with $N$ digits in the fractional part in the base $\pi$. The fifth column gives the systematic error according to Eq. (\[Eq:error\]). The case $N=1$ is emphasized since there is the difference 1 between $\Pi(\pi,1)$ by (\[Eq:pi:acrctg\]) and the approximation (\[Eq:hurbilketa\]). []{data-label="table_pi"}
The accuracy of the approximation to $\pi$ obtained by Galperin’s billiard can estimated by Eq. (\[Eq:error\]) which we treat as systematic error of the method.
It was demonstrated by Sinai [@SinaiBook], that in presence of the second wall, the motion is not ergodic for a rational angle (\[Eq:phi\]), i.e. when it can be written as $\varphi = 2\pi r/s$ with $r$ and $s$ some integer numbers.
Conclusions\[Sec:conclusions\]
==============================
To summarize, we have studied how the digits of the number $\pi$ are generated in a simple mechanical system consisting of one heavy ball, one light ball and a wall (Galperin’s billiard method). The number base $b$ and mantissa length $N$ define the mass ratio according to $M/m = b^{2N}$. We obtain for the first time, to our best knowledge, the complete explicit solution for the balls’ positions and velocities as a function of the collision number and time. Also we find that the adiabatic approximation works not only close to the return point but also for any ball-wall collision, even far from the wall. The action $I = P X / (2\pi)$ is an adiabatic invariant and its value is preserved during ball-wall collisions. The number $\pi$ is intimately related to the Galperin billiard and it has a number of “round” properties. We find that the portraits of the system in $(P,1/X)$ and $(V,1/X)$ coordinates have a shape close to a circular one. Another circle appears in $(V,v)$ coordinates and corresponds to the energy conservation law. Instead a hyperbolic shape appears in the $(X,t)$ plane. We derive a differential equation which describes the trajectory close to the point of return. Its solution is a parabola $X(t')/X_{min} = 1 + (V_0 t' / X_{min})^2$ with the width defined by the initial velocity $V_0$ and the minimal distance from the wall $X_{min}$ of the heavy particle. With a good precision the minimal and maximal values of the positions and velocities are approximated by $X_{min} = b^{-2N}X_0$ and $v_{max} = b^{2N}V_0$. We note that the behavior of the trajectories close to the return point is changed from a non analytic for few collisions to an analytic one (envelope in the shape of a parabola) for many collisions, and again a non analytic (of type $|x|$) in the limit of a large mass ratio.
Examples of integer bases $b$, including decimal, binary and ternary, are considered. We argue that smaller bases (for example, $b=2;3$) are the easiest to be realized in an experiment and show how the Galperin billiard can be generalized to finite-size balls (hard rods). We show that the dependence of a possible error in the last digit as a function of $b$ and $N$ has a complicated form, with the error disappearing in the limit of $b^N\to\infty$. We propose to treat the possible error in the last digit as a systematic error. In particular this resolves the problem of the correct number of obtained digits. We consider non-integer bases including an intriguing case of expressing $\pi$ in the base $\pi$. The generated expression is different from a finite number, $\pi = 1\times \pi^1$, and instead is given by an infinite representation, $\pi = 3 + 1/\pi^2 + 1/\pi^3+\cdots$. The difference between finite and infinite representation is similar to that of $1 = 0.999(9)$ in the decimal system. Finally we note that finite representation is not unique in the base of the golden number $\phi$.
Acknowledgment {#acknowledgment .unnumbered}
==============
The research leading to these results received funding from the MICINN (Spain) Grant No. FIS2014-56257-C2-1-P. MG has been partially supported by Juan de la Cierva-Formación Fellowship FJCI-2014-21229, the Russian Scientific Foundation Grant 14-41-00044 and the MICIIN/FEDER grants MTM2015-65715-P and MTM2016-80117-P (MINECO/FEDER, UE). MO acknowledges financial support from the National Science Foundation grant PHY-1607221 and the US-Israel Binational Science Foundation grant 2015616.
[33]{}ifxundefined \[1\][ ifx[\#1]{} ]{}ifnum \[1\][ \#1firstoftwo secondoftwo ]{}ifx \[1\][ \#1firstoftwo secondoftwo ]{}““\#1””@noop \[0\][secondoftwo]{}sanitize@url \[0\][‘\
12‘\$12 ‘&12‘\#12‘12‘\_12‘%12]{}@startlink\[1\]@endlink\[0\]@bib@innerbibempty @noop [**]{} (, ) @noop [****, ()]{} @noop [**]{} (, ) @noop [**]{} (, ) @noop [ ()]{} [****, ()](\doibase
10.1142/S0129183105007091) @noop [**]{} (, ) @noop [ ()]{} @noop [****, ()]{} @noop [**]{} (, ) @noop [**]{} (, ) @noop [**]{} (, ) @noop [****, ()]{} @noop [**]{} (, ) @noop [****, ()]{} @noop [****, ()]{} @noop [****, ()]{} @noop [****, ()]{} @noop [****, ()]{} @noop [****, ()]{} @noop [****, ()]{} @noop [****, ()]{} @noop [**]{} (, ) @noop [****, ()]{} @noop [****, ()]{} @noop [****, ()]{} @noop [**]{} (, , ) @noop [**]{} (, , ) @noop [“,” ]{} @noop [****, ()]{} @noop [**]{} (, )
[^1]: Generally, a notion of *involution* between two observables—vanishing of the Poisson bracket between them—requires a Hamiltonian reformulation of the laws of dynamics of the system. However, for the two-dimensional systems, the only zero Poisson bracket required is the one between the additional conserved quantity and the Hamiltonian; the latter is simply automatic.
[^2]: This can be shown, in particular, by observing that (a) the outgoing pair $(W_{\text{out}},\,w_{\text{out}})$ is an image of the incident pair, $(W_{\text{out}},\,w_{\text{out}})$, upon application of one of the elements of the group, and that (b) the condition (\[outgoing\_state\]) defines a particular *chamber* of this group. However, by construction, there is only one point of an orbit of a group per chamber.
| {
"pile_set_name": "ArXiv"
} | ArXiv |
---
abstract: 'A new theoretical framework is proposed for the description of baryon light-cone distribution amplitudes, based on the observation that their scale dependence to leading logarithmic accuracy is described by a completely integrable model. The physical interpretation is that one is able to find a new quantum number that distinguishes partonic components in the nucleon with different scale dependence.'
address: |
Institut für Theoretische Physik, Universität Regensburg,\
D-93040 Regensburg, Germany\
Vladimir.Braun@physik.uni-regensburg.de
author:
- 'V.M. Braun'
date: 19 June 2000
title: Baryon wave functions in QCD and integrable models
---
hep-ph/0011284
Introduction
============
The notion of baryon distribution amplitudes refers to the valence component of the Bethe-Salpeter wave function at small transverse separations and is central for the theory of hard exclusive reactions involving baryons [@earlybaryon]. As usual for a field theory, extraction of the asymptotic behavior (here: zero transverse separation) introduces divergences that can be studied by the renormalization-group (RG) method. The distribution amplitude $\phi$ thus becomes a function of the three quark momentum fractions $x_i$ and the scale that serves as a UV cutoff in the allowed transverse momenta. Solving the corresponding RG equations one is led to the expansion $$\begin{aligned}
\phi(x_i,Q^2) &=& 120 x_1 x_2 x_3 \sum_{N,q}\phi_{N,q}(Q_0^2)
P_{N,q}(x_i)
\left(\frac{\alpha_s(Q^2)}{\alpha_s(Q_0^2)}\right)^{\gamma_{N,q}/b}
\label{expansion}
\end{aligned}$$ where the summation goes over all multiplicatively renormalizable operators built of three quarks and $N$ derivatives. The polynomials $P_{N,q}(x_i)$ and anomalous dimensions $\gamma_{N,q}$ are obtained by the diagonalization of the mixing matrix for the three-quark operators $$\begin{aligned}
B_{k_1,k_2,k_3} &=& (D_+^{k_1} q) (D_+^{k_2} q)(D_+^{k_3} q); \quad
k_1+k_2+k_3 =N\end{aligned}$$ and $\phi_{N,q}(Q_0^2)$ are the corresponding (nonperturbative) matrix elements.
As well known, conformal symmetry allows one to resolve the mixing with operators containing total derivatives [@ER]–[@Nyeo] and requires that the ’eigenfunctions’ $P_{N,q}$ corresponding to different values of $N$ are mutually orthogonal with the weight function $ 120 x_1 x_2 x_3$ that plays the role of the asymptotic wave function. In difference to mesons, the conformal symmetry is not sufficient, however, to solve the RG equations: for each $N$ there exist $N+1$ independent operators with the same conformal spin which mix with each other. This mixing produces a nontrivial spectrum of anomalous dimensions, see Fig. 1, about which little was known until recently.
In what follows, I describe a new approach to the scale dependence of baryon distribution amplitude developed in [@BDM; @BDKM]. The main result is that one is able to identify the second summation index $q$ in (\[expansion\]) with an eigenvalue of a certain conserved charge. The physical interpretation is that one is able to find a new ‘hidden’ quantum number that distinguishes between partonic components in the proton with different scale dependence.
Hamiltonian formulation of the evolution equations
==================================================
In the usual formulation of the evolution equations the conformal invariance is not explicit. It can be made manifest, however, if the evolution kernels are rewritten in terms of the generators of the SL(2) collinear subgroup of conformal transformations. The scale dependence is different for helicity $\lambda=3/2$ and $\lambda=1/2$ operators and the corresponding evolution kernels can be written in the following compact form [@BDM; @BDKM]: $$\begin{aligned}
H_{3/2} &=& 2\left(1+\frac{1}{N_c}\right)\sum_{i<k}\Big[
\psi(J_{ik})-\psi(2)\Big]+\frac32 C_F\,,
\label{3/2}\\
H_{1/2} &=& H_{3/2}- 2\left(1+\frac{1}{N_c}\right)\left[
\frac{1}{J_{12}(J_{12}-1)}+
\frac{1}{J_{23}(J_{23}-1)}\right].
\label{1/2}\end{aligned}$$ Here $\psi(x)$ is the logarithmic derivative of the $\Gamma$-function and $J_{ik}, i,k =1,2,3$ are defined in terms of the two-particle Casimir operators of the $SL(2,R)$ group $$\begin{aligned}
J_{ik}(J_{ik}-1) &=& L_{ik}^2 \equiv (\vec L_i+\vec L_k)^2\,, \end{aligned}$$ with $\vec L_i$ being the group generators acting on the i-th quark. In the momentum fraction representation (\[expansion\]) the generators take the form [@BDKM] $$\begin{aligned}
L_{k,0} P(x_i) &=& (x_k\partial_k + 1) P(x_i)\,,\nonumber\\
L_{k,+} P(x_i) &=& - x_k P(x_i)\,,\nonumber\\
L_{k,-} P(x_i) &=& (x_k\partial_k^2 + 2\partial_k) P(x_i)\,.\end{aligned}$$ Solution of the evolution equations corresponds in this language to solution of the Schrödinger equation $$\begin{aligned}
H P_{N,q}(x_i) &=& \gamma_{N,q} P_{N,q}(X_i)\end{aligned}$$ with $\gamma_{N,q}$ being the anomalous dimensions. The $SL(2,R)$ invariance of the evolution equations implies that the generators of conformal transformations commute with the ‘Hamiltonians’ $$\begin{aligned}
[H,L^2]&=& [H,L_\alpha] = 0\,,\end{aligned}$$ where $L^2 = (\vec L_1+\vec L_2+\vec L_3)^2$ and $L_\alpha = L_{1,\alpha}+L_{3,\alpha}+L_{3,\alpha}$, so that the polynomials $P_{N,q}(x_i)$ corresponding to multiplicatively renormalizable operators can be chosen simultaneously to be eigenfunctions of $L^2$ and $L_0$: $$\begin{aligned}
L^2 P_{N,q} &=& (N+3)(N+2) P_{N,q}\,,\nonumber\\
L_0 P_{N,q} &=& (N+3) P_{N,q}\,,\nonumber\\
L_- P_{N,q} &=& 0\,.
\label{conform} \end{aligned}$$ The third condition in (\[conform\]) ensures that the operators do not contain overall total derivatives.
Main finding of [@BDM] is that the Hamiltonian $H_{3/2}$ possesses an additional integral of motion (conserved charge): $$\begin{aligned}
Q = \frac{i}{2}[L^2_{12},L^2_{23}] = i(\partial_1\!-\!\partial_2)
(\partial_2\!-\!\partial_3)(\partial_3\!-\!\partial_1) x_1 x_2 x_3
\,,
\quad [H_{3/2},Q] = 0\,. \end{aligned}$$ The evolution equation for baryon distribution functions with maximum helicity is, therefore, completely integrable. The premium is that instead of solving a Schrödinger equation with a complicated nonlocal Hamiltonian, it is sufficient to solve a much simpler equation $$\begin{aligned}
Q P_{N,q}(x_i) = q P_{N,q}(x_i)\,.
\label{Q}\end{aligned}$$ Once the eigenfunctions are found, the eigenvalues of the Hamiltonian (anomalous dimensions) are obtained as algebraic functions of $N,q$.
It is necessary to add that the Hamiltonian in (\[3/2\]) is known as the Hamiltonian describing the so-called $XXX_{s=-1}$ Heisenberg spin magnet. The same Hamiltonian was also encountered in the theory of interacting reggeons in QCD [@FK; @Lip].
Summary of results: Helicity $\lambda=3/2$ distributions
========================================================
The equation in (\[Q\]) cannot be solved exactly, but a wealth of analytic results can be obtained by means of the $1/N$ expansion [@K; @BDKM].
The general structure of the spectrum is illustrated in Fig. 2. It is easy to see that if $q$ is an eigenvalue of $Q$, then $-q$ is also an eigenvalue, whereas the Hamiltonian only depends on the absolute value $|q|$. It follows that all anomalous dimensions are double degenerate except for the lowest ones for each [*even*]{} $N$, corresponding to the solution with $q=0$. The corresponding eigenfunctions have a very simple form $$\begin{aligned}
x_1 x_2 x_3 P_{N,q=0}(x_i) &=& x_1(1-x_1)C^{3/2}_{N+1}(1-2x_1)
+x_2(1-x_2)C^{3/2}_{N+1}(1-2x_2)
\nonumber\\&&{}+x_3(1-x_3)C^{3/2}_{N+1}(1-2x_3)
\label{lowest}\end{aligned}$$ and the anomalous dimension is equal to $$\begin{aligned}
\gamma_{N,q=0} &=& \left(1+1/N_c\right)
\Big[4\psi(N+3)+4\gamma_E-6\Big]
+3/2 C_F\,.
\label{lowen}\end{aligned}$$ Furthermore, the eigenvalues of $Q$ lie on trajectories (see Fig. 2) corresponding to the semi-classically quantized soliton waves [@KK]. The corresponding trajectories for the anomalous dimensions have a rather peculiar form. Each of them can be considered as a separate partonic component in the nucleon wave function, in the same spirit as the leading-twist parton distribution functions in deep-inelastic scattering arises from the analytic continuation of the anomalous dimensions, giving rise to the Altarelli-Parisi splitting function. The asymptotic expansions for the charge $q$ and the anomalous dimensions at large $N$ are available to the order $1/N^8$ [@K; @BDKM] and give very accurate results. The algebraic structure of the spectrum is very complicated. As an example, I present the expression for the anomalous dimension $\gamma_{N,q}=(1+1/N_c)E_{N,q}+3/2 C_F$ as a function of the charge $q$: [@K] $$\begin{aligned}
E_{N,q} &=& 2 \ln 2 -6 +6\gamma_E + 2\mbox{\rm Re}\sum_{k=1}^3\psi(1+i\eta^3
\delta_k) + { O}(\eta^{-6})\,,
\label{disp}\end{aligned}$$ where $\eta = \sqrt{(N+3)(N+2)}$ and $\delta_k$, $k=1,2,3$ are the three roots of the cubic equation $$\begin{aligned}
2 \delta_k^3 - \delta_k - q/\eta^3.\end{aligned}$$ Accuracy of (\[disp\]) is excellent, as illustrated in Fig. 3.
Helicity $\lambda=1/2$ distributions
====================================
The additional term in $H_{1/2}$ spoils integrability but can be considered as a small correction for the most part of the spectrum [@BDKM]. Its effect on the two lowest levels is drastic, however. To illustrate this, consider the flow of energy levels for the Hamiltonian $H(\epsilon) = \sum_{i<k}\Big[
\psi(J_{ik})-\psi(2)\Big] - \epsilon \Big[1/L_{12}^2+1/L_{23}^2\Big]$ (cf. (\[3/2\]), (\[1/2\])) as a function of an auxiliary parameter $\epsilon$, see Fig. 4.
It is seen that the two lowest levels decouple from the rest of the spectrum and are separated from it by a finite mass gap. As shown in [@BDKM], this phenomenon can be interpreted as binding of the two quarks with opposite helicity and forming a scalar “diquark”. The effective Hamiltonian for the low-lying levels can be constructed and turns out to be a generalization of the famous Kroning-Penney problem for a particle in a $\delta$-function type periodic potential. The value of the mass gap between the lowest and the next-to-lowest anomalous dimensions at $N\to\infty$ can be calculated combining the small-$\epsilon$ and the large-$\epsilon$ expansions and is equal to $$\begin{aligned}
\Delta\gamma = 0.32\cdot(1+1/N_c)\end{aligned}$$ in agreement with the direct numerical calculations.
Further developments
====================
The approach developed in [@BDM; @BDKM] turns out to be general and is applicable to the analysis of all three-parton systems in QCD, albeit with some modifications. The extension to three three-gluon operators is straightforward but tedious, since elementary fields have a higher conformal spin [@Belitsky1]. Probably more interesting from both phenomenological and mathematical point of view are the applications to quark-antiquark-gluon systems. The corresponding evolution equations are integrable in the limit of large number of colors [@BDM] and reduce to a different type of integrable models - the so-called open chains. A rather detailed analysis of integrable quark-antiquark-gluon operators has been given in [@BDM; @Belitsky2; @DKM] and, most recently, a concrete phenomenological application of this method to the structure function $g_2(x,Q^2)$ was considered in work [@BKM00].
To summarize, a powerful mathematical framework has been developed for the description of the evolution of baryon distribution amplitudes. The mathematical structure of evolution equations is very elegant and reveals certain qualitative features of the distribution amplitudes. A lot of analytical results is obtained, in different limits. The formalism is general and was applied already to the other existing three-parton distributions. In a more general context, the integrability of evolution equations reveals an additional symmetry of QCD and its close relation to exactly solvable statistical models. Remarkably enough, the same symmetry has been observed in the studies of the Regge asymptotics of three-gluon distributions. All these features are not seen at the level of QCD Lagrangian and their origin has to be understood better.
Acknowledgments {#acknowledgments .unnumbered}
===============
The author is grateful to S. Derkachov, G. Korchemsky and A. Manashov for a very rewarding collaboration on the subject of this report. Special thanks are due to H.C. Pauli for the invitation to this conference.
[99]{} V.A. Avdeenko, V.L. Chernyak and S.A. Korenblit, Yad. Fiz. [**3**3]{} (1981) 481;\
G.P. Lepage and S.J. Brodsky, Phys. Rev. Lett. [**4**3]{} (1979) 545, 1625 (E);\
S.J. Brodsky, G.P. Lepage and A.A. Zaidi, Phys. Rev. [**D**23]{} (1981) 1152;\
S.J. Brodsky and G.P. Lepage, Phys. Rev. [**D**24]{} (1981) 2848;\
A.I. Milshtein and V.S. Fadin, Yad. Fiz. [**3**5]{} (1982) 1603.
A.V. Radyushkin, Preprint JINR-P2-10717, Jun 1977;\
A.V. Efremov and A.V. Radyushkin, Phys. Lett. [**B**94]{} (1980) 245; Teor. Mat. Fiz. [**42**]{} (1980) 147.
S.J. Brodsky et al., Phys. Lett. [**B**91]{} (1980) 239; Phys. Rev. [**D**33]{} (1986) 1881. Yu.M. Makeenko, Sov. J. Nucl. Phys. [**3**3]{} (1981) 440.
Th. Ohrndorf, Nucl. Phys. [**B198**]{} (1982) 26.
G.P. Lepage and S.J. Brodsky, Phys. Rev. Lett. [**4**3]{} (1979) 545; Erratum-ibid. [**4**3]{} (1979) 1625.
M.E. Peskin, Phys. Lett. [**8**8B]{} (1979) 128.
K. Tesima, Nucl. Phys. [**B**202]{} (1982) 523.
Su-Long Nyeo, Z. Phys. [**C**54]{} (1992) 615.
N.G. Stefanis, Acta Phys. Polon. [**B**25]{} (1994) 1777; Dissertation, JINR, 1997 (unpublished).
V.M. Braun, S.E. Derkachov and A.N. Manashov, Phys. Rev. Lett. [**8**1]{} (1998) 2020.
V.M. Braun, S.E. Derkachov, G.P. Korchemsky and A.N. Manashov, Nucl. Phys. [**B553**]{} (1999) 355.
L.D. Faddeev and G.P. Korchemsky, Phys. Lett. [**B**342]{} (1995) 311.
L.N. Lipatov, JETP Lett. [**5**9]{} (1994) 596.
G.P. Korchemsky, Nucl. Phys. [**B**443]{} (1995) 255; [*ibid.*]{} [**B**462]{} (1996) 333; [*ibid.*]{} [**B**498]{} (1997) 68; Preprint LPTHE-Orsay-97-62 \[hep-ph/9801377\].
G.P. Korchemsky and I.M. Krichever, Nucl. Phys. [**B**505]{} (1997) 387.
A. V. Belitsky, Nucl. Phys. [**B574**]{} (2000) 407.
A. V. Belitsky, Nucl. Phys. [**B558**]{} (1999) 259.
S. E. Derkachov, G. P. Korchemsky and A. N. Manashov, Nucl. Phys. [**B566**]{} (2000) 203.
V. M. Braun, G. P. Korchemsky and A. N. Manashov, Phys. Lett. [**B476**]{} (2000) 455.
| {
"pile_set_name": "ArXiv"
} | ArXiv |
---
abstract: 'We present the results of a narrow-band near-infrared imaging survey for Molecular Hydrogen emission-line Objects (MHOs) toward 26 regions containing high-mass protostellar candidates and massive molecular outflows. We have detected a total of 236 MHOs, 156 of which are new detections, in 22 out of the 26 regions. We use H$_2$ 2.12-$\mu$m/H$_2$ 2.25-$\mu$m flux ratios, together with morphology, to separate the signatures of fluorescence associated with photo-dissociation regions (PDRs) from shocks associated with outflows in order to identify the MHOs. PDRs have typical low flux ratios of $\sim$ 1.5 - 3, while the vast majority of MHOs display flux ratios typical of C-type shocks ($\sim$ 6-20). A few MHOs exhibit flux ratios consistent with expected values for J-type shocks ($\sim$ 3-4), but these are located in regions that may be contaminated with fluorescent emission. Some previously reported MHOs have low flux ratios, and are likely parts of PDRs rather than shocks indicative of outflows. We identify a total of 36 outflows across the 22 target regions where MHOs were detected. In over half these regions, MHO arrangements and fluorescent structures trace features present in CO outflow maps, suggesting the CO emission traces a combination of dynamical effects, which may include gas entrained in expanding PDRs as well as bipolar outflows. Where possible, we link MHO complexes to distinct outflows and identify candidate driving sources.'
author:
- 'Grace Wolf-Chase'
- Kim Arvidsson
- Michael Smutko
title: 'MHOs toward HMOs: A Search for Molecular Hydrogen emission-line Objects toward High-Mass Outflows '
---
INTRODUCTION
============
Outflows play an important role in the formation of stars across the entire stellar mass spectrum (e.g., Bally 2016); however, the nature of that role in the formation of stars $>$ 8 M$_{\odot}$ remains poorly understood, since massive stars form in highly clustered environments that contain a mix of objects in different evolutionary stages. Zhang et al. (2005; hereafter, ZHB05) detected a total of 39 molecular outflows towards 69 luminous [*IRAS*]{} point sources associated with dense molecular gas and far-infrared luminosities indicative of massive star formation. From observations of the CO J=2$\rightarrow$1 transition with a resolution of $\sim$ 30$^{\prime\prime}$, they presented maps and derived physical parameters for 35 of these outflows, showing that their mass, momentum, and energy range from one to a few orders of magnitude larger than typical values for outflows associated with low-mass young stellar objects (YSOs). Although nearly half of the outflows mapped show spatially resolved bipolar lobes, the remainder show complicated lobe morphology or little evidence of bipolarity, and in several cases, the outflow centroid is clearly offset from the [*IRAS*]{} point source position.
The H$_2$ 1-0 S(1) line at 2.12 $\micron$ is a powerful tracer of shocks in molecular outflows (Davis et al. 2010 and references therein; Smith 2012). It can be used to identify collimated outflows and candidate driving sources even in massive star-forming regions (e.g., Varricatt et al. 2010; hereafter VDR10), with the caveat that H$_2$ emission produced by shocks (Molecular Hydrogen emission-line Objects or MHOs) is typically interspersed with fluorescent emission from photo-dissociation regions (PDRs) associated with expanding regions. Although morphology of the H$_2$ emission is typically used to distinguish MHOs, H$_2$ 2.12 $\micron$/2.25 $\micron$ flux ratios provide a more robust method for separating the effects of shocks and fluorescence (Wolf-Chase et al. 2013; hereafter, WAS13). To investigate the relationship between massive CO outflows, MHOs, and their driving sources, we acquired 4.$^{\prime}$6 $\times$ 4.$^{\prime}$6 narrow-band near-infrared images at a resolution of $\sim$ 1$^{\prime\prime}$ toward 25 of the CO outflows mapped by ZHB05 and one outflow mapped by Zhang et al. (2007).
This paper is organized as follows. Details of the observations and data reduction are presented in §2. Table 1 lists the observations for all targets. Results and discussion of the survey are presented in §3. Table 2 indicates whether targets contain detected MHOs, catalogued regions, and catalogued massive young stellar objects (MYSOs). Table 3 lists the positions and fluxes of all MHOs for each target where MHOs were detected. Table 4 lists candidate outflows and driving sources. Our summary and conclusions are presented in §4. Images, results and discussion of individual regions containing MHOs are presented in Appendix A, except for Mol 121 & Mol 160, which were published previously (WAS13; Wolf-Chase et al. 2012, hereafter WSS12).
OBSERVATIONS AND DATA REDUCTION
===============================
We obtained H$_2$ 2.12-$\micron$, H$_2$ 2.25-$\micron$ and H$_2$r 2.13-$\micron$ (narrow-band continuum) images of 26 regions thought to contain massive YSOs, using the Near-Infrared Camera and Fabry-Perot Spectrometer (NICFPS: Vincent et al. 2003; Hearty et al. 2004) on the Astrophysical Research Consortium (ARC) 3.5-m telescope at the Apache Point Observatory (APO) in Sunspot, NM. All targets contain energetic molecular outflows identified through CO observations and thought be associated with massive YSOs (ZHB05 Table 2; Zhang et al. 2007). Table 1 lists the targets by [*IRAS*]{} designation (column 1) and Molinari number (Molinari et al. 1996, 1998, 2000, 2002), center position (columns 2 & 3), filter (column 4), date (column 5), total exposure time (column 6), and resolution (column 7) as determined by seeing conditions at the time of the observations. We used the narrowband filter centered on 2.13 $\micron$ to allow for continuum subtraction in the final images. All images were acquired with a 4.$^{\prime}$6 $\times$ 4.$^{\prime}$6 FOV and pixel scale of 0.$^{\prime\prime}$273 pixel$^{-1}$. Our basic data acquisition, reduction, and calibration methods are described in WSS12, and the methods we used to perform irregular aperture photometry on extended emission are presented in WAS13.
The target regions have varying brightness and morphologies, as well as varying background emission, which makes it impossible to apply one criterium for the spatial extent of all suspected MHOs. Rather, the morphologies of the regions used for aperture photometry were determined by signal-to-noise with respect to the background in the continuum subtracted images. To reduce the effect of the patchy nature of the extended background emission in continuum subtracted images, the images were smoothed using a 3-pixel Gaussian and the resulting background noise measured. The regions were then chosen to trace contours of multiples of the smoothed background noise. This could range from 10 times the background noise in the fainter regions to 60 times inside some of the brighter emission, but for most regions, 20 to 40 times the background noise contours were used. Whenever possible, the regions were chosen to trace emission that has approximately the same morphology in both 2.12 and 2.25-$\micron$ images.
[lllllcc]{} 00117$+$6412 (2) & 00 14 27.7 & +64 28 46 & H$_2$ 2.12 $\micron$ & 071218 & 1800 & 0.85\
& & & H$_2$ 2.25 $\micron$ & 071218 & 1800 &\
& & & H$_2$r 2.13 $\micron$ & 071218 & 1800 &\
00420$+$5530 (3) & 00 44 57.6 & +55 47 18 & H$_2$ 2.12 $\micron$ & 071119 & 1800 & 0.87\
& & & H$_2$ 2.25 $\micron$ & 071119 & 1800 &\
& & & H$_2$r 2.13 $\micron$ & 071119 & 1800 &\
05137$+$3919 (8) & 05 17 13.3 & +39 22 14 & H$_2$ 2.12 $\micron$ & 071203;070227;070204 & 4800 & 0.94\
& & & H$_2$ 2.25 $\micron$ & 071203 & 1800 &\
& & & H$_2$r 2.13 $\micron$ & 071203;070227;070204 & 4800 &\
05168$+$3634 (9) & 05 20 16.2 & +36 37 21 & H$_2$ 2.12 $\micron$ & 071218 & 1200 & 0.79\
& & & H$_2$ 2.25 $\micron$ & 071218 & 1200 &\
& & & H$_2$r 2.13 $\micron$ & 071218 & 1200 &\
05274$+$3345 (10) & 05 30 48.0 & +33 47 52 & H$_2$ 2.12 $\micron$ & 080115 & 1800 & 1.48\
& & & H$_2$ 2.25 $\micron$ & 080115 & 1800 &\
& & & H$_2$r 2.13 $\micron$ & 080115 & 1800 &\
05345$+$3157 (11) & 05 37 47.8 & +31 59 24 & H$_2$ 2.12 $\micron$ & 051224 & 3900 & 1.17\
& & & H$_2$ 2.25 $\micron$ & 061013 & 1800 &\
& & & H$_2$r 2.13 $\micron$ & 051224 & 3900 &\
05373$+$2349 (12) & 05 40 24.4 & +23 50 54 & H$_2$ 2.12 $\micron$ & 080319 & 1680 & 1.03\
& & & H$_2$ 2.25 $\micron$ & 080319 & 1680 &\
& & & H$_2$r 2.13 $\micron$ & 080319 & 1680 &\
06056$+$2131 (15) & 06 08 41.0 & +21 31 01 & H$_2$ 2.12 $\micron$ & 061010;061013 & 3000 & 0.84\
& & & H$_2$ 2.25 $\micron$ & 061013 & 1200 &\
& & & H$_2$r 2.13 $\micron$ & 061013 & 3000 &\
06584$-$0852 (28) & 07 00 51.0 & -08 56 29 & H$_2$ 2.12 $\micron$ & 070227;070327;080319 & 5400 & 1.21\
& & & H$_2$ 2.25 $\micron$ & 070327;080319 & 3600 &\
& & & H$_2$r 2.13 $\micron$ & 070227;070327;080319 & 5400 &\
18532$+$0047 (78) & 18 55 50.6 & +00 51 22 & H$_2$ 2.12 $\micron$ & 060612;060613 & 4500 & 1.38\
& & & H$_2$ 2.25 $\micron$ & 080611;080915 & 1800 &\
& & & H$_2$r 2.13 $\micron$ & 060613;080611;080915 & 4500 &\
19213$+$1723 (103) & 19 23 36.9 & +17 28 59 & H$_2$ 2.12 $\micron$ & 060513;060514 & 5400 & 0.98\
& & & H$_2$ 2.25 $\micron$ & & &\
& & & H$_2$r 2.13 $\micron$ & 060513 & 2700 &\
19368$+$2239 (108) & 19 38 58.1 & +22 46 32 & H$_2$ 2.12 $\micron$ & 060618 & 3240 & 0.87\
& & & H$_2$ 2.25 $\micron$ & 080530 & 1680 &\
& & & H$_2$r 2.13 $\micron$ & 060616;080530 & 3240 &\
19374$+$2352 (109) & 19 39 33.2 & +23 59 55 & H$_2$ 2.12 $\micron$ & 060606 & 2700 & 0.89\
& & & H$_2$ 2.25 $\micron$ & 080519 & 1800 &\
& & & H$_2$r 2.13 $\micron$ & 060606;080519 & 2700 &\
19388$+$2357 (110) & 19 40 59.4 & +24 04 39 & H$_2$ 2.12 $\micron$ & 080611 & 2400 & 1.12\
& & & H$_2$ 2.25 $\micron$ & 080611 & 2400 &\
& & & H$_2$r 2.13 $\micron$ & 080611 & 2400 &\
20050$+$2720 (114) & 20 07 06.7 & +27 28 53 & H$_2$ 2.12 $\micron$ & 080729;080915 & 2100 & 1.00\
& & & H$_2$ 2.25 $\micron$ & 080729;080915 & 2100 &\
& & & H$_2$r 2.13 $\micron$ & 080729;080915 & 2100 &\
20056$+$3350 (115) & 20 07 31.5 & +33 59 39 & H$_2$ 2.12 $\micron$ & 080915;081109 & 2400 & 0.71\
& & & H$_2$ 2.25 $\micron$ & 080915;081109 & 2400 &\
& & & H$_2$r 2.13 $\micron$ & 080915;081109 & 2400 &\
20188$+$3928 (121) & 20 20 39.3 & +39 37 52 & H$_2$ 2.12 $\micron$ & 081013 & 1740 & 1.21\
& & & H$_2$ 2.25 $\micron$ & 081013 & 1740 &\
& & & H$_2$r 2.13 $\micron$ & 081013 & 1740 &\
20220$+$3728 (123) & 20 23 55.7 & +37 38 10 & H$_2$ 2.12 $\micron$ & 110620 & 1950 & 1.34\
& & & H$_2$ 2.25 $\micron$ & 110620 & 1950 &\
& & & H$_2$r 2.13 $\micron$ & 110620 & 1950 &\
20278$+$3521 (125) & 20 29 46.9 & +35 31 39 & H$_2$ 2.12 $\micron$ & 081111;081112 & 1800 & 0.99\
& & & H$_2$ 2.25 $\micron$ & 081111;081112 & 1800 &\
& & & H$_2$r 2.13 $\micron$ & 081111;081112 & 1800 &\
20286$+$4105 (126) & 20 30 27.9 & +41 15 48 & H$_2$ 2.12 $\micron$ & 110613 & 1800 & 0.80\
& & & H$_2$ 2.25 $\micron$ & 110613 & 1800 &\
& & & H$_2$r 2.13 $\micron$ & 110613 & 1800 &\
21307$+$5049 (136) & 21 32 31.5 & +51 02 22 & H$_2$ 2.12 $\micron$ & 070105 & 1800 & 1.09\
& & & H$_2$ 2.25 $\micron$ && &\
& & & H$_2$r 2.13 $\micron$ & 070105 & 1800 &\
22172$+$5549 (143) & 22 19 09.0 & +56 04 45 & H$_2$ 2.12 $\micron$ & 050703;051011;051016;061202 & 4740 & 0.85\
& & & H$_2$ 2.25 $\micron$ & 051016 & 3600 &\
& & & H$_2$r 2.13 $\micron$ & 051011;051016;061202 & 4740 &\
22305$+$5803 (148) & 22 32 24.3 & +58 18 58 & H$_2$ 2.12 $\micron$ & 081109 & 1470 & 0.90\
& & & H$_2$ 2.25 $\micron$ & 081109 & 1470 &\
& & & H$_2$r 2.13 $\micron$ & 081109 & 1470 &\
22506$+$5944 (151) & 22 52 38.6 & +60 00 56 & H$_2$ 2.12 $\micron$ & 081111;081112 & 1800 & 1.00\
& & & H$_2$ 2.25 $\micron$ & 081111;081112 & 1800 &\
& & & H$_2$r 2.13 $\micron$ & 081111;081112 & 1800 &\
23314$+$6033 (158) & 23 33 44.3 & +60 50 30 & H$_2$ 2.12 $\micron$ & 071119 & 1800 & 0.93\
& & & H$_2$ 2.25 $\micron$ & 071119 & 1800 &\
& & & H$_2$r 2.13 $\micron$ & 071119 & 1800 &\
23385$+$6053 (160) & 23 40 53.2 & +61 10 21 & H$_2$ 2.12 $\micron$ & 050617 & 1800 & 0.75\
& & & H$_2$ 2.25 $\micron$ & 071121 & 1200 &\
& & & H$_2$r 2.13 $\micron$ & 071121 & 1200 &\
SURVEY RESULTS AND DISCUSSION
=============================
Table 2 identifies target regions (columns 1 & 2) as [*High*]{} (H) or [*Low*]{} (L), depending upon whether the associated [*IRAS*]{} source has infrared colors typical of an Ultra-Compact (UC) region or redder, respectively (Wood & Churchwell 1989; Palla et al. 1991). The designation (UC) indicates candidate UC regions based on the detection of centimeter-radio continuum emission at levels $>$ 1 mJy within 40$^{\prime\prime}$ of the [*IRAS*]{} position (Gómez-Ruiz et al. 2016). Columns 3 & 4 indicate whether a targeted region contains MHOs and, if so, whether collimated outflow(s) could be identified. Columns 5 & 6 list objects in the [*WISE*]{} catalog of Galactic regions within the targeted areas and their classifications by type (Anderson et al. 2014). Columns 7 & 8 list objects in the Red [*Midcourse Space Experiment*]{} ([*MSX*]{}: Egan et al. 2003) Source catalog (RMS: Lumsden et al. 2013) located within each targeted region and their type. In §3.1 we present our MHO results and in §3.2 we discuss their relationship to CO outflows. In §3.3 we discuss possible associations of MHOs with other signposts of massive star formation within surveyed regions, and in §3.4 we discuss identification of candidate outflow driving sources.
[llcclclr]{} 00117$+$6412 (2) & H(UC) & Y & Y & G118.964+01.893 & C & &\
00420$+$5530 (3)\* & L & Y & Y & G122.002-07.084 & C & &\
05137$+$3919 (8)\* & L & Y & Y & G168.066+00.819 & Q & G168.0627+00.8221 & YSO\
05168$+$3634 (9)\* & H & Y & Y & & & &\
05274$+$3345 (10)\* & H & Y & Y & G174.197-00.076 & Q & G174.1974-00.0763 & YSO\
05345$+$3157 (11)\* & L & Y & Y & G176.507+00.178 & C & &\
05373$+$2349 (12)\* & L & Y & Y & G183.720-03.665 & Q & G183.7203-03.6647 & YSO\
06056$+$2131 (15) & H(UC) & Y & Y & G189.030+00.780 & Q & G189.0307+00.7821 & YSO\
& & & & G189.032+00.809 & Q & G189.0323+00.8092 & YSO\
06584$-$0852 (28)\* & L & Y & N & G221.955-01.993 & Q & G221.9605-01.9926 & YSO\
18532$+$0047 (78) & H(UC) & N & …& G034.195-00.603 & Q & &\
& & & & G034.196-00.592 & K & &\
& & & & G034.204-00.584 & Q & &\
19213$+$1723 (103)\* & H(UC) & N & …& G052.098+01.042 & K & G052.0986+01.0417 & Diffuse Region\
19368$+$2239 (108) & H & Y & Y & G058.453+00.431 & Q & G058.4670+00.4360A & YSO\
& & & & G058.467+00.436 & Q & G058.4670+00.4360B & Region\
& & & & G058.477+00.427 & Q & &\
19374$+$2352 (109)\* & H(UC) & Y & N & G059.602+00.911 & K & G059.6032+00.9116A & Region\
& & & & G059.612+00.917 & G & G059.6032+00.9116B & Region\
& & & & & & G059.6032+00.9116C & Region\
& & & & & & G059.6032+00.9116D & Region\
19388$+$2357 (110)\* & H(UC) & Y & N & G059.829+00.671 & Q & G059.8329+00.6729 & YSO\
20050$+$2720 (114)\* & H & Y & Y & & & G065.7798-02.6121 & YSO\
20056$+$3350 (115)\* & H & N & & G071.312+00.828 & C & &\
20188$+$3928 (121)$^c$\* & H(UC) & Y & Y & G077.463+01.760 & C & G077.4622+01.7600A & YSO\
& & & & & & G077.4622+01.7600B & Region\
20220$+$3728 (123) & H(UC) & N & & G076.180+00.064 & Q & G076.1877+00.0974 & Region\
& & & & G076.187+00.097 & C & &\
& & & & G076.197+00.092 & C & &\
20278$+$3521 (125) & L & Y & Y & G075.155-02.087 & Q & &\
20286$+$4105 (126)\* & H & Y & N & G079.869+01.180 & C & G079.8749+01.1821 & Region\
21307$+$5049 (136)\* & L(UC) & Y & N & G094.263-00.414 & C & G094.2615-00.4116 & YSO\
22172$+$5549 (143)\* & L & Y & Y & & & G102.8051-00.7184A & YSO\
& & & & & & G102.8051-00.7184B & YSO\
& & & & & & G102.8051-00.7184C & YSO\
22305$+$5803 (148)\* & H & Y & Y & G105.509+00.230 & Q & G105.5072+00.2294 & YSO\
22506$+$5944 (151) & H & Y & Y & G108.603+00.494 & Q & G108.5955+00.4935A & YSO\
& & & & & & G108.5955+00.4935B & YSO\
& & & & & & G108.5955+00.4935C & YSO\
23314$+$6033 (158) & L & Y & N & G113.614-00.615 & G & G113.6041-00.6161 & Region\
23385$+$6053 (160)$^d$ & L & Y & N & G114.526-00.543 & C & &\
H$_2$ Emission
--------------
To choose our final list of MHOs, we excluded emission where the continuum-subtracted H$_2$ 2.12-$\micron$/H$_2$ 2.25-$\micron$ flux ratio is $<$ 3. Such emission is typically associated with fluorescence from photo-dissociation regions (PDRs) and is generally diffuse and extended. Some previously reported MHOs do not meet our flux ratio criterion, but are included in Table 3 for the sake of completeness. We used morphology as an additional criterion to exclude PDRs, but in no case did we find regions exhibiting PDR morphology (diffuse, extended arcs) where the H$_2$ 2.12-$\micron$/H$_2$ 2.25-$\micron$ flux ratio is $>$ 4. Since the vast majority of MHOs have flux ratios significantly greater, we are confident that this method has successfully identified the MHOs in our targets.
For each target region containing MHOs, Table 3 lists its [*IRAS*]{} source designation and Molinari number (column 1), identified MHOs (column 2), positions of the peak emission (columns 3 & 4), areas used in the aperture photometry (column 5), H$_2$ 2.12-$\micron$ line fluxes (column 6), H$_2$ 2.25-$\micron$ line fluxes (column 7), H$_2$ 2.12-$\micron$/H$_2$ 2.25-$\micron$ flux ratios (column 8), and comments relevant to certain catalog entries (column 9). Where the H$_2$ 2.25-$\micron$ line flux is below 3 times the noise (3$\sigma$) within an aperture, the H$_2$ 2.25-$\micron$ line is regarded as undetected and no value is given in column 7. In these cases, lower limits to the H$_2$ 2.12-$\micron$/H$_2$ 2.25-$\micron$ flux ratios are calculated using the 3$\sigma$ value for that aperture as an upper limit for the H$_2$ 2.25-$\micron$ line flux.
MHOs identified in Table 3 follow the numbering scheme used by the on-line catalog of these objects, which is currently hosted by the University of Kent[^1] and was initially published by Davis et al. (2010). In general, MHOs that appear to be associated with linear features are assigned the same number and given additional letter designations to identify individual knots. This is also the convention we use for MHOs that appear in groups. MHO numbers that are followed by “\_\#” indicate knots we have identified which we associate with previously identified MHO groups or outflows. This does not necessarily imply that all MHOs with the same given number are part of the same outflow. We discuss these associations on a case-by-case basis in the Appendix.
For the high excitation temperatures associated with J-shocks, one expects H$_2$ 2.12-$\micron$/2.25-$\micron$ $\sim$ 3 $-$ 4, and for the lower-excitation C-shocks, $\sim$ 6 $-$ 20. Above 20, the excitation temperature is $<$ 1000 K, and observable emission is not expected (Smith 1994, 1995; Smith et al. 2003). For the most part, our computed flux ratios are consistent with C-type shocks. A few MHOs with flux ratios that are more consistent with J-type shocks lie in the direction of PDRs and may include fluorescent emission. It is interesting to note that three of our surveyed regions, Mol 109, Mol 121 (WAS13), and Mol 126, exhibit one or more spots where the H$_2$ 2.12-$\micron$/2.25-$\micron$ flux ratio $<$ 1, which cannot be explained by either fluorescence or shocks.
****
[rrrrrrrrr]{} 00117$+$6412 (2) & 2967A & 00 14 33.6 & +64 28 39 & 3.341 & $6.83 \pm 0.52$ & $0.45 \pm 0.11$ & $15.1 \pm 3.7$ &\
& 2967B & 00 14 32.4 & +64 28 37 & 4.969 & $9.09 \pm 0.69$ & $0.57 \pm 0.14$ & $16.1 \pm 4.0$ &\
& 2967C & 00 14 30.9 & +64 28 35 & 1.165 & $1.42 \pm 0.12$ & $0.20 \pm 0.06$ & $7.1 \pm 2.0$ &\
& 2967D & 00 14 30.4 & +64 28 39 & 1.594 & $2.15 \pm 0.18$ & $0.22 \pm 0.07$ & $9.9 \pm 3.4$ &\
& 2967E & 00 14 28.8 & +64 28 34 & 1.405 & $3.73 \pm 0.29$ & & $ >19.5 $ &\
& 2967F & 00 14 28.8 & +64 28 27 & 4.609 & $6.84 \pm 0.53$ & $1.03 \pm 0.15$ & $6.6 \pm 1.1$ &\
& 2967G & 00 14 27.8 & +64 28 26 & 1.045 & $1.66 \pm 0.13$ & & $ >11.1 $ &\
& 2967H & 00 14 23.6 & +64 28 21 & 0.685 & $0.78 \pm 0.07$ & & $ >6.6 $ &\
& 2968A & 00 14 26.4 & +64 29 00 & 3.495 & $9.05 \pm 0.69$ & $0.99 \pm 0.14$ & $9.1 \pm 1.5$ &\
& 2968B & 00 14 26.5 & +64 29 05 & 1.422 & $2.38 \pm 0.19$ & $0.36 \pm 0.07$ & $6.7 \pm 1.4$ &\
Relationship of MHOs to Massive CO Outflows
-------------------------------------------
The targets for our survey were all chosen from sources associated with massive CO outflows listed in Table 2 of ZHB05, with the exception of Mol 10, which was mapped in CO by Zhang et al. (2007). We were able to identify MHOs in 22 of the 26 regions we observed (Table 3). Across these 22 regions, MHO detections fall into 3 morphological categories: (1) co-linear MHOs that define the position angle of one dominant outflow in the region; (2) multiple outflows along different position angles; and (3) isolated or scattered MHOs where no clear outflows or position angles can be identified. Table 4 lists candidate outflows (column 1) identified from MHO arrangements (column 2), possible driving sources (column 3), outflow position angles (column 4), and distinguishing MHO features (column 5).
In spite of the low resolution of the CO observations, there are several regions where chains of MHO detections follow the large-scale morphology of CO outflows presented in Figure 1 of ZHB05, including Mol 8, 9, 11, 12, 15, 109, 114 & 143. In some cases, both CO and H$_2$ emission appear to trace PDRs or other complex structures (e.g., Mol 2, 28, 126, & 136). This is not surprising, as CO emission has been associated with shells produced by expanding region bubbles identified from mid-infrared images (Dewangan et al. 2016; Churchwell et al. 2006). In a few cases, there is no clear relationship between the CO and H$_2$ emission (e.g., Mol 110, 125, 148, 151). There are several possible explanations for this: (1) While CO emission traces swept-up ambient material entrained in outflows, MHOs trace the current location of shocks, where the underlying wind is interacting with the ambient medium. It is possible that outflow axes change in time due to the precession of jets. (2) MHOs and CO emission may trace outflows from different sources. (3) Some differences may be explained by the size of the outflow and resolution of the observations. For example, Mol 151 displays a very compact ($<$ 1$^{\prime}$ in length) dramatic bipolar H$_2$ jet along an east-west direction, while the high-velocity CO emission shows little evidence of bipolarity except for a small east-west separation of $\sim$ 30$^{\prime\prime}$ between the peaks of the blue- and redshifted lobes, comparable to the resolution of the CO observations (ZHB05). (4) In some cases, extinction effects and proximity to the edges of molecular clouds may explain observed differences.
Association with Regions and YSOs
---------------------------------
Table 2 lists all sources in the [*WISE*]{} Catalog of Galactic Regions (Anderson et al. 2014) that are found within our targeted regions (Column 5). Column 6 indicates their classifications as known (K), grouped (G), candidate (C) or radio quiet (Q). K sources have measured Radio Recombination Lines (RRL) or H$\alpha$ emission that confirms their region status; C sources are spatially coincident with radio continuum emission, but do not have RRL or H$\alpha$ observations; G sources represent K & C sources that are associated positionally, typically within the radius of a PDR; and Q sources lack detected radio continuum emission and are presumably associated with pre-UC region objects or lower-luminosity intermediate-mass star-forming regions. Anderson et al. (2014) estimate that $\sim$95% of C sources are bona fide regions.
Ten of our target regions contain only Q sources (Table 2 - Mol 8, 10, 12, 15, 28, 108, 110, 125, 148, & 151), although two of these (Mol 15 & Mol 110) are identified as candidate UC regions by Gómez-Ruiz et al. (2016). All of these regions contain MHOs, and all but Mol 28 & 110 contain collimated outflows. Three of our targets with collimated MHOs have no entries in the [*WISE*]{} Catalog of Galactic Regions (Mol 9, 114, & 143). Of the four regions that harbor K or G sources (Mol 78, 103, 109, 158), two lack MHO detections entirely (Mol 78 & 103) and the other two (Mol 109 & 158) lack collimated flows. Although the three targets containing K sources all have [*IRAS*]{} colors associated with ‘High’ (H) objects, there is no obvious relationship between region class and [*IRAS*]{} color. The regions containing only Q sources represent a mix of H & L objects.
Table 2 also lists RMS sources found within our targeted regions (Column 7) and their type (Column 8). The RMS catalog is the largest statistically selected catalog of massive protostars and regions to date (Lumsden et al. 2013). It is thought to be complete for the detection of a B0 V star at the distance of the Galactic center, although inclusion in the catalog depends upon source detection by [*MSX*]{} and specific color criteria that may have excluded very young, compact objects containing a high fraction of ionized polycyclic aromatic hydrocarbons (PAHs) in thick PDRs (Kerton et al. 2015).
We additionally used infrared color criteria established by Koenig & Leisawitz (2014) and Fischer et al. (2016) to identify candidate Class 0 and Class I YSOs from the [*ALLWISE*]{} database, which combines data from the [*WISE*]{} cryogenic and [*NEOWISE*]{} post-cryogenic surveys (Wright et al. 2010; Mainzer et al. 2011). Class 0 and Class I YSOs are associated, respectively, with protostars in the main and late accretion phases of pre-main sequence evolution (Lada 1987; André et al. 1993).
Candidate Class I objects were required to satisfy all of the following criteria:
$$\begin{aligned}
W2-W3>2.0 \\
W2-W3<4.5 \\
W1-W2>0.46\times(W2-W3)-0.9 \\
W1-W2>-0.42\times(W2-W3)+2.2\end{aligned}$$
Class 0 candidates were required to satisfy both of these criteria:
$$\begin{aligned}
W2-W3<1.8\times(W3-W4)-6.5 \\
W1-W2>1\end{aligned}$$
W1, W2, W3, & W4 refer to the four [*WISE*]{} bands at 3.4 $\micron$, 4.6 $\micron$, 12 $\micron$, & 22 $\micron$, respectively.
Candidate Driving Sources
-------------------------
Assuming collimated outflows are generated by circumstellar disks, one would expect them to shut off once massive YSOs begin to ionize their surroundings. Since entries in the [*WISE*]{} Catalog of Galactic Regions were first identified by their characteristic mid-infrared “bubble” morphology, most of them are very large and clearly not good candidates for driving the collimated outflows identified in these regions. Four Q sources, which may be associated with younger or less luminous objects, do lie near the centers of outflows associated with Mol 8, 12, 15, & 148, but all are clustered with other candidate driving sources.
RMS YSOs can be identified as candidate driving sources for collimated MHO outflows in Mol 8, 12, 15, 143, 148, & 151. RMS YSOs may also drive MHO flows in Mol 10, 28, 110, 114, & 136, but the relationship is less clear cut, either because MHOs are either isolated detections or form linear arrangements offset from the YSO. (See Appendix A for further discussion.) We can also identify [*ALLWISE*]{} sources that fit Class 0/I YSO color criteria as candidate driving sources for outflows in 50% of the regions containing MHOs (11/22), with the caveat that candidates may contain multiple objects, since the resolution of [*WISE*]{} at 22 $\micron$ is 12$^{\prime\prime}$. Additionally, we note that some [*ALLWISE*]{} sources that satisfy Class 0/I color criteria are coincident with bright MHOs, which suggests that in some instances, these sources may be tracing shocked gas rather than YSOs. In any case, infrared colors alone cannot distinguish definitively between different classes of objects and further observations will be required to confirm the natures of these sources.
Table 4 presents 36 candidate outflows and driving sources, including cross-references to driving sources proposed in other publications. We identify candidate driving sources that fit [*WISE*]{} Class 0/I YSO color criteria (Koenig & Leisawitz 2014; Fischer et al. 2016) and/or are listed as YSOs in the RMS catalog for 26 of these outflows. We note that only 3 of the 36 ($\sim$ 8%) identified outflows (Mol 15 Flow 1, this paper and Mol 121 Flows 1 & 3 , WAS13) are associated with driving sources that are candidate UC regions; the remaining driving sources appear to be younger objects. This suggests that the jet phase shuts off shortly after the YSO has begun to ionize it surroundings; however, future high-resolution observations of the candidate UC regions, particularly radio recombination line data, would be particularly useful in establishing the evolutionary status of these objects. Seven of the candidate driving sources of outflows listed in Table 4 are both RMS YSOs and YSOs we identified from [*ALLWISE*]{} sources. Only in two cases (Mol 10 Flow 4, Mol 151 Flow 1) is the proposed driving source a RMS YSO lacking a counterpart identified from [*ALLWISE*]{}. Although evidence suggests that most of the collimated outflows identified in this study are driven by massive YSOs, much work remains to be done to establish source and outflow properties. NIR spectroscopy is needed to determine extinctions towards MHOs, which is essential for estimating MHO luminosities, and higher resolution observations are necessary to distinguish individual source contributions to bolometric luminosities.
[lclrr]{} Mol 2 Flow 1 & 2967A-H & MM2$^a$, & 75 & bright knots, complex morphology\
& & J001427.00+642819.3, & &\
& & J001425.05+642822.4, & &\
& & J001423.45+642824.9, & &\
& & J001422.48+642836.8 & &\
Mol 2 Flow 2 & 2968A-B & MM1$^a$? & 10 & bright knots, complex morphology\
Mol 3 Flow 1 & 2903 & J004457.30+554718.1 (D)\* & 58 & bow-shaped knot, faint bridge of emission\
Mol 3 Flow 2 & 2971-2974, 2976 & J004458.06+554656.7 (B)\* & -80 & curved chain of knots\
Mol 3 Flow 3 & 2969, 2970, 2900, 2975 & J004457.30+554718.1 (D)\*? & 25 & colinear knots\
Mol 8 Flow 1 & 1002,1003-1003\_2 & J051713,74+392219.6 (A,B)\* & 15-20 & bright bow-shaped knots, precessing jet?\
& & (G168.0627+00.8221) & & RMS counterpart\
Mol 8 Flow 2 & 1004\_2, 1004\_3 & J051713,74+392219.6 (A,B)\* & -20 & bright connected chain\
Mol 9 Flow 1 & 1055G & J052022.03+363756.5 & 60 & bright bow-shaped feature\
Mol 9 Flow2? & 1055A,C,E & J052022.03+363756.5 & 28 & faint knots\
Mol 10 Flow 1 & 1005 & MM-1/MM-2 (B)\* & 50 & bright jet\
Mol 10 Flow 2 & 1011, 1009, 1059? & MM-1/MM-2 & 5 & MHOs along “middle jet”$^b$\
Mol 10 Flow 3 & 1057, 1007, 1060? & …& -60 & MHOs along “long jet”$^c$\
Mol 10 Flow 4 & 1006 & G174.1974-00.0763 (A)\* & -90 & bright bow-shaped feature\
Mol 11 Flow 1 & 1015, 1015\_2, 1015\_3 & J053752.03+320003.9 & 64 & bright bow-shaped features\
Mol 11 Flow 2 & 1061A-F & J053752.03+320003.9? & -72 & linear chain of emission knots\
Mol 11 Flow 3$^c$ & 1018-1018\_3, 1016-1016\_3 & (A)\* & -49 & bright, complex morphology\
Mol 11 Flow 4 & 1062A & J053752.99+315934.8 & -50 & very bright jet feature\
Mol 12 Flow 1 & 738 group, & 742A? & J054024.23+235054.6 (A)\* & 31 & faint SW-NE chain\
& & (G183.7203-03.6647) & & RMS counterpart\
Mol 12 Flow 2 & 744A, 734 & 740 groups, 751 & J054024.79+235048.0 & -10 & S-shaped chain of knots\
Mol 12 Flow 3 & 739, 744B & J054024.23+235054.6 (A)\* & 63 &\
& & (G183.7203-03.6647) & & RMS counterpart\
Mol 12 Flow 4 & 742B, 742C, 737? & J054024.95+235220.6, & 95 & bright knots\
& & J054026.46+235222.0 & &\
Mol 15 Flow 1 & 1216A-F & J060840.45+213102.0 & -35 & chain of knots\
& & (G189.0307+00.7821) & & RMS counterpart\
Mol 108 Flow 1 & 2623A-D & J193859.37+224656.0 & 65 & chain of knots\
Mol 114 Flow 1 & 2608 & 2609 complexes & (C,F)\* & $\sim$ -87,-72 & curved chain of knots, bright bow\
Mol 121 Flow 1$^d$ & 864-865, 863, 891-892, 937 & Associated with Core D & 47 & bipolar MHOs\
Mol 121 Flow 2$^d$ & 867,938-939 & Associated with Core A & 90 & 867 bright bow feature\
Mol 125 Flow 1 & 949A-D & …& 68 & faint chain of knots\
Mol 126 Flow 1 & 961 & J203029+411558.6 (C)\* & -57 & “goldfish” bow, bright knot\
Mol 143 Flow 1 & 2765-2765\_2, 2766-2766\_2 & J221908.42+560501.2 (A)\* & 10 &\
& & (G102.8051-007184A) & & RMS counterpart\
Mol 143 Flow 2 & 2772A-B & J221908.42+560501.2 (A)\* & 40 &\
Mol 143 Flow 3 & 2773, 2776A-B & J221908.42+560501.2 (A)\* & -52 & linear arrangement of knots\
& & (G102.8051-007184B) & & RMS counterpart\
Mol 148 Flow 1 & 2777D-G & J223223.87+581859.7 (A)\* & 22 &\
& & (G105.5072+00.2294) & & RMS counterpart\
Mol 148 Flow 2 & 2777A-C & J223219.41+581750.7 & 73 & linear chain of 3 faint knots\
Mol 151 Flow 1 & 2778A-K & G108.5955+00.4935B & 85 & bright bow features A & B\
Mol 151 Flow 2 & 2779A-D & J225242.60+600041.5 & 10 & curved chain of knots\
Mol 160 Flow 1$^e$ & 2921-2922 & A$^f$ & & compact ‘bipolar’ knots\
SUMMARY AND CONCLUSIONS
=======================
1. Using NICFPS on the ARC 3.5-m telescope at the APO, we have detected a total of 236 MHOs, 156 of which are new detections, in 22 out of 26 regions associated with massive CO outflows.
2. We find an excellent agreement between predicted H$_2$ 2.12-$\micron$/2.25-$\micron$ flux ratios for shocks (H$_2$ 2.12-$\micron$/2.25-$\micron$ $>$ 3) and fluorescent emission (1 $<$ H$_2$ 2.12-$\micron$/2.25-$\micron$ $<$ 3) and morphology: MHOs are typically “knots”, bow-shaped or other compact features, while fluorescent emission has the morphology of bubbles (e.g., as seen in regions Mol 2, 10, 11, 109, 125, 126, 151), filaments or arcs (e.g., Mol 3, 15, 136, 158), or pillars (e.g., Mol 143). Based on low H$_2$ 2.12-$\micron$/2.25-$\micron$ flux ratios, we note that some previously reported MHOs may have been misidentified as such.
3. For MHOs with detected H$_2$ 2.25-$\micron$ emission, H$_2$ 2.12-$\micron$/2.25-$\micron$ flux ratios are typically between $\sim$ 6 and 20, consistent with C-type shocks. Where H$_2$ 2.12-$\micron$/2.25-$\micron$ $< \sim$ 5, the MHO is either faint or lies in the direction of a PDR. MHOs 1018 and 1018$\_3$ in Mol 11 are good examples of the latter. A few MHOs have H$_2$ 2.12-$\micron$/2.25-$\micron$ $>$ 20 (e.g., the bright bow features MHO 1015$\_2$ in Mol 11 and MHOs 2778 A&B in Mol 151), but in no case is H$_2$ 2.12-$\micron$/2.25-$\micron$ $>$ 30. Three of our survey targets, Mol 109, Mol 121 (WAS13), and Mol 126, exhibit one or more spots where the H$_2$ 2.12-$\micron$/2.25-$\micron$ flux ratio $<$ 1, which cannot be explained by either fluorescence or shocks.
4. MHO arrangements fall into three morphological categories: (1) co-linear MHOs that define the P.A. of one dominant outflow in the region; (2) multiple outflows along distinct P.A.s; and (3) isolated or clustered MHOs where no clear P.A. can be identified.
5. In over half the regions with MHO detections, MHO arrangements and fluorescent H$_2$ structures trace features present in low-resolution high-velocity CO maps presented in ZHB05, suggesting the CO emission traces a combination of complex dynamical effects that may include expanding regions as well as protostellar outflows.
6. All but three (Mol 9, 114, & 143) of our 26 target regions contain entries in the [*WISE*]{} Catalog of Galactic Regions (Anderson et al. 2014); however, the majority of these are radio quiet (Q) sources thought to be either massive objects in a pre-UC region phase, or intermediate-mass star-forming regions. In only four cases does a source lie near the center of a collimated MHO outflow, and in each of these cases, the source is clustered with other possible driving sources.
7. We identify candidate driving sources that fit [*WISE*]{} Class 0/I YSO criteria (Koenig & Leisawitz 2014; Fischer et al. 2016) and/or are listed as YSOs in the RMS catalog for 26 out of the 36 outflows we distinguish across our target regions.
8. Only 3 of the 36 outflows ($\sim$ 8%) identified in Table 4 (Mol 15 Flow 1, this paper and Mol 121 Flows 1 & 3 , WAS13) are associated with driving sources that are candidate UC regions; the remaining driving sources appear to be younger objects. This suggests that the jet phase shuts off shortly after the YSO has begun to ionize it surroundings. Future high-resolution observations of the candidate UC regions, particularly radio recombination line data, would be particularly useful in establishing the evolutionary status of these objects.
This research is based on observations obtained with the Apache Point Observatory 3.5-meter telescope, which is owned and operated by the Astrophysical Research Consortium. We particularly thank the 3.5-meter Observing Specialists and Al Harper for assistance in acquiring the data, as well as Michael Medford, who performed the original NICFPS data reduction. This research has made use of SAOImage DS9, developed by the Smithsonian Astrophysical Observatory; data products from the Wide-field Infrared Survey Explorer, which is a joint project of the University of California, Los Angeles, and the Jet Propulsion Laboratory/California Institute of Technology, funded by the National Aeronautics and Space Administration; and data products from the Midcourse Space Experiment. Processing of Midcourse Space Experiment data was funded by the Ballistic Missile Defense Organization with additional support from NASA Office of Space Science. This research has also made use of the NASA/ IPAC Infrared Science Archive, which is operated by the Jet Propulsion Laboratory, California Institute of Technology, under contract with the National Aeronautics and Space Administration. GW-C was funded in part through NASA’s Illinois Space Grant Consortium, and the authors gratefully acknowledge support from the Brinson Foundation grant in aid of astrophysics research at the Adler Planetarium. KA thanks Jamie Riggs for help in developing the near-infrared continuum subtraction procedure, [**and GW-C thanks Geza Gyuk for assistance converting Table 3 to machine-readable format.**]{}
INDIVIDUAL REGIONS: RESULTS AND DISCUSSION
==========================================
Figures 1 - 20 present images for all targets with MHO detections, with the exception of Mol 160 and Mol 121, which were included in previous papers (WSS12; WAS13). For each region, we present (a) a 3-color rgb image that combines continuum-subtracted H$_2$ 2.12-$\micron$ (blue), continuum-subtracted H$_2$ 2.25-$\micron$ (red), and an average of the continuum-subtracted H$_2$ 2.12-$\micron$ and H$_2$ 2.25-$\micron$ images (green); and (b) a continuum-subtracted H$_2$ 2.12-$\micron$ greyscale image with MHO numbers and regions (magenta) indicating the apertures used to compute fluxes listed in Table 3. The rgb images include positions of [*WISE*]{} Class 0/I candidates (labelled yellow circles indicating the resolution of [*WISE*]{} at 22 $\micron$), which were identified using the criteria presented in §3.3; RMS sources (orange +: YSO, diamond: region); and entries in [*WISE*]{} Catalog of Galactic Regions (purple boxes labelled according to type: C, G, K, or Q). [*IRAS*]{} source error ellipses (green) are included on the greyscale images. Maps of the CO outflows for all of the regions discussed below, except Mol 10, were presented in Figure 1 of ZHB05. CO maps of Mol 10 are presented in Zhang et al. (2007). We refer the reader to VDR10 for further discussion of regions marked with an asterisk below and in Tables 2 & 4.
IRAS 00117+6412 (Mol 2)
-----------------------
The brightest H$_2$ emission towards this target is due to fluorescence from a PDR centered on the [*IRAS*]{} source position and C-type region, G118.964+01.893 (Figures 1a & b). The CO outflow of ZHB05 shows N-S bipolarity. The blueshifted emission peaks at the position of the [*IRAS*]{} source and the redshifted emission peaks $\sim$ 30$^{\prime\prime}$ to the south. There is a steep drop-off in emission to the north and east, consistent with the location of G118.964+01.893 at the eastern border of the molecular cloud.
Palau et al. (2010; hereafter, PSB10) identified three intermediate-mass YSOs in this region. One can be associated with G118.964+01.893 (which they identify as a UC region). Two others are associated with millimeter sources, MM1 & MM2 (MM1 has multiple components). MM1 is located just west of the [*IRAS*]{} error ellipse ($\alpha$(2000) = 00$^h$14$^m$26.05$^s$, $\delta$(2000) = 64$^{\circ}$28$^{\prime}$43.7$^{\prime\prime}$) and MM2 is located $\sim$ 16$^{\prime\prime}$ to the south ($\alpha$(2000) = 00$^h$14$^m$26.31$^s$, $\delta$(2000) = 64$^{\circ}$28$^{\prime}$27.8$^{\prime\prime}$). PSB10 estimate the star powering the UC region to be $\sim$ 1 Myr and $\sim$ 6 M$_{\odot}$. MM1 is reported to be a Class 0/I source embedded in a $\sim$ 3 M$_{\odot}$ dust core with estimated luminosity of 400-600 L$_{\odot}$. PSB10 mapped a CO outflow along a NE-SW direction, centered on MM1, using CO J=2$\rightarrow$1 observations acquired with the Submillimeter Array (SMA: Ho et al. 2004). The blueshifted emission lies to the NE. MM2 is a deeply embedded object associated with H$_2$O maser emission, but not centimeter or near-infrared emission. PSB10 noted that MM2 appears to be a $\sim$ 1.6 M$_{\odot}$ Class 0 object, but curiously lacked CO outflow emission.
It seems likely that MM1 is the driving source of MHOs 2968A & B to the NE of this source, particularly since this is the direction of the CO outflow emission. We suggest that MM2 may drive the H$_2$ outflow defined by the string of MHO 2967 knots along a P.A. $\sim$ 75$^{\circ}$ to the south of the region; however, we identify additional Class I candidates (J001427.00+642819.3, J001425.05+642822.4, J001423.45+642824.9) and one Class 0 candidate (J001422.48+642836.8) that can’t be ruled out as driving sources, given the dispersion and morphology of the MHO knots. It is not clear whether these MHOs trace a single outflow and position angle. It is possible that PSB10 did not detect an outflow associated with MM2 because of the E-W orientation of the outflow traced by the MHOs and a lack of molecular material eastward of the UC region.
 
IRAS 00420+5530 (Mol 3)$^*$
---------------------------
We identify three Class I candidates (J004458.06+554656.7, J004457.30+554718.1, J004451.03+554639.9) and one Class 0 candidate (J004500.80+554627.0) in this region (Figures 2a & b). Four of the MHOs (2900-2903) were previously discovered by VDR10. One of these (MHO 2902) lies along a PDR that is probably associated with the [*IRAS*]{} source, which is coincident with J004457.30+554718. MHO 2902 has a very low H$_2$ 2.12-$\micron$/2.25-$\micron$ flux ratio of 2.9, so, at the very least, the PDR is likely to contribute to the flux caught by the aperture. A candidate region (G122.002-07.084) lies near a ridge of fluorescent emission in the southwest quadrant of the field. The CO high-velocity gas mapped by ZHB05 has a complex morphology with multiple peaks.
MHOs appear to lie along at least three distinct position angles in this region. The bow-shaped knot MHO 2903 appears to connect to J004457.30+554718.1 via a faint bridge of H$_2$ emission along a P.A. of $\sim$ 58$^{\circ}$. J004457.30+554718.1 coincides with VDR10 source “D”, H$_2$O masers, and a VLA source (Palla et al. 1991; Molinari et al. 2002, hereafter MTR02). The faint chain of knots (MHOs 2971-2974, 2976) along a P.A. of $\sim$ -80$^{\circ}$ appears to align with Class I candidate J004458.06+554656.7 (VDR10 source “B”), which is also associated with MM 2 (MTR02) and a mid-infrared source at $\alpha(2000) = 00^h44^m57.24^s$, $\delta(2000) = 55^{\circ}46^{\prime}50.2^{\prime\prime}$ that was identified using the MIRLIN camera (Ressler et al. 1994) on NASA’s 3-m Infrared Telescope Facility (IRTF) on Mauna Kea (J. O’Linger-Luscusk, private communication). H$_2$O masers are located between MHOs 2971 & 2973 (Harju et al. 1998). Although the bright knot (MHO 2901) appears to be associated with J004458.06+554656.7, it is not colinear with the fainter knots and its relationship to a specific outflow is unclear. Contours associated with overlapping red- and blue-shifted CO emission peaks extend in a general east-west direction roughly 15-30$^{\prime\prime}$ south of the [*IRAS*]{} source (ZHB05) consistent with the orientation of the MHO chain. MHOs 2969, 2970, 2900, and 2975 are approximately colinear with J004457.30+554718.1 along a P.A. of $\sim$ 25$^{\circ}$.
 
IRAS 05137+3919 (Mol 8)$^*$
---------------------------
We identify substructure in two outflows previously reported by VDR10 (Figures 3a & b). The outflow along a P.A. of $\sim$15-20$^{\circ}$ is defined by the bright, complex bow-shaped features 1002 & 1003, which are approximately equidistant from RMS YSO G168.0627+00.8221. As suggested by VDR10, our 2.12 $\micron$ image confirms that the jet driving this outflow may be precessing. Although the component of MHO 1003 furthest from the RMS YSO (1003\_2) lies along a P.A. of $\sim$20$^{\circ}$, the brighter, closer, component, MHO 1003, suggests a P.A. of $\sim$15$^{\circ}$. MHOs 1002 & 1003 are the brightest MHOs in this region, and follow the position angle of the CO outflow identified by ZHB05. Another outflow along a P.A. of $\sim$ -20$^{\circ}$ is defined by the MHO 1004 substructures 1004\_2 & 1004\_3. The morphology of the remaining MHO 1004 components is complex and it is not clear whether they (and MHO 1020) are associated with either outflow.
The two outflows intersect at the position of the RMS YSO / Class I candidate J051713.74+392219.6, which is also coincident with MM 1 (MTR02), the K sources VDR10 denote as “A” and “B”, and a mid-infrared source detected with MIRLIN (J. O’Linger-Luscusk, private communication, unpublished data). The fluxes derived from the MIRLIN observations (F(12.5 $\micron$) = 4.8 Jy; F(20.8 $\micron$) = 11.5 Jy; F(24.5 $\micron$) = 27.6 Jy) rise steeply in the mid-infrared and are comparable to [*WISE*]{} (F(12 $\micron$) = 4.4 Jy; F(22 $\micron$) = 18.6 Jy)and [*MSX*]{} (F(12 $\micron$) = 4.6 Jy; F(21 $\micron$) = 13.7 Jy) fluxes at similar wavelengths. Using the distance (11.5 kpc) and luminosity (2.25$\times 10^5$ L$_{\odot}$) derived from modeling the SED of this source (Molinari et al. 2008; VDR10), the total H$_2$ (2.12 $\micron$) luminosity, (L$_{2.12}$), calculated from summing the MHO fluxes in Table 3 and assuming no extinction, is 2.99 L$_{\odot}$. Following WAS13, and assuming the 2.12 $\micron$ luminosity is 5-10% of the total rovibrational H$_2$ emission, L$_{H_2}$ $\sim$ 29.9-59.8 L$_{\odot}$. Since the assumption of no extinction yields a lower limit to the true H$_2$ luminosity, this places the Mol 8 outflows at or above the linear fit to Log(L$_{H_2}$) vs. Log(L$_{bol}$) for high-mass YSOs presented in Caratti o Garatti (2015, Fig. 9).
 
IRAS 05168+3634 (Mol 9)$^*$
---------------------------
We identify nine Class I candidates, one Class 0 candidate (J052023.40+363745.9), and four sources that satisfy both Class 0 & Class I color criteria (J052023.04+363813.0, J052022.03+363756.5, J052010.31+363743.9, & J052010.76+363635.0) in this region (Figures 4a & b). VDR10 found no MHOs near the [*IRAS*]{} source position, but we note that the low-resolution CO outflow mapped by ZHB05 is centered near J052022.03+363756.5 and a dense core (BGPSv2\_G170.661-00.249) identified by the Bolocam Galactic Plane Survey (BGPS v2.1: Ginsburg et al. 2013), not on the [*IRAS*]{} source. Whereas the CO outflow of ZHB05 lies along a P.A. of $\sim$40 $^{\circ}$, the bright bow-shaped MHO 1055G lies along a P.A. of $\sim$ 60 $^{\circ}$ from J052022.03+363756.5. MHOs 1055 A, C, & E lie along a P.A. of $\sim$ 28$^{\circ}$. It is possible that J052022.03+363756.5 is the source of two outflows that produce the large-scale CO morphology, but this is unclear. Three faint MHOs (1055B, D, & F) lie to the southwest of the [*IRAS*]{} source and one to northwest (MHO 1056), but their associations are also unclear.
 
IRAS 05274+3345 (Mol 10)$^*$
----------------------------
VDR10 present a detailed background of this region, which contains multiple signs of star formation. We identify five [*ALLWISE*]{} sources in this region that satisfy Class I color criteria; two of these also satisfy Class 0 color criteria (J053049.21+334818.4, J053049.01+334746.7). The RMS YSO candidate G174.1974-00.0763 (VDR10 source ‘A’) lies close to the center of the [*IRAS*]{} error ellipse and is the brightest infrared source in this field (Figures 5a & b). At least three distinct outflows were identified from near-infrared observations (Chen et al. 2005) and via multiple spectral lines using the Submillimeter Array (Zhang et al. 2007). Chen et al. (2005) identified 28 MHOs in this region, which were not entered in the on-line MHO catalog (Davis et al. 2010). Several of these objects lie along three axes they identified as the “short”, “middle”, and “long” jets. The three outflows A, B, and C, identified by Zhang et al. (2007) along P.A.s=5$^{\circ}$, 35$^{\circ}$ & -60$^{\circ}$, respectively, intersect in the vicinity of the the hot cores they refer to as MM-1 and MM-2. These cores lie approximately at the center of a triangle defined by MHOs 1008, 1009, & 1010 (Figure 5b). Gómez-Ruiz et al. (2016) identified several methanol masers at 44 GHz in this region, which are clustered towards MHOs 1007 & 1008, as well as between MHOs 1008, 1009, & 1010.
Based on our 2.12-$\mu$m image (Figure 5b), we can make the following associations: MHO 1005 corresponds to the feature VDR10 identify as ‘1’ and Chen et al. (2005) associate with a feature they refer to as the ‘short jet’ (see Table 4, Mol 10 Flow 1). Similarly, MHOs 1006, 1007, 1008, 1009, 1010, & 1011, correspond to VDR10 features 2, 3, 4, 5, 6, & 7, respectively. Zhang et al. (2007) outflow ‘A’ (Chen et al. 2005 ‘middle jet’) appears as a faint bridge of emission connecting MHOs 1009 and 1011, and possibly also MHO 1059 (Figure 5b and Table 4, Mol 10 Flow 2). The ‘long jet’ identified by Chen et al. (2005), along a P.A. of $\sim$-60$^{\circ}$, connects MHOs 1057, 1007, and possibly 1060 (Figure 5b) along a line slightly south of MM-1 and MM-2 (see Table 4, Mol 10 Flow 3). Although it is tempting to associate Zhang et al. (2007) outflow ‘B’ with the ‘short jet’ detected by Chen et al. (2005), VDR10, and the present study, we note that outflow ‘B’ lies along a P.A. $\sim$ 35$^{\circ}$, while the jet associated with MHO 1005 lies along a P.A. of $\sim$ 50$^{\circ}$, so these features may not be tracing the same outflow. Outflow ‘C’ lies along the same P.A. as the long jet. Although this outflow may be associated with MM-1/MM-2, there are several near-infrared point sources (Figure 5a) between these sources and Class I candidate J053046.98+334748.3 that are viable driving source candidates, as is the RMS source, which is the probable driving source of the large-scale CO outflow mapped by Hunter et al. (1995). It is not clear whether MHO 1060 is associated with this outflow, but it lies along the same P.A. of -60$^{\circ}$.
The bow-shaped feature MHO 1006 lies directly west of the RMS source, along a P.A. of -90$^{\circ}$, and is likely driven by this source (Mol 10 Flow 4). MHO 1006 has the lowest 2.12/2.25 flux ratio of the MHOs we identified in this region and may be dominated by fluorescent emission from a compact PDR associated with with the RMS YSO/ Q-type object G174.1974-00.0763, which is in the [*WISE*]{} Catalog of Galactic Regions (Anderson et al. 2014). The associations of MHOs 1008, 1010 & 1058 with outflows in this region is unclear.
 
IRAS 05345+3157 (Mol 11)$^*$
----------------------------
The complex arrangements of MHOs clearly indicate multiple outflows in this region (Chen et al. 1999; Chen et al. 2003; VDR10). We identify ten [*ALLWISE*]{} sources that fit Class I color criteria, four of which also fit Class 0 criteria (J053753.90+320015.7, J053752.99+315934.8, J053752.11+320020.0, J053751.45+320021.5). There are at least two distinct epochs of star-formation (Figures 6a & b): (1) an infrared cluster that is centered on the [*IRAS*]{} position and is enveloped in a PDR (with no Class 0/I associations); and (2) younger objects to the northeast and northwest of the PDR, as indicated by the MHOs, Class 0/I candidates, and seven dense cores (Lee et al. 2011). Lee et al. (2011) suggested that two of the dense cores (Cores 1 & 3) are forming massive protostars. The low-resolution CO map of ZHB05 shows the molecular outflow to be centered at least 30$^{\prime\prime}$ to the northeast of the [*IRAS*]{} source, and elongated in a general ESE - WNW direction, with blueshifted outflow gas in the direction of the MHO 1061 knots.
The Class I candidate, J053752.03+320003.9, is coincident with Core 3. It is equidistant from the bright bow-shaped features MHO 1015 and 1015\_2, which define an outflow along a position angle of 64$^{\circ}$ (Table 4, Mol 11 Flow 1). VDR10 identified a source they refer to as ‘B’ at the southwest end of MHO 1015 as the possible driving source of this outflow, but our observations suggest J053752.03+320003.9 is the better candidate. The linear chain of knots MHO 1061 A-F (Table 4, Mol 11 Flow 2) also intersect Core 3 and may connect with the MHO 1063 knots to the southeast at a position angle of -72$^{\circ}$, but this is unclear. There are several [*WISE*]{} Class I candidates in the vicinity of the MHO 1061 knots that are also viable candidates. VDR10 suggested an outflow along a position angle of -49$^{\circ}$ (Table 4, Mol 11 Flow 3) connecting features we identify with the MHO 1018 and 1016 complexes. Their candidate driving source (‘A ’) coincides with a bright ‘orange’ point source on our rgb image (Figure 6A), which lies at the center of this outflow. The Class 0/I candidate, J053752.99+315934.8 (coincident with Core 1), lies at the northwestern end of MHO 1062A (Table 4, Mol 11 Flow 4). MHO 1062A is the second brightest MHO in this region. Although it aligns with the faint knots MHO 1062 B & C to the northwest, it is unclear whether these are part of this outflow. The remaining MHOs cannot be clearly linked to distinct outflows. MHO 1019 lies to the southwest of the infrared cluster and may be excited by a source at the edge of the PDR.
 
IRAS 05373+2349 (Mol 12)$^*$
----------------------------
Except for MHO 751, all of the MHOs in this region were previously detected (VDR10; Khanzadyan et al. 2011). We identify six [*ALLWISE*]{} sources that satisfy Class I color criteria and two that satisfy Class 0 (J054020.76+235031.0, J054026.46+235222.0) color criteria. The CO outflow mapped by ZHB05 is slightly elongated along a position angle of $\sim$ 31$^{\circ}$ about the [*IRAS*]{} source position. The Class I candidate J054024.79+235048.0, RMS YSO candidate G183.7203-03.6647, and Q-type region G183.720-03.665 are positionally coincident near the center of the [*IRAS*]{} error ellipse (Figures 7a & b).
Our observations suggest the presence of at least four outflows, as well as isolated knots not clearly connected to any outflows in this region. MHOs defining two of these outflows (Table 4, Mol 12 Flow 1 & Mol 12 Flow 3) intersect at the position of the [*IRAS*]{} source. Mol 12 Flow 1 is defined by the faint MHO 738 knots along a P.A. of $\sim$ 31$^{\circ}$, similar to the large-scale CO outflow. The faint knot in the northeast, MHO 742A, lies along this position angle and may be part of this outflow, but the brighter knots, MHOs 742B & C, lie along a distinctly different position angle of $\sim$ 95$^{\circ}$. Potential driving sources for this outflow (Table 4, Mol 12 Flow 4) include the Class I candidate J054024.95+235220.6 or the nearby Class 0 candidate J054026.46+235222.0. The faint emission knot to the west, MHO 737, may be part of this outflow. Mol 12 Flow 3 joins MHO 739 and MHO 744B along a P.A. of $\sim$ 63$^{\circ}$.
Mol 12 Flow 2 is defined by the S-shaped series of MHOs along a position angle of $\sim$ -10$^{\circ}$, which stretches from MHO 744A to MHO 740A, and includes the MHO 734 knots and MHO 751. Based on the arrangements of these MHOs, we suggest that Class I candidate J054024.79+235048.0 drives this outflow. This source is coincident with the brighter of two mid-infrared sources identified by MIRLIN (J. O’Linger-Luscusk, private communication). The differences in mid-infrared fluxes derived from the high-resolution MIRLIN observations (F(12.5 $\micron$) = 4.19, 1.37 Jy; F(20.8 $\micron$) = 7.48, 4.79 Jy; F(24.5 $\micron$) = 11.38, 8.98 Jy) , compared with [*MSX*]{} (F(12 $\micron$) = 4.65 Jy; F(21 $\micron$) = 12.27 Jy) and [*WISE*]{} (F(12 $\micron$) = 6.0 Jy; F(22 $\micron$) = 20.7 Jy) fluxes, suggest that multiple sources contribute to the mid- and far-infrared luminosity in this region. The remote MHO 735 nearly coincides with Class I candidate, J054019.59+235202.5, but since this is an isolated knot, it is not possible to identify an outflow.
 
IRAS 06056+2131 (Mol 15)
------------------------
The arrangement of MHOs in this region closely mirrors the complex morphology of the high-velocity CO emission mapped by ZHB05, which in turn clearly indicates there are multiple outflows (Figures 8a & b). We identify nine Class I candidates, two Class 0/I candidates (J060838.48+213043.8, J060842.88+213118.7), and one Class 0 candidate (J060836.97+213048.8). The western peak of the blueshifted CO emission of ZHB05 is centered on the [*IRAS*]{} source position, which is coincident with Class I candidate J060840.45+213102.0, RMS YSO G189.0307+00.7821, and Q-type region G189.030+00.780. The eastern blueshifted CO peak is centered $\sim$15$^{\prime\prime}$ south of Class I candidate, J060846.72+213144.0, which is coincident with RMS YSO G189.0323+00.8092 and Q-type region G189.032+00.809. The redshifted CO emission of ZHB05 displays a U-shaped morphology that roughly traces the MHO 1216, 1217, and 1218 complexes. Both regions lie near dense cores identified by BGPS. Mol 15 Flow 1 lies along a position angle of $\sim$ -35$^{\circ}$, including the MHOs in the 1216 group. This outflow is likely driven by RMS YSO G189.0307+00.7821/ Class I candidate J060840.45+213102.0. MHOs 1217A-C lie along a position angle of $\sim$ 88$^{\circ}$ degrees centered north of G189.0307+00.7821/ J060840.45+213102.0, and the MHO 1218 complex may be associated with an outflow produced by RMS YSO G189.0323+00.8092/ Class I candidate J060846.72+213144.0, but the complex arrangement of these MHOs makes it impossible to link these to specific outflows. Two other Class I candidates (J060843.63+213150.7, J060843.08+213147.2) are also potential driving sources.
 
IRAS 06584-0852 (Mol 28)$^*$
----------------------------
Most of the H$_2$ emission in this region is due to fluorescence, some of which may be associated with Q-type region G221.955-01.993 (Figures 9a & b). Interestingly, the morphology of the fluorescent emission closely parallels the complex CO high-velocity emission identified by ZHB05. VDR10 detected no MHOs in this region; however, we detect three isolated knots. MHO 3139 lies directly to the northwest of fluorescent H$_2$ that coincides with overlapping blue- and redshifted CO emission detected by ZHB05. MHO 3140 is coincident with Class I candidate, J070051.00-085629.8 (RMS YSO G221.9605-01.9926), and a double mid-infrared source identified by MIRLIN (O’Linger-Luscusk, private communication). The MIRLIN emission (F(12.5 $\micron$) = 5.27, 1.16 Jy; F(17.9$\micron$) = 9.98, 1.10 Jy; F(20.8 $\micron$) = 10.46, 1.38 Jy; F(24.5 $\micron$) = 11.32, $<$0.72 Jy) appears to trace the two stars VDR10 identify as reddened YSOs ‘A’ and ‘B’. Mol 3141 is a faint emission knot with no evident association.
 
IRAS 19368+2239 (Mol 108)
-------------------------
This region contains multiple signposts of star formation (Figures 10a & b), including three Q-type regions, a RMS source identified as an region (G058.4670+00.4360B), a RMS YSO (G058.4670+004360A), and twelve 44-GHz masers (Gómez-Ruiz et al. 2016). The CO outflow mapped by ZHB05 is complex and multi-lobed, suggesting multiple outflows in this region, with most of the high-velocity emission located north of the [*IRAS*]{} position (and RMS sources). We identified eight Class I candidates, two of which are positionally coincident with the RMS region (J193857.17+224624.5) and RMS YSO (J193856.89+224632.2). We detected a chain of four MHOs (Mol 108 Flow 1), along a position angle of $\sim$ 65$^{\circ}$, located north of the RMS YSO, but colinear with Class I candidate J193859.37+224656.0, which clearly shows multiple near-infrared components (see Figure 10a).
 
IRAS 19374+2352 (Mol 109)$^*$
-----------------------------
VDR10 detected three MHOs in this region (2600 - 2602). Two of these (2601 & 2602) are located along the periphery of the [*IRAS*]{} error ellipse, just south of a PDR associated with G059.612$+$00.917, a grouped (G-type) region (Figures 11a & b). Known (K-type) region G059.602$+$00.911, four RMS objects catalogued as regions, and a dense core detected by the APEX Telescope Large Area Survey of the Galaxy (ATLASGAL: Schuller et al. 2009) are clustered near the center of the CO outflow mapped by ZHB05, which lies to the southeast of the [*IRAS*]{} position. Gómez-Ruiz et al. (2016) detected two 44-GHz methanol masers just northeast of MHO 2600.
We identified three Class I candidates and three additional faint MHOs toward this target. Class I candidate J193934.64+235948.3 is positionally coincident with the ATLASGAL core, as well as an region catalogued in both the RMS survey and [*WISE*]{} catalog of galactic regions (Lumsden et al. 2013; Anderson et al. 2014). This source is particularly interesting since it exhibits “anomalous” H$_2$ emission, where the 2.12-$\micron$/2.25-$\micron$ ratio is less than one. We note that WAS13 detected such emission toward a deeply embedded source (DES: Yao et al. 2000) in Mol 121. The DES is thought to be a massive YSO driving a very young outflow. It is tempting to associate J193934.64+235948.3 with an outflow connecting MHOs 2600 -2602 along a P.A. of $\sim$ -40$^{\circ}$, but the faint chain of MHOs that extend to the southwest of MHO 2600 along a P.A. of $\sim$ 30$^{\circ}$ calls into question this interpretation. The CO outflow of ZHB05 shows multiple peaks, suggesting more than one outflow in this region. Faint MHO 2628 may be associated with Class I candidate J193933.43+235836.2, at the southern end of the field.
 
IRAS 19388+2357 (Mol 110)$^*$
-----------------------------
All three of the MHOs in this region (2613 - 2615) were previously identified by VDR10. The MHOs, RMS YSO G059.8329+00.6729, and two 44-GHz methanol masers (Gómez-Ruiz et al. 2016) are coincident with Class I candidate J194059.39+240443.9, to within the resolution of [*WISE*]{} at 22 $\micron$ (Figures 12a & b). MHOs 2614 & 2615 have low 2.12-$\micron$/2.25-$\micron$ flux ratios, suggesting these features are dominated by fluorescent emission; however, MHO 2613 has a flux ratio of 5.4, lower than typical values associated with C-type shocked emission, but higher than typical values for either fluorescent or J-type shocked emission. Although MHO 2613 is roughly elongated in the direction of the RMS YSO, the complex morphology of the H$_2$ emission in this region makes association with an outflow tenuous. Furthermore, the peak of the high-velocity CO emission mapped by ZHB05, which shows no hint of bipolarity, lies to the south of J194059.39+240443.9, closer to the center of the Q-type region G059.829+00.671. The CO emission may trace an outflow driven by a yet-to-be-identified source in this region, or it may trace the dynamics of an expanding region/PDR.
 
IRAS 20050+2720 (Mol 114)$^*$
-----------------------------
Bachiller et al. (1995) discovered an extremely high-velocity multipolar CO outflow in this region. The CO outflow map of ZHB05 likewise is oriented roughly E-W and is centered to the north of the [*IRAS*]{} position. VDR10 identified several MHOs, including a jet-like feature extending to the southeast of their candidate driving sources, ‘C’ and ‘F’. Our images (Figures 13a & b) reveal more components to the jet, which appears curved (Mol 114 Flow 1). VDR10 sources ‘C’ and ‘F’ lie close to a 44-GH maser (Gómez-Ruiz et al. 2016), which is located between MHOs 2608 & 2608\_6. Although the jet defined by the MHO 2608 group to the southeast of the candidate driving sources lies along a P.A. of $\sim$ -87$^{\circ}$, the direction of the jet to the northwest, defined by the MHO 2609 group, lies along a P.A. of $\sim$ -72$^{\circ}$. This curved jet is likely associated with the CO outflow along a P.A. of $\sim$-77$^{\circ}$, denoted “A” by Bachiller et al. (1995: Figure 3).
Although the bright [*WISE*]{} Class I candidate, J200706.50+272850.3 (coincident with VDR10 source ‘A’ and RMS YSO G065.7798-02.6121) appears to be the main contributor the the [*IRAS*]{} flux, this source(s) lies $\sim$ 10$^{\prime\prime}$ to the south of the MHOs that define Flow 1, making it an unlikely driving source candidate. We agree that VDR10 source ‘C’ or ‘F’ is more likely to drive the jet, and we note that the energetic parameters calculated by ZHB05 for the high-velocity CO emission (also located north of the [*IRAS*]{} position) do not suggest a particularly massive outflow. The bolometric luminosity of this outflow (3.88$\times10^2$ L$_{\odot}$) is the lowest of all the outflows in the ZHB05 study. There are also several isolated groups of MHOs in this region, which cannot be clearly linked to driving sources, and many YSO candidates that may be contributing to different outflows.
 
IRAS 20278+3521 (Mol 125)
-------------------------
Most of the H$_2$ emission in this region is due to a PDR associated with the Q-type [*WISE*]{} source, G075.155-02.087, which is positionally coincident with the [*IRAS*]{} source position (Figures 14a & b). A linear arrangement of four faint MHOs along a position angle of $\sim$ 68$^{\circ}$ defines Mol 125 Flow 1, but the axis is offset 10$^{\prime\prime}$ to the south of the [*IRAS*]{} source and no known objects within the mapped field can be identified as potential driving sources. We note that the center of the CO outflow identified by ZHB05 also lies about 10$^{\prime\prime}$ southwest of the [*IRAS*]{} source; however, the positional discrepancy is well within the $\sim$ 30$^{\prime\prime}$ resolution of the CO observations, and the CO emission shows no evidence of bipolarity, while the MHOs form a linear chain in the plane of the sky.
 
IRAS 20286+4105 (Mol 126)$^*$
-----------------------------
Most of the H$_2$ emission in this region traces fluorescence whose morphology suggests multiple PDRs (Figures 15a & b). The CO outflow mapped by ZHB05 is elongated to the west (and slightly north) of the [*IRAS*]{} source position, in the direction of extended fluorescent emission. VDR10 identified four YSOs in this region, which they labeled A-D. Source ‘A’ coincides with the position of G079.8749+01.1821, listed as a region in the RMS catalog ; source ‘C’ coincides with the [*WISE*]{} Class I candidate, J203029.51+411558.6; and source ‘D’ lies at the center of an arc that defines the southernmost PDR in our images. J203029.51+411558.6, and two 44-GHz masers $\sim$ 10$^{\prime\prime}$ to the southwest (Gómez-Ruiz et al. 2016), lie along the rim of the PDR associated with G079.8749+01.1821. It is of interest to note that there are three spots of anomalous H$_2$ emission in this region. One of these spots coincides with J203029.51+411558.6 (‘C’), which VDR10 noted was embedded in strong nebulosity. J203029.51+411558.6 also lies at the center of the ZHB05 CO outflow.
VDR10 suggested the bright emission feature they identify as ‘2’ (MHO 961) is driven by source ‘B’, which lies directly between G079.8749+01.1821 and MHO 961. In our images, MHO 961 has a curious “goldfish”-shaped morphology that points back toward J203029.51+411558.6 (‘C’), suggesting J203029.51+411558.6 is the likely source of MHO 961 as well as the CO outflow (Mol 126 Flow 1). J203029.51+411558.6 may also be the source of MHO 960, which appears to connect to a curving bridge of emission that points back toward this source; however, this is unclear due to the preponderance of fluorescent emission. MHOs 962 & 963 cannot be linked clearly to sources in this region.
 
IRAS 21307+5049 (Mol 136)$^*$
-----------------------------
MHO 882 (detected by VDR10) has a very low 2.12 $\micron$/ 2.25 $\micron$ flux ratio of 1.1, calling into question the nature of this object. VDR10 noted it lies just $\sim$ 3.7$^{\prime\prime}$ of a highly-reddened YSO they identify as ‘A’, which is coincident with RMS YSO G094.2615-00.4116 (Class I candidate J213230.61+510216.1). VDR10 identified ‘A’ with a cometary nebula that opens to the northwest. High-resolution ($\sim$ 6$^{\prime\prime}$) CO observations, obtained by Fontani et al. (2004) at the Owens Valley Radio Observatory (OVRO), suggest the RMS YSO (VDR10 ‘A’) is the source of a compact CO outflow. Our continuum-subtracted H$_2$ 2.12 $\micron$ images indicate a bipolar nebula with an hourglass shape centered roughly 16$^{\prime\prime}$ to the northeast of the YSO, close to the [*IRAS*]{} position and coincident, within positional accuracy, with the [*WISE*]{} C-type region, G094.263-00.414 (Figures 16a & b). The multi-peaked CO outflow mapped by ZHB05 is similar in direction (NW-SE) and extent to the bipolar nebula, which suggests the fluorescent H$_2$ emission may trace the periphery of a bipolar cavity produced by the CO outflow.
 
IRAS 22172+5549 (Mol 143)$^*$
-----------------------------
VDR10 noted that this is a known region; however, Molinari et al. (1998) did not find radio emission associated with the [*IRAS*]{} source and it does not appear in the [*WISE*]{} Catalog of Galactic regions (Anderson et al. 2014). Molinari et al. (2002) suggested that the O star HD211883, situated $\sim$ 2$^{\prime}$ north of a complex ridge of 3.6 cm emission, is the source of the ionization front. Our images (Figures 17a & b) indicate a PDR with ‘pillar’ morphology, suggesting the source of the region lies to the northwest of our imaged region, consistent with this interpretation. Three RMS YSOs (G102.8051-00.7184A, B, & C) are located near the apex of the pillar, roughly 20$^{\prime\prime}$ north of the [*IRAS*]{} position. MTR02 identified a dense core, via 1.3 and 3 mm continuum observations, that peaks close to the position of the southernmost RMS YSO, G102.8051-00.7184B.
ZHB05 detected multi-lobed CO emission in this region, suggesting more than one outflow. This is also evident in the $\sim$ 29$^{\prime\prime}$ CO map shown in Figure 14a of Fontani et al. (2004), whose higher-resolution ($\sim$ 6$^{\prime\prime}$) CO map (Figure 14c), obtained at OVRO, indicates a very compact ($\sim$ 20$^{\prime\prime}$) north-south outflow centered on the position of RMS YSO G102.8051-00.7184A ($\sim$ 5$^{\prime\prime}$ northeast of G102.8051-00.7184B). G102.8051-00.7184A & B are coincident with [*WISE*]{} Class I candidate, J221909.42+560501.2.
We identify three distinct jets along position angles $\sim$ 10, 40, & -52$^{\circ}$ (Mol 143 Flow 1, 2, & 3, respectively), which intersect at J221909.42+560501.2. G102.8051-00.7184A is the best candidate driving source for the jet along P.A. $\sim$ 10$^{\circ}$, which we identify with the compact N-S outflow of Fontani et al. (2004). G102.8051-00.7184A may also drive a jet along P.A. $\sim$ 40$^{\circ}$, but this is more difficult to distinguish due to the orientation of A & B. G102.8051-00.7184B likely drives the jet along P.A. $\sim$ 128$^{\circ}$. We note that the the high-velocity CO emission contours seen in the single-antenna maps of ZHB05 and Fontani et al. (2004) are consistent with outflows along position angles of 40 & -52$^{\circ}$.
 
IRAS 22305+5803 (Mol 148)$^*$
-----------------------------
We identified eight Class I candidates in this region. VDR10 found three clumpy emission features directly north of the [*IRAS*]{} source position, which coincides with the object we identify as RMS YSO G105.5072+00.2294 and [*WISE*]{} Class I candidate J223223.87+581859.7 (Figures 18a & b). The RMS YSO also coincides with the source VDR10 identify as ‘A’. The emission features are prominent in Figure 18b; however, they appear to be contaminated by fluorescent emission likely produced by a PDR associated with the Q-type object, G105.509+00.230. The objects VDR10 identify as ‘B’ and ‘C’ lie directly north of J223223.87+581859.7, towards the nebulous ‘white’ emission in Figure 18a. The MHOs we discovered in this region lie well outside of the image presented in Figure A48 of VDR10. MHOs 2777D-G lie along a NE-SW P.A. of $\sim$ 22$^{\circ}$ from the RMS YSO, similar in direction to the CO outflow mapped by ZHB05 (Mol 148 Flow 1). MHOs 2777A-C lie along a P.A. of $\sim$ 73$^{\circ}$, centered on [*WISE*]{} Class I candidate J223219.41+581750.7 (Mol 148 Flow 2).
 
 
 
IRAS 22506+5944 (Mol 151)
-------------------------
In this region, we identify a spectacular, compact jet that culminates in the bow-shaped MHOs 2278 A & B (Figures 19a & b). The jet is centered on RMS YSO, G108.5955+00.4935B (Mol 151 Flow 1). Another RMS YSO (G108.5955+00.4935A) is centered on the compact PDR just north of G108.5955+00.4935B and the jet. The location of G108.5955+00.4935B and its outflow on the periphery of the PDR is suggestive of triggered star formation, warranting further observations exploring the kinematics of this region. Gómez-Ruiz et al. (2016) discovered six 44-GHz masers in this region, all of which lie south of the jet axis. One of these masers is coincident with MHO 2778G. The proximity of this MHO to another RMS YSO (G108.5955+00.4935C) is suggestive of a possible outflow associated with this source. The curved chain of MHOs 2779A-D is roughly centered on [*WISE*]{} Class I candidate, J225242.60+600041.5, the likely driving source (Mol 151 Flow 2). We note that the extent of the outflows in this region are comparable in size to the resolution of the CO outflow map of ZHB05, precluding any meaningful comparison of morphology.
IRAS 23314+6033 (Mol 158)
-------------------------
The CO outflow mapped by ZHB05 in this region shows no hint of bipolarity. Both lobes peak at the western edge of the [*IRAS*]{} error ellipse, near the position of RMS region G113.6041-00.6161 and [*WISE*]{} Class I candidate, J233342.73+605025.4 (Figures 20a & b). These sources lie at the intersection of ridges of filamentary fluorescent emission. G113.6041-00.6161 is listed as a G-type (grouped) region in the [*WISE*]{} Catalog of Galactic Regions (Anderson et al. 2014). We identify a single MHO (MHO 2977) toward J233342.73+605025.4, which would be consistent with, though hardly conclusive of, an outflow oriented along our line of sight.
Anderson, L. D., Bania, T. M., Balser, D. S., et al. 2014, ApJS, 212, 1 André, P., Ward-Thompson, D., & Barsony, M. 1993, ApJ, 406, 122 Bachiller, R., Fuente, A., & Tafalla, M. 1995, ApJL, 445, L51 Bally, J. 2016, ARA&A, 54, 491 Caratti o Garatti, A., Stecklum, B., Linz, H., Garcia Lopez, R., & Sanna, A. 2015, A&A, 573, 82 Chen, Y., Yao, Y., Yang, J., et al. 1999, AJ 117, 446
Chen, Y., Yao, Y., Yang, J., & Zeng, Q. 2005, ApJ, 629, 288 Chen, Y., Yao, Y., Yang, J., Zeng, Q., & Sato, S. 2003, A&A, 405, 655 Churchwell, E., Povich, M. S., Allen, D., et al. 2006, , 649, 759 Davis, C. J., Gell, R., Khanzadyan, T., Smith, M. D., & Jenness, T. 2010, A&A, 511, A24 Dewangan, L. K., Baug, T., Ojha, D. K., et al. 2016, ApJ, 826, 27 Egan, M. P., Price, S. D., Kraemer, K. E., et al. 2003, Air Force Research Laboratory Technical Report AFRL-VS-TR-2003-1589 Fischer, W. J., Padgett, D. L., Stapelfeldt, K. L., & Sewilo, M. 2016, ApJ, 827, 96 Fontani, F., Cesaroni, R., Testi, L., et al. 2004, A&A, 424, 179 Ginsburg, A., Glenn, J., Rosolowsky, E. et al., 2013, ApJS, 208, 14 Gómez-Ruiz, A. I., Kurtz, S. E., Araya, E. D., Hofner, P., & Loinard, L. 2016, ApJS, 222, 18 Harju, J., Lehtinen, K., Booth, R. S., & Zinchenko, I. 1998, A&A, 132, 211 Hearty, F. R., Morse, J., Beland, S., et al., 2004, SPIE, 5492, 1623 Ho., P. T. P., Moran, J. M., & Lo, K. Y. 2004, ApJL, 616, L1-L6 Hunter, T. R., Testi, L., Taylor, G. B., Tofani, G., Felli, M., & Phillips, T. G. 1995, A&A, 302, 249 Kerton, C. R., Wolf-Chase, G., Arvidsson, K., Lintott, C. J., & Simpson, R. J. 2015, ApJ, 799, 153 Khanzadyan, T., Movsessian, T. A., Davis, C. J., Magakian, T. Y., Gredel, R., & Nikogossian, E. H. 2011, MNRAS, 418, 1994 Koenig, X. P., & Leisawitz, D. T. 2014, ApJ, 791, 131 Kurtz, S., Churchwell, E., & Wood, D. O. S. 1994, ApJS, 91, 659
Lada, C. J. 1987, IAUS, 115,1 Lee, K. I., Looney, L. W., Klein, R., & Wong, S. 2011, MNRAS, 415, 2790 Lumsden, S. L., Hoare, M. G., Urquhart, J. S., et al. 2013, ApJS, 208, 11 Mainzer, A., Bauer, J., Grav, T., et al. 2011, ApJ, 731, 53 Molinari, S., Brand, J., Cesaroni, R., & Palla, F. 1996, A&A, 308, 573
Molinari, S., Brand, J., Cesaroni, R., & Palla, F. 2000, A&A, 355, 617
Molinari, S., Brand, J., Cesaroni, R., Palla, F., & Palumbo, G. G. C. 1998, A&A, 336, 339 Molinari, S., Testi, L., Rodríguez, L. F., & Zhang, Q. 2002, ApJ, 570, 758 (MTR02)
Molinari, S., Pezzuto, S., Cesaroni, R., Brand, J., Faustini, F., & Testi, L. 2008, A&A, 481, 345 Palau, A., Sánchez-Monge, A., Busquet, G., et al. 2010, A&A, 510, 5
Palla, F., Brand, J., Cesaroni, R., Comoretto, G., & Felli, M. 1991, A&A, 246, 249 Ressler, M., E., Werner, M. W., Van Cleve, J., & Chou, H. A. 1994, Experimental Astronomy, 3, 277 Schuller, F., Menten, K. M., Contreras, Y., et al. 2009, A&A, 504, 415 Smith, M. D. 1994, A&A, 289, 256 Smith, M. D. 1995, A&A, 296, 789 Smith, M. D., Khanzadyan, T., & Davis, C. J. 2003, MNRAS, 339, 524 Smith, M. D. 2012, Astrophysical Jets and Beams (Cambridge: Cambridge University Press), pp. 116 - 117
Varricatt, W. P., Davis, C. J., Ramsay, S., & Todd, S. P. 2010, MNRAS, 404, 661 (VDR10) Vincent, M. B., Morse, J., Beland, S., et al., 2003, SPIE, 4841, 367 Wolf-Chase, G., Arvidsson, K., Smutko, M., & Sherman, R. 2013, ApJ, 762, 87 (WAS13) Wolf-Chase, G., Smutko, M., Sherman, R., Harper, D. A., & Medford, M. 2012, ApJ, 745, 116 (WSS12) Wood, D. O. S., & Churchwell, E. 1989, ApJ, 340, 265
Wright, E. L., Eisenhardt, P. R. M., Mainzer, A. K., et al. 2010, AJ, 140, 1868 Yao, Y., Ishii, M., Nagata, T., Nakaya, H., & Sato, S. 2000, ApJ, 542, 392
Zhang, Q., Hunter, T. R., Beuther, H., et al. 2007, ApJ, 658, 1152 Zhang, Q., Hunter, T. R., Brand, J., et al. 2005, ApJ, 625, 864 (ZHB05)
[^1]: <http://astro.kent.ac.uk/~df/MHCat/>
| {
"pile_set_name": "ArXiv"
} | ArXiv |
---
author:
- Ward Homan
- Leen Decin
- Alex de Koter
- Allard Jan van Marle
- Robin Lombaert
- Wouter Vlemmings
bibliography:
- 'wardhoman\_biblio.bib'
date: 'Received <date> / Accepted <date>'
subtitle: Analytical approach to embedded spiral geometries
title: 'Simplified models of stellar wind anatomy for interpreting high-resolution data'
---
Introduction
============
Near the end of their lives, low- and intermediate-mass stars ascend the asymptotic giant branch (AGB), where they develop a dense molecular envelope interspersed with (sub-)microscopic dust particles. The rate at which these stars expel their outer envelopes is high, from a few times $10^{-8} \msoy$ to about $10^{-4} \msoy$ [e.g. @DeBeck2010], and often varies in strength on a range of timescales [e.g. @Olofsson2010; @Maercker2012]. The winds are thought to be driven by a combination of pulsations and radiation pressure on dust grains, although the fundamentals of this mechanism have turned out to be an elusive problem. So far, empirical models and hydrodynamic simulations of these outflows have focused mainly on describing isolated AGB stars featuring spherical outflows. Recent observations with high-resolution telescopes, however, have revealed a rich spectrum of structural complexities. These include bipolar structures [e.g. @Balick2013], arcs [e.g. @Decin2012; @Cox2012], shells [e.g. @Mauron2000], clumps [e.g. @Bowers1990], spirals [e.g. @Mauron2006; @Mayer2011; @Maercker2012; @Kim2013], tori [e.g. @Skinner1998], and bubbles [@Ramstedt2014]. Several processes may be responsible for the formation of these structures. Shells and arcs may be caused by temporal variations in the mass-loss rate and/or expansion velocity [@Maercker2014 and references therein]. Aspherical structures may be the result of non-isotropic mass-loss [e.g. @Ueta2006], systemic motion relative to the local interstellar medium [e.g. @Decin2012], magnetic fields [e.g. @Sanchez2013; @VanMarle2014], or binarity [@Soker1997; @Huggins2007]. The latter need not be surprising as the multiplicity frequency of the progenitors of AGB stars is above 50 percent [@Raghaven2010; @Duchene2013]. A complex morphology of the circumstellar envelope (CSE) may have an effect on the strength and shape of spectral lines as well. Exotic profile shapes may result from deviations from sphericity, which may not be recovered using spherical models [@DeBeck2012]. Alternatively, the lines may have shapes typical for spherical flow but strengths that are deviant. In the latter case, spherical models may lead to erroneous determinations of the overall wind properties.
In the present paper we attempt to broaden our understanding of embedded stellar wind morphologies by developing simplified mathematical models of stellar wind structres. Here, we focus on the spiral structure, for which very convincing direct evidence [@Mauron2006; @Maercker2012], and theoretical background exists [@Theuns1993; @Soker1994; @Mastrodemos1999; @Kim2011; @Mohamed2012]. Hydrodynamical simulations indeed show that the influence of a binary companion on its direct surroundings can form Archimedean spirals embedded in the outflow, along the orbital plane of the system. A few distinct mechanisms are known to produce such a morphology. The existence of a binary companion inside the CSE of the mass-losing star can cause local perturbations which, after propagation through the CSE, develop spiral patterns. Local companion-wind interactions cause the wind material near the companion to be gravitationally funneled into an enhanced density region (a gravity wake) resulting in a flattened ( i.e. only barely extending away from the orbital plane) spiral. Additionally, the presence of this binary companion will cause the mass-losing star to wobble around the common centre of mass of the binary system. This reflex motion alters the local densities, ultimately forming an Archimedean spiral that is comprised of concentric shells, when observed edge-on (for visualisation, see Fig \[SpiralEdge\]) [@Kim2011; @Kim2012; @Kim2013]. Here, the specific geometrical properties of the resulting spiral have been found to be determined by the strength of the local gravity field with respect to the outflow velocity [@Kim2012]. Radial models of shocks generated by a pulsating mass-losing source, that are locally enhanced by interaction with a binary companion have also shown that not only single, but also multiple spiral branches can form in the CSE [@Wang2012]. In this case the properties of the spiral depend strongly on the ratio between the pulsation period of the star and the binary period. Finally, if the binary contains two mass-losing stars, the colliding winds at the interaction zone may form a spiral pattern [@Stevens1992; @Walder2003; @VanMarle2011]. Here the geometrical properties of the spiral are determined primarily by the velocity of the slower wind and the particular orbital motion of the binary.
In this paper, we make no assumptions on the spiral formation mechanism. Instead, we assume its presence, and show how these structures are manifested in the observables by means of 3D radiative transfer calculations. The stellar wind properties are described analytically, and its emission is modelled with [LIME]{}, a fully three-dimensional non-local-thermodynamical-equilibrium (NLTE) radiative transfer code [@Brinch2010]. Arguably, an exclusively analytical description of the stellar wind properties might not be very physically consistent. However, this approach allows us to calculate a considerable grid of models, which at the present time would be virtually impossible to achieve via consistent radiative-hydrodynamical calculations. We present the results both in spectral form, exhibiting the manifestation of the spiral structure in the molecular emission lines, and in spatial form, providing a reference for high spatial resultion data from telescopes like *ALMA*.
This paper is organised as follows: In Sect. 2 we describe the mathematical structure of the used analytical expression in detail, followed by the computational tools which have been used to produce the results. In Sect. 3 we give an overview of the explored parameter space, and of the assumptions we have made in this paper. Section 4 presents the results of the radiative transfer, both in spectral and in spatial format. Finally, in Sect. 5, we make an effort to relate the emission to intrinsic geometrical properties, and we show the effect of ALMA antenna configurations on the characteristics of the intrinsic emission.
Numerical procedure
===================
Spiral geometrical model
------------------------
The following mathematical descriptions are formulated in the spherical coordinate system, $$\begin{aligned}
r &=& \sqrt{x^2+y^2+z^2} \\
\mathcal{\theta} &=& \arccos{\left(\frac{z}{r} \right)} \\
\mathcal{\phi} &=& \arctan{ \left(\frac{y}{x} \right)},\end{aligned}$$ representing radial, azimuthal and equatorial distance respectively. The Archimedean spiral is the solution of the equation $r = b\phi$, and is characterised by a constant distance ($d = 2\pi b$) between consecutive turnings. The analytical properties of the assumed spiral model are described by a Gaussian distribution, such that sharp cut-offs of the wind properties described by this geometrical pattern are avoided. In dimensionless units this distribution is given by
$$\label{spir}
S(r,\theta,\phi) = \exp\left[-\frac{(r-r_0(r,\theta,\phi))^2}{2\sigma_r(r,\theta,\phi)^2} -\frac{(\theta-\theta_0(r,\theta,\phi))^2}{2\sigma_\theta(r,\theta,\phi)^2} \right],$$
where $\sigma_r$ and $\sigma_\theta$ are Gaussian widths in radial and azimuthal space. In the following subsections, we describe the specific geometrical properties of this model both radially and azimuthally, and provide an in-depth explanation of the functional parameters present in Eq. \[spir\]. The connection between $S(r,\theta,\phi)$ and the distributions of stellar wind properties (density, temperature) are outlined in Sect. 3.
### Geometry in the orbital plane
The parameters describing the geometry in the orbital plane are given below and are illustrated in Fig. \[SpiralFaceOn\]:
- $r_0(r,\theta,\phi) = b(\phi-\phi_0)$ describes the exclusive $\phi$-dependent outward evolution of the spiral windings. With $\phi > 0$ and $\phi_0 > 0$, $\phi_0$ corresponds to the position angle, that is, the angle with respect to a reference point from which the spiral originates.
- $\sigma_r(r,\theta,\phi) = \sigma_r$ describes the Gaussian width of the spiral arm in the orbital plane.
![Face-on view of the spiral model by taking a slice through the orbital plane, qualitatively depicting the meaning of the geometrical parameters. The colour scheme used is qualitative, and serves to demonstrate the Gaussian nature of the $S(r,\theta,\phi)$ function. \[SpiralFaceOn\]](1.eps)
### Geometry in the meridional plane
The parameters describing the geometry in the meridional plane, which is perpendicular to the orbital plane, are:
- $\theta_0(r,\theta,\phi) = \pi/2 $, chosen such as to keep the Archimedean spiral symmetrical relative to the orbital plane.
- $\sigma_\theta(r,\phi,\theta) = \alpha$ determines the angular height of the spiral, perpendicular to and symmetrical around the orbital plane. It is, in effect, a measure for the extent by which the spiral fans out away from the orbital plane. Its mathematical equivalence to the Gaussian standard deviation in Eq. \[spir\] implies that, in case of a density spiral, 68.2% of the total mass in the spiral can be found within the volume bound by $\alpha$. The choice of an angle as the parameter to quantify the height of the spiral is justified by the geometrical implications of an exclusively radial outflow. If $\alpha$ is small, then the bulk of the material is confined into a narrow spiral close to the orbital plane. If caused by binary interactions, this would correspond to a case where the spiral is formed by a strong gravitational wake of the companion. For larger $\alpha$, the spiral windings become more extended shell-like structures. Such a morphology would be a reasonable representation of the effect of a wobbling mass-losing star. In the most extreme case, which is for an opening angle of $\alpha = 180^\circ$, one is left with a dense winding of shells. A visualisation of the effect of $\alpha$ on the edge-on view of the spiral in shown in Fig. \[SpiralEdge\].
![Cross-section through the centre of, and perpendicular to the orbital plane. The effect of an increase in $\alpha$ on the edge-on view of the spiral model, with $\alpha$ the angle between the two dashed lines. A narrow spiral with an opening angle of $\alpha=10^\circ$ (left panel), an intermediate spiral with opening angle $\alpha = 35^\circ$ (middle panel), and a shell spiral with $\alpha = 180^\circ$ (right panel) are shown. The red horizontal dotted line represents the location of the edge-on orbital plane. The colour scheme used is qualitative, and serves to demonstrate the Gaussian nature of the $S(r,\theta,\phi)$ function. \[SpiralEdge\]](2.eps)
### Additional considerations
Hydrodynamical simulations by @Kim2011 show the existence of morphological substructures within the overall spiral shape. These models show this substructure to consist of two adjacent spirals. When seen edge-on (as in Fig. \[SpiralEdge\]) a frontal (curved) wake, and a rear (flat) wake can be identified. Such details can also be modelled with the previously developed analytical expression. The frontal spiral can be modelled by adopting Eq. \[spir\]. To model the rear portion of the spiral substructure, Eq. \[spir\] needs to be transformed from spherical coordinates to cylindrical coordinates, by substituting the spherical radius ($r=\sqrt{x^2+y^2+z^2}$) with the cylindrical radius ($r=\sqrt{x^2+y^2}$), and the spherical height ($\theta$) with the cylindrical height ($z$). By subsequently correctly choosing the ratio between the $b$-parameters of both spiral components, a resulting spiral pattern consistent with hydrodynamical results can be constructed.
We recall the probability of observing the orbital plane of a binary system face-on or edge-on. We define the face-on position to be an orientation for which the normal on the orbital plane deviates by at most ten degrees from the line of sight. Similarly, we define the edge-on position of the orbital plane to be the orientation for which the normal deviates by at most ten degrees from being perpendicular to the line of sight. With these definitions, we find that the probability of a face-on orientation is 1.5%, compared to 17.3% for edge-on. One is thus more than ten times more likely to observe a binary system edge-on than face-on.
We also make a note on our definition of position angle, which, in effect, is a rotation around the z-axis (or a rotation over the angle $\phi$). However, the rotational operations over $\theta$ and $\phi$ are not commutative. If the spiral is rotated over $\phi$ first, then the Archimedean spiral will remain in the x,y-plane. Its effect will be a face-on rotation, revolving the start of the spiral around the origin of the coordinate system. This effect is trivial, and is ignored in this paper. If, on the other hand, the spiral is rotated around $\theta$ prior to a manipulation of $\phi$, then the inclined spiral will be pivoted around the z-axis. Under such an operation, the normal on the Archimedean spiral plane will precess around the z-axis. We refer to angular manipulation of the latter case as operations on the position angle.
Radiative transfer
------------------
To generate the 3D intensity channel maps of simulated emission of the analysed spiral models, the Non-Local-Thermodynamical-Equilibrium (NLTE) full-3D radiative transfer code [ LIME]{} [@Brinch2010] was used. For a technical overview of the inner workings of the code, see @Brinch2010.
The code was originally designed for modelling cool molecular clouds. For the modelling of stellar winds, however, the central mass-losing star is a non-negligible source of energy. We have implemented the option to position an arbitrary number of stars at locations of choice. Radiative interaction of every grid point with these ’stars’ is forced at least once per iteration. The stars emit as black bodies. It is important to note that the code is non-dynamical, meaning it exclusively calculates the radiative transfer for the input parameters.
We modelled CO emission, for which the spectroscopic CO data of the LAMDA database [@Schoier2005] were used. The collisional rates were taken from @Yang2010.
Synthetic ALMA simulations
--------------------------
To produce simulations of *ALMA* observations of the emission of the spiral wind models we used the Common Astronomy Software Applications, [CASA]{}, post-processing package [@McMullin2007]. It consists of a collection of [C++]{} tools, managed by a [python]{} wrapper code.
The specific tools used to simulate the observations are the ‘simobserve’ and ‘clean’ functions. The simobserve task converts any model image (corresponding to the true sky brightness distribution) into virtual observations. For that it is necessary to specify quantities such as integration time, antenna configuration, frequency, and astrometrical and atmospherical parameters. These are used to calculate the noise contributions in uv-space. Subsequently, the task creates the Fourier transform of the model image and projects this onto this grid. The clean task converts your visibilities from uv-space back into real coordinates, to simulate how *ALMA* would see the input model. Finally, it performs a deconvolution of this ‘dirty image’ to get rid of the sidelobe structure, yielding a 3D datacube of a synthetical *ALMA* observation of the intrinsic 3D emission.
Model assumptions
=================
We consider a singular mass-losing system, located in the centre of the model. Though likely to be formed by binary interactions, we attempt to make no assumptions on the formation of the spiral structure in this paper. For all intents and purposes, this singular mass-losing system represents the binary system.
The 3D density field consist of two components. One spherical outflow component, for which mass conservation implies that $$\rho_{\rm HO}(r) = \dot{M}_{\rm HO}/(4\pi r^2 {\rm v}(r)).$$ $\dot{M}_{\rm HO}$ is the rate at which mass is lost in the homogeneous outflow. Hydrodynamical models show the outflow velocity of the spiral to be comparable to the intrinsic wind speed [@Kim2012; @Kim2013]. We express the velocity of both the spherical and the spiral wind components by a beta-law, $${\rm v}(r) = {\rm v}_\infty(1-R_0/r)^\beta,$$ where ${\rm v}_\infty$ is the terminal wind velocity, and $R_0$ the dust condensation radius. ${\rm v}(r)$ is thus exclusively radial in nature, and is the key to understanding why the spiral is bound by the opening angle $\alpha$. In addition to ${\rm v}(r)$ a constant turbulent velocity field ${\rm v}_{turb}$ exists throughout the wind.
This homogeneous density field is superimposed by a spiral density enhancement, $$\rho_{\rm spiral}(r,\theta,\phi) = \rho_0(r,\theta,\phi)\,S(r,\theta,\phi)$$ where $S(r,\theta,\phi)$ is described in Eq. \[spir\]. $\rho_0(r,\theta,\phi) = \rho_{\rm max}(2\pi b/r)^2$ describes the variation of the local Gaussian density peak, where $\rho_{\rm max}$ is a constant, and will be determined later. Supported by mass conservation in a radial outflow, we assumed a radial dependence of $\rho_0$ proportional to $1/r^2$. The resulting overall density profile of the spiral wind is $\rho_{\rm wind}(r,\theta,\phi)=\rho_{\rm HO}(r)+\rho_{\rm spiral}(r,\theta,\phi)$.
The temperature distribution throughout the wind follows the same twofold format as the density structure. Its main trend is a radial power law, $$T_{\rm HO}(r) = T_*(R_*/r)^\epsilon,$$ which describes the temperature of the homogeneous outflow. $T_*$ and $R_*$ are the stellar effective temperature and stellar radius respectively. The temperature of the spiral, $$T_{\rm spiral}(r,\theta,\phi) = T_0(r,\theta,\phi)\,S(r,\theta,\phi),$$ is superposed onto this background, with a chosen distribution maximum of $T_0(r,\theta,\phi)=T_0=\rm 60K$, corresponding to a value of the same order of magnitude as established from hydrodynamical calculations [@Kim2012]. The resulting overall temperature profile of the spiral wind is $T_{\rm wind}(r,\theta,\phi)=T_{\rm HO}(r)+T_{\rm spiral}(r,\theta,\phi)$.
Though not expected to affect the CO emission much, dust was taken into account in calculating the radiative-transfer, with the dust density distribution following the gas density distribution. The specific dust input parameters are presented in Sect. 2.2.4.
Additional effects such as scattering, magnetic fields, or rotation are not taken into account.
Outflow Parameters
------------------
We conducted a parameter study to asses the effect of changes in the free parameters of our model. Here, we present the applied parameter values. The general stellar and dust properties, as well as some overall characteristics of the CSE were taken from @DeBeck2012, making the high mass-loss case of the parameter study similar to the carbon-rich system. The bias towards C-rich objects reflects that most spirals were observed in such environments [@Morris2006; @Maercker2012; @Decin2015].
[ l l ]{}\
$T_*$ & $2330\ K$\
$L_*$ & $11300\ \lso$\
Distance & $150 {\rm\ pc}$\
\
$v_\infty$ & $14.5 \kms$\
$v_{turb}$ & $1.5 \kms$\
$\beta$ & $0.4$\
$R_0$ & $2.0 \rst$\
$\epsilon$ & $0.5$\
${\rm CO/H}_2$ & $6.0\times 10^{-4}$\
\
$2\pi b$ & $270 {\rm\ AU}$\
$\sigma_r$ & $20 {\rm\ AU}$\
$T_0$ & $60 {\rm\ K}$\
\
Amorphous Carbon & 53%\
Silicon Carbide & 25%\
Magnesium Sulfide & 22%\
Gas/Dust & 100\
Table \[fix\] gives an overview of the fixed parameters of the radiative-transfer models. These are the parameters which are either deemed to have a predictable or even trivial effect on the observables, or which bring about a global influence which is beyond the scope of this paper. The fixed spiral parameters are chosen such that a spiral with sufficiently contrasting features can be clearly identified in the observables.
The variable parameters are presented in Table \[var\], and the motivation behind their choice is described below.
- The total mass-loss of the system has profound effects on the emission in terms of absolute strength and optical depth effects. Therefore we simulated CSEs for two extreme mass-loss rates, one very high and one very low.
- The contrast between the density of the spiral and the homogeneous outflow brings about changes in relative emission strengths, which can easily be recognised in the observables. We define a mass contrast between both wind elements, using the parameter
$$\Sigma = \frac{\dot{M}_{\rm spiral}}{\dot{M}_{\rm HO}} = \frac{M_{\rm spiral}}{M_{\rm HO}},$$
with $\dot{M}_{\rm spiral}+\dot{M}_{\rm HO}=\dot{M}_{\rm total}$. $\dot{M}_{\rm spiral}$ represents the portion of the material confined to the boundaries of the spiral structure, flowing outward per unit time. $\Sigma$ thus quantifies the way in which the total mass lost by the star is distributed over the spherical outflow and spiral density enhancement components. $\Sigma \to 0$ corresponds to $\dot{M}_{\rm HO} \to \dot{M}_{\rm total}$, $\Sigma \to \infty$ to $\dot{M}_{\rm spiral} \to \dot{M}_{\rm total}$. A particular choice of $\Sigma$ will result in a specific value for $\rho_{\rm max}$. In principle, this parameter should also affect the value of $T_0$. We have chosen not to let it depend on $\Sigma$ because the relation between both is not clear to us.
- Two different spiral heights were simulated: one narrow spiral, with a small opening angle, and one spiral having a maximally large opening angle, referred to as a shell spiral.
- The inclination i is defined as the angle that tilts the view of the model between face-on, where the orbital plane lies perpendicular to the line of sight, and edge-on, where the orbital plane lies along the line of sight. Each model is observed under six different inclination angles, evenly spaced between this face-on and edge-on view.
Parameter Values Labels
-------------------------------------- ------------------------------------------------------------- --------------------
$1.5 \times 10^{-5}$ H (‘High’)
$1.5 \times 10^{-7}$ L (‘Low’)
$\forall \hspace{2mm} n \in \{1\} : m \in \{2,5,10,100\}$ Sd2,Sd5,Sd10,Sd100
$\forall \hspace{2mm} m \in \{1\} : n \in \{1,2,5,10,100\}$ S1,S2,S5,S10,S100
$10^\circ$ N (‘Narrow’)
$180^\circ$ S (‘Shell’)
Inclination $ \theta = (n/5)(\pi/2)$ $ n \in \{0,1,2,3,4,5\}$ i=0,18,36,54,72,90
Results
=======
We present the general findings of the parameter study in two parts. First, we exhibit the general trends and effects of the parameters on the spectral lines. Second, we show the effect of identical parameter simulations on the intensity maps. In order to investigate these 3D data we make use of the so-called position velocity (PV) diagram. The three dimensional data consists of two linearly independent angular dimensions, representing the two angular coordinates in the plane of the sky, and one velocity dimension. The PV diagram is, in effect, a slice through the 3D data at an arbitrary angular axis in the plane of the sky (thus preserving the velocity axis). It is thus a 2D plot of the emission along this chosen axis, versus velocity. In principle any slit width can be chosen. If the slit width is larger than one singular data pixel, then the emission is collapsed onto each other by summing up the emission with identical PV coordinates. For a condensed overview of the overall wind morphology, it is useful to make PV diagrams with a maximal slit width, namely the full size of the datacube. This is referred to as a wide-slit PV diagram. Additionally, it is advantageous to construct two different wide-slit PV diagrams of the 3D data, choosing any set of linearly independent (and thus perpendicular) angular dimensions oriented such as to maximally exploit the asymmetry of the data. This also counteract additional projection effects brought about by the position angle of the system. Our PV diagrams were constructed following these instructions. The axes along which the asymmetry of the models appears strongest are labelled as X and Y. Much easier to interpret than the full 3D data, they provide the user with clearly correlated structural trends of the complete wind, which in turn provides strong clues on possible geometries harboured by the wind.
All the shown data will are of the emission of the ground vibrational CO rotational transition J=3-2, unless otherwise stated. All results were continuum subtracted. In this section, the PV diagrams that best exhibit the asymmetry of the data (along X and Y) are referred to as PV1 and PV2 respectively. The spectral data were modelled as seen with a 20 arcsecond beam. (The quality of most images shown below is strongly enhanced when viewed on screen.)
Reference model
---------------
Fig. \[base\_model\] shows both the spectral line and the PV diagrams of an optically thin narrow spiral, with a total mass-loss = $1.5\times10^{-7}\msoy$ and $\Sigma\,=\,1$. This is considered the reference model, and will be referred to as such. Every model below shows the effect of the change of a singular parameter with respect to this model.
Spectral aspect
---------------
### Narrow spiral
**Inclination**: Fig. \[L\_N\_S1\_long\_line\] shows the effect of the inclination angle under which the spiral is seen. The appearance of the narrow spiral is very dependent on its orientation with respect to the observer. We therefore expect the inclination to have a strong effect on the observables. The homogeneous background wind is indifferent to these same transformations. The face-on view (${\rm i} = 0$) of the spectal line clearly shows the dual nature of the CSE. The spiral wind produces a sharp and narrow peak in velocity space. The width of this peak depends strongly on the height of the spiral. It is much narrower than the terminal velocity of the wind. This is because the spiral only hardly extends away from the $v=0 \kms$ Doppler plane. Because the wind is optically thin, higher local densities generate more emission, which is why the peak extends above the lower plateau, generated by the spherical wind. As the inclination angle increases, the contribution of the spiral decreases and widens, whilst the contribution of the homogeneous background wind remains unchanged, as expected. This evolution of the central peak is explained by the fact that the total emission is smeared out over a wider velocity range. At the highest inclinations, a double-peaked profile emerges in the spiral wind feature. Additionally, its width becomes comparable to the width of the spherical outflow, ultimately concealing the dual nature of the system. As a side note, the position angle has no effect on the spectral lines.
{width="90.00000%"}
**Mass contrast:** The main effect of the mass contrast is on the relative heigths of the spectral features generated by the spiral and spherical winds, as seen in Fig. \[L\_N\_Sd5\_line\]. For $\Sigma > 1$, the contribution of the spherical wind rapidly diminishes and disappears from the resultant profile, and only the spectral feature generated by the spiral wind remains. For $\Sigma \sim 1$ we find that the relative sizes of both contributions are comparable. When $\Sigma$ becomes very small, the spherical wind dominates the spectal feature.
![Spectral lines of a narrow spiral in the low mass-loss wind, in which the total amount of mass contained in the homogeneous outflow is five times the amount in the spiral ($\Sigma = 1/5$), seen face-on (left panel) and edge-on (right panel). \[L\_N\_Sd5\_line\]](7.eps)
Figure \[L\_N\_Sd5\_line\] shows a model with $\Sigma = 1/5$. The face-on case shows both a decreased contribution of the spiral wind and an increased contribution of the spherical wind relative to the reference model. Viewed edge-on, the spiral wind’s double-peak characteristics (visible in the $\Sigma = 1$ case) are reduced, leaving a flat-topped resultant spectral profile.
**Total mass-loss:** The first and most important effect of an increase in the total mass-loss is an increase in flux, visible in Fig. \[H\_N\_S1\_line\]. For low-inclination angles the peak due to the contribution of the spiral wind broadens with respect to the face-on reference model. This is due to the increased emission in the Gaussian tails of the density distribution of the narrow spiral. For high inclinations (nearing an edge-on view) the double-peaked aspect of the spectral line is less pronounced as the spiral becomes optically thick and relative brightening due to its long column depths in the line of sight direction (at maximum and minimum velocity) is suppressed.
![Spectral lines of a narrow spiral in the high mass-loss wind, in which the total amount of outstreaming mass is equally divided between both wind components, seen face-on (left panel) and edge-on (right panel). \[H\_N\_S1\_line\]](8.eps)
### Shell spiral
In the case of the shell spiral, the spiral and spherical wind components possess virtually identical widths in velocity space, which makes both components essentially indistinguishable from one-another. This is due to the broadening effect of the opening angle $\alpha$ on the emission feature of the spiral wind. For high values of $\alpha$ both wind components will be of comparable width. For this reason we refrain from showing this in figures. In Sect. 4.3 we discuss the integrated line fluxes for various combinations of mass-loss in spiral and spherical outflow to assess the potential for erroneous estimates of the total mass-loss rate if assuming a spherical wind only.
### Other CO lines
Different rotational transitions of CO (besides CO J=3-2) can provide additional information. In Fig. \[ladder\] the effect of rotational CO transitions ranging from J=3-2 to J=38-37 (probing the inner parts of the wind) on the line shapes of the reference model is shown. Hydrodynamical simulations of the regions close to the mass-losing star show that the sphericity of the outflow is completely destroyed locally by the binary interactions. The effect of a well-behaved analytical spherical outflow close to the centre of the spiral is thus unrealistic. The spiral shape, however, is recognisable on this scale. Therefore, to properly show how the spectral lines are affected by the spiral, we omitted the contribution of the spherical wind.
![The effect on the shape of the spectral lines as a function of CO transition. The first row shows the line shapes for a face-on narrow spiral, the second row for an edge-on narrow spiral, and the third row for a shell spiral. To focus the attention on the shapes of the emission lines, the peak strengths have been normalised, and the spherical wind contributions have been omitted. \[ladder\]](9.eps)
In the case of the face-on narrow spiral, the spectral lines show no recognisable changes in the overall characteristics. However, other viewing angles show a very characteristic evolution in the line shapes of the narrow spiral, with a progressively receding red or blue wing. This is explained as follows. For the higher CO transitions the emitting region, probing the most central regions of the spiral, will merely contain an incomplete portion of the first revolution of the spiral. Depending on whether the part of the spiral contibuting to the line flux is predominantly approaching or receding, emission is more concentrated in the blue or the red wing. It is likely that the most central regions of the binary system will show additional complexities, as this region harbours all the mechanics that form the spiral. These effects are not taken into account, and may cause additional effects on the overall shape of the high CO transition lines. The shell spiral shows an evolution towards a triangular line shape. However, this is not dissimilar from the expected evolution of a homogeneous outflow, as the higher CO transitions probe the acceleration region of the wind.
Spatial aspect
--------------
In this section we present the spatial counterpart of the above results. Generally, the two-fold nature of the wind is clearly recognisable in the PV diagrams. The spiral is manifested as a sequence of emission bands, whilst the contribution of the spherical wind is a large circular shape with a strongly enhanced central bar, representing the warm and dense regions near the star. This elongated feature approximately reaches up to the terminal velocity of the wind, and is visible in most PV diagrams below.
### Narrow spiral
**Inclination:** The appearance of the narrow spiral is dependent on its orientation with respect to the observer, thus the velocity channel maps will be sensitive to global angular transformations. We recognise the central bar generated by the spherical wind in Fig. \[L\_N\_S1\_long\_pv\]. The main feature, however, is the strongly periodic pattern seen in both PV1 and PV2, generated by the spiral. The face-on view of the narrow spiral generates two identical PV diagrams, with a narrow column of periodic emission bands. The bands decrease in brightness for increased distance to the orbital plane (positioned at an angular deviation of 0 arcsec), as a result of the combined $1/{r^2}$ density dependence and local temperature. As the inclination increases, the primary contribution to PV1 gradually twists into an S-shaped feature, tightening around the central contribution, until it becomes a narrow, horizontal dumbell-like feature. In PV2 the evolution is different, because of the strong asymmetry of the edge-on narrow spiral. Its evolution stretches the initial shape sideways, ultimately forming an ellipsiodal pattern of periodic emission bands for the highest inclination values.
{width="90.00000%"}
**Mass contrast:** As the mass contrast varies the emission contribution between the spiral and spherical winds changes accordingly. In the case of $\Sigma = 1/5$, as seen in Fig. \[L\_N\_Sd5\_pv\], the contribution of the spherical wind is strongly enhanced, with a decreased relative emission of the spiral wind. However, the general structure of the PV diagrams remains unchanged.
![Position-velocity diagrams of a narrow spiral in the low mass-loss wind, in which the total amount of mass contained in the homogeneous outflow is five times the amount in the spiral, seen face-on and edge-on. The top row shows PV1, the bottom row PV2. \[L\_N\_Sd5\_pv\]](12.eps)
**Total mass-loss:** In the high mass-loss regime the PV diagrams show an overall increase in emission compared to the low mass-loss regime, as seen when comparing Fig. \[H\_N\_S1\_pv\] with Fig. \[L\_N\_S1\_long\_pv\]. The apparent widening of the features of the spiral wind is due to non-negligible emission coming from the tails of the Gaussian density profile; emission which, in the low mass-loss case, barely contributed to the PV features. Optical depth effects keep the emission peak from increasing by the same factor as the wing emission.
![Position-velocity diagrams of a narrow spiral in a high mass-loss wind for $\Sigma = 1/5$, seen face-on (left panels) and edge-on (right panels). The top row shows PV1, the bottom row PV2. \[H\_N\_S1\_pv\]](13.eps)
### Shell spiral
The primary and spherical wind components possess identical spectral widths, and their general tendency to invariance under angular transformations translates into particularly similar PV1 and PV2 diagrams. This strong charateristic is a reliable diagnostic, assisting with the differentiation between the shell and narrow spiral.
**Inclination:** Unlike for the spectral lines, the PV diagrams do reveal information on the inclination of the shell spiral, be it only for higher inclinations. Fig. \[L\_S\_S1\_pv\] shows that for low inclinations both PV1 and PV2 are virtually indistinguishable. Higher inclinations show the appearance of gaps in the emission bands of PV1. The gaps appear at i=54 around + and - 12 km/s and follow these bands as i increases, before reaching a vertial position at i=90. Under an iclination of $i=0^\circ$, any chosen PV slit (or axis along the plane of the sky) will be parallel to the orbital plane of the system. This means that the resulting PV diagram will be invariant to the particular choice of axis along which to construct the PV diagram. This explains why PV1 and PV2 are identical. However, at an inclination of $i=0^\circ$, and when the axes are chosen to best show the asymmetry of the system, the orbital plane of the model will be perpendicular to one, whilst remaining parallel to the other axis. The axis parallel to the orbital plane thus again produces an identical PV diagram as before, whilst the other will enhance the features along the dimension perpendicular to the orbital plane, which, in case of the shell spiral, contains the gaps.
{width="90.00000%"}
**Mass contrast:** The $\Sigma = 1/5$ case is displayed in Fig. \[L\_S\_Sd5\_pv\], and shows how the relative contribution of the spherical wind intensifies as the contribution of the spiral wind fades. The general structure of the PV diagrams remains unchanged.
![Position-velocity diagrams of a shell spiral in a low mass-loss windfor$\Sigma = 1/5$, seen face-on (left panels) and edge-on (right panels). The top row shows PV1, the bottom row PV2. \[L\_S\_Sd5\_pv\]](15.eps)
**Total mass-loss:** A strong overall increase in emission, as seen when comparing Fig. \[H\_S\_S1\_pv\] with Fig. \[L\_S\_S1\_pv\]. The general structure of the PV diagrams remains unchanged.
![Position-velocity diagrams of a shell spiral in a high mass-loss wind for $\Sigma=1$, seen face-on (left panels) and edge-on (right panels). The top row shows PV1, the bottom row PV2. \[H\_S\_S1\_pv\]](16.eps)
Discussion
==========
Constraining the geometry
-------------------------
For our specific model parameters and our specific choice of PV axes, we find that PV2 readily provides the angular width and size of the gap between the emission bands, and thus the value of $b$. A slice through $v=0\, \kms$ of PV2 allows to constrain the geometrical parameters in the orbital plane of the spiral. We refer to these diagrams as Slice Profiles (SPs) below. A number of such SPs are shown in Fig. \[slice\]. The top row exhibits the SP as they are, the bottom row shows the right half of the SP, overplotted with the geometrical model. We find as a general rule that the emission peaks systematically coincide with the density peaks. The angular distance between the spiral windings is thus readily available from the SPs of the PV diagrams. This distance can, if caused by binary interactions, directly be related to local physics according to @Kim2012 via the relation $$\Delta r_{\rm arm} = \left( \langle V_w \rangle + \frac{2}{3} V_p\right) \times \frac{2 \pi r_p}{V_p},$$ with $\langle V_w \rangle$ the wind velocity, $V_p$ the orbital velocity, and $r_p$ the orbital radius of the primary. The orbital period $T_p$ is given by the last term.
The angular width of the emission bands, however, does not relate one-on-one to the spiral geometry and is thus more difficult to deduce from actual observations.
The plots labelled A in Fig. \[slice\] present the SP for a low mass-loss wind harbouring a narrow spiral seen edge-on, with $\Sigma=1$. As expected, the spiral geometry tightly fits the emission characteristics. Additionally, a smooth bell-shaped background feature is visible, which is generated by the homogeneous component of the wind. Finally, the peak strength of the spiral-induced emission spikes clearly changes as a function of offset. This is due to a combination of the model specific radial dependence of the spiral density and of the temperature law, as the latter defines the line-contribution regions.
For more complex SPs, such as those for the shell spiral (labelled C), the emission peaks are not as narrow and smooth because the emission is smeared out over the offset dimension. Nevertheless, the presence of the spiral is still unmistakable. In this case, only the top portion of the peak, which ranges from approximately the emission plateau strictly to the left of the spiral peak to the plateau strictly to its right, should be used to determine the geometry. Additionally, the bell-shaped background curve is still recognisable, but as a result of the nature of the shell spiral, the smeared-out emission somewhat conceals its shape.
In the case of high mass-loss, the Gaussian tails of the density distribution contribute considerably to the SPs, while optical depth effects prevent the highest density regions from augmenting their emission by as much. This results in a decreased peak to background emission contrast. The emission from the tails broadens the emission zones significantly, which therefore no longer directly traces the width of the spiral arms. This effect is inherent to the mathematical properties of the Gaussian density distribution. Sharp cut-offs of the spiral density would not result in such a broadening. The bell-shaped emission feature of the smooth background of the A models has become much more prominent compared to the emission peaks generated by the spiral. Its shape has also changed to triangular. These two differences can be asigned to optical depth effects, which suppress emission from high-density regions.
It is necessary to note that these SPs should not be overinterpreted as their overall and absolute morphology is extremely sensitive to the specificities of the geometrical and radiative transfer model. Additionally, superimposed noise and diffraction effects will blur the small features in the SP.
A 1.
B 1.
C 1.
A 2.
B 2.
C 2.
### Constraining the inclination
The dependence of the emission signature on the incination is characterised by the formation of an S-shaped feature in PV1 as inclination inceases. For a narrow spiral the S-shape is unmistakable. The shell spiral exhibits this S-shape in the form of gaps in the emission bands. We constructed an analytical expression that fits the overall morphological trend of the inclination dependence of the PV1 diagram. This relation is given by $$\label{fit}
S(v) = -A ln\left[\frac{B}{v+\left(\frac{B}{2}\right)}-1\right].$$ Exhibited as the wide green curve in Fig. \[PVfit\], left panel, this expression allows for the determination of the angle that the emission feature makes with the velocity axis, around $v=0 \kms$, by using $$\frac{dS(v=0)}{dv}=\frac{4A}{B}.$$ This is referred to as the skew of the diagram, and was determined for a range of inclinations. We related the measured skew to the actual inclination angle to determine how they are related. The resulting curve (Fig. \[PVfit\], right panel) shows a clear trend.
Impact of the spiral structure on the total line strengths
----------------------------------------------------------
The signature of a spiral can quite easily be discerned in PV diagrams. However, the effect on the CO spectral lines is much less outspoken particularly so for extended (large $\alpha$) spirals and high mass-loss rate systems. Here, we investigate the effect of a spiral structure on the frequency-integrated line strength of CO transitions. We discuss this integrated line strength relative to the case in which all of the gas is in a spherical outflow to assess the error we would make in constraining the mass-loss if we applied a model for a spherical wind to a system in which part (or all) of the material leaves the star in a spiraling outflow.
Figure \[f2r\] presents the total line flux normalised to the case of a spherical flow for six different CO transitions, from J=2-3 to 38-37. The total flux ratio is plotted versus $\Sigma$. Models for a range of inclinations are presented at each value of $\Sigma$. The four basic configurations are labelled with different colours (blue for the shell spiral and red for the narrow spiral) and symbols (triangles for the low-$\dot{M}$ model and squares for the high-$\dot{M}$ model). The green horizontal bar represents a typical error margin of line-data calibrations, approximately 20 percent.
For low values of $\Sigma$ the bulk of the matter is in the spherical outflow and all ratios tend towards unity, as expected. The seeming failure of convergence to unity for high J levels originates in the fact that the elevated temperature of the spiral is not present in the spherical reference models. For the J=3-2 transition, optical depth effects do not play a dominant role, safe for the high-$\dot{M}$ narrow spiral models. For these models the flux ratio drops because of self-shielding. Note that, generally, the low-$\dot{M}$ models tend to have an increased flux ratio and the high-$\dot{M}$ models a decreased flux ratio. The former is an effect of density. If the material is concentrated in a spiral structure collisional excitations are more important causing the upper level of the transition to become overpopulated. This may increase the flux ratio by up to a factor of three. However, densities never reach values that strongly affect the populations of very high levels, with $J \gtrsim 30$. Note also that the effect of inclination is not that important (compared to calibration uncertainties) for the low mass-loss case as the medium remains optically thin (in the line of sight). In the high-$\dot{M}$ case the line ratio drops, the more so if most of the material is in a narrow spiral structure (high $\Sigma$) and the inclination is high (approaching an edge-on view). In that case the flux may drop by up to a factor of ten. If the spiral fans out to well above the plane of symmetry (i.e. high $\alpha$) the drop in flux is not that dramatic, simply because of less self-shielding compared to the narrow spiral.
The models discussed here present a wide range in $\Sigma$, that may not all be realistic. The blue line in Fig. \[f2r\] represents the mass ratio $\Sigma$ that was derived for the wind structure of the binary AGB system by @Kim2012. It shows that for this particular system one would not derive a significantly different total mass-loss rate if one would have applied a spherical outflow model.
{width="90.00000%"}
We used the analytical expression derived by @DeBeck2010, relating integrated line flux to mass-loss, to estimate the error on derived mass-loss rates provided prior knowledge on flux ratios. They showed that the analytically derived mass-loss scales as the integrated line flux to a certain exponent, $\dot{M} \sim [ \int F_{\nu} d\nu ]^{\gamma}$, where $\gamma$ depends on the specific CO transition, and on the optical thickness of the spectral line. For the CO J=3-2 transition they found values for $\gamma$ to be 1.14 in the optically thin regime, and 0.61 in the optically thick regime. This translates, in the optically thin regime, into an uncertainty on the mass loss which is approximately a factor 1.3 greater than the uncertainty on the line flux (a flux uncertainty of a factor ten transflates into an uncertainty in deduced mass loss of a factor thirteen). For optically thick spectral lines, the uncertainty on the mass-loss is approximately a factor of 2.5 smaller than the uncertainty on the line flux (a flux uncertainty of a factor ten transflates to an uncertainty in deduced mass loss of a factor four). Note that the results of @DeBeck2010 were derived by explicitly calculating the energy balance throughout the parameter study, as opposed to assuming a fixed temperature structure as in our models.
It is worth noting that difficulties with the 1D modelling of an abundance of spectral lines, like for example a combination of ground-based and unresolved PACS lines, are usually solved by forcing the model to have a peculiar and possibly somewhat misshapen temperature profile. The need for such actions might indicate the presence of a complex geometrical structure in the observed stellar wind.
Single-dish simulations
-----------------------
Spectral features of single-dish observations can be heavily influenced by the telescope beam profile. To simulate how the intrinsic spectral lines of our models are affected by a beam profile, we covered our specific intensity images with a Gaussian filter, centred onto the location of the central source. In Fig. \[beam\] the effect of Gaussian filters with different full-width at half-maximum (FWHM) values on the overall shape of the spectral lines is shown. The adopted values for the FWHM are $n/20 \ {\rm for}\ n \in \{1,2,3,...,10\}$ of the total image size, which is 20 arcsec. It is important to note that, for the clarity of the figures, the Gaussian filter was normalised. This operation scales the flux by the normalisation factor, allowing us to properly demonstrate the effect on the line shape without overlapping the lines. Plot A shows the effect of an ever narrower beam on a face-on narrow spiral. The prominent central feature remains, but as the beam size diminishes, a double-peaked profile forms around this central peak, created by the strongly resolved spherical wind. Plot B presents the effect of beam size on an edge-on narrow spiral. When fully resolved, the spectral line shows a very pronounced double-peak, which could be incorrectly identified as a detached shell. Finally, plot C shows the same effect for a face-on shell spiral. As a result of its similarity to a homogeneous outflow, the gradual evolution to a double-peaked profile is expected. However, for an extremely resolved shell spiral, a small bump appears in the centre of the spectal line, which is some remaining numerical artefacts.
ALMA simulations
----------------
The functions simobserve and the clean of [CASA]{} generated these synthetical *ALMA* obsevations.
[ l l ]{}
\
Pixel size & 0.04”\
Field size & 20”\
Peak flux & Taken from [LIME]{} output\
Transition & CO 3-2 (345.76599 GHz)\
Antenna configurations & All available in cycle2\
Pointing & Single\
Channel width & 1.15MHz, centred on rest freq.\
PWV & 0.913 mm\
Thermal noise & standard\
Temperature & 269 K\
Integration time & 10 min on-source\
In Fig. \[antenna\], we present the effect of the seven different antenna configurations offered in ALMA Cycle2[^1] on the face-on reference model, for which the fixed geometrical parameters can be found in Table \[fix\]. For these model-specific geometrical parameters (combined with a fixed distance of 150 pc), we find that the resolution of the C34-1 configuration is too coarse. It only poorly samples the spiral structure. Additionally, the largest recoverable scale for the C34-1 configuration is large enough to nicely see the spherical wind. Combined, these effects show a strong bias towards the spherical wind. The C34-7 configuration has the opposite effect, with a resolution high enough to detect the spiral structure, but a largest recoverable scale that is too small for a proper sampling of the spherical wind. Its flux contribution is completely lost. The C34-3 and C34-4 antenna configurations combine the advantages of both extremes, making them the optimal setup with which to simulate observations of this spiral model. A direct consequence of using this best-fit configuration is that, as a general tendency, the overall structure of the PV diagrams remains relatively unaffected by the observation. We thus conclude that for this specific model, an extended configuration with a maximum baseline of about 500m and a minimum baseline about 20m produces images of the higest quality. These baselines correspond to angular resolutions of approximately 0.32-0.4 arcseconds, which, as expected, is similar to the angular size of the minimal length-scale of our models (the spiral arm thickness) of approximately 0.26 arcseconds. We would like to emphasise that these results are very model dependent, with the angular width of the smallest scale of the object being the most influential on the quality of the observations. The angular resolution $\Delta \theta$ of a configuration with a maximum baseline $L_{max}$ can be estimated with the following relation $$\Delta \theta \simeq \frac{61800}{L_{max} \nu},$$ with $\nu$ the observing frequency in GHz. Keeping the value of $\Delta \theta$ close to, but lower than the smallest expected angular size of the observed features of interest (in the case of the spiral structure the value $\sigma_{r}/{\rm distance}$) should ensure a good quality observation.
The compact array simulations are not shown. The resolution of the compact array is too low to be able to discern any useful features in the PV diagrams.
{width="90.00000%"}
A. intrinsic emission
B. C34-1
C. C34-4
D. C34-7
Fig. \[ALMAmod\] presents the effect of the C34-1,4 and 7 antenna configurations on the PV diagrams of the face-on high and low mass-loss narrow spiral models (labelled by HN and LN respectively), and the high and low mass-loss shell spiral models (labelled by HS and LS respectively). An immediate conclusion from a preliminary inspection of the figures is that the lower resolution configuration provides considerably more information on the object than configurations with resolutions which are too high. The longest baseline configurations resolve out all the large-scale information, leaving only an exceptionally biased focus on the smallest scale, largest emission features.
From these simulations we conclude that it is impossible, without any prior knowledge on the geometrical parameters of the spiral structure, to determine the optimal antenna configuration with which to observe the object. This being said, two scenarios can be thought of that can facilitate the configuration determination. Either hydrodynamical simulations predict a preferred spiral arm thickness, inter-spiral-winding-distance, or relation between both, in which case an optimal (longest) interferometry baseline can be calculated. Or theoretical models provide no such information, in which case it is advised to observe with a combination of configurations.To determine the best-fit antenna configuration the observed wind needs to be probed with at least the two most extreme maximum baselines.
Conclusions
===========
Using the 3D radiative transfer code [LIME]{}, we have conducted a large-scale parameter study to investigate the effect of geometrical and global wind properties of AGB outflows containing spiral structures that are embedded in a spherical outflow. Considering the CO v=0 J=3-2 transition, a singular spectral line generally conceals the dual nature of the wind. Only for limited combinations of parameters (inclination and spiral width) is its dual nature recognisable. However, when comparing different rotational transitions of CO, a characteristical evolution of the line shape is percieved as one progresses to higher transitions for the non-face-on narrow spiral models. The peculiarity of the (evolution of the) resulting line shapes shows strong evidence for the presence of an embedded spiral.
The [LIME]{} output also allowed us to investigate the 3D emission of the CO v=0 J=3-2 transition throughout velocity space. We found that the best tools for analysing these images are the wide-slit PV diagrams, which intensify the correlated structural trends in the emission. Using these PV diagrams, we consistently found a strong signature of the spiral in the data, from which most of the geometrical properties can be recovered. It seems that one specific parameter, the mass- or density contrast, cannot be deduce by only analysing the emission, that is without resorting to detailed radiative transfer efforts. We compared the integrated flux of the 3D emission with an equivalent 1D model (where all the material resides in the spherical outflow component) to explore the uncertainties such a 1D misinterpretation brings about. We found that generally the errors on derived mass-loss do not exceed a factor of a few. Only in extreme cases can the difference between the embedded spiral flux and the exclusively spherical flux reach factors of up to ten, translating into uncertainties on derived mass losses exceeding an order of magnitude (for optically thin lines), or up to a factor of four (for optically thick lines).
Finally, we have simulated cycle 2 *ALMA* observations of the model emission with [CASA]{}. In addition to the importance of recognising interferometric artefacts, we conclude that no preferred antenna configuration exists, since the required resolutions depend strongly on the spiral geometry. To promote spiral detection the source of interest should be observed with a range of different maximum-baseline configurations.
W.H. acknowledges support from the Fonds voor Wetenschappelijk Onderzoek Vlaanderen (FWO). A.J.v.M. acknowledges support from FWO, grant G.0277.08, KU Leuven GOA/2008/04 and GOA/2009/09. R.L. acknowledge supports from the Belgian Federal Science Policy Office via the PRODEX Programme of ESA. W.V. acknowledges support from the Swedish Research Council (VR), Marie Curie Career Integration Grant 321691 and ERC consolidator grant 614264. In addition, we would also like to express sincere thanks to Michiel Hogerheijde and Christian Brinch for their support with the [LIME]{} code, as well as to Markus Schmalzl of the Allegro node ALMA support community for his assistance with [CASA]{}.
[^1]: http://almascience.eso.org/documents-and-tools/cycle-2/alma-technical-handbook
| {
"pile_set_name": "ArXiv"
} | ArXiv |
---
abstract: 'We study a contextual bandit setting where the learning agent has access to sampled bandit instances from an unknown prior distribution ${\mathcal{P}}$. The goal of the agent is to achieve high reward on average over the instances drawn from ${\mathcal{P}}$. This setting is of a particular importance because it formalizes the *offline optimization* of bandit policies, to perform well on average over anticipated bandit instances. The main idea in our work is to optimize *differentiable* bandit policies by policy gradients. We derive reward gradients that reflect the structure of our problem, and propose contextual policies that are parameterized in a differentiable way and have low regret. Our algorithmic and theoretical contributions are supported by extensive experiments that show the importance of baseline subtraction, learned biases, and the practicality of our approach on a range of classification tasks.'
author:
- |
Branislav Kveton\
Google Research\
Martin Mladenov\
Google Research\
Chih-Wei Hsu\
Google Research\
Manzil Zaheer\
Google Research\
Csaba Szepesvári\
DeepMind / University of Alberta\
Craig Boutilier\
Google Research
bibliography:
- 'References.bib'
title: 'Differentiable Meta-Learning in Contextual Bandits'
---
| {
"pile_set_name": "ArXiv"
} | ArXiv |
---
abstract: 'We study the renormalization group flow of the ${\rm O}(N)$ nonlinear sigma model in arbitrary dimensions. The effective action of the model is truncated to fourth order in the derivative expansion and the flow is obtained by combining the non-perturbative renormalization group and the background field method. We investigate the flow in three dimensions and analyze the phase structure for arbitrary $N$. While a nontrivial fixed point is present in a reduced truncation of the effective action and has critical properties which can be related to the well-known features of the ${\rm O}(N)$ universality class, one of the fourth order operators destabilizes this fixed point and has to be discussed carefully. The results about the renormalization flow of the models will serve as a reference for upcoming simulations with the Monte-Carlo renormalization group.'
author:
- Raphael Flore
- Andreas Wipf
- Omar Zanusso
title: |
Functional renormalization group of the nonlinear sigma model\
and the ${\rm O}(N)$ universality class.
---
Introduction
============
Nonlinear sigma models (NLSM) are of great interest since they appear as effective models for many quantum systems in various branches of physics, ranging from solid state physics (e.g. quantum Hall effect) to particle physics (e.g. theory of light mesons) [@quHall; @BrezinZinn; @Delamotte; @quAntiferro; @Pelissetto; @Zakrzewski; @Novikov; @Leutwyler]. They are particularly useful in the description of theories whose symmetries are broken below a certain scale. From a perturbative point of view, NLSM are considered as fundamental theories only in two dimensions, while in higher dimensions they are mainly used as effective theories, since the coupling constant has negative mass dimension and the theories are not perturbatively renormalizable. However, there is the possibility that in $d=3$ dimensions these models are non-perturbatively renormalizable, i.e. “asymptotically safe” [@weinberg]. The concept of asymptotic safety has mainly been applied in order to provide a non-perturbative renormalization of gravity [@Reuter:1996cp]. It is known that NLSM and gravity exhibit many structural similarities, since both are described by non-polynomial interactions and have the same power counting behavior. Further, both have been shown to be non-perturbatively renormalizable in a large-$N$ expansion (see [@largeN] and references therein). NLSM can therefore serve as interesting laboratory to test and develop techniques for the more involved theory of gravity.
A promising approach to explore the non-perturbative quantization of asymptotically safe theories is provided by the “functional renormalization group” (FRG) [@wetterich; @Reuter:1996cp]. It describes the scale dependence of a quantum field theory and in particular of its effective action by means of an exact functional equation. In order to treat theories with nontrivial target spaces like the NLSM, some advanced background techniques have been developed within the FRG framework, which still require a deeper understanding. Nonlinear O($N$) models provide a useful testing ground for these conceptual issues, because they attracted a lot of attention within statistical field theory and, as a result, their critical properties are well-known.
The FRG approach is a suitable tool to investigate the phase structure of physical systems. A second order phase transition, for instance, is related to a fixed point of the renormalization group flow. While we will explicitly study the nonlinear O($N$) model, it is expected that they possess the same critical properties as their linear counterparts. The hypothesis of universality states that two short-range theories with the same dimension and the same symmetries belong to the same universality class. This expectation is strongly supported by many explicit computations for $d=3$, see [@Zinn; @Ballesteros; @Butera; @Pelissetto; @Chan] and references therein. While Monte Carlo simulations yield the critical exponents within an O($N$)-invariant formulation, there is further need of non-perturbative analytical studies that are manifestly covariant and can confirm the critical properties of the nonlinear models.
Understanding the phase structure of field theories with curved target spaces and advancing the techniques developed to deal with such theories within the FRG are the main motivations for the present work. We will study the nonlinear O($N$)-models in a background field expansion [@Codello:2008qq; @Percacci:2009fh] and compute the effective action in a truncation that includes all terms of fourth order in the derivatives (see also [@Fabbrichesi:2010xy] for some related computation). We will derive the flow in arbitrary dimensions and then focus on the interesting case $d=3$. We will employ a manifestly covariant background field method and improve the formalism used in [@Percacci:2009fh] by taking care of the specific scaling of the fluctuation fields in a fully non-perturbative way.
The three dimensional linear sigma model has been examined in the FRG framework for a truncation that contains all operators up to the fourth order in the derivative expansion [@Canet:2003qd]. The results in [@Canet:2003qd] confirm the existence of a second order phase transition corresponding to a fixed point of the RG flow, and show that the fourth order truncation provides a significant improvement over the second order (local potential) approximation, in terms of precision with which the critical exponents and the anomalous dimension are computed, as compared to other methods. In our computations of the nonlinear model, the inclusion of fourth order operators also improves the sensitivity to the critical properties in comparison to the second order truncation studied in [@Codello:2008qq], such that the critical exponent of the correlation length at the nontrivial fixed point qualitatively agrees with the expectation for the O$(N)$ universality class. However, one particular operator of fourth order destabilizes the system and no fixed point can be found for the full fourth order truncation. Whether this represents a physical property of the nonlinear model is a delicate question, since it would imply a departure of the critical behavior of the nonlinear model from that of the linear one. It is not yet possible to give a conclusive answer to this issue, but we will argue that the nontrivial fixed point may reappear, once higher order operators are taken into account.
Another important motivation of this article is to provide explicit results for a direct comparison between the FRG approach and the widely used ab-initio lattice approach to field theories. An accompanying analysis of the renormalization group flow of various nonlinear O($N$) models on the lattice by means of the Monte Carlo Renormalization Group (MCRG) method will be presented elsewhere [@upcoming]. It will be interesting to see how these two non-perturbative methods complement each other with respect to the FP structure and critical properties.
This article is structured in the following way: Section \[background field method\] is devoted to a covariant description of the O($N$) models with curved target spaces and the development of the background field expansion. In section \[functional rg\] we will recall the FRG approach and show how to implement the background field method. Our truncation of the effective action is discussed in section \[higher derivative model and beta functions\] and the running of the remaining couplings is derived. Section \[on phase transition\] presents an analysis of the renormalization flow and the critical properties of the model. The conclusions are contained in section \[conclusion\].
Geometry of the model {#background field method}
=====================
A field of the ${\rm O}(N)$ nonlinear sigma model is a map $\varphi: \Sigma \to S^{N-1}$ from spacetime $\Sigma$ to the unit-sphere in $\mathbb{R}^N$. In the present work we will choose for $\Sigma$ the Euclidean space $\mathbb{R}^d$. It is convenient to regard the components $\varphi^a$ as coordinates on the target space $S^{N-1}$. The field space of all these maps is denoted by ${\cal M}\equiv\{\varphi:\mathbb{R}^d\to S^{N-1}\}$.
The target space is the homogeneous space $S^{N-1}={\rm O}(N)/{\rm O}(N-1)$ with ${\rm O}(N)$ as its isometry group. It is equipped with a unique Riemannian metric $h_{ab}$ that is ${\rm O}(N)$-invariant. The microscopic action for the model is $$\label{Caction}
S[\varphi] = \frac{1}{2}\zeta \int d^d x ~ h_{ab}(\varphi) {\partial_{\mu} \varphi}^a {\partial^{\mu} \varphi}^b\,,$$ where $\zeta$ is a coupling constant. Since the fields $\varphi^a$ have the status of coordinates, it is natural to regard them as dimensionless, such that $\zeta$ has mass-dimension $[\zeta]=d-2$.
The action is invariant under arbitrary reparametrizations $\varphi\to\varphi'(\varphi)$ of the fields, provided that the metric $h_{ab}$ transforms as a symmetric $2$-tensor. In addition the model admits the ${\rm O}(N)$ isometries on the sphere as symmetries. These isometries are generated by vector fields $K^a_i(\varphi)$ which satisfy a generalized angular momentum algebra, $$\left[K_i,K_j\right]=-f_{ij\ell} K_\ell\,,$$ where $f_{ij\ell}$ are the structure constants of the Lie algebra of the rotation group. The infinitesimal symmetries generated by the $K_i$ are nonlinear $$\label{symmetry}
\varphi^a \to \varphi^a+\epsilon^i K^a_i(\varphi).$$
From the invariant metric $h_{ab}$ on the sphere one obtains the unique Levi-Civita connection $\Gamma_a{}^b{}_c$ and the corresponding Riemann tensor $R_{abcd}=h_{ac}h_{bd}-h_{ad}h_{bc}$, Ricci tensor $R_{ab}=(N-2)h_{ab}$ and scalar curvature $R=(N-1)(N-2)$.
The Levi-Civita connection on the sphere is used to construct ${\rm O}(N)$-covariant derivatives of the pull-backs of tensors on the sphere. For example, given a pull-back of a vector on the sphere, its covariant derivative is $$\label{pull-back connection}
\nabla_\mu v^a \equiv \partial_\mu v^a + \partial_\mu \varphi^b \Gamma_b{}^a{}_c v^c\,.$$ The pull-back covariant derivative $\nabla$ will be used extensively throughout this work. Note that $\partial_\mu \varphi^a$ transform as vectors, while the coordinates $\varphi^a $ are scalars. We also define the square of the covariant derivative $\Box \equiv \delta^{\mu\nu}\nabla_\mu\nabla_\nu$ and its Laplacian $ \Delta=-\Box$.
In order to construct expansions of invariant functionals like such that the nonlinear symmetries are maintained, we will employ a covariant background field expansion. We therefore promote the metric $h_{ab}(\varphi)$ on the sphere to a metric $h_{ab}(\varphi)$ on field space ${\cal M}$ where trivial spacetime indices have been suppressed for brevity. In a similar manner, the Levi-Civita connection, the curvature tensors and the Laplacian can be promoted to ${\cal M}$ as well [@vilkovisky]. It is emphasized that the expansion variable ought to possess well-defined transformation properties both in the background field $\varphi^a$ and in the field $\phi^a$. It would for instance be a particularly hard task to construct ${\rm O}(N)$ covariant functionals in terms of the difference $\phi^a-\varphi^a$ of two points in field space, as it transforms neither like a scalar, nor like a vector under isometries.
For $\phi$ being in a sufficiently small neighborhood of $\varphi$, there exists a unique geodesic in ${\cal M}$ connecting $\varphi$ and $\phi$. We then construct the “exponential map” $$\label{exponential map}
\phi^a = {\rm Exp}_{\varphi} \xi^a = \phi^a(\varphi,\xi)\,.$$ Here $\xi$ is an implicitly defined vector that belongs to the tangent space of ${\cal M}$ at $\varphi$. Also, $\xi$ is a bi-tensor, in the sense that it has definite transformation properties under both $\varphi^a$ and $\phi^a$ transformations and for this reason we shall write it often as $\xi^a=\xi^a(\varphi,\phi)$. It transforms as a vector under the ${\rm O}(N)$ transformations of $\varphi^a$ and as a scalar under those of $\phi^a$. By construction, the norm of $\xi$ equals the distance between $\varphi$ and $\phi$ in ${\cal M}$ [@ketov]. The definite transformation properties make $\xi$ a candidate to parametrize any expansion of functionals like around a background $\varphi$. The correspondence is invertible, therefore in we introduced the notation $\phi^a=\phi^a(\varphi,\xi)$ to be used when $\phi^a$ is understood as a function of $\xi^a$. One always has to keep in mind that the relation between $\xi^a$ and $\phi^a$ is nonlinear.
This background field expansion can easily be applied to any functional $F[\phi]$, $$\label{expansion}
F[\phi]= F[\phi(\varphi,\xi)] = F[\varphi,\xi]\,,$$ where by abuse of notation we indicated with the same symbol the functional when it is a function of the single field $\phi$, or of the couple $\{\varphi,\xi\}$. Even though it may not be evident from the way the r.h.s. of is written, a functional like $F[\phi]$ will never be a genuine function of two fields, but rather a function of the single combination $\phi^a(\varphi,\xi)$, and thus is called “single-field” functional. There may be, however, functionals of $\varphi$ and $\xi$ independently, and those are called “bi-field” functionals. As an example, we define the “cutoff-action” $$\label{cutoff}
\Delta S_k[\varphi,\xi] = \frac{1}{2} \int d^dx ~ \xi^a {\cal R}^k_{ab}(\varphi) \xi^b\,,$$ where ${\cal R}^k_{ab}(\varphi)$ is some symmetric $2$-tensor operator constructed with $\varphi^a$. The cutoff action will play an important role in the next section, where we will also specify its properties. The functional is still invariant under transformations of the background field $\varphi^a$ as well as the field $\phi^a$; however, for general ${\cal R}^k_{ab}(\varphi)$, there is no evident way to recast it as a functional of the single field $\phi^a$ [@reuter_background_works].
We now construct an expansion of functionals like , viewed as functions of the pair $\{\varphi,\xi\}$, in powers of $\xi$. This can be achieved in a fully covariant way by introducing the affine parameter $\lambda\in [0,1]$ that parametrizes the unique geodesic connecting $\varphi$ and $\phi$ [@ketov; @mukhi]. Let $\varphi_\lambda$ be this geodesic with $\varphi_0=\varphi$ and $\varphi_1=\phi$. Let us also define the tangent vector to the geodesic $\xi_\lambda = d\varphi_\lambda/d\lambda$ at the generic point $\varphi_\lambda$. We have that is equivalent to $$\begin{aligned}
\frac{d \xi^a_\lambda}{d\lambda}
&=&
-\Gamma_b{}^a{}_c(\varphi_\lambda) ~ \xi^b_\lambda \xi^c_\lambda\,,
\\
\xi^a_0
&=&
\xi^a\,.\end{aligned}$$ We can rewrite the differential equation satisfied by the tangent vectors by introducing the derivative along the geodesic $\nabla_\lambda\equiv \xi^a_\lambda\nabla_a$ to find that $ \nabla_\lambda \xi^a_\lambda = 0 $. By construction $\nabla_\lambda h_{ab}=0$. Another important property is the relation to the pull-back connection $$\nabla_\lambda \partial_\mu \varphi^a_\lambda = \nabla_\mu \xi^a_\lambda\,.$$ The commutator of the pull-back of the covariant derivative $\nabla_\lambda$ with $\nabla_\mu$ can be computed on the pull-back of a general tangent vector $v^a$ $$[\nabla_{\lambda},\nabla_{\mu}]v^a = R_{cd}{}^a{}_b(\varphi_\lambda) \xi^c\partial_\mu\varphi_\lambda^d v^b\,,$$ and is extensively needed in the covariant expansion. Let us now use the covariant derivative along the geodesic to perform the expansion of . Viewing a general functional $F[\phi]$ as limit $\lambda\to 1$ of $F[\varphi_\lambda]$ we expand the latter in powers of $\lambda$ around $\lambda=0$. One can show that $$\label{expansion2}
F[\phi]
=
\sum_{n \ge 0} \frac{1}{n!} \frac{d^n}{d\lambda^n} \left.F[\varphi_\lambda]\right|_{\lambda=0}
=
\sum_{n \ge 0} \frac{1}{n!} \nabla_\lambda^n \left.F[\varphi_\lambda]\right|_{\lambda=0}\,,$$ where we used the fact that $F[\phi]$ is a scalar function of $\phi$. The r.h.s. yields an expansion in powers of $\xi^a$ of the form $$F[\phi] = \sum_{n \ge 0} F^n_{(a_1,\dots,a_n)}[\varphi]\xi^{a_1}\dots\xi^{a_n}\,.$$ As an example we give the first few terms of the expansion of the microscopic action , $$\begin{aligned}
\label{Caction_expansion}
S[\phi] &=&
S[\varphi]
+ \zeta \int \! d^dx ~ h_{ab}\partial_\mu \varphi^a\nabla^\mu \xi^b
\nonumber
\\
&
+& \frac{\zeta}{2} \int \! d^dx ~ \Bigl(\nabla_\mu \xi^a\nabla^\mu \xi_a + R_{abcd}\partial_\mu\varphi^b\partial^\mu\varphi^c\xi^a\xi^d\Bigr)
\nonumber
\\
&
+& {\cal O}(\xi^3)\,.\end{aligned}$$
Functional RG {#functional rg}
=============
We define the scale-dependent average effective action of the nonlinear ${\rm O}(N)$ sigma model [@Codello:2008qq; @Percacci:2009fh] by the functional integral $$\begin{aligned}
\label{erge1}
e^{-\Gamma_k[\varphi,\bar{\xi}]} &=& \int \! D\xi \, \mu[\varphi] \, e^{-S[\phi]+\frac{\delta\Gamma_k}{\delta \bar{\xi}^a}[\varphi,\bar{\xi}] (\bar{\xi}^a-\xi^a)}\nonumber\\
&&\hskip1.7cm\times\, e^{-\Delta S_k[\varphi,\bar{\xi}-\xi]}\,,\end{aligned}$$ with density of the covariant measure $\mu[\varphi] = {\rm Det}\, h(\varphi)^{1/2}
$. Averages are defined through the path integral on the r.h.s. of , i.e. $\bar{\xi}^a\equiv\left<\xi^a\right> $. Using the relation $\bar{\xi}^a = \xi^a(\varphi,\bar{\phi})$ one can obtain $\bar{\phi}^a$ that plays the role of full average field of the model. The definition differs from the usual definitions of effective action only by the presence of the functional $\Delta S_k$ defined in . This cutoff action is chosen in such a way that the average effective action $\Gamma_k[\varphi,\bar{\xi}]$ interpolates between the classical action at the scale $k=\infty$ and the full effective action at $k=0$. It is chosen to be quadratic in $\xi^a$ with a $\varphi$-dependent kernel ${\cal R}^k_{ab}$ that regularizes the infrared contributions of the fluctuations $\xi$. This kernel will be called “cutoff kernel” from now on. The reference scale $k$ distinguishes between infrared energy scales $k'\lesssim k$ and ultraviolet ones $k'\gtrsim k$ and the cutoff action leads to a scale dependent effective action in the following way: The propagation of infrared modes in the path integral is suppressed, such that only the ultraviolet modes are integrated out and we are left with a scale-dependent average effective action $\Gamma_k$ for the remaining infrared modes. In order to provide for an interpolation between the classical and the effective action, the cutoff-kernel has to fulfill the two conditions $\lim_{k\rightarrow 0}{\cal R}^{k}_{ab}[\varphi]=0$ and $\lim_{k\rightarrow \infty}{\cal R}^{k}_{ab}[\varphi]=\infty$.
Due to the presence of the cutoff term, $\Gamma_k[\varphi,\bar{\xi}]$ is genuinely a bi-field functional [@pawlowski; @reuter_background_works], $$\hat{\Gamma}_k[\varphi,\bar{\phi}] = \Gamma_k[\varphi,\bar{\xi}(\varphi,\bar{\phi})]\,.$$ This observation is important in order to understand that the only way to construct a single field effective action is to set $\varphi=\bar{\phi}$ or equivalently $\bar{\xi}=0$ and consider $$\bar{\Gamma}_k[\bar{\phi}] = \hat{\Gamma}_k[\bar{\phi},\bar{\phi}] = \Gamma_k[\bar{\phi},0]\,.$$ The limit $k\to 0$ of $\bar{\Gamma}_k[\bar{\phi}]$ coincides with the well known effective action introduced by deWitt [@dewitt].
The very useful feature of the definition is that $\Gamma_k[\varphi,\bar{\xi}]$ satisfies a functional renormalization group equation [@wetterich] $$\label{erge2}
k\partial_k \Gamma_k[\varphi,\bar{\xi}] = \frac{1}{2}{\rm Tr}\left(
\frac{k\partial_k{\cal R}_k[\varphi]}
{\Gamma_k^{(0,2)}[\varphi,\bar{\xi}]+{\cal R}_k[\varphi])}\right)\,.$$ The functional differential equation is equivalent to the definition once an initial condition $\Gamma_\Lambda[\varphi,\bar{\xi}]$ for some big UV-scale $\Lambda$ is specified, which accounts for the renormalization prescription and the inclusion of counter terms.
Having an equation like at our disposal, it is possible to investigate properties of the quantum field theory under consideration, without having to explicitly compute the path integral . In particular we can investigate whether or not the theory admits a second order phase transition for some value of its coupling constants. It is well known that, from a renormalization group perspective, the critical behavior is dictated by a the fixed points of the renormalization group flow.
Higher derivative model and beta functions {#higher derivative model and beta functions}
==========================================
It has been stressed in the previous section that the flow equation is non-perturbative in nature such that we can explore nontrivial features of the renormalization group flow. For this purpose we must find an ansatz for the scale-dependent effective action that includes all relevant operators. We use a covariant expansion of the effective action in orders of derivatives and take into account all possible operators containing up to four derivatives. We furthermore consider the split $$\begin{aligned}
\label{split}
\Gamma_k[\varphi,\xi] &=& \Gamma^{\rm s}_k[\phi(\varphi,\xi)]+\Gamma^{\rm b}_k[\varphi,\xi]\,,\end{aligned}$$ where we dropped the overline on the arguments $\xi$ and $\phi$ for notational simplicity. Some comments are in order: We have already stressed that due to the presence of the cutoff action in the definition the functional $\Gamma_k[\varphi,\xi]$ depends on the two fields $\{\varphi,\xi\}$ separately, and not only on the combination $\phi(\varphi,\xi)$. In the split we collected the contribution that can actually be written as functional of the “full” field $\phi$ into $\Gamma^{\rm s}_k[\phi(\varphi,\xi)]$, and parametrized the rest as $\Gamma^{\rm b}_k[\varphi,\xi]$, with $\Gamma^{\rm b}_k[\varphi,0]=0$ [@reuter_background_works].
As an ansatz for $\Gamma^{\rm s}_k[\phi(\varphi,\xi)]$, we use the most general local action up to fourth order in the derivatives [@Percacci:2009fh; @bk] $$\begin{aligned}
\label{Eaction1}
\Gamma^{\rm s}_k[\phi]
&=&
\frac{1}{2} \int d^d x~\Bigl(
\zeta_k h_{ab} {\partial_{\mu} \phi}^a {\partial^{\mu} \phi}^b
+ \alpha_k h_{ab} \Box \phi^a \Box \phi^b
\nonumber\\
&&
+ T_{abcd}(\phi) {\partial_{\mu} \phi}^a {\partial^{\mu} \phi}^b {\partial_{\nu} \phi}^c {\partial^{\nu} \phi}^d\Bigr)\,.\end{aligned}$$ The isometries of the model are respected only if the tensor $T_{abcd}$ is invariant. In the simple case of the ${\rm O}(N)$-model, there exists a unique (up to normalization) invariant $2$-tensor $h_{ab}$ and all higher rank invariant tensors are constructed from $h_{ab}$. According to the symmetries $T_{abcd}=T_{((ab)(cd))}$ that can be deduced trivially from , we see that the most general parametrization of $T_{abcd}$ reads $$\begin{aligned}
\label{Ttensor}
T_{abcd}
&=&
L_{1,k} h_{a(c}h_{d)b} +L_{2,k} h_{ab}h_{cd}\,.\end{aligned}$$ Using this parametrization in , we realize that a total of four couplings has been introduced $\{\zeta_k,\alpha_k,L_{1,k},L_{2,k}\}$. They parametrize the set of operators that we include in our truncation and encode the explicit $k$-dependence of $\Gamma^{\rm s}_k$.
Now we need a consistent ansatz for $\Gamma^{\rm b}_k[\varphi,\xi]$. It is important to dress the $2$-point function of the field $\xi^a$ correctly, because it is the second derivative w.r.t. $\xi$ which determines the flow . We choose $$\begin{aligned}
\label{Eaction2}
\Gamma^{\rm b}_k[\varphi,\xi]
&=&
\Gamma^{\rm s}_k[\phi(\varphi,Z_k^{1/2}\xi)]-\Gamma^{\rm s}_k[\phi(\varphi,\xi)]
\nonumber\\
&&
+Z_k\frac{m_k^2}{2}\int d^dx ~ h_{ab}\xi^a\xi^b\,.\end{aligned}$$ This choice includes a mass term for the fluctuation fields as well as a nontrivial wave function renormalization of these fields $\xi^a \rightarrow Z_k^{1/2} \xi^a$, which takes into account the possibility that the fields $\varphi^a$ and $\xi^a$ may have a different scaling behavior. In a first step beyond the covariant gradient expansion we assume that the wave function renormalization only depends on the scale $k$. Later we shall see that the wave function renormalization enters the flow solely via the anomalous dimension $\eta_k=-k\partial_kZ_k/Z_k$ of the fluctuation field. Contrary to $Z_k$ the square mass $m^2_k$ enters the flow equation directly. It is the most direct manifestation of the fact that $\Gamma_k[\varphi,\xi]$ is a function of the two variables separately. While one could add many other covariant operators to $\Gamma^b_k$, we first want to investigate the effects of these simple structures. Besides the different expansion of the flow equation, the ansatz for $\Gamma^{\rm b}_k[\varphi,\xi]$ represents the main conceptual departure of our computation from the one in [@Percacci:2009fh], where the approximation $Z_k=1$ and $m_k^2=0$ was employed. It will become clear in the following that the fields $\varphi^\alpha$ and $\xi^\alpha$ possess rather different wavefunction renormalizations, therefore making the inclusion of the relative factor $Z_k^{1/2}$ a necessary ingredient for a consistent truncation.
Now we can plug the ansatz into the flow equation . Projecting the r.h.s. of the flow equation onto the same operators that appear in the ansatz for $\Gamma_k$ it is possible to determine the non-perturbative beta functions of the model. In order to proceed we must specify the cutoff kernel appearing in . We want it to be a function of $\varphi$ solely through the Laplacian and it should otherwise be proportional to the metric $h_{ab}$ $$\begin{aligned}
\label{cutoff profile}
{\cal R}^k_{ab}[\varphi] &=& Z_k h_{ab} R_k[\Delta]\,.\end{aligned}$$ The cutoff is specified through the non-negative “profile function” $R_k[z]$ which can be regarded as a momentum-dependent mass. The choice ensures that we are coarse-graining the theory relative to the modes of the covariant Laplacian $\Delta$. In order to serve as a good cutoff the function $R_k[z]$ has to be monotonic in the variable $z$ and such that $R_k[z]\simeq 0$ for $ z \gtrsim k^2$. The wave function renormalization of the $\xi^a$ field has been used in as an overall parametrization. It is convenient to compute the scale derivative of already at this stage. We obtain $$\begin{aligned}
\label{cutoff profile derivative}
k\partial_k{\cal R}^k_{ab}[\varphi] &=& Z_k h_{ab} \left(k\partial_kR_k[\Delta]-\eta R_k[\Delta]\right)\,.\end{aligned}$$ For the sake of convenience we will compute the beta functions of the two sets of couplings $\{\zeta_k,\alpha_k,L_{1,k},L_{2,k}\}$ and $\{Z_k,m^2_k\}$ in two separate steps. We begin the computation of the flow of $\Gamma^{\rm s}_k[\phi(\varphi,\xi)]$ by considering the limit $\xi\to 0$ of . The result is a flow equation of the form $$\begin{aligned}
\label{erge background}
k\partial_k \Gamma^{\rm s}_k[\varphi]
&=&
\frac{1}{2}{\rm Tr}\left(\frac{k\partial_k{\cal R}_k[\varphi]}{
\Gamma_k^{(0,2)}[\varphi,0]+{\cal R}_k[\varphi]}\right)
\nonumber\\
&=&
\frac{1}{2}{\rm Tr}\,\{G_k{}^a{}_a \left(k\partial_kR_k[\Delta]-\eta R_k[\Delta]\right)\}
\,,\end{aligned}$$ where the dependence on $Z_k$ enters via the modified propagator $ G_k{}^{ab}$ which is the inverse of $(Z_k^{-1}\Gamma_k^{(0,2)}[\varphi,0] +{\cal R}_k[\Delta])_{ab}$. It shows how the fluctuations $\xi$ drive the flow of the couplings $\{\zeta_k,\alpha_k,L_{1,k},L_{2,k}\}$. The modified propagator is computed from using (\[Eaction1\],\[Eaction2\]) and reads $$\begin{aligned}
\label{modified propagator}
G_k
&=& \left(P_k[\Delta] +\Sigma \right)^{-1}
\\
P_k[\Delta]
&=&
\alpha_k \Delta^2 + \zeta_k \Delta + m^2 \mathbb{I} + R_k[\Delta]
\\
\Sigma
&=&
B^{\mu\nu} \nabla_\mu \nabla_\nu + C^\mu\nabla_\mu + D\,,\end{aligned}$$ where indices in the tangent space to ${\rm O}(N)$ have been suppressed for brevity. The matrices $B^{\mu\nu}$, $C^\mu$ and $D$ are endomorphisms in the tangent space. In the following we will need the explicit form of two of them: $$\begin{aligned}
\label{operators}
B_{ab}^{\mu\nu}
&=&
2 \delta^{\mu\nu} (\alpha_k R_{acbd}- T_{abcd}) \partial_{\rho} \varphi^c \partial^{\rho} \varphi^d\nonumber\\
&&- 4 T_{acbd} {\partial^{\mu} \varphi}^c {\partial^{\nu} \varphi}^d
\nonumber\\
D_{ab}
&=&
-\zeta_k R_{acbd}\partial_{\rho} \varphi^c \partial^{\rho} \varphi^d - \alpha_k R_{acbd} \Box \varphi^c \Box \varphi^d \nonumber\\
&& + (\alpha_k R_{acde}R_{bfg}{}^{e}+2R_{e(ab)f}T^e{}_{gcd})\nonumber\\
&& ~ \times \partial_{\rho} \varphi^c \partial^{\rho} \varphi^d \partial_{\sigma}\varphi^f \partial^{\sigma}\varphi^g \,.\end{aligned}$$ Each term in $B^{\mu\nu}$ and $D$ consists of at least two derivatives of the field $\varphi^a$. This implies that a Taylor expansion of in $\Sigma$ is possible, because in our truncation ansatz we are interested only in terms up to fourth order in derivatives [@rgmachine]. The tensor $C^\mu$ contains three derivatives of $\varphi^a$ and thus can be ignored in our truncation. Thus the expansion reads $$\label{rhsExp}
G_k \!= \! P_k^{-1}\! - \!P_k^{-1}\Sigma P_k^{-1}\! +\! P_k^{-1}\Sigma P_k^{-1}\Sigma P_k^{-1}\! +{\cal O}(\partial^6)\,.$$ Inserting this expansion into and using the cyclicity of the trace we obtain $$\begin{aligned}
\label{erge background expansion}
k\partial_k \Gamma^{\rm s}_k[\varphi]
&=&
\frac{1}{2}{\rm Tr}\,f_1[\Delta]
-\frac{1}{2}{\rm Tr}\,\Sigma f_2[\Delta]
\nonumber\\
&&
+\frac{1}{2}{\rm Tr}\,\Sigma^2 f_3[\Delta]
+{\cal O}(\partial^6)
\,,\end{aligned}$$ where we defined the functions $$\begin{aligned}
f_l[z]\equiv \frac{k\partial_kR_k[z]-\eta R_k[z]}{P_k[z]^l}\,.\end{aligned}$$ In we also used the fact that we can commute $\Sigma$ and $P_k[\Delta]$ in the third term in the expansion, because their commutator leads to terms of order ${\cal O}(\partial^6)$. The traces appearing in are computed using off-diagonal heat kernel methods [@off-diagonal]. To outline the general procedure briefly we consider the traces $$\label{general trace}
{\rm Tr}\,(\nabla_{\mu_1}\dots\nabla_{\mu_r} f[\Delta])\,$$ which transform as tensors under isometries. We are interested in the particular cases $0\leq r\leq 4$ and $f[\Delta]=f_l[\Delta]$ for some $l=1,2,3$. Introducing the inverse Laplace transform ${\cal L}^{-1}[f](s) $ of $f[z]$, we rewrite as $$\label{heat kernel}
\int_0^{\infty} ds\,{\cal L}^{-1}[f](s) \, {\rm Tr}\,(\nabla_{\mu_1}\dots\nabla_{\mu_r} e^{-s \Delta})\,.$$ The expansion of the trace $ {\rm Tr}\,(\nabla_{\mu_1}\dots\nabla_{\mu_r} e^{-s \Delta})$ is obtained from the off-diagonal heat kernel expansion (the case $r=0$ yields the trace of the heat kernel itself). This expansion is an asymptotic small-$s$ expansion that corresponds, for dimensional and covariance reasons, to an expansion in powers of the curvature and covariant derivative. Thus we have $$\begin{aligned}
\label{heat kernel expansion}
{\rm Tr}\,(\nabla_{\mu_1}\dots\nabla_{\mu_r} e^{-s \Delta})
=
\sum_{n=0}^\infty \frac{B_{\mu_1\dots\mu_r ,n}}{(4\pi s)^{d/2}}s^{\frac{2n-[r]}{2}}\,,\end{aligned}$$ where $[r]=r$ if $r$ is even and $[r]=r-1$ if $r$ is odd. The coefficients $B_{\mu_1\dots\mu_r,n} $ contain a number of powers of the derivatives of the field that increases with $n$, thus only a finite number of them is needed to compute the traces with ${\cal O}(\partial^4)$ accuracy [@off-diagonal]. The final step in evaluating is the s-integration. We define $$\begin{aligned}
Q_{n,l} = \frac{1}{(4\pi)^{d/2}}\int_0^{\infty}ds\, s^{-n} {\cal L}^{-1}[f_l](s)\,,\end{aligned}$$ which we will denote as “$Q$-functionals” that can be rewritten (for positive $n$) as Mellin transforms of $f_l(z)$ $$\begin{aligned}
\label{Qfunctionals}
Q_{n,l} = \frac{1}{(4\pi)^{d/2}\Gamma[n]}\int_0^{\infty} dz\, z^{n-1} f_l[z].\end{aligned}$$ In the heat kernel expansion, the $Q$-functionals play a similar role that is played by the regularized Feynman diagrams in perturbation theory. With the help of the heat kernel method we now expand the right hand side in the flow equation and obtain $$\begin{aligned}
\label{erge background expansion2}
k\partial_k\Gamma_k^{\rm s}[\varphi]
=&
\frac{1}{2}{\rm tr}\! \int \! d^d x \Bigl\{
\frac{1}{12} Q_{\frac{d}{2}-2,1} \left[\nabla_\mu,\nabla_\nu\right]^2
\nonumber\\
&+ \frac{1}{2} Q_{\frac{d}{2}+1,2} B^{\mu}{}_{\mu}
- \frac{1}{2} Q_{\frac{d}{2},2} B^{\mu\nu}\left[\nabla_\mu,\nabla_\nu\right]
\nonumber\\
&+ \frac{1}{2}Q_{\frac{d}{2}+2,3}\Bigl(B^{(\mu\nu)}B_{\mu\nu}+\frac{1}{2}(B^{\mu}{}_{\mu})^2\Bigr)
\\
&- Q_{\frac{d}{2},2} D - Q_{\frac{d}{2}+1,3}B^{\mu}{}_{\mu} D
+ Q_{\frac{d}{2},3} D^2 \Bigr\}\,\nonumber\end{aligned}$$ whereby the tensor fields $B$ and $D$ are given in . The beta functions for $\{\zeta_k,\alpha_k,L_{1,k},L_{2,k}\}$ are defined as their log-$k$-derivatives and are denoted by $\{\beta_\zeta,\beta_\alpha,\beta_{L_1},\beta_{L_2}\}$. We extract them by evaluating both sides of the flow equation at $\xi = 0$. With the left hand side reads $$\begin{aligned}
\label{lhs flow background}
k\partial_k \Gamma^{\rm s}_k[\varphi]
&=&
\frac{1}{2} \int d^d x~\Bigl(
\beta_\zeta {\partial_{\mu} \varphi}^a {\partial^{\mu} \varphi}_a
+ \beta_\alpha \Box \varphi^a \Box \varphi_a
\nonumber\\
&
& \hskip-10mm +\,\beta_{L_1} ({\partial_{\mu} \varphi}^a {\partial^{\mu} \varphi}^b)^2 + \beta_{L_2} ({\partial_{\mu} \varphi}^a {\partial^{\mu} \varphi}_a)^2
\Bigr)\,,\end{aligned}$$ and a comparison with the right hand side in yields the beta functions $$\begin{aligned}
\label{beta functions}
&\beta_\zeta
=\zeta_k (N-2) Q_{\frac{d}{2},2} +(N-2) d \alpha_k Q_{\frac{d}{2}+1,2}
\nonumber\\
&~~ + L_{1,k}(N+d)Q_{\frac{d}{2}+1,2}
- L_{2,k}((N-1)d+2)Q_{\frac{d}{2}+1,2}
\nonumber\\
&\beta_\alpha
=(N-2) Q_{\frac{d}{2},2} \alpha_k
\,.\end{aligned}$$ The remaining beta functions $\beta_{L_1}$ and $\beta_{L_2}$ are quite long and given in the appendix . It suffices to know that $\beta_{L_1},\,\beta_{L_2}$ are quadratic polynomials of $L_1$ and $L_2$ and that the coefficients of these polynomials depend nontrivially on both the parameters $N$ and $d$, and on the other two couplings. The combined limit $\{d=4,\zeta_k= 0,\eta=0,m^2_k=0\}$ of $\{\beta_{L_1},\beta_{L_2}\}$ is known to yield a universal result which can be compared with the results of chiral perturbation theory in the case $ S^3\simeq {\rm SU(2)}$ [@Percacci:2009fh; @hasenfratz]. We checked that we can reproduce the perturbative results in this limit. The results of [@Percacci:2009fh] are recovered fully in the limit $d=4$ and for general $S^{N-1}$, if a chiral perturbative expansion is performed to $1$-loop.
Let us finally outline the computation of the flow of $Z_k$ and $m^2_k$. The simplest setting to perform this is a vertex expansion of the flow in powers of the field $\xi^a$. We first notice that for a constant background field $\varphi^a_c$ the ansatz for the effective action reduces to $$\begin{aligned}
\label{fluctuations action}
\Gamma_k[\varphi_c,\xi]
=&
\frac{Z_k}{2}\int d^dx \Bigl\{\zeta_k h_{ab} \nabla_\mu\xi^a\nabla^\mu\xi^b + \alpha_k h_{ab} \Box\xi^a\Box\xi^b
\nonumber\\
&
+ m_k^2 h_{ab} \xi^a \xi^b
+\frac{1}{3} \zeta_k Z_k R_{abcd}\xi^a \xi^d {\nabla_{\mu} \xi}^c {\nabla^{\mu} \xi}^b
\nonumber\\
&
+\frac{4}{3} \alpha_k Z_k R_{abcd} \xi^a {\nabla_{\mu} \xi}^b {\nabla^{\mu} \xi}^d \Box \xi^c
\nonumber\\
&
+ \frac{1}{3} \alpha_k Z_k R_{abcd} \xi^a \Box \xi^b \Box \xi^c \xi^d \\
&
+ Z_k T_{abcd} {\nabla_{\mu} \xi}^a {\nabla^{\mu} \xi}^b {\nabla_{\nu} \xi}^c {\nabla^{\nu} \xi}^d
\Bigr\} + {\cal O}(\xi^6)\,.\nonumber\end{aligned}$$ In this particular limit the pull-back connection becomes trivial: $\nabla_\mu=\partial_\mu$ and $\Box=\partial^2$. This observation is particularly useful, since we can now easily perform the computations in momentum space. We define $ \xi^a(x) = \int d^dq\, e^{\imath q x} \xi^a_q$ and obtain from the $2$-point function for incoming momentum $p^\mu$: $$\begin{aligned}
\Gamma^{(0,2)}_k[\varphi_c,0]_{p,-p}
&=
Z_k (\alpha_k p^4 + \zeta_k p^2 +m_k^2)\,.\end{aligned}$$ We also compute its scale derivative $$\begin{aligned}
\label{two point function flow}
k\partial_k\Gamma^{(0,2)}_k[\varphi_c,0]_{p,-p}
&=&
Z_k \Bigl((\beta_\alpha -\eta \alpha_k) p^4\\
&&\hskip-.5cm +\, (\beta_\zeta-\eta \zeta_k) p^2
+(\beta_{m^2} -\eta m_k^2) \Bigr)\,.\nonumber\end{aligned}$$ On the other hand, the quantity $k\partial_k\Gamma^{(0,2)}_k[\varphi_c,0]_{p,-p}$ can be computed from by applying two functional derivatives w.r.t. $\xi^a$, taking the limit $\varphi^a=\varphi^a_c={\rm const.}$ and transforming to momentum space. After these manipulations, the flow equation reduces to $$\begin{aligned}
\label{two point function flow rhs}
k\partial_k\Gamma_k^{(0,2)}[\varphi_c,0]_{p,-p}&&\\
&&\hskip-2cm =
-\frac{1}{2 Z_k} {\rm Tr} f_2(q^2) \Gamma_k^{(0,4)}[\varphi_c,0]_{p,-p,q,-q}\,.\nonumber\end{aligned}$$ The momentum space $4$-point vertex function $\Gamma_k^{(0,4)}[\varphi_c,0]$ is obtained from and has to be traced over two of its four indices, while the $3$-point function vanishes for $\varphi^a={\rm const.}$ therefore playing no role in our computation. The trace that appears in consists of an internal trace on the tangent space of the model, that involves two of the four indices of the $4$-vertex, and a momentum space integral $\int d^dq/(2\pi)^d$. The final result is a long expression that depends solely on $p^2$. Comparing the power $p^n$ with $n=0,2,4$ of with those of and dividing both sides by $Z_k$, we can determine the coefficients $$\begin{aligned}
\label{fluctuations system}
\beta_\alpha -\eta \alpha_k
&=
\frac {1}{3}(N-2)\alpha_k Q_{\frac{d}{2},2}
\nonumber \\
\beta_\zeta\,-\eta \zeta_k
\,&=
\frac{1}{3} (N-2) \zeta_k Q_{\frac{d}{2},2}
-\Bigl((dN-d+2) L_{2,k}
\nonumber \\
&~~~ +(N+d) L_{1,k}
- (N-2) d \alpha_k\Bigr) Q_{\frac{d}{2}+1,2}
\nonumber \\
\beta_{m^2} -\eta m_k^2
\,&=
\frac{1}{12}(N-2)d(d+2) \alpha_k Q_{\frac{d}{2}+2,2}
\nonumber \\
& ~~~ +\frac{1}{6}(N-2)d \zeta_k Q_{\frac{d}{2}+1,2}\,.\end{aligned}$$ As anticipated, there is no explicit dependence on $Z_k$, because it is a redundant parameter.
One interesting feature of the method arises at this point: Using , we can solve the system of equations in terms of the two unknown quantities $\{\eta,\beta_{m^2}\}$. For a solution to exist, one equation of must be redundant and it is a nontrivial check of our computation, at this stage, that this actually holds true. The final result for the anomalous scaling reads $$\begin{aligned}
\label{anomalous scaling}
&\eta=\frac{2}{3}(N-2) Q_{\frac{d}{2},2}\,.\end{aligned}$$
${\rm O}(N)$ phase diagram {#on phase transition}
==========================
We will now analyze in more detail the structure of the $\beta$-functions and the resulting phase diagram. For this purpose we focus on three spacetime dimensions as a particularly interesting case which has been intensively studied [@Pelissetto; @Ballesteros; @Butera]. Concerning the two dimensional case, we just want to mention that our computation reproduces the well-known statement that the theory has no nontrivial fixed point, but rather is asymptotically free [@Polyakov].
In order to evaluate the Q-functionals and hence the explicit running of the couplings in three dimensions, we have to choose a specific regulator that fulfills the requirements described in section \[functional rg\]. We decided to choose a modified “optimized cutoff” [@OptReg]: $$\begin{aligned}
\label{optimized}
R_k[z] = \left(\zeta_k (k^2-z)+ \alpha_k (k^4-z^2)\right)\Theta(k^2-z) \end{aligned}$$ Note that the $k$-subscript of the couplings will be suppressed in the following. This choice of regulator enables us to calculate the Q-functionals explicitly, $$\begin{aligned}
\label{Q-optimized}
Q_{n,l} &=& \frac{k^{2n+2}}{(4\pi)^{d/2}\,\Gamma(n)}\,\Bigl(
\frac{ (2n+2-\eta+\partial_t)\zeta }{n(n+1)(\zeta k^2 + \alpha k^4 + m^2)^l}\nonumber\\
&&\hskip10mm+\frac{2 k^2 (2 n + 4 -
\eta+\partial_t)\alpha}{n(n+2)(\zeta k^2 + \alpha k^4 + m^2)^l} \Bigr) \,,\end{aligned}$$ but it renders the system of differential equations rather involved, since the derivatives $\partial_t \alpha\equiv k\partial_k\alpha = \beta_{\alpha}$ and $\partial_t \zeta = \beta_{\zeta}$ also appear on the r.h.s. of the flow equation. We note that the $Q$-functionals possess a threshold-like structure due to the presence of the mass $m^2$. Further, they are linear in the anomalous scaling $\eta$ and in the beta functions.
Under the condition , the systems and contain the beta functions of the theory in an implicit form. In order to determine their explicit form it is necessary to solve together the two systems in terms of the quantities $\{\beta_\zeta,\beta_\alpha,\beta_{L_1},\beta_{L_2},\eta,\beta_{m^2}\}$. Fortunately, the joint system is linear in these quantities, because the $Q$-functionals are. As a final step we rewrite the result in terms of the dimensionless couplings $\tilde{\zeta} = k^{2-d} \zeta$, $\tilde{\alpha}=k^{4-d}\alpha$, $\tilde{L}_1=k^{4-d}L_1$, $\tilde{L}_2=k^{4-d}L_2$ and $\tilde{m}^2=k^{-d}m^2$. and obtain the beta functions $\{\beta_{\tilde{\zeta}},\beta_{\tilde{\alpha}},\beta_{\tilde{L}_1},
\beta_{\tilde{L}_2},\beta_{\tilde{m}^2}\}$, which are involved rational functions and hence not given here explicitly.
Now we are ready to study the phase diagrams and the critical properties that arise from these flow equations. We will proceed in a systematic way, by including more and more operators in our truncation. The simplest truncation that only contains the coupling $\zeta$ was already studied in [@Codello:2008qq]. Their work points to the existence of a nontrivial fixed point in dimensions larger than two and hence to the possibility of non-perturbative renormalizability.
We want to add the coupling $\alpha$ and the related fourth-order operator. The corresponding renormalization group flow is depicted in Figure \[alphaflow\] (for the case $N=3$) and confirms the nontrivial fixed point found in the simpler truncation. The critical couplings are $\tilde{\zeta}^* = 16(N-2)/(45\pi^2)$ and $\tilde{\alpha}^*=0$.
![The flow of the couplings for the truncation with two couplings $\alpha$ and $\zeta$ for $N=3$. The arrows point toward the ultraviolet. The removed region lies beyond an unphysical singularity which is introduced by the choice of the cutoff. []{data-label="alphaflow"}](N3_ZetaAlpha.pdf){width="48.00000%"}
Note that the arrows of the flow point into the direction of increasing $k$, i.e. towards the ultraviolet. It is interesting to note that the coupling $\alpha$ belongs to an IR-irrelevant operator and vanishes at the fixed point. In fact, already the simple structure of $\beta_{\alpha}$ that is obtained from reveals that $\alpha$ has to vanish for every possible fixed point, as the flow of the dimensionless coupling $\tilde{\alpha}$ reads in $d=3$: $$\begin{aligned}
\beta_{\tilde{\alpha}} = \tilde{\alpha} + (N-2)~Q_{\frac{3}{2},2}~\tilde{\alpha}\end{aligned}$$ Since $Q_{\frac{3}{2},2}$ is strictly positive for any reasonable regulator, the only possible fixed point value is $\tilde{\alpha}=0$. This statement remains true when we include the couplings $L_1,L_2$. Further, it is a general feature of the system that at any FP for which $\tilde{\alpha}=0$, the critical $\alpha$ direction in phase space is decoupled from the others and therefore cannot contribute significantly to the critical properties of the other couplings. There is also a fixed point for $\lambda = 1/\alpha = 0$, but this is a trivial one whose critical exponents are equal the canonical mass dimensions of the operators. It is the three-dimensional analogue of the fixed point in four dimensions which is discussed in [@Percacci:2009fh].
The result $\tilde{\alpha}^*=0$ agrees with an alternative computation of the effective action of the nonlinear ${\rm O}(N)$-model up to fourth order, which is presented in [@Chan], and in which a term $\propto \partial^2\phi\partial^2\phi$ is not generated, either. However, it is very likely that an extension of the truncation to the sixth order in derivatives and an inclusion of operators like e.g. $\Box \phi_a \Box \phi^a \partial_{\mu} \phi^b
\partial^{\mu}\phi_b$ will affect the running of $\alpha$ and shift the position of the fixed point. This discussion will be relevant when we compare our results with the renormalization flows obtained by the Monte Carlo Renormalization Group (MCRG), that will be studied in an upcoming work [@upcoming]. At this point we just mention the important fact that both non-perturbative methods agree on the structure of the flow diagram, as it is depicted in Figure \[alphaflow\], and hence on the existence of the nontrivial fixed point and on the number of relevant directions at this fixed point. But they differ in the position of the fixed point, which has a positive $\tilde{\alpha}^*$ in case of the MCRG. This is in fact not surprising, since the position of the fixed point is not universal and because in the lattice calculations the higher order operators affect the flow, even though the applied RG procedure keeps track only of a truncated operator space, cf. [@upcoming] for more details.
Since $\alpha$ is not generated in this truncation, the system of two couplings effectively reduces to the one-parameter truncation that was investigated in detail in [@Codello:2008qq]. While the critical value $\tilde{\zeta}^* = 16(N-2)/(45\pi^2)$ depends linearly on $N$, the critical exponent $\tfrac{d}{d\tilde{\zeta}}\beta_{\tilde{\zeta}}|_{\tilde{\zeta}^*}$ is independent of $N$: it is $-16/15$ for all $N$. In this sense our computation is reminiscent of the one-loop large-$N$ calculations [@Zinn], apart from the small deviation of our critical exponent from the large-$N$ value $-1$. It is interesting to note that the running of $\zeta$ in the one-parameter truncation of the nonlinear model agrees exactly with the running that can be derived by means of the FRG if one regards the nonlinear model as the limit of a linear model with infinitely steep potential [@Codello:2008qq]. However, while already a simple truncation of the linear model reproduces reasonable results for the $N$-dependent critical exponents of the ${\rm O}(N)$-universality class, this $N$-dependence is apparently suppressed by taking the limit.
Let us briefly recall what is known about the critical properties of the nonlinear ${\rm O}(N)$ model in three dimensions: The second order phase transition of the model is described by the nontrivial fixed point and its IR-relevant direction. The scaling of the critical coupling is directly related to the scaling of the correlation length: $$\begin{aligned}
\nu = -\frac{1}{\Theta^*},\end{aligned}$$ where $\Theta^*$ denotes the eigenvalue of the stability matrix evaluated at the fixed point, which corresponds to the relevant direction. The critical properties of linear and nonlinear ${\rm O}(N)$ models are intensively studied, see [@Zinn; @Pelissetto; @Ballesteros; @Butera], and it is generally believed that both theories belong to the same universality class. While there are Monte Carlo simulations of the nonlinear model which manifestly implement the nonlinearity of the target space and confirm this equivalence [@Ballesteros], the analytic calculations rely either on an explicit breaking of the symmetry and/or on an embedding of the target space in a linear space. An important motivation for the analytic approach presented here is to implement the isometries of the target space in every step of the calculation.
In order to become sensitive to the $N$-dependence of the critical exponent $\nu$, we must include higher order operators in our manifestly covariant flow equation. This agrees with the finding of [@Codello:2008qq] that the derived $\beta$-functions of sigma models with different symmetric target spaces coincide in a simple truncation. They differ only if one takes higher order operators into account, whose number and structure depends strongly on the specific type of model.
Hence we increase the truncation and include the operator $L_1 (h_{ab}\partial_{\mu} \phi^a \partial_{\nu} \phi^b)^2$. The resulting flow of the couplings $\tilde{\zeta}$ and $\tilde{L}_1$ of the ${\rm O}(3)$ model is depicted in Figure \[Lflow\], where we have set the irrelevant coupling $\tilde{\alpha}$ to $\tilde{\alpha}^*=0$.
![The flow of the couplings $\tilde{\zeta}$ and $\tilde{L}_1$ towards the UV for $N=3$. We have set $\tilde{\alpha}=\tilde{\alpha}^*=0$. The fixed point with one irrelevant direction is at $\tilde{\zeta}^*=0.059$ and $\tilde{L}_1^*=-0.013$.[]{data-label="Lflow"}](N3_ZetaL1){width="48.00000%"}
It contains the nontrivial fixed point which was already discovered in the leading order truncation and which has only one relevant direction. The fixed point exists for all $N$, and while the critical value of $\tilde{L}_1$ is almost independent of $N$ and close to $-0.013$, the fixed point value $\tilde{\zeta}^*$ is an involved expression in $N$ which for $N=3$ attains the values $0.059$. It increases with $N$ such that it approaches a linear function with a slope of roughly $0.036$ for large $N$. Similarly as for the leading order truncation the flow diagram has qualitatively the same structure as the diagram obtained by the MCRG [@upcoming]. The operator corresponding to $L_1$ is irrelevant in both approaches and its critical value $\tilde{L}_1^*$ is small in comparison to $\tilde{\zeta}^*$.
Actually there are additional fixed points in the truncation with coupling $\zeta,\alpha$ and $L_1$, some with negative $\tilde{\zeta}^*$ and one with quite large values of $\tilde{L}_1^*$ and $\tilde{\zeta}^*$. These could be artifacts of our choice of the cutoff functions which may develop singularities for negative couplings. We could not relate the additional fixed points to known critical properties of sigma models and their physical relevance remains unclear. We will therefore focus on the fixed point depicted in Figures \[alphaflow\] and \[Lflow\].
As anticipated, the inclusion of fourth order operators renders the exponent $\nu$ sensitive to the dimension of the target space. The $N$-dependence of the exponent is depicted in Figure \[nuComp\], while the numerical values are given in Table \[nuTab\].
![The critical exponent $\nu$ as function of $N$, computed in the truncation $\{\zeta,\alpha,L_1\}$. Depicted are the results of various approximations in comparison with average values from the literature.[]{data-label="nuComp"}](nuComparison){width="45.00000%"}
The values in the third row denoted by “full system” refer to calculations with the truncation $\{\zeta, \alpha, L_1\}$, in which the anomalous scaling $\eta$ of the fluctuation fields $\xi$ is taken into account. If one sets $Z\equiv 1$, one obtains the values in the second row of Table \[nuTab\]. If in addition one neglects the $k$-derivative of the couplings in $k\partial_k{\cal R}^k$ on the right hand side of the flow equation, that amounts to an adiabatic approximation, then we obtain the values in the first row of Table \[nuTab\]. At the fixed point the $k$-derivative of the couplings vanish such that the approximation with $Z=1$ and the cruder adiabatic approximation yield the same fixed point couplings.
[|c||c|c|c|c|c|c|]{} $N$ & 3 & 4 & 6 & 8 & 10 & 20\
adiabatic approx. & 0.824 & 0.924 & 0.969 & 0.981 & 0.987 & 0.995\
with $Z=1$ & 0.654 & 0.756 & 0.802 & 0.815 & 0.820 & 0.828\
full system & 0.704 & 0.833 & 0.895 & 0.912 & 0.920 & 0.931\
literature & 0.710 & 0.747 & 0.790 & 0.830 & 0.863 & 0.934\
Since $\nu(N)$ is a rather involved and long expression, we tabulated only some selected values in the rows of Table \[nuTab\]. For a comparison we added the last row which contains the values taken from the vast literature about the critical properties of the ${\rm O}(N)$ universality class. For $N=3$ and $N=4$ the values are taken from the review [@Pelissetto], which contains the results of many independent computations of which we took the non-biased mean values. For $N>5$ we took the mean values of the results in [@Butera; @Kleinert; @Antonenko], which have been obtained by a high-temperature expansion, a strong-coupling expansion and six-loop RG expansion including a Pade-Borel resummation. The corresponding values deviate from each other by less than two percent.
The three truncations of the flow equation presented in this work yield a critical exponent $\nu$ whose $N$-dependence roughly agrees with the results in the literature. The values obtained in the adiabatic approximation deviate considerably from the references values for small $N$, but show the correct large-$N$ asymptotic. If one takes the running of the couplings in $k\partial_k{\cal R}^k$ into account, the results for small $N$ improve significantly, especially if one neglects the wave function renormalization. In this case, however, $\nu(N)$ approaches for large $N$ the value $5/6$ instead of the correct value $1$. If in addition one includes the wave function renormalization, one obtains a critical exponent $\nu(N)$ which is closer to the reference value than the adiabatic result and whose asymptotic behavior is better behaved as in the approximation with $Z=1$. The deviation from the best-known value is maximal for $N=5$, where it is $14\%$, and the asymptotic value is $15/16$ instead of $1$. This is in fact the value we found in the reduced truncation with just one coupling and agreement originates from $\lim_{N\to\infty} (\tilde{L}^*_1/\tilde{\zeta}^*)=0$.
It is not clear to what extent the inclusion of the wave function renormalization improves the situation: on one hand a running $Z$ improves the asymptotic of $\nu(N)$ for large $N$ and on the other hand $Z=1$ yields more accurate results for small $N$. We included the wave function renormalization mainly for conceptual reasons since the background and fluctuation fields are treated differently in the FRG formalism and hence may possess different renormalization properties. This is taken into account by admitting a running of $Z$.
So far we did not consider the last contribution to the functional $\Gamma_k^b[\varphi,\xi]$ in containing the mass parameter $m_k^2$. We included it in order to examine if terms that go beyond the ansatz of a “single-field functional” can improve the accuracy of the results. If we consider a truncation with couplings $\{\zeta,\alpha,L_1,m^2\}$ we find the same fixed point as before with slightly modified critical values and a positive mass parameter. However, the results for the critical exponent get worse rather than better and are close to the values of the adiabatic approximation. This is a surprising finding and certainly requires a better understanding. For this purpose, the effects of higher-order terms in $\Gamma_k^b[\varphi,\xi]$ ought to be studied.
Let us finally include the remaining operator with four derivatives $L_2(h_{ab}\partial_{\mu}\phi^a\partial^{\mu}\phi^b)^2$. Although it is of the same order as the operators with couplings $L_1$ and $\alpha$, it changes the flow such that there is no nontrivial fixed point for the system with couplings $\{\zeta,\alpha,L_1,L_2\}$. This statement holds true for all possible modifications of the flow, i.e. in the adiabatic approximation, in the approximation with $Z=1$ and even if we include a mass parameter. In fact, already in the cruder truncation $\{\zeta,\alpha,L_2\}$ there is no nontrivial fixed point and it seems as if the renormalization of the coupling $L_2$ is not well-balanced.
One may wonder why the operator corresponding to $L_2$ destabilizes the renormalization group flow. In the computation of the full effective action, the renormalization of an operator of a given order is always affected by operators of higher order. These contributions are lost in a truncation in which the higher order operators are neglected. In the present case the beta functions of $L_1$ and $L_2$ are quadratic functions, see in the appendix, and the coefficients of the polynomials must be fine-tuned such that both beta functions vanish. We checked that the beta function of $L_2$, when evaluated at the fixed point for the subsystem consisting of all other couplings, is nearly zero. Thus we expect that the inclusion of higher order terms will slightly modify the flow in a way that one recovers the fixed point and the information about the phase transition of the ${\rm O}(N)$ model, that we already detected in the truncation $\{\zeta,\alpha,L_1\}$.
However, there could be more subtle explanations why the flow of the operator $(h_{ab}\partial_{\mu}\phi^a\partial^{\mu}\phi^b)^2$ does not lead to a stable fixed point. We only want to mention two possibilities: There were arguments brought forward recently [@bologna] that the regularization procedure of the functional renormalization group given by the introduction of $\Delta S_k$ may naturally require a corresponding modification of the path integral measure, which in turn provides an additional term to the flow equation of the average effective action. While this term yields only a renormalization of the vacuum energy in theories with linearly realized symmetries, it can affect the renormalization of nontrivial operators in theories whose path integral measure is field-dependent.
The second possibility is that, in order to find a stable fixed point for the full system, one has to enlarge the truncation as dictated by hidden symmetries involving the background and the fluctuation fields. To this day the background field method is the most effective way to deal with nontrivial field-space geometries in the framework of the FRG. Nevertheless, further studies maybe needed to understand better which truncations in terms of background and fluctuation fields ought to be chosen in order to maintain the full reparametrization invariance of the theory. An ansatz that is based on the so-called Nielsen identities was presented recently in the context of gravity [@pawlowski] and it could be interesting to apply this approach to the nonlinear sigma model.
#### A smooth $\alpha$-independent cutoff. {#a-smooth-alpha-independent-cutoff. .unnumbered}
The destabilization of the flow induced by the $L_2$-term does not seem to depend on a specific choice of regulator. For example, we used the alternative regulator $R_k[z]= k^{d+2}/z$ and confirmed the existence of nontrivial fixed points as well as the $N$-dependence of $\nu$ in the truncation $\{\zeta,\alpha,L_1\}$, but the fixed point still disappears if one includes $L_2$. It was pointed out in [@Morris:2005ck] that there are concerns about the application of an optimized cutoff like in the study of the linear sigma model beyond the local potential approximation. This motivates us to briefly discuss our system using a smooth cutoff like the exponential one $$R_k[z]= A Z_k \frac{z}{{\rm e}^{z/k^{2}}-1}\,,$$ where $A$ is an external parameter that can be tuned following the so-called “principle of minimum sensitivity” [@Canet:2003qd]. For a generic choice of $A$ it is possible to evaluate numerically the $Q$-functionals and therefore determine the beta functions of the theory with arbitrary precision. Using the exponential cutoff, we studied the complete system of beta functions at $N=3,4,5,10,100,1000$, including $L_2$, in the parameter range $A\in (0,10]$, but could not stabilize the nontrivial fixed point. Resorting to the approximation $L_2=0$, a nontrivial FP with one attractive direction appears as a function of the two parameters. The numerical results for the scaling exponent $\nu$ of the correlation length are in qualitative agreement with the optimized cutoff results for every value bigger than $A\simeq 0.18$, that we analyzed in the parameter space with accuracy $\Delta A=0.01$. Since $\nu$ is computed as a numerical function of $A$ in the form $\nu(A)$, one could apply the principle of minimum sensitivity, i.e. finding the best value for $A$ as a local minimum $A^\star$ of $\nu(A)$. However, the result seems to indicate that such a minimum does not exists, since the function $\nu(A)$ appears to be (slowly) monotonically increasing to our accuracy.
Conclusion
==========
In this article we applied the functional renormalization technique based on a scale dependent effective action to investigate the renormalization group flow of the nonlinear ${\rm O}(N)$-model. The flow is formulated in a manifestly reparametrization invariant way, so that the results do neither depend on any specific choice of coordinates on the target sphere, nor on an implicit embedding of the nonlinear model into the linear one. Since the symmetries of the theory are realized nonlinearly, we adopted a geometric formulation where a background (base-point) dependence is introduced in order to maintain the covariance of the model. The background field is used to construct a quadratic infrared cutoff term for the fluctuations, whose purpose is to allow us to effectively integrate out the ultraviolet modes while simultaneously respecting the symmetries of the model. The main achievement is the construction of a scale-dependent effective action, that is ${\rm O}(N)$ invariant for both the transformations of the background and the quantum field. The consistency of our formalism was underlined by the appearance of nontrivial relations between the renormalization flow of background and fluctuation operators.
The model has been studied using a truncation ansatz that includes all possible covariant operators up to fourth order in the derivatives. The beta functions of the theory are identified with the derivatives of the couplings w.r.t. the logarithm of the RG scale. They provide important informations about the phase diagram of the model and we concentrated our discussion in particular on the fixed points of the renormalization flow, because they are associated to second order phase transitions. The scaling of the correlation length is a universal property and is directly related to the critical exponent corresponding to the relevant direction. Our explicit calculations focused on three dimensions, in which the critical exponents of the model are well-studied and provide a rich literature that can serve as a reference for comparison.
Our investigation did not immediately point to the existence of such universal behavior, because we did not find a suitable fixed point for the full fourth order system. However, we discovered that in the restricted subspace of couplings, where one of the higher derivative couplings is set to zero ($L_{2}=0$), a fixed point for all $N$ emerges. It exhibits one relevant direction, which is already present in an one-parameter truncation, and it is such that the inclusion of further couplings ($L_1$ and $\alpha$) only adds irrelevant directions. The critical value of the coupling $\alpha$ is identically zero in a fourth-order expansion. However, the one-parameter truncation is not sensitive to the $N$-dependence of the critical exponent $\nu$, which requires higher-order operators. The results we obtain for $\nu$ in the truncation $\{\zeta,\alpha,L_1\}$ agree qualitatively with the pre-existing literature, but show some numerical difference that is likely due to the limited truncation ansatz. We tested the presence of this fixed point against various approximations and choices of the coarse-graining scheme, to find that it is a very stable result.
The renormalization properties of the nonlinear ${\rm O}(N)$ model have also been studied by means of the Monte Carlo Renormalization Group and will be presented in detail elsewhere [@upcoming], where a discretized version of the action is considered as lattice action. The quantitative comparison of the results of FRG and MCRG is a delicate topic that has to be addressed with care. In fact, one must bear in mind that our truncation ansatz is *not* a bare action [@Ellwanger:1997tp]. The nature of the “mass” regularization of FRG, as opposed to that of the lattice which is mass-independent, is expected to produce a nontrivial relation between the renormalized couplings of the two methods [@Manrique:2008zw]. However, we do expect the phase diagrams of FRG and MCRG to share the same qualitative properties (namely the existence of a fixed point with one UV-attractive direction). The MCRG flows that are computed for the fourth-order expansion of the model are (apart from the coupling $L_2$) in good qualitative agreement with the results presented here [@upcoming].
As mentioned, a nontrivial fixed point appears in our calculations only if one coupling, $L_{2}$, is neglected. We strongly believe that this is not a pathology of the computation, but rather it is due to the limited truncation considered and the quadratic structure of the computed beta functions. However, we also admit the possibility that the operator corresponding to the coupling $L_{2}$ may have a special role in the nature of this phase transition. Therefore further investigations are required in order to obtain a deeper understanding.
Acknowledgments
===============
We would like to thank Holger Gies, Jan Pawlowski, Martin Reuter, Bj[ö]{}rn Wellegehausen and Daniel K[ö]{}rner for useful discussions and the latter two for the collaboration on an accompanying numerical project on the renormalization group flow of sigma models. RF and OZ benefited from discussions with Alessandro Codello, Maximilian Demmel and Roberto Percacci. This work has been supported by the DFG Research Training Group “Quantum and Gravitational Fields” GRK 1523. The research of OZ is supported by the DFG within the Emmy-Noether program (Grant SA/1975 1-1).
Beta functions of $L_1$ and $L_2$
=================================
In this appendix we give the explicit form of the beta functions of the couplings $L_1$ and $L_2$. These have been omitted from the main text because of their size. Together with they complete the system of beta functions of the couplings of . We computed:
$$\begin{aligned}
\label{bL}
\beta_{L_1}
=& [ (2N-5) L_1 + 2 L_2 - \alpha] Q_{\frac{d}{2},2}
+ \Bigl[(2(N+4) + 4 d + d^2) L_1^2 + 8 L_2^2 + 4 (d+2) L_2 \alpha+ d(d+2) \alpha^2
\nonumber\\
&
+ 2 L_1 (2 (d+6) L_2 + (d^2 + 3 d +2) \alpha)\Bigr] Q_{\frac{d}{2}+2,3}
+ 2 [(d+1)L_1 + 2 L_2 + d \alpha]\zeta Q_{\frac{d}{2}+1,3} + \zeta^2 Q_{\frac{d}{2},3} + \frac{1}{6} Q_{\frac{d}{2}-2,1} \,, \nonumber
\\
\beta_{L_2}
=
&
\Bigl[ (d^2 (N-1) + 2 d (N+1)+12)L_2^2
+ (N + 2 d + 6) L_1^2 - 2 (d+2)(2 + d (N-2)) L_2 \alpha
\nonumber\\
&
+ 2 (d^2 + 2 d + 4 + N(d+2))L_1 L_2
- 2(d+2)(N-1 + d) \alpha L_1
+ d (d+2) (N-3) \alpha^2\Bigr] Q_{\frac{d}{2}+2,3}
\nonumber\\
&
-2 [L_2 ((N-2)d+2) + L_1 (N-1 + d)
- (N-3)d\alpha] \zeta Q_{\frac{d}{2}+1,3}
+ [L_1 +2(N-3) L_2- (N-3) \alpha] Q_{\frac{d}{2},2}
\nonumber\\
&
+(N-3) \zeta^2 Q_{\frac{d}{2},3}
-\tfrac{1}{6} Q_{\frac{d}{2}-2,1}
\,.\end{aligned}$$
[99]{}
A. Pelissetto and E. Vicari, Phys. Rept. [**368**]{}, 549 (2002) \[cond-mat/0012164\]. Phys. Rev. Lett. [**51**]{} (1983) 1915 H. Levine, S. B. Libby and A. M. M. Pruisken, Phys. Rev. Lett. [**51**]{}, 1915 (1983) \[Erratum-ibid. [**52**]{}, 1254 (1984)\]. E. Brezin and J. Zinn-Justin, Phys. Rev. Lett. [**36**]{}, 691 (1976). B. Delamotte, D. Mouhanna and M. Tissier, Phys. Rev. B [**69**]{}, 134413 (2004) \[cond-mat/0309101\]. E. Manousakis and R. Salvador, Phys. Rev. B [**40**]{}, 2205 (1989). W. J. Zakrzewski, In \*Storrs 1988, Proceedings, 4th Meeting of the Division of Particles and Fields of the APS\* 790-793.
V. A. Novikov, M. A. Shifman, A. I. Vainshtein and V. I. Zakharov, Phys. Rept. [**116**]{}, 103 (1984) \[Sov. J. Part. Nucl. [**17**]{}, 204 (1986)\] \[Fiz. Elem. Chast. Atom. Yadra [**17**]{}, 472 (1986)\]; A. D’Adda, P. Di Vecchia and M. Luscher, Nucl. Phys. B [**152**]{}, 125 (1979). J. Gasser and H. Leutwyler, Annals Phys. [**158**]{}, 142 (1984). S. Weinberg, In [*General Relativity: An Einstein centenary survey*]{}, ed. S. W. Hawking and W. Israel, pp.790–831, Cambridge University Press (1979).
M. Reuter, Phys. Rev. D [**57**]{} (1998) 971 \[hep-th/9605030\]. I. Y. .Arefeva, E. R. Nissimov and S. J. Pacheva, Commun. Math. Phys. [**71**]{}, 213 (1980); R. Percacci, Phys. Rev. D [**73**]{}, 041501 (2006) \[hep-th/0511177\]. C. Wetterich, Phys. Lett. B [**301**]{}, 90 (1993). L. -H. Chan, Phys. Rev. D [**36**]{}, 3755 (1987). J. Zinn-Justin, Int. Ser. Monogr. Phys. [**113**]{}, 1 (2002). H. G. Ballesteros, L. A. Fernandez, V. Martin-Mayor and A. Munoz Sudupe, Phys. Lett. B [**387**]{}, 125 (1996) \[cond-mat/9606203\]. P. Butera and M. Comi, Phys. Rev. B [**56**]{}, 8212 (1997) \[hep-lat/9703018\]. A. Codello and R. Percacci, Phys. Lett. B [**672**]{} (2009) 280 \[arXiv:0810.0715 \[hep-th\]\]. R. Percacci and O. Zanusso, Phys. Rev. D [**81**]{}, 065012 (2010) \[arXiv:0910.0851 \[hep-th\]\]. M. Fabbrichesi, R. Percacci, A. Tonero and O. Zanusso, Phys. Rev. D [**83**]{}, 025016 (2011) \[arXiv:1010.0912 \[hep-ph\]\]; M. Fabbrichesi, R. Percacci, A. Tonero and L. Vecchi, Phys. Rev. Lett. [**107**]{}, 021803 (2011) \[arXiv:1102.2113 \[hep-ph\]\]; F. Bazzocchi, M. Fabbrichesi, R. Percacci, A. Tonero and L. Vecchi, Phys. Lett. B [**705**]{}, 388 (2011) \[arXiv:1105.1968 \[hep-ph\]\]. L. Canet, B. Delamotte, D. Mouhanna and J. Vidal, Phys. Rev. B [**68**]{}, 064421 (2003) \[arXiv: 0302227 \[hep-th\]\]; D. F. Litim, D Zappala, Phys. Rev. D [**83**]{}, 085009 (2011) \[arXiv: 1009.1948 \[hep-th\]\]
R. Flore, D. K[ö]{}rner, B. Wellegehausen and A. Wipf, [*to be published*]{}
G. A. Vilkovisky, In \*Christensen, S.M. ( Ed.): Quantum Theory Of Gravity\*, 169-209
L. Alvarez-Gaume, D. Z. Freedman and S. Mukhi, Annals Phys. [**134**]{} (1981) 85;\
S. V. Ketov, Berlin, Germany: Springer (2000) 420 p.
E. Manrique and M. Reuter, Annals Phys. [**325**]{}, 785 (2010) \[arXiv:0907.2617 \[gr-qc\]\]; E. Manrique, M. Reuter and F. Saueressig, Annals Phys. [**326**]{}, 463 (2011) \[arXiv:1006.0099 \[hep-th\]\]. S. Mukhi, Nucl. Phys. B [**264**]{}, 640 (1986). J. M. Pawlowski, hep-th/0310018; J. M. Pawlowski, Annals Phys. [**322**]{} (2007) 2831 \[hep-th/0512261\]; I. Donkin and J. M. Pawlowski, arXiv:1203.4207 \[hep-th\]. B. S. DeWitt, Int. Ser. Monogr. Phys. [**114**]{}, 1 (2003); C. P. Burgess and G. Kunstatter, Mod. Phys. Lett. A [**2**]{}, 875 (1987) \[Erratum-ibid. A [**2**]{}, 1003 (1987)\]. I. L. Buchbinder and S. V. Ketov, Teor. i Matem. Fiz. [**77**]{} 42-50 (1988); I. L. Buchbinder and S. V. Ketov, Fortsch. Phys. [**39**]{}, 1 (1991). D. Benedetti, K. Groh, P. F. Machado and F. Saueressig, JHEP [**1106**]{}, 079 (2011) \[arXiv:1012.3081 \[hep-th\]\]. Y. Decanini and A. Folacci, Phys. Rev. D [**73**]{} (2006) 044027 \[gr-qc/0511115\]; D. Anselmi and A. Benini, *JHEP* [**0710** ]{} (2007) 099, \[[arXiv:0704.2840]{}\]; K. Groh, F. Saueressig and O. Zanusso, \[[arXiv:1112.4856]{}\]. P. Hasenfratz, Nucl. Phys. B [**321**]{}, 139 (1989). A. M. Polyakov, Phys. Lett. B [**59**]{}, 79 (1975). D. F. Litim, Phys. Rev. D [**64**]{}, 105007 (2001) \[hep-th/0103195\]. S. A. Antonenko and A. I. Sokolov, Phys. Rev. E [**51**]{}, 1894 (1995) \[hep-th/9803264\]. H. Kleinert, Phys. Rev. D [**60**]{} (1999) 085001 \[hep-th/9812197v2\].
G. P. Vacca and L. Zambelli, Phys. Rev. D [**83**]{}, 125024 (2011) \[arXiv:1103.2219 \[hep-th\]\]. T. R. Morris, JHEP [**0507**]{}, 027 (2005) \[hep-th/0503161\]. U. Ellwanger, Z. Phys. C [**76**]{}, 721 (1997) \[hep-ph/9702309\]. E. Manrique and M. Reuter, Phys. Rev. D [**79**]{}, 025008 (2009) \[arXiv:0811.3888 \[hep-th\]\]; E. Manrique and M. Reuter, PoS CLAQG [**08**]{}, 001 (2011) \[arXiv:0905.4220 \[hep-th\]\].
| {
"pile_set_name": "ArXiv"
} | ArXiv |
---
abstract: 'Klein foams are analogues of Riemann surfaces for surfaces with one-dimensional singularities. They first appeared in mathematical physics (string theory etc.). By definition a Klein foam is constructed from Klein surfaces by gluing segments on their boundaries. We show that, a Klein foam is equivalent to a family of real forms of a complex algebraic curve with some structures. This correspondence reduces investigations of Klein foams to investigations of real forms of Riemann surfaces. We use known properties of real forms of Riemann surfaces to describe some topological and analytic properties of Klein foams.'
author:
- 'Sabir M. Gusein-Zade, Sergey M. Natanzon [^1]'
title: Klein foams as families of real forms of Riemann surfaces
---
Moscow State Lomonosov University, Faculty of mechanics and mathematics,\
Moscow, GSP-1, 119991, Russia\
E-mail: sabir@mccme.ru
National Research University Higher School of Economics, 20 Myasnitskaya Ulitsa, Moscow 101000, Russia\
ITEP 25 B.Cheremushkinskaya, Moscow 117218, Russia ,\
e-mail: natanzons@mail.ru
Introduction
============
A Klein foam (a seamed surface) is constructed from Klein surfaces by gluing segments on their boundaries. Klein foams first appeared in mathematical physics (string theory, A-models) [@B; @KR; @R] (see also [@MMN]) and have applications in mathematics: [@MV; @AN1; @AN3] etc.
A Klein surface is an analog of a Riemann surface for surfaces with boundaries and non-orientable ones [@AG; @N1]. One can consider a Klein foam as an analog of a Riemann surface for surfaces with one-dimensional singularities. Further on we consider only compact Riemann and Klein surfaces.
Following [@CGN] we also require existence of a dianalytic map from a Klein foam to the complex disc. According to [@CGN] such function exists if and only if the corresponding topological foam is strongly oriented. According to [@N] this condition makes it possible to extend the 2D topological field theory with Klein surfaces [@AN] to Klein foams.
In this paper we prove, that a Klein foam is a collection of real forms of a complex algebraic curve. Any Klein surface is the quotient $S/\tau$, where $\tau :S\rightarrow S$ is an antiholomorphic involution of a Riemann surface $S$. In terms of algebraic geometry the pair $(S,\tau)$ is a complex algebraic curve with an involution of complex conjugation. Thus it is a real algebraic curve [@AG]. We will say also that $\tau$ is a real form of $S$. The fixed points of $\tau$ correspond to the boundary $\partial(S/\tau)$ of the quotient $S/\tau$ and to real points of the corresponding real algebraic curve. The fixed points form closed contours that are called *ovals* [@N4]. In what follows we consider only real algebraic curves with real points, that is with non-empty $\partial(S/\tau)$.
A general Riemann surface has not more that one antiholomorphic involution. However, there exist Riemann surfaces with several antiholomorphic involutions [@Na1; @Na2; @Na3]. Here we prove that the Klein foams are in one-to-one correspondence with the equivalence classes of the families of real forms of Riemann surfaces, i.e. of the collections $\{\widehat{S}, G, (G_1,\widehat{\tau}_1),\ldots, (G_r,\widehat{\tau}_r) \}$ consisting of: 1) a compact Riemann surface $\widehat{S}$; 2) a finite subgroup $G$ of the group ${{\rm Aut}\,}(\widehat{S})$ of holomorphic automorphisms of $\widehat{S}$; 3) real forms $\widehat{\tau}_1,\ldots, \widehat{\tau}_r$ of $\widehat{S}$ such that $\widehat{\tau}_i G\widehat{\tau}_i=G$; 4) subgroups $G_i\subset G$ ($i=1,2,\ldots, r$) generating $G$ and such that $\widehat{\tau}_i G_i\widehat{\tau}_i=G_i$.
A collection of real forms of a Riemann surface has some non-trivial “collective” properties. Initially these properties were observed in [@Na5; @Na6; @Na7; @Na4]. These results were developed and refined in a long series of publications (see [@BCGG] and references therein). We use these properties of real forms to describe some non-trivial combinatorial properties of analytic structures of Klein foams. In particular, we give bounds for the number of non-isomorphic Klein surfaces in a Klein foam and for the total number of ovals in a Klein foam.
Klein foams
===========
A Klein foam is a topological foam with an analytic structure. A topological foam is obtained from surfaces with boundaries by gluing some segments on the boundaries [@R; @AN1]. For our goal we modify this definition considering instead of a surface with boundary a surface without boundary, but with an involution such that the corresponding quotient is homeomorphic to the first one. This change of the definition does not change the notion itself.
A [*generalized graph*]{} is a one-dimensional space which consist of finitely many vertices and edges, where edges are either segments connecting [**different**]{} vertices or isolated circles without vertices on them. A pair of vertices may be connected by several edges.
A [*normal topological foam*]{} $\Omega$ is a triple $(S,\Delta,\varphi)$, where
- $S = S(\Omega)$ is a closed (i.e. compact without boundary, but usually disconnected) oriented surface with a reversing the orientation involution with the fixed point set being a closed curve $L\neq\emptyset$, that is finite disconnected union of simple contours;
- $\Delta=\Delta(\Omega)$ is a generalized graph;
- $\varphi=\varphi_{\Omega}: L\to\Delta$ is [*the gluing map*]{}, that is, a map such that: (a) $\mbox{Im\,}\varphi=\Delta$; (b) on each connected component of $L$, $\varphi$ is a homeomorphism on a circle in $\Delta$; (c) for an edge $l$ of $\Delta$, any connected component of $S$ contains at most one connected component of $\varphi^{-1}(l\setminus\partial l)$; (d) (the normality condition) for $\check{\Omega}=S\cup_{\varphi}\Delta$ (the result of the gluing of $S$ along $\Delta$) and for each vertex $v$ from the set $\Omega_b$ of vertices of the graph $\Delta$, its punctured neighbourhood in $\check{\Omega}$ is connected.
A triple $(S,\Delta,\varphi)$ which satisfies all the properties above but (c) will be called a [*topological pseudofoam*]{}. The same terminology will be applied to analytic and Klein foams defined below.
Normal topological foams arise naturally in the theory of Hurwitz numbers [@AN1; @AN3]. A foam $\Omega=(S,\Delta,\varphi)$ will be called [*connected*]{} if $\check{\Omega}$ is connected. In what follows topological foams are assumed to be normal and connected. Let $\Omega_{b}$ be the set of vertices of the graph $\Delta$.
A *morphism* $f$ of topological foams $\Omega^{\prime}\rightarrow \Omega^{\prime\prime}$ ($\Omega^{\prime}=(S^{\prime}, \Delta^{\prime}, \varphi^{\prime})$, $\Omega^{\prime\prime}=(S^{\prime\prime},\Delta^{\prime\prime}, \varphi^{\prime\prime})$) is a pair $(f_{S},f_{\Delta})$ of (continuous) maps $f_{S}:S^{\prime}\to S^{\prime \prime}$ and $f_{\Delta}: \Delta^{\prime}\to\Delta^{\prime\prime}$ such that $f_{S}$ is an orientation preserving ramified covering commuting with the involutions on $S^{\prime}$ and $S^{\prime\prime}$, $\varphi^{\prime\prime}\circ f_{S} = f_{\Delta}\circ \varphi^{\prime}$ and $f_{\Delta}|_{\Delta^{\prime}\setminus\Omega^{\prime }_{b}}$ is a local homeomorphism $\Delta^{\prime}\setminus\Omega^{\prime}_{b}$ on $\Delta^{\prime\prime}\setminus\Omega^{\prime\prime}_{b}$.
An [*analytic foam*]{} is a topological foam $\Omega=(S,\Delta,\varphi)$, where $S$ is a compact Riemann surface, the involution of which is antiholomorphic. A morphism $f$ of analytic foams $\Omega^{\prime}\rightarrow \Omega^{\prime\prime}$ ($\Omega^{\prime}=(S^{\prime}, \Delta^{\prime}, \varphi^{\prime})$, $\Omega^{\prime\prime}=(S^{\prime\prime},\Delta^{\prime\prime}, \varphi^{\prime\prime})$) is a morphism $(f_{S},f_{\Delta})$ of the corresponding topological foams such that $f_{S}$ is complex analytic.
The simplest analytic foam is $\Omega_{0}=(\overline{\mathbb{C}}, S^1, Id)$, where $\overline{\mathbb{C}}=\mathbb{C}\cup\infty$ is the Riemann sphere with the involution $z\mapsto \bar{z}$. An *analytic function* on an analytic foam $\Omega$ is a morphism of $\Omega$ to $\Omega_{0}$.
A *Klein foam* is an analytic foam $\Omega=(S,\Delta,\varphi)$ admitting an everywhere locally non-constant analytic function. This condition appeared first in [@CGN] and is equivalent to the condition of being “strongly oriented” for the corresponding topological foam. The category of strongly oriented topological foams [@N] is a subcategory of topological foams that allows the topological field theory to be extended to the Klein topological field theory [@AN].
A Klein (pseudo)foam $\Omega$ will be called *compressed* if the coincidence of $F(x)$ and $F(x')$ for two points $x$ and $x'$ of $\Delta_{\Omega}$ and for all analytic functions $F$ on $\Omega$ implies that $x=x'$. For any Klein foam $\Omega$ there exists a unique compressed Klein pseudofoam $\Omega'$ with a morphism $\Psi:\Omega\rightarrow \Omega'$ such that $\Psi_S$ is an isomorphism. This follows from the following fact. In [@CGN Theorem 2.1], it was shown that there exists a connected Riemann surface $\check{S}$ with an antiholomorphic involution $\check{\tau}$ and an analytic nowhere locally constant maps $f:\check{\Omega}\to \check{S}$, commuting with the involutions on $S$ and on $\check{S}$, such that the corresponding morphism of the foams establishes an isomorphism between the fields of analytic functions on ${\Omega}$ and on $\check{S}$ respectively. One can see that $\Omega'$ is obtained from $\Omega$ by gluing all the points $x$ and $x'$ of $\Delta(\Omega)$ with $f(x)=f(x')$. (In particular the graph $\Delta_{\Omega'}$ coincides with the set of the real points of $\check{S}$ cut into edges by the ramification points of $f$.) We shall say that $\Omega'$ is the *compressing* of $\Omega$.
Two Klein foams $\Omega_1$ and $\Omega_2$ will be called *weakly isomorphic* if their compressings are isomorphic.
Real forms.
===========
An [*equipped family of real forms*]{} of a Riemann surface $\widehat{S}$ is a collection $\{\widehat{S}, G, (G_1,\widehat{\tau}_1),\ldots, (G_r,\widehat{\tau}_r) \}$ consisting of:
- a compact Riemann surface $\widehat{S}$;
- antiholomorphic involutions (i.e. real forms) $\{\widehat{\tau}_1,\ldots, \widehat{\tau}_r\}$ of $\widehat{S}$;
- a finite subgroup $G$ of the group of holomorphic automorphisms of ${{\rm Aut}\,}(S)$ such that $\widehat{\tau}_i G\widehat{\tau}_i=G$ for each $i=1,2,\ldots, r$;
- subgroups $G_i\subset G$ such that $\widehat{\tau}_i G_i\widehat{\tau}_i=G_i$ for $i=1,2,\ldots, r$ and $G$ as a group is generated by $G_1$, …, $G_r$;
- the inclusion $G_i\subset G$ generates a morphism of Klein surfaces $S/G_i\rightarrow S/G$ that is a homeomorphism on any oval of $S/G$.
Two equipped families of real forms of Riemann surfaces $$\{\widehat{S}, G, (G_1,\widehat{\tau}_1),\ldots, (G_r,\widehat{\tau}_r)\}\quad \text{and} \quad
\{\widehat{S}, G^\prime, (G_1^\prime,\widehat{\tau}_1^\prime),\ldots, (G_r^\prime,\widehat{\tau}_r^\prime)\}$$ are equivalent if there exists $h\in{{\rm Aut}\,}{\widehat{S}}, h_i\in G, l_i\in G_i$, such that $$hGh^{-1}=G',\quad hh_iG_ih_i^{-1}h^{-1}=G_i', \quad hh_il_i\widehat{\tau}_il_i^{-1}h_i^{-1}h^{-1}=\widehat{\tau}_i'\,.$$
Two equipped families of real forms of Riemann surfaces $$\{\widehat{S}, G, (G_1,\widehat{\tau}_1),\ldots, (G_r,\widehat{\tau}_r)\}\quad \text{and} \quad
\{\widehat{S}', G', (G_1^\prime,\widehat{\tau}_1^\prime),\ldots, (G_r^\prime,\widehat{\tau}_r^\prime)\}$$ are equivalent if there exists an analytic isomorphism $H:\widehat{S}\to\widehat{S}^\prime$ such that $$\{\widehat{S}',HGH^{-1}, (HG_1H^{-1},H\widehat{\tau}_1H^{-1}),\ldots, (HG_rH^{-1},H\widehat{\tau}_rH^{-1})\}$$ and $$\{\widehat{S}', G^\prime (G_1^\prime,\widehat{\tau}_1^\prime),\ldots, (G_r^\prime,\widehat{\tau}_r^\prime)\}$$ are equivalent.
Main theorem.
=============
\[T1\] There exists a natural one-to-one correspondence between the classes of weakly isomorphic Klein foams and the equivalence classes of equipped families of real forms of Riemann surfaces.
Consider a Klein foam $\Omega=(S,\Delta,\varphi)$ with $S$ consisting of connected components $S_i$, $i=1,\ldots, r$. Let $\tau_i$ be the restriction of the (antiholomorphic) involution to $S_i$.
In [@CGN Theorem 2.1], it was shown that there exists a connected Riemann surface $\check{S}$ with an antiholomorphic involution $\check{\tau}$ and analytic nowhere locally constant maps $f:\check{\Omega}\to \check{S}$, commuting with the involutions on $S$ and on $\check{S}$, such that the corresponding morphism of the foams establishes an isomorphism between the fields of analytic functions on ${\Omega}$ and on $\check{S}$ respectively. Consider the restrictions $f_i: S_i\to \check{S}$ of $f$ to $S_i$.
Let $\check{S}^\circ$ be the surface $\check{S}$ without all the critical values of the maps $f_i$ and $S^{\circ}= f^{-1}(\check{S}^\circ)$. Consider an uniformization $U\rightarrow U/\check{\Gamma}=\check{S}^\circ$. (Excluding trivial cases we assume that $U$ is the upper half-plane in ${{\mathbb C}}$ and $\Gamma$ is a Fuchsian group.)
The restriction $f^\circ_i: S^\circ_i\to \check{S}^\circ$ of $f_i$ to $S^\circ_i= S_i\cap S^\circ$ is a covering without ramification. Thus there exists a subgroup $\Gamma_i\subset\check{\Gamma}$, that uniformises $S^\circ_i= U/\Gamma_i$ and generates the map $f^\circ_i$ by the inclusion $\Gamma_i\subset\check{\Gamma}$. This subgroup $\Gamma_i$ is well-defined up to conjugation in $\Gamma$. Let $\sigma_i$ be a lifting to $U$ of the involution $\tau_i$ to $S^\circ_i$, such that $\sigma_i\Gamma_i\sigma_i=\Gamma_i$. This $\sigma_i$ is well-defined up to conjugation in $\Gamma_i$.
The subgroup $\Gamma_i$ has a finite index in $\Gamma$ and there are only finitely many different subgroups of $\check{\Gamma}$ conjugate to $\Gamma_i$. Let us consider the intersection of all the subgroups $\Gamma_i$ and all their conjugates: $$\hat{\Gamma}=\bigcap_{j=1}^r\bigcap_{a\in\check{\Gamma}}a^{-1}\Gamma_j a\subset\check{\Gamma}\,.$$
Since all the involutions $\tau_i$ give one and the same involution $\check{\tau}$ on the surface $\check{S}$, one has $\sigma_i\sigma_j \in\check{\Gamma}$ for all $i$ and $j$. Thus $\sigma_ia\sigma_j\in \check{\Gamma}$ for any $a\in\check{\Gamma}$ and $$\begin{aligned}
\sigma_i\hat{\Gamma}\sigma_i&=&
\sigma_i\left(\bigcap_{j=1}^r\bigcap_{a\in\check{\Gamma}}a^{-1}\Gamma_j a\right)\sigma_i=\\
&{\ }&\bigcap_{j=1}^r\bigcap_{a\in\Gamma}
\sigma_ia^{-1}\sigma_j\Gamma_j \sigma_ja\sigma_i=
\bigcap_{j=1}^r\bigcap_{b\in\Gamma}b^{-1}\Gamma_j b=\hat{\Gamma}\,.\end{aligned}$$
Consider now the Riemann surface $\hat{S}^\circ = U/\hat{\Gamma}$. The involution $\sigma_i$ generates an involution $\hat{\tau}^0_i$ on $\hat{S}^\circ$. The inclusion $\hat{\Gamma}\subset\check{\Gamma}$ generates a covering $\Phi^\circ: \hat{S}^\circ\rightarrow \check{S}^\circ$. The inclusion $\check{S}^\circ\subset\check{S}$ generates a branching covering of Riemann surfaces $\hat{S}\rightarrow\check{S}$. The involution $\hat{\tau}^0_i$ generates an antiholomorphic involution $\hat{\tau}_i$ on $\hat{S}$. Moreover $\hat{\Gamma}$ is a normal subgroup of $\check{\Gamma}$ and thus the subgroup $\check{\Gamma}/\hat{\Gamma}$ in a natural way is isomorphic to a subgroup $G\subset{{\rm Aut}\,}(\hat{S})$ of the group of biholomorphic automorphisms of $\hat{S}$.
The subgroups $\Gamma_i$ from this construction also have an interpretation in terms of the Riemann surface $\hat{S}$. Let us consider the group ${{\rm Aut}\,}^*(\hat{S})$ of biholomorphic and antiholomorphic automorphisms of $\hat{S}$. All the involutions $\hat{\tau}_i$ belong to ${{\rm Aut}\,}^*(\hat{S})$. Moreover, the involution $\sigma_i$ preserves the subgroup $\Gamma_i\subset\check{\Gamma}$. The homomorphism $\phi: \check{\Gamma}/\hat{\Gamma} \rightarrow{{\rm Aut}\,}(\hat{S})$ maps the subgroup $\Gamma_i$ to a subgroup $G_i\subset G$, such that $\hat{\tau}_iG_i\hat{\tau}_i=G_i$.
Thus we have constructed an equipped family of real forms of a compact Riemann surface corresponding to a Klein foam $\Omega=(S,\Delta,\varphi)$. This family depends on the choice of the uniformizing subgroups $\Gamma_i$ and of the pull-backs $\sigma_i$. This means that the family is defined up to equivalence.
Since the compressing of a real foam depends only on the maps $f_i$, the weak equivalence class of the foam $\Omega$ is determined only by the equivalence class of the corresponding equipped family of real forms of a Riemann surface.
One can see that there are finitely many Klein foams corresponding to an equivalence class of equiped families of real forms of a Riemann surface. They are determined by some additional combinatorial data.
Some topological and analytical properties of Klein foams.
==========================================================
Consider a Klein foam $$\Omega=(S,\Delta,\varphi), \quad \text{where} \quad S=\coprod\limits_{i=1}^r(S_i,\tau_i)\,.$$ Denote by $g_i$ and $k_i$ the genus of $S_i$ and the number of ovals of $S_i$ (i.e. the number of the connected components of $\partial(S_i/\tau_i)$) respectively.
We say that $(S_i,\tau_i)$ and $(S_j,\tau_j)$ are *foam-equivalent* if the images under $\varphi$ of the fixed point sets of $\tau_i$ and $\tau_j$ coincide. Otherwise we say that $(S_i,\tau_i)$ and $(S_j,\tau_j)$ are *foam-different*.
Recall that Klein surfaces $(S',\tau')$ and $(S'',\tau'')$ are isomorphic if there exists a biholomorphic map $h:S'\to S''$ such that $\tau''=h\tau'h^{-1}$. We say that a real algebraic curve $(S',\tau')$ *covers* $(S'',\tau'')$ if there exists a holomorphic map $h:S'\to S''$ such that $\tau''h=h\tau'$. We say that Klein surfaces $(S',\tau')$ and $(S'',\tau'')$ are *weakly equivalent* if there exists a Klein surface $(S^\ast,\tau^\ast)$ covering both $(S',\tau')$ and $(S'',\tau'')$.
Consider the equipped family of real forms $$F=\{\hat{S}, G, (G_1,\hat{\tau}_1),\dots, (G_r,\hat{\tau}_r) \}$$ corresponding to $\Omega$. The main topological invariants of the family are the genus $\hat{g}$ of $\hat{S}$, the numbers of the ovals $\hat{k}_i= |\pi_0(\partial(\hat{S}/\hat{\tau}_i))|$ and the cardinalities $|G_i|$ of the subgroups. From the definitions it follow that $(S_i,\tau_i)=(\hat{S}/G_i, \hat{\tau}_i/G_i)$. Therefore $$g_i\leq\frac{\hat{g}-1}{|G_i|}+1 \quad \text{and} \quad k_i\leq\hat{k}_i\,.$$
Let us describe some properties of foams related with their genuses. These properties depend not only on the topological properties of a foam $\Omega=(S,\Delta,\varphi)$, but also on a holomorphic structure on it. Further on we assume that $\hat{g}>1$. The topological classification of the equipped families of real forms for $\hat{g}\leq 1$ is more simple, but requires other methods.
Let $\Omega=(S,\Delta,\varphi)$ be a Klein foam such that $S$ contains the union $\coprod\limits_{i=1}^r S_i$ with all $(S_i,\tau_i)$ foam-different. Then $\sum\limits_{i=1}^r k_i\leq 42(\hat{g}-1)$.
The involutions $\hat{\tau}_i$ corresponding to foam-different Klein surfaces are different. On the other hand, the sum of of the numbers of all the ovals of all the real forms of a Riemann surface of genus $\hat{g}>1$ is at most $ 42(\hat{g}-1)$ [@Na6; @Na4]. Thus $\sum\limits_{i=1}^r \hat{k}_i\leq 42(\hat{g}-1)$.
Let $\Omega=(S,\Delta,\varphi)$ be a Klein foam such that $S$ contains the union $\coprod\limits_{i=1}^r S_i$ with all $(S_i,\tau_i)$ not weakly equivalent. Then $r\leq2(\sqrt{\hat{g}}+1)$ for any $\hat{g}>1$ and $r\leq4$, if $\hat{g}$ is even.
The involutions $\hat{\tau}_i$ corresponding to not weakly equivalent Klein surfaces are not conjugate in $\hat{G}$. On the other hand, the number of conjugation classes of involution on a Riemann surface of genus $\hat{g}$ is at most $2(\sqrt{\hat{g}}+1)$ for any $\hat{g}>1$ [@Na5] and at most $4$ if $\hat{g}$ is even [@GI].
Let $\Omega=(S,\Delta,\varphi)$ be a Klein foam such that $S$ contains the union $\coprod\limits_{i=1}^r S_i$ with all $(S_i,\tau_i)$ not weakly equivalent. In addition assume that either $r=3, 4$ or all the surfaces $\hat{S}/G_i$ are orientable. Then $\sum\limits_{i=1}^r k_i\leq 2\hat{g}-2+2^{r-3}(9-r)<2\hat{g}+30$.
The corresponding bounds for the number of involutions on a Riemann surface are proved in [@Na4; @Na7; @Na8; @Na9].
Partially supported by the grants NSh–5138.2014.1, RFBR–13-01-00755 (S.G.-Z.), RFBR–13-02-00457 (S.N.).
Keywords: foam, real forms.
[99]{} A.Alexeevski, S.Natanzon. Noncommutative two-dimentional topological field theory and Hurwitz numbers for real algebraic curves. Selecta Math. (N.S.) v.12 (2006), no.3-4, 307–377 (first appeared in arXiv math/0202164).
A.Alexeevski, S.Natanzon. The algebra of bipartite graphs and Hurwitz numbers of seamed surfaces. Izv. Math. v.72 (2008), no.4, 627–646.
A.Alexeevski, S.Natanzon. Hurwitz numbers for regular coverings of surfaces by seamed surfaces and Cardy–Frobenius algebras of finite groups. Amer. Math. Sos. Transl., ser.2, v.224 (2008), 1–25.
N.L.Alling, N.Greenleaf. Foundation of the Theory of Klein Surfaces. Lect. Not. Math. v.219 (1971), 117 pp.
J.C.Baez. An introduction to spin foam models of $BF$ theory and quantum gravity. Geometry and quantum physics (Schladming, 1999), 25–93, Lecture Notes in Phys., v,543, Springer, Berlin, 2000.
E.Bujalance, F.J.Cirre, J.M.Gamboa, G.Gromadski. Symmetries of compact Riemann surfaces. Lecture Notes in Mathematics, Springer-Verlag, Berlin–Heidelberg, v.2007 (2010), 155 p.
A.F.Costa, S.M.Gusein-Zade, S.M.Natanzon. Klein foams. Indiana Univ. Math. J. v.60 (2011), no.3, 985–995.
G.Gromadski, M.Izquierdo. Real forms of a Riemann surface of even genus. Proc.Amer.Math.Soc. v.126 (12) (1998), 3475–3479.
M.Khovanov, L.Rozansky. Topological Landau-Ginzburg models on the world-sheet foam. Adv. Theor. Math. Phys. v.11 (2007), no.2, 233–259.
M.Mackaay, P.Vaz. The foam and the matrix factorization $sl_3$ link homolgies are equivalent. Algebr. Geom. Topol. 8 (2008), no.1, 309-342.
A.D.Mironov, A.Morozov, S.M. Natanzon. Hurwitz theory avatar of open-closed strings. The European Physical Journal C - Particles and Fields, v.73 (2013), no.2, 1–10.
S.M.Natanzon. On the order of a finite group of homeomorphisms of a surface onto itself and the number of real forms of a complex algebraic curve. Dokl. Akad. Nauk SSSR, 242:4 (1978), 765–768 (Russian); English translation in: Soviet Math. Dokl., 19:5 (1978), 1195–1199.
S.M.Natanzon. Automorphisms of the Riemann surface of an M-curve. Funktsional. Anal. i Prilozhen., 12:3 (1978), 82–83 (Russian); English translation in: Functional Anal. Appl., 12:3 (1978), 228–229.
S.M.Natanzon. Automorphisms and real forms of a class of complex algebraic curves. Funktsional. Anal.i Prilozhen., 13:2 (1979), 89–90 (Russian); English translation in: Functional Anal. Appl., 13 (1979), 148–150
S.M.Natanzon. On the total number of ovals of real forms of a complex algebraic curve. Uspekhi Mat.Nauk, 35:1 (1980), 207–208 (Russian); English translation in: Russian Math.Surveys, 35:1 (1980), 223–224.
S.M.Natanzon. On the total number of ovals of four complex-isomorphic real algebraic curves. Uspekhi Mat. Nauk, 35:4 (1980), 184.
S.M.Natanzon. Lobachevskian geometry and automorphisms of complex M-curves. Sel. Math. Sov., 1:1 (1981), 81–99.
S.M.Natanzon. Finite groups of homeomorphisms of surfaces and real forms of complex algebraic curves. Trudy Moscow Math. Obshch., v.51 (1988), 3–53 (Russian); English translation in: Trans. Moscow Math. Soc., 1989, 1–51.
S.M.Natanzon. A Harnack-type theorem for a family of complex-isomorphic real algebraic curves. Uspekhi Mat. Nauk, 52:6 (1997), 173–174 (Russian); English translation in: Russian Math.Surveys, 52:6 (1997), 1314–1315.
S.M.Natanzon. Geometry and algebra of real forms of complex curves. Math. Zeit. v.243 (2003), 391–407.
S.M.Natanzon. Cyclic foam topological field theory. J. Geom. Phys. v.60 (2010), no.6-8, 874–883.
S.M.Natanzon. Klein surfaces, Russian Math. Surveys. v.45 (1990), no.6, 53–108.
S.M.Natanzon. Moduli spaces of real algebraic curves. Proc. Moscow Math. Soc. v.37 (1978), 219–253.
S.M.Natanzon. Moduli of Rieman Sufaces, Real Algebraic Curves and its Superanalogs. Translations of Math. Mon., v.225, Amer. Math. Soc. (2004), 159pp.
L.Rozansky. Topological A-models on seamed Riemann surfaces. Adv. Theor. Math. Phys., v.11 (2007), no.4, 517–529.
Partially supported by the grants NSh–5138.2014.1, RFBR–13-01-00755 (S.G.-Z.), RFBR–13-02-00457 (S.N.).
Keywords: foam, real forms.
[^1]: AMS 2010 Math. Subject Classification: 30F50, 57M20, 14H30, 14H37.
| {
"pile_set_name": "ArXiv"
} | ArXiv |
---
abstract: |
Two types of Poisson pencils connected to classical R-matrices and their quantum counterparts are considered. A representation theory of the quantum algebras related to some symmetric orbits in $sl(n)^*$ is constructed. A twisted version of quantum mechanics is discussed.
[**Mathematics Subject Classification (1991):**]{} 17B37, 81R50.
author:
- |
D. Gurevich\
ISTV, Université de Valenciennes, 59304 Valenciennes, France\
J.Donin\
Departement of Mathematics, Bar-Ilan University, 52900\
Ramat-Gan,Israel\
V.Rubtsov\
ITEP, Bol.Tcheremushkinskaya 25, 117259 Moscow, Russia
title: Two types of Poisson pencils and related quantum objects
---
[*Dedicated to Alain Guichardet with regards and friendship.*]{}
Introduction
============
There are (at least) two reasons for Poisson pencils (P.p.) to be currently of great interest. First, they play a very important role in the theory of integrable systems (in the so-called Magri–Lenart scheme). Second, they appear as infinitesimal (quasi-classical) objects in the construction of certain quantum homogeneous spaces. Roughly speaking, we can say that the P.p. arising in the framework of the latter construction are of the first type and those connected to integrable systems belong to the second type (while the integrable systems themselves are disregarded).
The main characteristic that joins together these two classes of P.p. is that a classical R-matrix participates in their construction. Let us recall that by a classical R-matrix on a simple Lie algebra $g$ one means a skew-symmetric $(R\in \wedge^2(g))$ solution of the classical Yang–Baxter equation $$[R^{12},R^{13}]+[R^{12},R^{23}]+[R^{13},R^{23}]=a\varphi,\, a\in k,$$ where $\varphi$ is a unique (up to a factor) ad-invariant element belonging to $\wedge^3(g))$. In what follows we deal with the “canonical” classical R-matrix $$R=\frac12\sum X_{\alpha}\wedge X_{-\alpha}$$ where $\{H_{\alpha},\, X_{\alpha},\, X_{-\alpha}\}$ is a Cartan–Weyl–Chevalley base of a simple Lie algebra $g$ over the field $k={\bf C}$. We assume that a triangular decomposition of the Lie algebra $g$ is fixed. The field $k={\bf R}$ is also admitted, but in this case we consider the normal real forms of the Lie algebras corresponding to this triangular decomposition.
Thus, the R-matrix (1) enters the constructions of both type of P.p. under consideration. However, the constructions and properties of these types are completely different. Moreover, the methods of quantizing these two types of P.p. are completely different, as well as the properties of the resulting associative algebras. We will use the term [*associative pencils*]{} (a.p.) of the first (second) type for the families of these quantum algebras arising from the first (second) type P.p.
The ground object of the second type P.p. is a quadratic Poisson bracket (P.b.) of Sklyanin type. By this type we mean either the famous Sklyanin bracket[^1] or its various analogues: the elliptic Sklyanin algebras in the sense of [@S], the so-called second Gelfand–Dikii structures, etc. (Although we restrict ourselves to finite-dimensional Poisson varieties, we would like to note it is not reasonable to consider these structures as the infinite-dimensional analogues of P.p. on symmetric orbits in $g^*$, as it was suggested in [@KRR]. The principal aim of the present paper is to make the difference between the two classes of P.p. related to classical R-matrices more transparent.)
All these P.b. (further denoted by $\{\,\,,\,\,\}_2$) are quadratic and their linearization gives rise to another (linear) P.b. (denoted by $\{\,\,,
\,\,\}_1$) compatible with the initial one. As a result we have a P.p. $$\{\,\,,\,\,\}_{a,b}=a\{\,\,,\,\,\}_1+b\{\,\,,\,\,\}_2$$ generated by these two brackets.
Now let us describe the corresponding quantum objects. We construct them in two steps. First, we quantize the Sklyanin bracket and get the well-known quadratic “RTT=TTR” algebra. Then by linearizing the determining quadratic relations (or more precisely, by applying a shift operator to the algebra, cf. below) we get a (second type) a.p. which is the quantum counterpart of the whole P.p. These second type P.p. and their quantum counterparts are considered in Section 2.
Let us note that the shift operator mentioned above does not give rise to any meaningful deformation and it is reduced to a mere change of base.
We are interested in P.p. and their quantum counterparts connected to symmetric orbits in $g^*$, where $g$ is a simple Lie algebra. It should be pointed out that the latter P.p. are not any reduction of the second type P.p. Let us say a little bit more about the origin of these P.p. on symmetric orbits.
It is well known that any orbit ${{\cal O}_{x}}$ of any semisimple element in $g^*$ can be equipped with the reduced Sklyanin (sometimes called Sklyanin–Drinfeld) bracket. One usually considers the reduction procedure for the compact forms of simple Lie algebras, but it is also valid for normal ones (the R-matrix for compact forms differs from (1) by the factor $\sqrt{-1}$).
It is worth noting that the symmetric orbits in $g^*$ admit a complementary nice property: the reduced Sklyanin bracket is compatible with the Kirillov–Kostant–Souriau (KKS) bracket. This fact was shown in [@KRR], [@DG2] (in [@KRR] it was also shown that only symmetric orbits possess this property).
Moreover, as shown in [@DG2], both components $\{\,\,,\,\,\}_{{\epsilon}},\,{\epsilon}=
l,\,r$ of the Sklyanin bracket (cf. footnote 1) become Poisson after being reduced to ${{\cal O}_{x}}$ and one of them (say, $\{\,\,,\,\,\}_{r}$ coincides up to factor with the KKS one if we identify ${{\cal O}_{x}}$ with right coset $G/H$).
Since the brackets $\{\,\,,\,\,\}_{{\epsilon}},\,{\epsilon}=l,r$ are always compatible, we have a (first type) P.p. (2) on a symmetric orbit ${{\cal O}_{x}}\in {g}^*$ with $\{\,
\,,\,\,\}_{1}=\{\,\,,\,\,\}_l^{{\rm red}}$ and $\{\,\,,\,\,\}_{2}=\{\,\,,\,\,
\}_{r}^{{\rm red}}$ (“red” means reduced on ${{\cal O}_{x}}$)[^2].
The difference between these two types of P.p. also manifests itself on the quantum level. One usually represents quantum analogues of the Sklyanin bracket reduced to a semisimple orbit in terms of “quantum reduction” by means of pairs of quantum groups or their (restricted) dual objects. However, in this way it is not possible to represent the whole a.p. arising from the first type P.p. under consideration. We discuss here another, more explicit, way to describe the corresponding quantum a.p. Namely, we represent them as certain quotient algebras.
In virtue of the results of the paper [@DS1], a formal quantization of the P.p. under consideration exists on any symmetric orbit. However, it is not so easy to describe the quantum objects explicitly or, in other words, to find systems of equations defining them. We consider here two ways to look for such a system in the case when the orbit ${{\cal O}_{x}}$ is a rank 1 symmetric space in $sl(n)^*$, i.e., that of ${SL(n)/S(L(n-1)\times L(1))}$ type.
The first way was suggested in [@DG1] (but there the system of equations was given in an inconsistent form). It uses the fact that such an orbit can be described by a system of quadratic equations (cf. Section 3). The second way consists in an attempt to construct a representation theory of the first type quantum algebras (cf. Section 4) and to compute the desired relations in the modules over these algebras. This approach is valid in principle for any symmetric orbit in $g^*$ for any simple Lie algebra $g$. This enables us to treat this type of quantum algebras from the point of view of twisted quantum mechanics (Section 5). This means that quantum algebras and their representations are objects of a twisted (braided) category. (We consider the term “twisted” as more general, keeping the term “braided” for nonsymmetric categories).
Completing the Introduction, we would like to make two remarks. First, let us note that there is a number of papers devoted to q-analogues of special functions of mathematical physics. In fact they deal with quantum analogues of double cosets. Meanwhile, the problem of finding an explicit description of quantum symmetric spaces (in particular, orbits in $g^*$) is disregarded (in some sense the latter problem is more complicated). If the usual q-special functions arise from a one-parameter deformation of the classical objects, our approach enables us to consider their analogues arising from the two-parameter deformation. This approach will be developed in a joint paper of one of the author (D.G.) and L.Vainerman.
Second, we would like to emphasize that we consider the present paper as an intermediate review of the papers [@DG1], [@DGR], [@G2] where the present topic (i.e., the problem of the explicit description of a.p. arising from quantization of certain P.p. associated with the “canonical” classical R-matrices) was discussed. Meanwhile, this topic is still in progress.
In what follows ${U_q(g)}$-Mod stands for the category of all finite-dimensional ${U_q(g)}$-modules and their inductive limits. The parameter $q$ is assumed to be generic (or formal, when we speak about flatness of deformation, in fact we do not distinguish these two meanings).
We dedicate this paper to our friend, professor Alain Guichardet. We acknowledge his warm hospitality at Ecole Polytechnique for a long time and benefited a lot from our numerous discussions and conversations.
This paper was finished during the visit of V.R. to the University Lille-1. He is thankful to his colleagues from the group of Mathematical Physics for their hospitality. The work of V.R. was supported partially by the grants RFFI-O1-01011, INTAS-93-2494 and INTAS-1010-CT93-0023.
Second type Poisson and associative pencils
===========================================
Let us fix a simple algebra $g=sl(n),\,n\geq 2$ and consider the corresponding quadratic Skyanin bracket extended on the space ${\rm Fun}({\rm
Mat}(n))$ which is the algebra of polynomials on the matrix elements $a_i^j$. More precisely, this bracket is determined by the following multiplication table $$\{a_{k}^{i},a_{k}^{j}\}_{2}=a_{k}^{i}\,a_{k}^{j},\,\,
\{a_{i}^{k},a_{j}^{k}\}_{2}=a_{i}^{k}\,a_{j}^{k},\,\,i<j;$$ $$\{a_{i}^{l},a_{k}^{j}\}_{2}=0,\,\,
\{a_{i}^{j},a_{k}^{l}\}_{2}=2\,a_{i}^{l}\,a_{k}^{j},
\,\,i<k,\,j<l.$$
Let us linearize the bracket ${\{\,\,,\,\,\}_2}$. To do this we introduce a shift operator ${\rm Sh}_{h}$ which sends $a_{i}^{j}$ to $a_{i}^{j}+h \delta_{i}^{j}$ (we extend this operator onto the whole algebra ${\rm Fun}({\rm Mat}(n))$ by multiplicativity). Then applying this operator to r.h.s. of the above formulas and taking the linear terms in $h$, we get a linear bracket with by the following multiplication table $$\{a_{k}^{i},a_{k}^{j}\}_{1}=\delta_{k}^{i}\,a_{k}^{j}+a_{k}^{i}\,
\delta_{k}^{j},\,\,
\{a_{i}^{k},a_{j}^{k}\}_{1}=\delta_{i}^{k}\,a_{j}^{k}+a_{i}^{k}\,
\delta_{j}^{k},\,\,i<j;$$ $$\{a_{i}^{l},a_{k}^{j}\}_{1}=0,\,\,
\{a_{i}^{j},a_{k}^{l}\}_{1}=2\,(\delta_{i}^{l}\,a_{k}^{j}+a_{i}^{l}\,
\delta_{k}^{j}),
\,\,i<k,\,j<l.$$
The brackets ${\{\,\,,\,\,\}_{1,2}}$ are compatible.
[*Proof.*]{} This follows immediately from the following fact: the r.h.s. of the formulas for the bracket ${\{\,\,,\,\,\}_2}$ does not contain any summand of the form $a_i ^i a_j^j$. This implies that the images of r.h.s. elements under the operator ${\rm Sh}_{h}$ do not contain any term quadratic in $h$. The details are left to the reader. $\Box$
[*It is often more convenient to take, instead of $\{a_i^j
\}$, the base consisting of $a_i^j,\,i\not=j,\,\,a_1^1-a_i^i,\,i>1$ and $a_0=
\sum a_i^i$. Then the mentioned property can be reformulated as follows: the r.h.s. of the above formulas do not contain the term $(a_0)^2$ (with respect to these new generators the operator ${\rm Sh}_h$ acts nontrivially only on $a_0$).*]{}
It is easy to see that the bracket ${\{\,\,,\,\,\}_1}$ can be represented in the form $$\{a_{i}^{j},a_{k}^{l}\}_{1}=\{{\bf R}(a_{i}^{j}), a_{k}^{l}\}_{gl}+
\{a_{i}^{j}, {\bf R}(a_{k}^{l})\}_{gl}.$$ Here $\{\,\,,\,\,\}_{gl}$ is the linear bracket corresponding to the Lie algebra $gl(n)$ (namely, $\{a_{i}^{j},a_{k}^{l}\}_{gl}=a_{i}^{l}\delta_{k}^
{j}-a_{k}^{j}\delta_{i}^{l}$) and ${\bf R}:W\to W$ is an operator defined in the space $W={\rm Span}(a_i^j)$ as follows ${\bf R}(a_{i}^{j})={\rm sign}(j-
i)\,a_{i}^{j}$ (we assume that ${\rm sign}(0)=0$).
Thus, the space $W$ is equipped with two Lie algebra structures. The first one is $gl(n)$ and the second one corresponds to the Poisson bracket $\{\,\,,
\,\,\}_{1}$ and therefore we have a double Lie algebra structure on the space $W$ according to the terminology of [@S-T] (our construction coincides with [@S-T] up to a factor).
Let us consider now the quantum analogue of the above P.p.
Let $U_q(g)$ be the Drinfeld-Jimbo quantum group corresponding to $g$ and ${\cal R}$ be the universal quantum R-matrix. Consider the linear space $V$ of vector fundamental representation $\rho : g\to {\rm End}(V)$ of the initial Lie algebra $g$. Denote by $\rho_q$ the corresponding representation of the quantum group ${U_q(g)}$ to ${\rm End}(V)$ (we assume that $\rho_q\to \rho$ when $q\to 1$). Then the operator $S:{V^{\otimes 2}}\to{V^{\otimes 2}}$ defined as follows $S=\sigma\rho^{{\otimes}2}({\cal R})$ where $\sigma$ is the flip $(\sigma
(x{\otimes}y)=y{\otimes}x)$ satisfies the quantum Yang–Baxter equation (QYBE) $$S^{12}S^{23}S^{12}=S^{23}S^{12}S^{23},\,\, S^{12}=S{\otimes}{\rm id},\, S^{23}={\rm id}{\otimes}S$$ and possesses two eigenvalues (we recall that $g=sl(n)$).
Let us identify now the space $W$ with $V{\otimes}V^*$ and equip it with the operator $S_W=S{\otimes}(S^*)^{-1}:{W^{\otimes 2}}\to{W^{\otimes 2}}$, where $S^*$ is defined with respect to the pairing $$\langle x\otimes y, a\otimes b\rangle =\langle x,a\rangle\langle y,b\rangle,
\,\,x,y\in V,\,\,a,b\in V^{*}.$$
Let us fix a base $\{a_i\}\in V$. Let $\{a^i\}\in V^*$ be the dual base and set $a_i^j=a_i\otimes a^j$. Then we have the following explicit form for the operator $S_W$: $$S_{W}(a_{i}^{k}{\otimes}a_{j}^{l})=S^{mn}_{ij}{S^{-1}}^{kl}_{pq}
(a_{m}^{p}{\otimes}a_{n}^{q}) \,\,{\rm where}\,\,
S(a_{i}{\otimes}a_{j})=S_{ij}^{kl}a_{k}{\otimes}a_{l}.$$
It is easy to see that the operator $S_W$ also satisfies the QYBE and possesses 1 as an eigenvalue. Then it is natural to introduce deformed analogues of symmetric and skew-symmetric subspaces of the space ${W^{\otimes 2}}$ as follows $$I_{-}^{q}={\rm Im}(S_{W}-{\rm id}),\,\,I_{+}^{q}={\rm Ker}(S_{W}-{\rm id}).$$ Let us describe explicitly the operator $S$ and the spaces $I_{\pm}$: $$S(a_i\otimes a_j)=(q-1)\delta_{i,j}a_i\otimes a_j+a_j\otimes a_i+
\sum_{i<j}(q-q^{-1})a_i\otimes a_j,$$ $$I_-^q={\rm Span}(a_{k}^{i}a_{k}^{j}- qa_{k}^{j}a_{k}^{i},\,
a_{i}^{k}a_{j}^{k}-qa_{j}^{k}a_{i}^{k},\, i<j;$$ $$a_{i}^{l}a_{k}^{j}-a_{k}^{j}a_{i}^{l},\,a_{i}^{j}a_{k}^{l}-a_{k}^{l}a_{i}^{j}
-(q-q^{-1})a_{k}^{j}a_{i}^{l},\, i<k,j<l)$$ and $$I_+^q={\rm Span}((a_i^k)^2,\,q a_{k}^{i}a_{k}^{j}+a_{k}^{j}a_{k}^{i},\,
qa_{i}^{k}a_{j}^{k}+a_{j}^{k}a_{i}^{k},\, i<j;$$ $$a_{i}^{j}a_{k}^{l}+a_{k}^{l}a_{i}^{j},\,
a_{i}^{l}a_{k}^{j}+a_{k}^{j}a_{i}^{l}+(q-q^{-1})a_{i}^{j}a_{k}^{l},\,
i<k,j<l).$$
Let us introduce also the quotient algebra $A_{0,q}=T(W)/\{I_{-}^{q}\}$. In what follows $T(W)$ denotes the free tensor algebra of the space $W$ and $\{I\}$ denotes the two-sided ideal generated by the set $I\subset T(W)$ (here $I=I_-^q$ is a subspace of the space ${W^{\otimes 2}}$).
It is well known (cf. [@RTF]) that the quantum analogue ${\rm Fun}_q(SL
(n))$ of the space ${\rm Fun}(SL(n))$ is the quotient algebra of $A_
{0,q}$ over the ideal generated by the element $\det_q-1$ ($\det_q$ is the so-called “quantum determinant”). Moreover, the latter quotient algebra can be endowed with a Hopf structure. However, we are interested rather in the algebra ${A_{0,q}}$ itself. This algebra is quadratic and is a flat deformation of the classical counterpart, namely, of the symmetric algebra ${\rm Sym}(W)$ of the space $W$ (cf. [@DS2] for details). Let us note that the skew-symmetric algebra of $W$ has also an evident quantum analogue $T(W)/\{I_+^q\}$.
Now we want to introduce two parameter deformation of the algebra ${\rm Sym}
(W)$ quantizing the whole P.p. under consideration. This deformation can be realized by means of the same shift operator as above. Applying this operator to $I^q_-$ we obtain the following space $$J_{h,q}={\rm Span}(a_{k}^{i}a_{k}^{j}- qa_{k}^{j}a_{k}^{i}-
h(\delta_{k}^{i}\,a_{k}^{j}+a_{k}^{i}\,\delta_{k}^{j}),$$ $$a_{i}^{k}a_{j}^{k}-qa_{j}^{k}a_{i}^{k}-
h(\delta_{i}^{k}\,a_{j}^{k}+a_{i}^{k}\,\delta_{j}^{k}),\,\,i<j;\;
a_{i}^{l}a_{k}^{j}-a_{k}^{j}a_{i}^{l},$$ $$a_{i}^{j}a_{k}^{l}-a_{k}^{l}a_{i}^{j}-(q-q^{-1})a_{k}^{j}a_{i}^{l}-
hm\,(\delta_{i}^{l}\,a_{k}^{j}+a_{i}^{l}\,\delta_{k}^{j}),\, i<k,j<l),$$ where $m=1+q^{-1}$ (we have realized here a substitution $h(q-1)\to h$).
Let us introduce the algebra $A_{h,q}=T(W)/\{J_{h,q}\}$. It is evident that this algebra is a flat deformation of the initial commutative algebra $A_{0,
1}$ since the passage from the algebra ${A_{0,q}}$ to that ${A_{h,q}}$ is trivial from the deformation point of view and reduces to a change of a base. Thus, we have constructed the a.p. ${A_{h,q}}$, which is a quantum analogue of the P.p. under consideration (it is easy to see that quasi-classical term of this two parameter deformation is just the above second type P.p.).
A similar method can be applied to quantize the P.p. generated by the elliptic Sklyanin P.b [@S]. This bracket (denoted also by $\{\,\,,\,\,\}_2$) is determined by the following multiplication table $$\{S_{1}\,S_{0}\} = 2J_{23}S_{2}S_{3},\;\, \{S_{1}\,S_{2}\} = -2S_{0}S_{1}$$ and their cyclic permutations with respect to the indices $(1,2,3)$ with some elliptic functions $J_{ij}$ (cf. [@S]). The linearization of this bracket defined by the shift operator $S_0\to S_0+h,\, S_i\to S_i, i\not=0$ ($S_0$ plays here the role of the element $a_0$ from Remark 1) gives rise to the linear one corresponding to the Lie algebra $g=so(3)\oplus k$. The brackets are compatible (by the same reason as above).
To quantize the P.p. generated by them, it suffices to quantize the bracket ${\{\,\,,\,\,\}_2}$ and apply the shift operator. It is well known (cf. [@S]) that a quantum analogue of the bracket ${\{\,\,,\,\,\}_2}$ is the quotient algebra $T(W)/\{I\}$ where $W={\rm Span}(S_0,....,S_3)$ and $\{I\}$ is two-sided ideal in $T(W)$ generated by the elements $$S_1 S_0-S_0 S_1+i J_{23}(S_2 S_3 +S_3 S_2),\,\,
S_1 S_2 -S_2 S_1 -i(S_0 S_3 + S_3 S_0)$$ and their cyclic permutations with some elliptic functions $J_{ij}$.
We leave to the reader the explicit description of the resulting two parameter quantum a.p.
The algebra $T(W)/\{I\}$ originally defined by E. Sklyanin is an elliptic analogue of the symmetric algebra of the space $W$. Unfortunately, we do not know any natural elliptic analogue of the skew-symmetric one (it seems very plausible that such analogue does not exist).
[*There are other quantum analogues of symmetric and skew-symmetric algebras of the space $W$ defined by the so-called reflection equations (RE) $$Su_1Su_1=u_1Su_1S,\; u_1=u{\otimes}1, \;u=(u_i^j)$$ where $S$ is a solution of (3). Conjecturally this “RE algebra” is a flat deformation of its classical counterpart (assuming $S$ to be of Hecke type). At least a necessary condition for flatness of a deformation, i.e., the existence of a P.b. as a quasiclassical term of the deformation, is satisfied. The mentioned bracket is also quadratic and admits a linearization. These two brackets are also compatible. We get a quantization of the P.p. generated by them by applying to the above algebra a shift operator (for details the reader is referred to [@I], [@IP], [@G2]).*]{}
Rank 1 quantum orbits
=====================
In what follows we consider the P.p. and their quantum counterparts connected to symmetric orbits. These P.p. were described in the Introduction: they are generated by the KKS bracket and by the so-called R-matrix bracket.
We want to describe the corresponding quantum a.p. explicitly by means of a system of equations. In the present section we recall the method suggested in [@DG1] to look for such systems. Unfortunately, it is valid only for rank 1 symmetric spaces, namely, those of ${SL(n)/S(L(n-1)\times L(1))}$ type.
Let us fix a Lie algebra $g=sl(n)$ and an element $x={\omega}\in{h}^*$, where $h\subset g=sl(n)$ is the Cartan subalgebra (a triangular decomposition of $g$ is assumed to be fixed). We consider ${\omega}$ as an element of ${g}^*$ extending it by zero to the nilpotent subalgebras. Let ${{\cal O}_{{\omega}}}$ be its orbit in ${g}^*$.
It is well known that this orbit is symmetric iff ${\omega}(h_i)=0$ for all $i\not
=i_0$ where $\{h_i=e_{i,i}-e_{i+1,i+1},\,i=0,....,n-1\}$ is the standard basis in $h$. The rank of this symmetric space is $rk({{\cal O}_{{\omega}}})=\min(i_0,n-
i_0)$. We assume here that $rk({{\cal O}_{{\omega}}})=1$. So we have $i_0=1$ or $i_0=n-1$. To the sake of concreteness we set $i_0=1$.
Thus, we have a family ${{\cal O}_{{\omega}}}$ of such orbits parametrized by the value ${\omega}(h_1)\in k$ ($k={\bf C}$ or $k={\bf R}$ corresponding to the case under question). Let us represent the algebra ${\rm Fun}({{\cal O}_{{\omega}}})$ as a quotient algebra of ${\rm Sym}(g)={\rm Fun}({g}^*)$. It is not difficult to show that this algebra can be described by a system of equations quadratic in generators of the space $g$.
Let us describe this system explicitly. Consider the space $V=g$ as an object of the category $g$-Mod. Let ${V^{\otimes 2}}=\oplus {V_{\beta}}$ be a decomposition of ${V^{\otimes 2}}$ into a direct sum of isotypic $g$-modules, where $\beta$ denotes the highest weight (h.w.) of the corresponding module. Let $\alpha$ be h.w. of $g$ as a $g$-module.
Then in the above decomposition there is a trivial module $V_0$ ($\beta=0$), a module $V_{2\alpha}$ of the h.w. $\beta=2\alpha$, and two irreducible modules isomorphic to $g$ itself. One of them belongs to the space $I_+
\subset {V^{\otimes 2}}$ of symmetric tensors and the other one to the space $I_-\subset
{V^{\otimes 2}}$ of skew-symmetric tensors. We will use the notation $V_+\,(V_-)$ for the former (latter) of them.
Note that a similar decomposition is valid for other simple Lie algebras, but for them there exists only one module isomorphic to $g$. It belongs to the skew-symmetric part and we keep the notation $V_-$ for it.
Let us denote by $C$ the generator (called [*Casimir*]{}) of the module $V_0$. Then the orbit ${{\cal O}_{{\omega}}}$ is given by the following system of equations $$I_-=0; {V_{\beta}}=0\,\,\forall\,{V_{\beta}}\subset I_+,\,\,
{V_{\beta}}\not\in\{V_0,\,V_+,\,V_{2\alpha}\};\;C=c_0,\, V_+=c_1 g$$ with some factors $c_0$ and $c_1$ (the latter relation is a symbolic way to write the system which arises if we choose some bases in $g$ and $V_+$ and equate the highest weight element of $V_+$ to that one of $g$ times a factor $c_1$ and consider all the relations that follow).
It is not difficult to find the values of the factors $c_i,\,i=0,1$. It suffices to substitute “the point” ${\omega}$ to this system. (Let us note that in [@DG1] this system was given in a inconsistent form, where the condition $c_1=0$ was assumed.)
A similar procedure can be realized in the category ${U_q(g)}$-Mod. More precisely, we equip the space $V$ with the structure of a ${U_q(g)}$-module deforming the initial $g$-module structure on $V$(there exists a regular way to convert any finite-dimensional $g$-module into a ${U_q(g)}$ one, cf., i.e., [@CP]). Using the comultiplication in the quantum algebra ${U_q(g)}$, we equip the space ${V^{\otimes 2}}$ with an ${U_q(g)}$-module structure and decompose it as above into a direct sum of irreducible ${U_q(g)}$-modules. We will denote these components by ${V_{\beta}^q}$. By $C_q$ we denote a generator of the module $V^q_0$ (we call $C_q$ a [*braided Casimir*]{}).
Then we can impose a similar system of equations by replacing all ${V_{\beta}}$ participating in the system (4) by their quantum (or q-) analogues ${V_{\beta}^q}$ and $C$ by $C_q$. The only problem is: what are the proper values of the factors $c_i(q), \, i=0,1$, which now depend on $q$.
In [@DG1] the following way to find out a proper system of equations was suggested. Let us consider the data $(V,\,I^q, \,\nu_0,\,\nu_1)$, where $I^q=
{V^{\otimes 2}}\setminus V^q_{2\alpha}$ and $\nu_i,\, i=0,1$ are two ${U_q(g)}$-morphisms defined as follows $\nu_0:V^q_0\to k\;(\nu_0(C_q)=c_0)$ and $\nu_1:V^q_+\to
c_1 g$ (all other components of ${V^{\otimes 2}}\setminus V^q_{2\alpha}$ are sent by $\nu_i$ to zero).
Let us also consider the graded quadratic algebra $T(V)/\{I^q\}$ and its filtered analogue $T(V)/\{{I^q_{c_0,c_1}}\}$, where $\{{I^q_{c_0,c_1}}\}$ the ideal generated by the elements $$I^q_-,\, {V_{\beta}}\subset I_+^q,\,\,
{V_{\beta}}\not\in\{V_0,\,V_+,\,V_{2\alpha}\},\;C-c_0(q),\, V_+^q-c_1(q)g$$ (see the end of this Section for the definition of the spaces $I^q_{\pm}$).
Let us note that the algebra $T(V)/\{I^q\}$ is the quantum analogue of the function space on the cone and it is a flat deformation of its classical counterpart. The flatness can be shown, e.g., in the way suggested in [@DG2], where an intertwining operator of the initial commutative product and the deformed one is given. Moreover, the algebra $T(V)/\{I^q\}$ is Koszul for a generic $q$ since it is so for the case $q=1$ by virtue of [@B].
Let us assume now that the above data satisfy the following system $$\begin{aligned}
&&{\rm Im}(\nu_1{\otimes}{\rm id}-{\rm id}{\otimes}\nu_1)(I{\otimes}V\bigcap V{\otimes}I)\subset
I,\\
&&(\nu_1(\nu_1{\otimes}{\rm id}- {\rm id} {\otimes}\nu_1)+
\nu_0 {\otimes}{\rm id} - {\rm id} {\otimes}\nu_0)(I {\otimes}V\bigcap V{\otimes}I)=0,\\
&&\nu_0 (\nu_1{\otimes}{\rm id}- {\rm id} {\otimes}\nu_1)(I {\otimes}V\bigcap V{\otimes}I)=0.\end{aligned}$$ Then by virtue of the PBW theorem in the form of [@BG] we can conclude that the algebra $T(V)/\{{I^q_{c_0,c_1}}\}$ is a flat deformation of the algebra $T(V)/
\{I^q\}$.
Let us note that the above conditions represent a more general form of the Jacobi relation connected to deformation theory. Thus, if the above form of the Jacobi identity is fulfilled, the algebra $T(V)/\{{I^q_{c_0,c_1}}\}$ is a flat deformation of the orbit ${{\cal O}_{{\omega}}}$ (more precisely, of the function algebra on it).
So, the proper quantities $c_i(q),\,i=0,1$, if they exist, can be found from the above equations. However, a priori it is not clear why such quantities exist. Let us assume here that they exist and denote by ${A_{0,q}}$ the corresponding algebra $T(V)/\{{I^q_{c_0,c_1}}\}$. We are interested in its further deformation.
To do this, we begin by discussing the following question: what is the deformational quantization of the KKS bracket? We want to represent the latter quantum object also as a quotient of the enveloping algebra $U(sl(n))
_h$ The index $h$ here means that we have introduced a factor $h$ in the bracket in the definition of the enveloping algebra, i.e., $$U(g)_h=
T(g)/\{xy-yx-h[x,y]\}.$$
There are some factors $c_i(h),\,i=0,1$, now depending on $h$ such that the quotient algebra ${A_{h,0}}=U(sl(n))_h/\{J\}$, where the ideal $\{J\}$ is generated by the family of elements from (4) lying in $I_+$ but with new $c_i(h)$, is a flat deformation of the initial algebra corresponding to the case $h=0$. These factors can be also found by means of the above form of Jacobi identity.
This approach can also be applied in order to get the quantum algebras corresponding to the whole P.p. under consideration, since these algebras are quadratic as well. The only problem is: what is the proper quantum analogue of the algebra $U(sl(n))_h$? Or, in other words, what is a consistent way to introduce a quantum analogue of the Lie bracket?
We will introduce a q-generalization of the ordinary Lie bracket by means of the following
[*Let $g$ be a simple Lie algebra equipped with a ${U_q(g)}$-module structure and ${V_{\beta}^q}$ be the irreducible modules in the category ${U_q(g)}$-Mod entering its tensor square. We call [*q-Lie bracket*]{} the operator $[\,\,,\,\,]_q:{V^{\otimes 2}}\to V$ defined as follows $$[\,\,,\,\,]_q|_{{V_{\beta}}^q}=0\mbox{ for all }{V_{\beta}}^q\not=V_-^q$$ and $ [\,\,,\,\,]_q:V_-^q\to V$ is a ${U_q(g)}$-isomorphism.*]{}
Let us observe that in this way the q-Lie bracket is defined up to a factor. We fix this factor and introduce the q-analogue of the algebra $U(g)_h$ as follows $$U(g)_{h,q}=T(V)/\{{\rm Im}({\rm id}-h[\,\,,\,\,]_q)I_-^q\}.$$ The space $V=g$ equipped with this bracket will be denoted by ${\overline{g}}$ and called [*braided Lie algebra*]{}[^3]. We will also use the notation ${U(\overline{g})}$ for the algebras $U(g)_{1,q}$. Let us note that this is another, as compared with the quantum group ${U_q(g)}$, q-analogue of the enveloping algebra $U(g)$, but the deformation $U(g)\to U(g)_{h,
q}$ is not flat except for the $sl(2)$ case. However, some quotient algebras of this algebra are flat deformations of their classical counterparts.
Let us return to the case $g=sl(n)$ and introduce the first type a.p. ${A_{h,q}}$ as the quotient algebra of $U(sl(n))_{h,q}$ by the ideal $$\{C_q-c_0,\,V_+^q-c_1V,\,I^q_+\setminus(V^q_{2\alpha}\oplus V_0^q\oplus
V_+^q)\}$$ with some factors $c_i(h,q),\,i=0,1$.
In order to look for the consistent factors $c_i(h,q)$, we must only modify the above morphism $\nu_1$ on the “q-skew-symmetric” subspace $I^q_-$ by setting $\nu_1: V_-^q\to h g$ (all other components are still sent by $\nu
_1$ to zero) and verify the above form of Jacobi identity. This provides us with the factors $c_i(h,q)$ ensuring the flatness of the deformation of the function algebra on the orbit ${{\cal O}_{{\omega}}}$ under consideration.
Let us emphasize that the existence of the proper factors $c_i(h,q)$, as well as that of the above factors $c_i(q)$, can be deduced from the paper [@DS1], where a formal quantization of the P.p. under question was considered. In the next Section we discuss another way to look for appropriate factors $c_i(h,q)$.
It remains only to note that the quasiclassical term of two-parameter deformation ${\rm Fun}({{\cal O}_{{\omega}}}) \to {A_{h,q}}$ is just the above P.p. on the orbit ${{\cal O}_{{\omega}}}$ (cf. [@DG1]).
It is worth to note that the product in the resulting quantum algebras ${A_{h,q}}$ is ${U_q(sl(n))}$-invariant. This means that the following property $$X\mu(a{\otimes}b)=\mu(X_1(a){\otimes}X_2(b)),\; a,b\in{A_{h,q}},\; X\in{U_q(sl(n))}$$ is satisfied. (Hereafter we use Sweedler’s notation $X_1{\otimes}X_2$ for ${\Delta}(X)$.) This property follows immediately from the construction of the algebras. This algebra is the inductive limit of finite-dimensional ${U_q(sl(n))}$-modules (moreover, it is multiplicity free, i.e., the multiplicity of each irreducible ${U_q(g)}$-module is $\leq 1$). Thus, this algebra belongs to the category ${U_q(sl(n))}$-Mod.
Let us note that a particular case of the a.p. ${A_{h,q}}$ is the algebra ${A_{0,q}}$ arising as the result of quantization of the only R-matrix bracket. This algebra is commutative in the category ${U_q(sl(n))}$-Mod in the following sense. This category is balanced (see [@CP] for the definition). Moreover, for any two finite-dimensional objects $U$ and $V$ of this category there exists an involutive operator ${\widetilde{S}}:U{\otimes}V\to V{\otimes}U$ such that it is a ${U_q(g)}$-morphism, it commutes with $S$ and it is a deformation of the flip.
Then $I^q_{\pm}={\rm Im}({\rm id}\pm{\widetilde{S}})$ with ${\widetilde{S}}:{V^{\otimes 2}}\to{V^{\otimes 2}}$. Using the fact that the algebra ${A_{h,q}}$ can be decomposed into a direct sum of finite-dimensional objects of the category ${U_q(sl(n))}$-Mod we can extend the operator ${\widetilde{S}}$ onto ${A_{h,q}}^{{\otimes}2}$. Then the above mentioned commutativity of the algebra ${A_{0,q}}$ means that the multiplication operator $\mu$ satisfies the following relation $\mu=\mu {\widetilde{S}}$. A proof of this fact can be obtained from [@DS1]. Strictly speaking, just the algebra ${A_{0,q}}$ is the quantum analogue of the orbit under consideration.
Let us note that the above method can be also applied to find the equations describing quantum algebras arising from similar P.p. on certain nilpotent orbits in ${g}^*$ (see the Introduction). Thus, such a.p. related to the highest weight element orbits in ${g}^*$ can be defined by means of the above ideal ${I^q_{c_0,c_1}}$, but with $c_0(h,q)=0$ and with a suitable $c_1(h,q)$.
Modules for first type quantum algebras
=======================================
In the previous section we discussed the first step of quantization procedure. The resulting object of this step is an a.p. represented as a quotient algebras over some suitable ideal. Now we want to consider the second step of quantization, consisting in an attempt to represent the above algebras in certain linear spaces. At this step the difference between two types of quantum algebras (a.p.) under question becomes clearer.
Moreover, this step provides us with another way to look for consistent factors $c_i(h,q)$. Briefly speaking, this method reduces to the computation of the factor $c_i(h,q)$ on the image of the first type algebras into the space ${\rm End}(V)$.
To do this, we recall a natural way to equip the space of endomorphisms of a ${U_q(g)}$-module with a ${U_q(g)}$-module structure. Let $U^q$ be a (finite-dimensional) ${U_q(g)}$-module and $\rho_q:{U_q(g)}\to {\rm End}(U^q)$ be the corresponding representation. Let us introduce the representation ${\rho^{{\rm End}}_q}:{U_q(g)}\to{\rm End}({\rm End}(U^q))$ by putting $${\rho^{{\rm End}}_q}(a)M=\rho(a_1)\circ M\circ\rho(\gamma(a_2)),\,a\in{U_q(g)}, \,M\in {\rm End}
(U^q).$$ We denote the matrix product by $\circ$, while $\gamma$ is the antipode in ${U_q(g)}$.
We deal with a coordinate representation of module elements. We consider the endomorphisms as matrices and their action is the left multiplication by these matrices.
Let us note that this way of equipping ${\rm End}(U^q)$ with a ${U_q(g)}$-module structure is compatible with the matrix product in it in the following sense: $${\rho^{{\rm End}}_q}(a)(M_1\circ M_2)={\rho^{{\rm End}}_q}(a_1)M_1\circ{\rho^{{\rm End}}_q}(a_2)M_2.$$ This means that $\circ: {\rm End}(U^q)^{{\otimes}2}\to {\rm End}(U^q)$ is a ${U_q(g)}$-morphism.
Now we will introduce a useful notion for constructing a representation theory of the algebras under consideration.
[*Let ${\overline{g}}$ be a braided Lie algebra. We say that a ${U_q(g)}$-module $U^q$ is a [*braided module*]{} or, more precisely, a [*braided $\,{\overline{g}}$-module*]{} if it can be equipped with a structure of a ${U(\overline{g})}$-module and the representation $\rho:U({\overline{g}})\to {\rm
End}(U^q)$ is a ${U_q(g)}$-morphism. We also say that the classical counterpart $U=U^1$ of the ${U_q(g)}$-module $U^q$ allows braiding.* ]{}
A natural way to construct braided modules is given by the following
Let $U^q$ be a ${U_q(g)}$-module. If the decomposition ${\rm
End}(U^q)=\oplus V^q_{\gamma}$ of the ${U_q(g)}$-module ${\rm End}(U^q)$ into the direct sum of irreducible ${U_q(g)}$-modules is such that
[1.]{} it does not contain modules isomorphic to ${V_{\beta}^q}\subset I_-^q$ apart from those isomorphic to $V^q$, where by $V^q$ we denote $g=V$ equipped with the ${U_q(g)}$-module structure,
[2.]{} the multiplicity of the module $V^q$ is $1$,
then $U^q$ can be equipped with a braided module structure [(]{}briefly, braided structure[)]{}.
[*Proof*]{}. By the assumption, there exists a unique ${U_q(g)}$-submodule in ${\rm End}(U^q),$ isomorphic to $V^q$. Consider a ${U_q(g)}$-morphism defined up to a factor $$\rho: V^q\to {\rm End}(U^q).$$ This map is an almost representation of the braided Lie algebra ${\overline{g}}$ in the sense of the following
*We say that a map (5) is an [*almost representation*]{} of the braided Lie algebra ${\overline{g}}$ if it is a ${U_q(g)}$-morphism and the following properties are satisfied*
1\. $\circ{\rho^{\otimes 2}}({V_{\beta}^q})=0$ for all ${V_{\beta}^q}\subset I_-^q$ apart from that $V^q_-$,
2\. $\circ{\rho^{\otimes 2}}V_-^q=\nu\rho[\,\,,\,\,]_qV^q_-$ with some $\nu\not=0$.
In a more explicit form these conditions can be reformulated as follows. If the elements $\{b^{i,j}_{k,\beta}u_iu_j, 1\leq k\leq\dim\, {V_{\beta}^q}\}$ form a basis of the space ${V_{\beta}^q}\subset I_-^q$ and similarly the elements $\{b^{i,
j}_{k,-}u_iu_j, 1\leq k\leq\dim\, V^q_-\}$ form a basis of the space $V_-^q$ then $$b^{i,j}_{k,\beta}\rho(u_i)\rho(u_j)=0\,\,{\rm if}\, \,{V_{\beta}^q}\not=V_-^q\, \,{\rm
and}\,\, b^{i,j}_{k,-}\rho(u_i)\rho(u_j)=\nu b^{i,j}_{k,-}\rho([u_i,u_j]_q).$$
Let us complete the proof. The image of the composed map $\circ{\rho^{\otimes 2}}(V^q_-)$ is isomorphic to the ${U_q(g)}$-module $V^q$ since $\rho$ and $\circ$ are ${U_q(g)}$-morphisms. Such a module in the decomposition of ${\rm End}(U^q)$ is unique, therefore the image of the space $[\,\,,\,\,]_q V^q_-=V^q$ with respect to the morphism $\rho$ coincides with the previous one (it suffices to show that the images of the highest weight element of the module $V_-^q$ with respect to both operators coincide up to a factor). This gives the second property of Definition 2 (the property that $\nu\not=0$ for a generic $q$ follows from the fact that this is so for $q=1$).
The first property of the Definition follows from the fact that ${\rm End}
(U^q)$ does not contain any modules isomorphic to ${V_{\beta}^q}\subset I^q_-,\, {V_{\beta}^q}\not=V_-^q$. Finally, changing the scale, i.e., considering the map $\rho_
{\nu}=\nu^{-1}\rho$ instead of $\rho$, we get a representation of the algebra ${U(\overline{g})}$. This completes the proof. $\Box$
A natural question arises: how many ${U_q(g)}$-modules can be converted into braided ones or, in other words, how many $g$-modules allow braiding? The answer to this question for the $sl(n)$-case is given by the following proposition, which can be proved by straightforward computations using Young diagram techniques.
Let $\omega$ be a fundamental weight of the Lie algebra $sl(n)$. Then the $sl(n)$-modules ${V_{k\omega}}$ [*(*]{}for any nonnegative integer $k$[*)*]{} allow braiding. In other words, their q-analogues ${V_{k\omega}}^q$ are braided modules.
For other simple Lie algebras, $g$ it seems very plausible that a similar statement is valid for fundamental weights $\omega$ such that their orbits in ${g}^*$ are symmetric (in the $sl(n)$-case all fundamental weights satisfy this condition). Note that all orbits of such type have been classified by E. Cartan (cf. [@KRR]).
Let us now discuss how the braided modules can be used to find the above factors $c_i(h,q)$.
Once more we set $g=sl(n)$ and we take as fundamental the weight ${\omega}$ such that one ${\omega}(h_1)=1,\, {\omega}(h_i)=0,\, i>1$. Then we consider the $U(sl
(n))$-module ${V_{k\omega}^q}$ (which is in fact ${V_{k\omega}}$ equipped with a representation $\rho_q :{U_q(g)}\to {\rm End}({V_{k\omega}}^q)$). Let us realize a braiding of the modules, i.e., construct a $U_q(g)$-morphism $$\rho:U(g)_{h,q}\to {\rm End}({V_{k\omega}^q})$$ in the way described above for the algebra $U({\overline{g}})=U(g)_{1,q}$.
The representation $\rho$ is factorized to a representation of the algebras ${A_{h,q}}$ with certain $c_i(h,q)$. [*(*]{}This means that $\rho({V_{\beta}^q})=0$ if ${V_{\beta}^q}\subset I^q_+\setminus(V^q_{2\alpha}
\oplus V_0^q \oplus V_+^q)\}$, $\rho(C_q)=c_0(h,q) {\rm id}$, and $\rho(V_+
^q)=c_1(h,q)\rho(V^q)$.[*)*]{}
[*Proof.*]{} It suffices to show that the modules belonging to $I^q_+
\setminus(V^q_{2\alpha}\oplus V_0^q\oplus V_+^q)\}$ are not represented in ${\rm End}({V_{k\omega}}^q)$ for any $k$ and that multiplicity of the modules $V^q_0$ and $V_+^q$ in ${\rm End}({V_{k\omega}}^q)$ is 1. This can be done by straightforward calculations by means of Young diagram techniques.
Finally, the factors $c_i(h,q)$ are defined by these relations. Now if we want to find the relations defining the “quantum orbits” (i.e. algebras corresponding to the case $h=0$) we pass to the limits $k\to\infty$ and $h\to
0$ in such a way that $c_0(h,q)$ has a limit (denoted by $c_0(q)$). By virtue of [@DS1] $c_1(h,q)$ also has a limit (denoted by $c_1(q)$). These two constants are just the factors that we are looking for. $\Box$
It is interesting to emphasize that to find the system of equations describing a commutative algebra in the category ${U_q(g)}$-Mod in the framework of this approach, we looking first for the system corresponding to the noncommutative algebra ${A_{h,q}}\,(h\not=0)$.
On “twisted quantum mechanics”
==============================
In this section we discuss the so-called twisted quantum mechanics. Roughly speaking, it is a quantum mechanics in twisted categories. Our aim is to consider the above quantum algebras from this point of view.
It is well known (De Wilde–Lecomte–Fedosov) that any symplectic Poisson structure can be quantized and the result of quantization can be treated as an operator algebra ${\rm End}(V)$ in a complex Hilbert space $V$. The space ${\rm End}(V)$ is equipped in this case with a trace and a conjugation (involution) possessing the usual properties $${\rm tr}(A\circ B)={\rm tr}(B\circ A),\;{\rm tr}\,A^*={\rm tr}\, A,\;
(A\circ B)^*=B^*\circ A^*.$$
The quantum observables are identified with Hermitian (self-adjoint) operators in this space and they form a linear space over ${\bf R}$ closed with respect to the bracket $i[\,\,,\,\,]$.
If we consider a super-version of quantum mechanics, this bracket must be replaced by its super-analogue. The notions of ordinary trace and conjugation operators must be also replaced by their super-counterparts; this leads, in particular, to a modification of the notion of self-adjoint operators. Meanwhile, the observables that play the role of Hamiltonians must be even (since only for an even operator $H$ the equation $dA/dt=i[H,A]$ is consistent) and such an operator is self-adjoint in the usual sense iff it is super-self-adjoint. Moreover, the super-bracket with such an operator becomes the ordinary Lie bracket.
What is a proper analogue of this scheme in a twisted category? If a category under question is symmetric, i.e., the Yang–Baxter twist $S$ is involutive $(S^2={\rm id})$ in it, the corresponding generalization was suggested in [@GRZ]. In this case “S-analogues” of Lie bracket, trace and conjugation operators introduced there satisfy the following version of the above relations $${\rm tr}(A\circ B)={\rm tr}\circ S(A{\otimes}B),\;{\rm tr}\,A^*={\rm tr}\, A,\;
(A\circ B)^*=\circ(*{\otimes}*)S(A{\otimes}B).$$
Let us recall a construction of a conjugation operator in this case (assuming $V$ to be finite-dimensional). Let us suppose that we can identify $V$ and $V^*$ (this means that there exists a pairing $\langle\,\,,\,\,\rangle:{V^{\otimes 2}}\to
k$, which is a morphism of the category). Then ${\rm End}(V)$ can be identified with ${V^{\otimes 2}}$ and the conjugation $*:{\rm End}(V)\to{\rm End}(V)$ is just the image of the operator $S:{V^{\otimes 2}}\to {V^{\otimes 2}}$ under this identification. Therefore the conjugation satisfies the following relation $(*{\otimes}{\rm id})S=
S({\rm id}{\otimes}*)$.
More precisely, we consider the space $g=V$ over the field $k={\bf R}$ and assume all matrix elements of $S$ to be real. This implies that if we introduce an S-Lie bracket in ${\rm End}(V)$ by $$[A,B]=A\circ B-\circ S(A{\otimes}B)$$ its structural constants are real as well.
Let us extend this bracket to the space $V_{{\bf C}}=V{\otimes}{\bf C}$ by linearity. Let $*: V_{{\bf C}}\to V_{{\bf C}}$ be a conjugation, i.e., an involutive $(*^2={\rm id})$ operator such that $(\lambda z)^*=\overline{
\lambda}z^*,\,\lambda\in {\bf C},\, z\in V_{{\bf C}}$. Assume also that the relation $(*{\otimes}{\rm id})S=S({\rm id}{\otimes}*)$ and the third relation (6) are satisfied. Then it is easy to see that this conjugation is compatible with the S-Lie bracket in the following sense $$*[A,B]=-[A^*,B^*].$$
It is not so evident what is the proper definition of a conjugation operator, say, in the algebra ${\rm End}(V)$, where $V$ is an object of a twisted but nonsymmetric category. Now possessing a q-analogue of the Lie bracket, we can try to define the compatibility of such an operator with the twisted structure by means of relation (7). (Let us note that the operator $S$ does not enter explicitly in it and formula (7) is in some sense universal.)
Let us consider a space $V_{{\bf C}}$ equipped with a bracket $[\,\,,\,\,]:V_{{\bf C}}^{{\otimes}2}\to V_{{\bf C}}$.
[*We say that a conjugation $*$ is [*compatible*]{} with this bracket if the relation (7) is satisfied.*]{}
The odd elements with respect to this involution [*(*]{}i.e., elements such that $z^*=-z$[*)*]{} form a subalgebra, i.e., the element $[a,\,b]$ is odd if $a$ and $b$ are.
[*Proof*]{}. It is obvious. $\Box$
Therefore the space of “$*$-even” operators is closed with respect to the bracket $i[\,\,,\,\,]$.
In [@DGR], we have classified all such conjugations for ${U_q(sl(2))}$-case. Let us reproduce the final result here, but first represent the multiplication table for q-Lie bracket in some base $\{u,v,w,\}$: $$[u,u]=0,\ [u,v]=-q^2Mu,\ [u,w]=({q+q^{-1}})^{-1}Mv,$$ $$[v,u]=Mu,\ [v,v]=(1-q^2)Mv,\ [v,w]=-q^2Mw,$$ $$[w,u]=-({q+q^{-1}})^{-1}Mv,\ [w,v]=Mw,\ [w,w]=0,$$ Here $M$ is an arbitrary real factor (cf. [@G2] for details).
For a real $q\not=1$ there exist only two conjugations in the space $V_{{\bf C}}$ compatible with q-Lie bracket, namely, that $a^*=
-\overline a$ for any $a\in V_{{\bf C}}$ and the following one $u^*=u,\,
v^*=-v,\,w^*=w$.
Although these conjugations are rather trivial they are, together with quantum traces (their construction in ${\rm End}(V)$, where $V$ is an object of a rigid category, is well known) the ingredients of twisted quantum mechanics in the sense of the following informal definition.
[*We say that an associative algebra is a [*subject of twisted quantum mechanics*]{} if it belongs to a twisted category, it is represented in the space ${\rm End}(V)$ equipped with a twisted Lie bracket, a trace and a conjugation as above and the representation map is a morphism in this category.*]{}
Unfortunately, we cannot give a final axiom system of twisted quantum mechanics for a noninvolutive $S$ (though it seems very plausible that the above relations between the operators under consideration are still valid in the category ${U_q(g)}$-Mod if we replace $S$ by ${\widetilde{S}}$). However, we want to point out the principal difference between the twisted version of quantum mechanics and its classical version: the twisted (quantum) trace must occur in calculations of partition functions.
In the above sense the second type quantum algebras represented in some linear spaces in spirit of the paper [@VS] are not subjects of twisted quantum mechanics. Though the algebra ${A_{0,q}}={\rm Sym}(W)$ itself belongs to the twisted category generated by the space $W$, its modules constructed in [@VS] for $su(2)$ case do not belong to this category.
As for the second type algebras ${A_{h,q}}$ with $h\not=0$, they differ nonessentially from the algebras ${A_{0,q}}$.
This is not the case for the first type algebra ${A_{h,q}}$. The representation theory of the algebras ${A_{h,q}}$ for generic $h$ and that for the case $h=0$ are completely different. From this point of view $h=0$ is a singular point and we disregard it.
Let us note that our first type quantum algebras are subjects of twisted quantum mechanics arising from the quantization of Poisson brackets. However, it is possible to consider similar objects connected to nondeformational solutions of the QYBE (cf. [@G1], [@GRZ]).
Concluding the paper, we would like to formulate two problems: that of calculating the above factors $c_i(h,q)$ (or more generally, giving an exact description of the two parameter deformation of all symmetric orbits in ${g}^*$ for any simple Lie algebra $g$) and the problem of generalizing our approach to infinite-dimensional Lie algebras.
[DGM]{}
R.Börgvad [*Some homogeneous coordinate rings that are Koszul algebras*]{}, alg-geom/951011.
A.Braverman, D.Gaitsgory [*Poincaré-Birkhoff-Witt theorem for quadratic algebras of Koszul type*]{}, hep-th/9411113.
V.Chari , A.Pressley [*A guide to Quantum Groups*]{}, Cambrige University press, 1994
J.Donin, D.Gurevich [*Quantum orbits of R-matrix type*]{}, Lett. Math. Phys. 35 (1995), pp. 263–276.
J.Donin, D.Gurevich [*Some Poisson structures associated to Drinfeld-Jimbo R-matrices and their quantization*]{}, Israel Math. Journal 92 (1995), pp.23-32
J.Donin , D.Gurevich, S. Majid [*R-matrix brackets and their quantization*]{} Ann. Inst. Henri Poincaré 58 (1993), pp. 235-246
J.Donin, D.Gurevich, V.Rubtsov [*Quantum hyperboloid and braided modules*]{}, Proceedings of French-Belgium meeting, 1995, Reims, to appear
J.Donin , S.Shnider [*Quantum symmetric spaces*]{} Journal of pure and applied algebra 100 (1995), pp. 103-115
J. Donin, S.Shnider [*Quasi-associativity and flatness criteria for quadratic algebra deformation*]{}, Israel Math. J., to appear
D.Gurevich [*Algebraic aspects of the quantum Yang-Baxter equation*]{}, Leningrad Math.J. 2 (1991), pp.801–828.
Gurevich D. [*Braided modules and reflection equations*]{}, Publications of Banach Center, Warsaw, to appear
D.Gurevich, D.Panyushev [*On Poisson pairs associated to modified R-matrices*]{}, Duke Math. J. 73 (1994), pp.249–255.
D.Gurevich, V.Rubtsov, [*Quantization of Poisson pencils and generalized Lie algebras*]{}, Teor. i Mat. Phys. 103 (1995), pp. 476–488.
D.Gurevich , V.Rubtsov, N.Zobin [*Quantization of Poisson pairs: R-matrix approach*]{}, Journ.Geom.and Phys. 9 (1992), pp.25–44
A.Isaev [*Interrelation between Quantum Groups and Reflection Equation (Braided) Algebras*]{}, Lett. Math. Phys. 34 (1995), pp. 333–341.
A.Isaev, P.Pyatov [*Covariant differential complex on quantum linear groups*]{}, J. Phys. A: Math. Gen. 28 (1995), pp. 2227–2246.
S.Khoroshkin, A.Radul, V.Rubtsov[*Families of Poisson structures on Hermitean Symmetric Spases*]{}, Comm. Math. Phys. 153 (1993), pp.299-315
N.Reshetikhin, L.Takhtadzhyan, L. Faddeev [*Quantization of Lie groups and Lie algebras*]{}, Leningrad Math.J. 1 (1990), pp.S193–226
M. Semenov-Tian-Shansky [*What is classical r-matrix?*]{}, Funct. Anal. Appl. 17 (1883), pp.S259–272
E.Sklyanin [*Some algebraic structures connected with the Yang-Baxter equation*]{} Funct. Anal. Appl. 16 (1982), pp.S263-270
L.Vaksman, Y.Soibelman [*Algebra of functions on the quantum qroup $su(2)$*]{}, Funct. Anal. Appl. 22 (1988), pp.S170-181
[^1]: Let us specify that this bracket is given by $$\{\,\,,\,\,\}_S=\{\,\,,\,\,\}_l-\{\,\,,\,\,\}_r,\;\{f,g\}_{{\epsilon}}=
\mu\langle\rho_{{\epsilon}}^{{\otimes}2}(R),df{\otimes}dg\rangle,\,{\epsilon}=l,r,$$ where $\rho_{l}\,(\rho_{r}):g\to {\rm Vect}({\rm Mat}(n))$ is a representation of the Lie algebra $g$ into the space of right (left) invariant vector fields on ${\rm Mat}(n)$, $R$ is the classical R-matrix (1) corresponding to $sl(n)$ and $\langle\,\,,\,\,\rangle$ is the pairing between vector fields and differentials extended to the their tensor powers. Hereafter $\mu$ is the multiplication in the algebra under consideration. (Let us note that for other simple Lie groups $G$ analogous P.b. being extended to ${\rm Mat}(n)$ are no longer Poisson.)
[^2]: Let us note that similar (first type) P.p. exists on certain nilpotent orbits in $g
^*$. These pencils are generated by the KKS bracket and the so-called R-matrix bracket defined by $$\{\,\,,\,\,\}_{R}=\mu\langle\rho^{{\otimes}2}(R),df{\otimes}dg\rangle,\;\rho={\rm ad}^*.$$ In fact the R-matrix bracket is just that $\{\,\,,\,\,\}_{l}^{{\rm red}}$ for any such orbit but only for the symmetric ones the brackets KKS and $\{\,\,,
\,\,\}_{r}^{{\rm red}}$ are proportional to each other. The reader is referred to [@GP] where all orbits in $g^*$ possessing the above P.p. are classified.
[^3]: Other definitions of q-counterparts of the Lie algebras have been suggested recently in [@DH] and [@LS]. It is very plausible that they are equivalent to ours.
| {
"pile_set_name": "ArXiv"
} | ArXiv |
---
abstract: 'Let $M=\mathcal{W}\cup_\mathcal{T} \mathcal{V}$ be an amalgamation of two compact 3-manifolds along a torus, where $\mathcal{W}$ is the exterior of a knot in a homology sphere. Let $N$ be the manifold obtained by replacing $\mathcal{W}$ with a solid torus such that the boundary of a Seifert surface in $\mathcal{W}$ is a meridian of the solid torus. This means that there is a degree-one map $f\colon M\to N$, pinching $\mathcal{W}$ into a solid torus while fixing $\mathcal{V}$. We prove that $g(M)\ge g(N)$, where $g(M)$ denotes the Heegaard genus. An immediate corollary is that the tunnel number of a satellite knot cannot be smaller than the tunnel number of its pattern knot.'
address: |
Department of Mathematics\
Boston College\
Chestnut Hill, MA 02467
author:
- Tao Li
title: 'Heegaard genus, degree-one maps, and amalgamation of 3-manifolds'
---
[^1]
Introduction {#Sintro}
============
Degree-one maps are fundamental objects in topology. For 3-manifolds, such maps have close relations with the geometrization of 3-manifolds as well as many topological properties, e.g. see [@BW; @HWZ; @Rong; @So; @WZ]. It has been known for a long time that maps of nonzero degree between surfaces are standard [@E]. However, many important questions remain open for maps between 3-manifolds. One of the most fundamental questions on degree-one maps between 3-manifolds is the relation between their Heegaard genera.
\[Qgenus\] Let $M$ and $N$ be closed orientable 3-manifolds and suppose there is a degree-one map $f\colon M\to N$. Then $g(M)\ge g(N)$, where $g(M)$ is the Heegaard genus of $M$.
Conjecture \[Qgenus\] is an old and difficult question in 3-manifold topology. It implies the Poincaré Conjecture: If a closed 3-manifold $N$ is homotopy equivalent to $S^3$, since a homotopy equivalence is a degree-one map, Conjecture \[Qgenus\] implies that $0=g(S^3)\ge g(N)$ and $N$ must be $S^3$.
There is a general strategy of proving Conjecture \[Qgenus\] dated back to Haken and Waldhausen. Suppose that $f\colon M\to N$ is a degree-one map. Let $N=\mathcal{W}_N\cup \mathcal{V}_N$ be a minimal genus Heegaard splitting of $N$, where $\mathcal{W}_N$ and $\mathcal{V}_N$ are genus-$g$ handlebodies. Let $\mathcal{V}=f^{-1}(\mathcal{V}_N)$ and $\mathcal{W}=f^{-1}(\mathcal{W}_N)$. By a theorem of Haken [@H1] and Waldhausen [@W1] (also see [@RW]), after some homotopy on the degree-one map $f$, we may assume that $f|_{\mathcal{V}}\colon \mathcal{V}\to \mathcal{V}_N$ is a homeomorphism. So we have $M=\mathcal{W}\cup \mathcal{V}$, where $\partial \mathcal{W}=\partial \mathcal{V}=\mathcal{W}\cap \mathcal{V}$, $\mathcal{V}$ is a genus-$g$ handlebody, and $f|_{\mathcal{V}}$ is a homeomorphism.
Consider $\mathcal{W}$, $\mathcal{W}_N$ and the map $f|_\mathcal{W}\colon \mathcal{W}\to \mathcal{W}_N$. Let $\mathcal{T}=\partial \mathcal{W}=\partial \mathcal{V}$. For any compressing disk $D$ in the handlebody $\mathcal{W}_N$, after some homotopy on $f|_\mathcal{W}$, we may assume that $f^{-1}(D)$ is an incompressible surface in $\mathcal{W}$. Thus there is a collection of $g$ non-separating simple closed curves $\gamma_1,\dots,\gamma_g$ in $\mathcal{T}$, such that
1. $\mathcal{T}-\cup_{i=1}^g \gamma_i$ is connected and
2. each $\gamma_i$ is the boundary of an incompressible surface in $\mathcal{W}$ and these incompressible surfaces are disjoint.
Note that, given any 3-manifold $\mathcal{W}$ with connected genus-$g$ boundary, if $\mathcal{W}$ satisfies these two conditions, one can always construct a degree-one map $f\colon \mathcal{W}\to \mathcal{W}_N$ by pinching each incompressible surface into a disk and pinching the complement of the incompressible surfaces into a 3-ball. So we can formulate the question in Conjecture \[Qgenus\] as a question about surgery.
\[Qgenus3\] Suppose $M=\mathcal{W}\cup_\mathcal{T} \mathcal{V}$ with $\mathcal{T}=\partial \mathcal{W}=\partial \mathcal{V}=\mathcal{W}\cap \mathcal{V}$ a genus-$g$ surface. Suppose $\mathcal{W}$ satisfies the two conditions above. Let $N$ be the closed 3-manifold obtained by replacing $\mathcal{W}$ with a genus-$g$ handlebody $H$ such that each $\gamma_i$ in the conditions above bounds a disk in $H$. Then $g(M)\ge g(N)$.
Let $M$ and $N$ be the 3-manifolds in Conjecture \[Qgenus3\]. There is a degree-one map $f\colon M\to N$ pinching $\mathcal{W}$ into a handlebody while fixing $\mathcal{V}$. Thus Conjecture \[Qgenus3\] follows from Conjecture \[Qgenus\]. Conversely, if Conjecture \[Qgenus3\] holds, then by the theorem of Haken [@H1] and Waldhausen [@W1] explained earlier, Conjecture \[Qgenus\] holds. So the two conjectures are equivalent and a key to understanding a degree-one map between two closed 3-manifolds is the pinching map $f\colon \mathcal{W}\to \mathcal{W}_N$ which pinches $\mathcal{W}$ into a handlebody.
In this paper, we study the genus one case, i.e. the case that $\mathcal{T}$ is a torus. Let $\mathcal{W}$ be the exterior of a knot in a homology sphere. The Seifert surface of the knot gives a non-separating surface in $\mathcal{W}$. Hence $\mathcal{W}$ satisfies the two conditions above. We prove that Conjecture \[Qgenus3\] holds if $\mathcal{W}$ be a knot exterior in a homology sphere.
\[Ttorus\] Let $M=\mathcal{W}\cup_\mathcal{T} \mathcal{V}$, where $\mathcal{W}$ is the exterior of a knot in a homology sphere and $\mathcal{T}$ is a torus. Let $N=\widehat{\mathcal{T}}\cup_\mathcal{T} \mathcal{V}$ be the manifold obtained by replacing $\mathcal{W}$ with a solid torus such that the boundary of a Seifert surface in $\mathcal{W}$ is a meridian of the solid torus. Then $g(M)\ge g(N)$.
As mentioned above, the degree-one map $f\colon \mathcal{W}\to \widehat{\mathcal{T}}$ from $\mathcal{W}$ to a solid torus $\widehat{\mathcal{T}}$ extends to a degree-one map from $M$ to $N$. Thus Theorem \[Ttorus\] gives some evidence for Conjectures \[Qgenus\] and \[Qgenus3\]. Moreover, Theorem \[Ttorus\] has some interesting corollaries on the tunnel numbers of satellite knots. Let $k$ be a satellite knot in $S^3$. So there is a knotted solid torus $V\subset S^3$ such that $k$ is a nontrivial knot in $V$. The core curve of the solid torus $V$ is the companion knot for $k$. If we re-embed $V$ into an unknotted solid torus in $S^3$, then the image of $k\subset V$ becomes a knot $k'$ in $S^3$ and $k'$ is called the pattern knot for $k$. Note that we can view $\mathcal{W}=S^3\setminus{\mathrm{int}}(V)$ as a knot exterior. If we pinch $S^3\setminus{\mathrm{int}}(V)$ into a solid torus, then the resulting manifold is still $S^3$ and $k$ becomes the pattern knot $k'$ after the pinching operation. Thus the following is an immediate corollary of Theorem \[Ttorus\].
\[C1\] The tunnel number of a satellite knot is larger than or equal to the tunnel number of its pattern knot.
If $k=k_1\# k_2$ is a connected sum of two knots, then a swallow-follow torus is an essential torus in $S^3\setminus N(k)$. So we may view $k$ as a satellite knot with $k_1$ as its companion and $k_2$ its pattern knot. Thus Corollary \[C1\] implies the following theorem of Schirmer [@Schi].
\[C2\] Let $k_1$ and $k_2$ be knots in $S^3$, then $t(k_1\# k_2)\ge\max\{t(k_1), t(k_2)\}$, where $t(k)$ denotes the tunnel number of a knot $k$.
By a theorem in [@L18], there are knots $k_1$ and $k_2$ with $t(k_1\# k_2)=t(k_1)$, so the inequalities in Corollary \[C1\] and Corollary \[C2\] are sharp.
I would like to thank Trent Schirmer for helpful conversations. Some of the arguments in this paper are repackaging of arguments in [@Schi].
Generalized Heegaard splittings {#SGHS}
===============================
For any topological space $X$, we use $\overline{X}$, ${\mathrm{int}}(X)$, $|X|$, and $N(X)$ to denote the closure, interior, number of components, and a small neighborhood of $X$ respectively. If $X$ is a surface, $g(X)$ denotes the genus of $X$, and if $X$ is a 3-manifold, $g(X)$ denotes the Heegaard genus of $X$. Moreover, the genus of a disconnected surface is defined to be the sum of the genera of its components. Throughout the paper, we use $I$ to denote the unit interval $[0,1]$.
Since Heegaard genus is additive under connected sum [@H], we only need to consider the case that $\mathcal{W}$ and $\mathcal{V}$ in Theorem \[Ttorus\] are irreducible. If $\mathcal{W}$ or $\mathcal{V}$ is a solid torus, then Theorem \[Ttorus\] holds trivially. So we suppose neither $\mathcal{W}$ nor $\mathcal{V}$ is a solid torus. As $\mathcal{W}$ and $\mathcal{V}$ are irreducible, this means that the torus $\mathcal{T}$ is incompressible, which implies that $M$ is also irreducible.
A Heegaard splitting $M=H_1\cup H_2$ can be viewed as a handle decomposition of $M$ where $H_1$ is the union of $0$- and $1$-handles and $H_2$ is the union of $2$- and $3$-handles. Scharlemann and Thompson [@ST] observed that one can rearrange these handles and obtain a sequence of nice surfaces. This process is called untelescoping. These surfaces give rise to a decomposition of $M$ into a collection of submanifolds $\mathcal{N}_0,\dots, \mathcal{N}_m$ along surfaces $\mathcal{F}_1,\dots,\mathcal{F}_m$, where $\mathcal{F}_i$ may not be connected. Each $\mathcal{N}_i$ has a Heegaard surface $\mathcal{P}_i$ that decomposes $\mathcal{N}_i$ into two compression bodies $\mathcal{A}_i$ and $\mathcal{B}_i$, see Figure \[Funtel\] for a schematic picture.
The 1-handles of $H_1$ are rearranged as the 1-handles of the $\mathcal{A}_i$’s and the 2-handles of $H_2$ are rearranged as the 2-handles of the $\mathcal{B}_i$’s. This decomposition is called a [**generalized Heegaard splitting**]{}. Note that the Heegaard splittings of the $\mathcal{N}_i$’s can be amalgamated along the $\mathcal{F}_i$’s into the original Heegaard splitting $M=H_1\cup H_2$, see [@Sch]. The genus of the generalized Heegaard splitting is defined to be the genus of splitting $M=H_1\cup H_2$.
[untel.eps]{} (10,-5)[$\mathcal{N}_0$]{} (58,20)[$\dots\dots$]{} (37,-5)[$\mathcal{N}_1$]{} (85,-5)[$\mathcal{N}_m$]{} (24,8)[$\mathcal{F}_1$]{} (51,8)[$\mathcal{F}_2$]{} (71,8)[$\mathcal{F}_m$]{} (14,15)[$\mathcal{P}_0$]{} (40,15)[$\mathcal{P}_1$]{} (87,15)[$\mathcal{P}_m$]{} (4,25)[$\mathcal{A}_0$]{} (32,25)[$\mathcal{A}_1$]{} (77,25)[$\mathcal{A}_m$]{} (17,25)[$\mathcal{B}_0$]{} (43,25)[$\mathcal{B}_1$]{} (90,25)[$\mathcal{B}_m$]{} ![[]{data-label="Funtel"}](untel.eps "fig:"){width="3in"}
Scharlemann and Thompson [@ST] proved that if the Heegaard splitting $M=H_1\cup H_2$ is unstabilized, then one can rearrange the handles so that each $\mathcal{F}_i$ is incompressible in $M$ and each $\mathcal{P}_i$ is a strongly irreducible Heegaard surface of $\mathcal{N}_i$. See [@CG; @He; @L8; @L4] for the definition and some properties of strongly irreducible Heegaard splittings.
Suppose $M=H_1\cup H_2$ is a minimal genus Heegaard splitting of $M$. By [@ST], we may assume each $\mathcal{F}_i$ is incompressible and each $\mathcal{P}_i$ is strongly irreducible. Let $\Sigma$ be the union of all the surfaces $\mathcal{F}_i$’s and $\mathcal{P}_i$’s in the untelescoping. Suppose $M=\mathcal{W}\cup_\mathcal{T} \mathcal{V}$ and $\mathcal{T}$ is an incompressible torus. A theorem of Bachman-Schleimer-Sedgwick [@BSS Lemma 3.3 and Remark 3.4] (also see [@L4; @L14]) says that one can isotope $\Sigma$ to intersect $\mathcal{T}$ nicely.
\[LBSS\] Let $M=\mathcal{W}\cup_\mathcal{T} \mathcal{V}$ and suppose $\mathcal{T}$ is an incompressible torus. Let $\Sigma$ be the a collection of incompressible and strongly irreducible surfaces in the decomposition above. Then $\Sigma$ can be isotoped so that either
1. $\mathcal{T}$ is a component of some $\mathcal{F}_i$, or
2. $\Sigma$ transversely intersects $\mathcal{T}$, and each component of $\Sigma\cap \mathcal{W}$ and $\Sigma\cap \mathcal{V}$ is an essential or a strongly irreducible surface in $\mathcal{W}$ and $\mathcal{V}$ respectively.
\[Rtorus\] If $\mathcal{T}$ is a component of $\Sigma$, then the minimal genus Heegaard splitting of $M$ is an amalgamation of Heegaard splittings of $\mathcal{W}$ and $\mathcal{V}$, see [@Sch]. In particular, $g(M)=g(\mathcal{W})+g(\mathcal{V})-g(\mathcal{T})=g(\mathcal{W})+g(\mathcal{V})-1\ge g(\mathcal{V})$, see [@La2; @L4; @L14]. Since $N=\widehat{\mathcal{T}}\cup_{\mathcal{T}}\mathcal{V}$ is obtained by a Dehn filling on $\mathcal{V}$, we have $g(\mathcal{V})\ge g(N)$. Hence $g(M)\ge g(N)$ and Theorem \[Ttorus\] holds. Thus, to prove Theorem \[Ttorus\], we only need to consider the case that $\Sigma$ transversely intersects $\mathcal{T}$, i.e. the second conclusion of Lemma \[LBSS\].
The general strategy for the proof of Theorem \[Ttorus\] is to build a sequence of surfaces in the solid torus $\widehat{\mathcal{T}}$ according to $\Sigma\cap\mathcal{W}$, such that these surfaces merge with $\Sigma\cap\mathcal{V}$ and yield a generalized Heegaard splitting of $N$ with the same genus.
By Lemma \[LBSS\] and Remark \[Rtorus\], we may assume that the intersection of the torus $\mathcal{T}$ with each compression body in the generalized Heegaard splitting consists of a collection of incompressible annuli. These annuli divide each compression body into submanifolds with certain properties. We describe these properties in the next section.
Relative compression bodies {#Srelative compression body}
===========================
A compression body is a manifold obtained by adding 2- and 3-handles on the same side of $F\times I$, where $F$ is a closed and orientable surface. Next, we allow the surface $F$ to have boundary and add some additional structures.
\[Drelative compression body\] Let $F$ be a compact connected and orientable surface. Let $X$ be the 3-manifold obtained by adding 2- and 3-handles to $F\times I$ along $F\times\{0\}$. Denote $(\partial F)\times I$ by $\partial_v^0 X$. There are two sets of disjoint annuli in $F\times\{1\}$, denoted by $\partial_v^+ X$ and $\partial_v^-X$ (it is possible that $\partial_v^\pm X=\emptyset$), with the following properties:
1. For each annulus $A$ in $\partial_v^-X$, there is a compressing disk $D_A$ such that $\partial D_A\subset F\times\{1\}$, $A\cap\partial D_A$ in a single essential arc in $A$, and $\partial D_A$ does not intersect any other annulus in $\partial_v^-X$. We call $D_A$ a **dual disk** of $A$.
2. Each annulus $A$ in $\partial_v^\pm X$ is assigned a numerical order, denoted by $o(A)$. A dual disk $D_A$ of an annulus $A$ in $\partial_v^-X$ may intersect annuli in $\partial_v^+X$ but it intersects only annuli with order smaller than $o(A)$.
We call the manifold $X$ described above a **relative compression body**. Let $\partial_v X=\partial_v^0X\cup\partial_v^+X\cup\partial_v^-X$ and we call $\partial_vX$ the **vertical boundary** of $X$. Let $\partial_h^+X$ be the closure of $F\times\{1\}\setminus(\partial_v^+X\cup\partial_v^-X)$ and let $\partial_h^-X$ be the closure of $\partial X\setminus(F\times\{1\}\cup \partial_v^0X)$. Denote $\partial_hX=\partial_h^+X\cup\partial_h^-X$. We call $\partial_hX$ the **horizontal boundary** of $X$. Clearly $\partial X=\partial_hX\cup\partial_vX$. Denote $\partial^+X=\partial_h^+X\cup\partial_v^+X$ and call $\partial^+X$ the **positive boundary** of $X$. Similarly, denote $\partial^-X=\partial_h^-X\cup\partial_v^-X$ and call $\partial^-X$ the **negative boundary** of $X$. Thus $\partial X=\partial^+X\cup\partial^-X\cup\partial_v^0X$.
\[Rproperty\] Below are some basic properties of a relative compression body. These properties are similar to the properties of a compression body.
1. Let $\Delta$ be a compressing disk for $\partial_h^+X$. By cutting and pasting, we can construct a set of dual disks for annuli in $\partial_v^-X$ that are disjoint from $\Delta$. This means that if we compress $X$ along $\Delta$, the resulting manifold is still a relative compression body.
2. Although we do not require all the dual disks to be disjoint in Definition \[Drelative compression body\], similar to Property (1), we can perform cutting and pasting on the dual disks and obtain a set of disjoint dual disks for the annuli in $\partial_v^-X$.
3. In Definition \[Drelative compression body\], each annulus $A$ in $\partial_v^-X$ has a dual disk $D_A$. As shown in Figure \[Fmono\](a), by connecting two parallel copies of $D_A$ using a band around $A$, we get a compressing disk $D$ of $X$ with $\partial D\subset\partial^+X$, such that $D$ cuts off a collar neighborhood of $A$ in $X$, see the shaded disk of Figure \[Fmono\](b) for a picture of $D$. Thus, similar to a compression body, if we maximally compress the positive boundary $\partial^+X$, the resulting manifold is a product neighborhood of the negative boundary $\partial^-X=\partial_h^-X\cup\partial_v^-X$.
4. As a converse to Property (3), there is a core graph $G$ properly embedded in $X$ connecting all the components of $\partial^-X$ (similar to a core graph of a handlebody or compression body), such that $X\setminus N(G)$ is a product neighborhood of $\partial^+X$. Moreover, for any nontrivial subgraph $G'$ of a core graph, $X\setminus N(G')$ is a relative compression body.
5. If we add a 1-handle to $\partial_h^+X$ or add a 2-handle to $\partial_h^-X$, the resulting manifold is still a relative compression body. Moreover, if we add a 2-handle along any component of $\partial_v^-X$, the resulting manifold is still a relative compression body.
6. Let $A$ be a component $\partial_v^0 X$. If one adds a 2-handle to $X$ along $A$, then the resulting manifold is still a relative compression body. Conversely, since maximal compressing $\partial^+X$ results in a product neighborhood of $\partial^-X$, for each component of $\partial_h^-X$, there is a vertical arc $\beta$ (of the product structure above) connecting this component of $\partial_h^-X$ to $\partial_h^+X$, such that $X\setminus N(\beta)$ is a relative compression body and the annulus around $\beta$ is a component of $\partial_v^0(X\setminus N(\beta))$.
7. Let $\gamma$ be a properly embedded $\partial$-parallel arc in $X$. Suppose $\partial\gamma\subset\partial_h^+ X$ and $\gamma$ is parallel to an arc $\beta$ in $\partial^+X$. So $\gamma\cup\beta$ bounds a disk. Let $\mathcal{N}(\gamma)=\Delta\times I$ be a tubular neighborhood of $\gamma$ in $X$ and denote $A_\gamma=(\partial\Delta)\times I$. Then the closure of $X\setminus \mathcal{N}(\gamma)$ is a relative compression body, if we set $A_\gamma$ as a component of $\partial_v^-(X\setminus \mathcal{N}(\gamma))$ and assign $A_\gamma$ an order higher than the order of any other annulus (the disk bounded by $\gamma\cup\beta$ determines a dual disk for $A_\gamma$).
[monogon.eps]{} (26,-3)[(a)]{} (73,-3)[(b)]{}
\[Lconnected\] For any relative compression body $X$, $\partial^+X$ is connected.
In Definition \[Drelative compression body\], $\partial^+X$ is the closure of $F\times\{1\}\setminus\partial_v^-X$ and the surface $F$ is connected. The existence of the dual disks for each annulus in $\partial_v^-X$ implies that $\partial_v^-X$ is non-separating in $F\times\{1\}$. Thus $\partial^+X$ is connected.
\[Dstack\] Let $X_1,\dots, X_n$ be a collection of relative compression bodies. Suppose that we can glue $X_1,\dots, X_n$ together by identifying some components of $\partial_h^+X_1,\dots, \partial_h^+X_n$ in pairs and some components of $\partial_h^-X_1,\dots, \partial_h^-X_n$ in pairs via surface homeomorphisms. We would like to emphasize that we only identify positive (resp. negative) boundary to positive (resp. negative) boundary and do not mix positive and negative boundaries. The resulting manifold $\widehat{X}$ is called a **stack** of relative compression bodies, or simply a stack. The vertical boundaries $\partial_v X_1,\dots,\partial_vX_n$ are glued into a collection of annuli and tori in $\partial \widehat{X}$. We call the union of these annuli and tori the vertical boundary of $\widehat{X}$ and denote it by $\partial_v\widehat{X}$. Let $\partial_h\widehat{X}$ be the closure of $\partial\widehat{X}\setminus\partial_v\widehat{X}$, and we call $\partial_h\widehat{X}$ the horizontal boundary of $\widehat{X}$. So $\partial_h\widehat{X}$ consists of components of $\partial_h^\pm X_1,\dots, \partial_h^\pm X_n$ that are not identified to other components.
The components of $\partial_h^\pm X_1,\dots, \partial_h^\pm X_n$ that are identified to other components become surfaces properly embedded in $\widehat{X}$, and we call these surfaces [**horizontal surfaces**]{} in $\widehat{X}$. Let $P$ be a horizontal surface in $\widehat{X}$. We say $P$ is a positive (resp. negative) horizontal surface if $P$ lies in the positive (resp. negative) boundary of some $X_i$.
\[Dgood\] Let $M=\mathcal{W}\cup_\mathcal{T} \mathcal{V}$ be an amalgamation of two compact 3-manifolds $\mathcal{W}$ and $\mathcal{V}$ along a torus $\mathcal{T}$. Let $\Sigma$ be a collection of surfaces in $M$ transversely intersecting $\mathcal{T}$. We say that $\Sigma$ is in **good position** with respect to $\mathcal{T}$ if
1. $\Sigma$ divides $\mathcal{W}$ and $\mathcal{V}$ into two stacks of relative compression bodies, where $\Sigma\cap \mathcal{W}$ and $\Sigma\cap \mathcal{V}$ are the horizontal surfaces in the respective stacks,
2. positive (resp. negative) horizontal surfaces in $\mathcal{W}$ are glued to positive (resp. negative) horizontal surfaces in $\mathcal{V}$ along $\mathcal{T}$.
3. if an annulus $A\subset\mathcal{T}$ is shared by two relative compression bodies $X\subset\mathcal{W}$ and $Y\subset\mathcal{V}$, then $A\subset\partial_v^\pm X$ if and only if $A\subset\partial_v^\mp Y$, and $A$ has the same order in both $X$ and $Y$.
\[Ldivide\] Let $M=\mathcal{W}\cup_\mathcal{T} \mathcal{V}$ be an amalgamation of two compact irreducible 3-manifolds $\mathcal{W}$ and $\mathcal{V}$ along an incompressible torus $\mathcal{T}$. Let $\Sigma$ be the collection of incompressible and strongly irreducible surfaces in an untelescoping of a Heegaard splitting of $M$ as in section \[SGHS\]. Then, after isotopy, either
1. $\mathcal{T}$ is isotopic to a component of $\Sigma$, or
2. $\Sigma$ is in good position with respect to $\mathcal{T}$.
By Lemma \[LBSS\], we may assume $\mathcal{T}$ intersects each compression body in a collection of incompressible annuli. Let $X$ be a compression body in the untelescoping. Consider the set of annuli $\mathcal{T}\cap X$. First, note that no annulus in $\mathcal{T}\cap X$ can have both boundary circles in $\partial_-X$. To see this, if $A$ is a component of $\mathcal{T}\cap X$ with $\partial A\subset\partial_-X$, then since $A$ is incompressible, for any compressing disk $D$ of $\partial_+X$, one can isotope $D$ so that $D\cap A=\emptyset$. Recall that if one maximally compresses $\partial_+X$, the resulting manifold is a product $\partial_-X\times I$. As $\partial A\subset\partial_-X$, this means that $A$ must be a $\partial$-parallel annulus, a contradiction to Lemma \[LBSS\]. Thus each annulus in $\mathcal{T}\cap X$ has at least one boundary curve in $\partial_+X$. If an incompressible annulus has one boundary circle in $\partial_+X$ and the other boundary circle in $\partial_-X$, we call $A$ a spanning annulus (or vertical annulus) in $X$.
Let $A_1,\dots, A_n$ be the annuli in $\mathcal{T}\cap X$ with both boundary curves in $\partial_+X$, and let $C_1,\dots, C_m$ be the spanning annuli in $\mathcal{T}\cap X$ with one boundary curve in $\partial_+X$ and the other boundary curve in $\partial_-X$. Since $X$ is a compression body and $\partial A_i\subset\partial_+X$, each $A_i$ is $\partial$-compressible in $X$. Thus we can assign a $\partial$-compressing disk $\Delta_i$ to each $A_i$, with $\partial\Delta_i=\alpha_i\cup\beta_i$, $\beta_i\subset\partial_+X$, and $\alpha_i$ being an essential arc in $A_i$. Note that $\Delta_i$ may intersect other annuli in $\mathcal{T}\cap X$, but we require that $\Delta_i$ has minimal intersection with $\mathcal{T}\cap X$ among all $\partial$-compressing disks for $A_i$, for each $i$.
For each annulus $E$ in $\mathcal{T}\cap X$, if $E\cap\Delta_i$ contains a closed curve, then since $E$ is incompressible, this curve must be trivial in both $E$ and $\Delta_i$. By compressing $\Delta_i$ along an innermost such closed curve, we get a new $\partial$-compressing disk for $A_i$ with fewer intersection curves with $E$. Similarly, if $E\cap\Delta_i$ contains an arc that is $\partial$-parallel in $E$, then by performing a $\partial$-compression on $\Delta_i$ along an outermost such arc, we get a new $\partial$-compressing disk for $A_i$ with fewer intersection arcs with $E$. Since $\Delta_i$ is chosen to have minimal intersection with $\mathcal{T}\cap X$, this implies that $\Delta_i\cap E$ consists of arcs essential in $E$ for every annulus $E$ in $\mathcal{T}\cap X$ and for each $i$. Since $\partial\Delta_i\cap\partial_-X=\emptyset$, this implies that $\Delta_i\cap C_k=\emptyset$ for all $i$, $k$.
For each arc $\delta$ in $\Delta_i\cap A_k$, $\delta$ is an arc in $\Delta_i$ with both endpoints in $\beta_i$. Moreover, the subarc of $\beta_i$ between the two points of $\partial\delta$, together with $\delta$, bounds a subdisk of $\Delta_i$, which we denote by $\Delta_\delta$. We say that $A_k$ is *coherent* with $\Delta_i$ if (1) these subdisks $\Delta_\delta$ for all the arcs of $\Delta_i\cap A_k$ are non-nested in $\Delta_i$ and (2) each subdisk $\Delta_\delta$ is parallel to the $\partial$-compressing disk $\Delta_k$ of $A_k$.
These $\partial$-compressing disks $\Delta_i$ may be chosen so that, for every $k$, $A_k$ is coherent with all the $\Delta_i$.
Consider an annulus $A_i$ and its $\partial$-compressing disk $\Delta_i$. We say that $A_i$ is outermost if the $\partial$-compressing disk $\Delta_i$ is disjoint from all other annuli in $\mathcal{T}\cap X$. Note that we can choose these $\partial$-compressing disks $\Delta_i$’s so that there is at least one outermost annulus. To see this, suppose $A_i$ is not outermost, then the intersection of $\Delta_i$ with $\mathcal{T}\cap X$ is a collection of arcs with endpoints in $\beta_i$. Let $\delta$ be an outermost such intersection arc in $\Delta_i$. So $\delta$ is an essential arc of some annulus $A_j$ and $\delta$ cuts off a subdisk $\Delta'$ in $\Delta_i$. Note that $\Delta'$ is a $\partial$-compressing disk for $A_j$ and $\Delta'$ is disjoint from all other annuli in $\mathcal{T}\cap X$. By choosing $\Delta_j=\Delta'$, we see that $A_j$ is an outermost annulus.
Without loss of generality, suppose $A_1$ is outermost. As shown in Figure \[Fmono\](a), by connecting two parallel copies of $\Delta_1$ using a band around $A_1$, we get a compressing disk $D_1$ for $\partial_+X$, see the shaded region in Figure \[Fmono\](b) for a picture of $D_1$. Since $\Delta_1$ is disjoint from other annuli in $\mathcal{T}\cap X$, after a small perturbation, we may assume $D_1$ is disjoint from $\mathcal{T}\cap X$. Note that $D_1\cap\Delta_1=\emptyset$. If $D_1$ intersects other $\Delta_i$ ($i=2,\dots,n$), consider an arc $\delta'$ in $D_1\cap(\bigcup_{i=2}^n\Delta_i)$ that is outermost in $D_1$. Suppose $\delta'$ is an arc in $\Delta_i$. Then we can perform a $\partial$-compression on $\Delta_i$ along the subdisk of $D_1$ cut off by $\delta'$. This $\partial$-compression on $\Delta_i$ produces a new $\partial$-compressing disk $\Delta_i'$ for $A_i$. Since $D_1$ is disjoint from $\mathcal{T}\cap X$, $\Delta_i'$ still has minimal intersection with $\mathcal{T}\cap X$ and we can replace $\Delta_i$ with $\Delta_i'$. After finitely many such $\partial$-compressions, we get a new set of $\partial$-compressing disks, which we still denote by $\Delta_1,\dots \Delta_n$, such that $D_1\cap\Delta_i=\emptyset$ for all $i$.
As illustrated in Figure \[Fmono\](b), if we compress $X$ along $D_1$, $A_1$ becomes a $\partial$-parallel annulus in the resulting compression body. Thus, if a $\partial$-compressing disk $\Delta_i$ intersects $A_1$, then the subdisks of $\Delta_i$ cut off by $A_1$ are all parallel copies of $\Delta_1$. This means that $A_1$ is coherent with every $\Delta_i$.
We compress $X$ along $D_1$ and get a new compression body $X_1$. Now we can ignore the annulus $A_1$ (since $A_1$ is $\partial$-parallel in $X_1$) and consider $A_2,\dots, A_n$ and their $\partial$-compressing disks $\Delta_2,\dots, \Delta_n$ in $X_1$. We can find an annulus (without considering $A_1$) that is outermost in $X_1$ and repeat the argument above. Therefore, we can inductively conclude that each $A_k$ is coherent with every $\Delta_i$ for all $k$.
We define an order for each $\Delta_i$ as follows. If $\Delta_i$ is disjoint from other annuli of $\mathcal{T}\cap X$, i.e. $A_i$ is outermost, then set the order $o(\Delta_i)$ to be $0$. Suppose $\Delta_i$ intersects other annuli of $\mathcal{T}\cap X$. Then the intersection consists of arcs with endpoints in $\beta_i\subset\partial\Delta_i$. Each arc, together with a subarc of $\beta_i$, bounds a subdisk of $\Delta_i$. Roughly speaking, the order $o(\Delta_i)$ is the maximal number of such subdisks that are nested to one another. More precisely, for any point $x\in {\mathrm{int}}(\beta_i)$, we draw an arc in $\Delta_i$, denoted by $\gamma_x$, connecting $x$ to $\alpha_i$ ($\alpha_i= A_i\cap\partial\Delta_i$) and count the number of intersection points of ${\mathrm{int}}(\gamma_x)$ with the annuli $\mathcal{T}\cap X$, i.e. the number $|{\mathrm{int}}(\gamma_x)\cap \mathcal{T}|$. First define $o(x)$ to be the minimal number of such intersection points amount all arcs connecting $x$ to $\alpha_i$. This means that if $\Delta_i$ intersects other annuli of $\mathcal{T}\cap X$, there is a point $x\in \beta_i$ with $o(x)\ge 1$. Define the order of $\Delta_i$ to be $o(\Delta_i)=\max_{x\in\beta_i}\{ o(x)\}$. We define the order of the annulus $A_i$ to be $o(A_i)=o(\Delta_i)$. Furthermore, for each annulus $A_i$, we assign a normal vector pointing into the $\partial$-compressing disk $\Delta_i$. We assign $\pm$-signs to the two sides of $A_i$ so that this normal vector pointing from the plus-side to the minus-side (i.e. $\Delta_i$ is on the minus-side of $A_i$).
The annuli of $\mathcal{T}\cap X$ divide $X$ into submanifolds and let $N$ be one of these submanifolds. We will show next that each $N$ is a relative compression body. First note that $\partial N$ has 3 parts:
1. the annuli from $\mathcal{T}\cap X$, which we denote by $\partial_vN$,
2. a subsurface of $\partial_-X$, i.e. $\partial N\cap\partial_-X$, denoted by $\partial_h^-N$, and
3. a subsurface of $\partial_+X$, i.e. $\partial N\cap\partial_+X$, denoted by $\partial_h^+N$.
We also divide the annuli in $\partial_vN$ also into 3 types: $\partial_v^0N$, $\partial_v^+N$ and $\partial_v^-N$, where
1. $\partial_v^0N$ consists of annuli connecting $\partial_+X$ to $\partial_-X$, i.e. $\partial_v^0N$ corresponds to the spanning annuli $C_1,\dots, C_m$,
2. $\partial_v^+N$ consists of annuli $A_i$ with normal direction pointing out of $N$, and
3. $\partial_v^-N$ consists of annuli $A_i$ with normal direction pointing into $N$.
Set the order of each annulus $A_i$ in $\partial_v^\pm N$ to be the order $o(A_i)$ defined for $\mathcal{T}\cap X$ above. Each $\partial$-compressing disk $\Delta_i$ is cut into a collection of subdisks by $\mathcal{T}\cap X$, and let $\Delta_i'$ be the subdisk that contains the arc $\alpha_i=\partial\Delta_i\cap A_i$. So, if $N$ is on the minus side of $A_i$, then the minus side of $A_i$ is an annulus of $\partial_v^-N$, $\Delta_i'\subset N$, and $\Delta_i'$ is the dual disk of $A_i$ in $N$. By the claim above, we may assume each $A_k$ is coherent with each $\Delta_i$. Thus the definition of $o(A_i)$ above implies that the dual disk $\Delta_i'$ satisfies the conditions in Definition \[Drelative compression body\].
Since each annulus of $\partial_v^-N$ has a dual disk, to prove that $N$ is a relative compression body, it suffices to show that, for each region $N$, after a sequence of compressions on $\partial^+N$, we obtain a product neighborhood of $\partial^-N$. We prove this using the fact that $X$ is a compression body, which means that if we maximally compress $\partial_+X$, we obtain a product neighborhood of $\partial_-X$.
Start with $\partial$-compressing disks of order $0$. Let $\Delta_i$ be a $\partial$-compressing disk of order $0$, i.e. $A_i$ is an outermost annulus. Let $\widehat{\Delta}_i$ be the disk obtained by connecting two parallel copies of $\Delta_i$ using a band around $A_i$, as shown in Figure \[Fmono\](a). As in the proof of the claim, we may assume that, after isotopy, $\widehat{\Delta}_i$ does not intersect the $\partial$-compressing disks.
Instead of considering $N$ itself, we consider all the components of $\overline{X\setminus \mathcal{T}}$ at the same time. Denote the two components of $\overline{X\setminus \mathcal{T}}$ on the plus and minus sides of $A_i$ by $N_+$ and $N_-$ respectively. So the two sides of annulus $A_i$ can be viewed as annuli in $\partial_v^+N_+$ and $\partial_v^-N_-$. Moreover $\widehat{\Delta}_i\subset N_-$ and $\widehat{\Delta}_i$ is a compressing disk for both $\partial_h^+ N_-$ and $\partial_+X$.
Now compress $N_-$ along $\widehat{\Delta}_i$. As in the proof of the Claim, after the compression along $\widehat{\Delta}_i$, $A_i$ becomes a $\partial$-parallel annulus in the resulting manifold. So the compressing disk $\widehat{\Delta}_i$ divide $N_-$ into a collar neighborhood of $A_i$, which we denote by $E_i$, and a submanifold with fewer vertical boundary components. Enlarge $N_+$ by including this solid torus $E_i$ into $N_+$ and then delete the annulus $A_i$. Since $E_i$ is a collar neighborhood of $A_i$, this operation does not really change $N_+$. The effect of this operation on $\partial N_+$ is equivalent to merging $A_i$ from $\partial_v^+N_+$ into $\partial_h^+N_+$. We do these operations on all the disks $\Delta_i$ of order 0. Note that the compressions on $N_-$ are also compressions on $\partial_+X$, and these compressions change $X$ into a new (possibly disconnected) compression body, and we can consider the remaining annuli $A_j$’s in the new compression body.
Next, consider $\partial$-compressions disks $\Delta_j$ of order $1$. Since the annuli of order $0$ are deleted, the disks $\Delta_j$ of order $1$ do not intersect other annuli. Thus we can apply the same operations using the disks $\Delta_j$ of order 1, i.e., first compress the compression body along the disk illustrated in Figure \[Fmono\], and then remove the annuli $A_j$ of order 1. We can inductively repeat this operation using these $\partial$-compressing disks. After all these compressions and deleting the annuli $A_k$’s, the remaining annuli in the resulting (possibly disconnected) compression body are a collection of spanning annuli, i.e. the annuli $C_1,\dots C_m$. So we can perform more compressions disjoint from the spanning annuli $C_i$’s and change the compression body into a product $\partial_-X\times I$. By restricting these operations on each component $N$ of $\overline{X\setminus \mathcal{T}}$, we see that if one maximally compresses $\partial^+N$, the resulting manifold is $\partial^- N\times I$. Therefore, $N$ satisfies all the requirements of Definition \[Drelative compression body\] and is a relative compression body. Moreover, it follows from our construction that all the conditions in Definition \[Dgood\] are satisfied.
Next, we prove a converse to Lemma \[Ldivide\].
\[Lmerge\] Let $\widehat{\mathcal{W}}$ and $\widehat{\mathcal{V}}$ be compact 3-manifolds with torus boundary. Suppose there are collections of horizontal surfaces dividing $\widehat{\mathcal{W}}$ and $\widehat{\mathcal{V}}$ into stacks of relative compression bodies. Suppose the boundary curves of these horizontal surfaces of $\widehat{\mathcal{W}}$ and $\widehat{\mathcal{V}}$ divide the tori $\partial \mathcal{W}$ and $\partial \mathcal{V}$ into the same number of annuli $w_1,\dots,w_k$ and $v_1\dots, v_k$ respectively. Let $W_i$ and $V_i$ be the relative compression bodies in the stacks $\widehat{\mathcal{W}}$ and $\widehat{\mathcal{V}}$ containing $w_i$ and $v_i$ respectively (it is possible that $W_i=W_j$ and $V_i=V_j$ for different $i$, $j$). Suppose that
1. $w_i\subset\partial_v^\pm W_i$ if an only if $v_i\subset\partial_v^\mp V_i$, and $o(w_i)=o(v_i)$ for all $i$.
2. $w_i\subset\partial_v^0 W_i$ if and only if $v_i\subset\partial_v^0 V_i$.
Let $M$ be the closed 3-manifold obtained by gluing $\widehat{\mathcal{W}}$ to $\widehat{\mathcal{V}}$ and identifying $w_i$ to $v_i$ for all $i$, and suppose positive (resp. negative) horizontal surfaces in $\widehat{\mathcal{W}}$ are glued to positive (resp. negative) horizontal surfaces of $\widehat{\mathcal{V}}$. Then the union of the horizontal surfaces of $\widehat{\mathcal{W}}$ and $\widehat{\mathcal{V}}$ divides $M$ into a collection of compression bodies.
The proof of this lemma is similar to the latter part of the proof of Lemma \[Ldivide\]. First note that the positive and negative horizontal surfaces in $\widehat{\mathcal{W}}$ and $\widehat{\mathcal{V}}$ match up and yield a collection of positive and negative surfaces in $M$. The relative compression bodies in $\widehat{\mathcal{W}}$ and $\widehat{\mathcal{V}}$ also match up and become a collection of regions between these positive and negative surfaces. Denote these regions by $\mathcal{N}_1,\dots \mathcal{N}_k$. We use $\partial_+\mathcal{N}_i$ and $\partial_-\mathcal{N}_i$ to denote the unions of the components of $\partial \mathcal{N}_i$ that are positive and negative surfaces respectively. By the definition of relative compression body, $\partial_+\mathcal{N}_i\ne\emptyset$ for each $i$.
Our goal is to show that each $\mathcal{N}_i$ is a compression body and $\partial_+\mathcal{N}_i$ and $\partial_-\mathcal{N}_i$ are its plus and minus boundaries respectively.
We view $M=\widehat{\mathcal{W}}\cup_{\mathcal{T}}\widehat{\mathcal{V}}$, where $\mathcal{T}= \partial\widehat{\mathcal{W}}=\partial\widehat{\mathcal{V}}$ is a torus in $M$. Let $X_1,\dots,X_m$ be the relative compression bodies of the stacks $\widehat{\mathcal{W}}$ and $\widehat{\mathcal{V}}$ that lie in $\mathcal{N}_i$. For each annulus $A$ of $\mathcal{T}\cap\mathcal{N}_i$, the two sides of $A$ are viewed as annuli in $\partial_v X_1,\dots,\partial_vX_m$. By the hypotheses, if one side of $A$ is a component of $\partial_v^\pm X_i$, then the other side of $A$ must be a component of $\partial_v^\mp X_j$ with the same order, similar to the conditions in Definition \[Dgood\].
We first consider the annuli in $\partial_v^-X_j$ ($j=1,\dots,m$) and their dual disks. Define the order $o(\Delta_i)$ of each dual disk $\Delta_i$ to be the order of the corresponding annulus. Suppose $\Delta_1,\dots,\Delta_k$ are all the dual disks in the $X_i$’s and suppose $o(\Delta_1)\le o(\Delta_2)\le\cdots\le o(\Delta_k)$. Since $\Delta_1$ has the smallest order among all dual disks in $\mathcal{N}_i$, and by the hypothesis that $o(w_j)=o(v_j)$ for all $j$ (i.e. the orders of the two sides of the same annulus are the same), $\partial\Delta_1$ does not meet any other annulus in $\mathcal{T}\cap\mathcal{N}_i$. Hence, by connecting two parallel copies of $\Delta_1$ using a band as in Figure \[Fmono\](a), we obtain a compressing disk $D_1$ for $\partial_+\mathcal{N}_i$ and $D_1$ is disjoint from the torus $\mathcal{T}$. Similar to the proof of Lemma \[Ldivide\], we may assume $D_1\cap\Delta_j=\emptyset$ for all $j$. Moreover, we can compress $\partial_+\mathcal{N}_i$ along $D_1$ and then push the resulting annulus that is parallel to $A_1$ to the other side of the torus $\mathcal{T}$. Since $A_1$ is eliminated from the compression body after this operation, $\partial\Delta_2$ does not meet any other annulus and we can repeat this operation using $\Delta_2$. So we successively repeat this operation using the disks $\Delta_2,\dots,\Delta_k$, and denote the resulting relative compression bodies by $X_1',\dots, X_m'$. So, after we finish these compressions and isotopies, $\partial_v^\pm X_i'=\emptyset$ for all $i$.
Similar to the proof of Lemma \[Ldivide\], we can perform more compressions on each $\partial_h^+X_i'$ changing the manifold into a product $\partial_h^-X_i'\times I$. These products match up along $\mathcal{T}$ and yield a product $\partial_-\mathcal{N}_i\times I$. This means that if we maximally compress $\partial_+\mathcal{N}_i$ in $\mathcal{N}_i$, the resulting manifold is $\partial_-\mathcal{N}_i\times I$. Hence each $\mathcal{N}_i$ is a compression body.
We end this section with certain conditions that describe the boundary of a relative compression body.
\[Dadmissible\] Let $S$ be a closed orientable surface. Suppose $S$ contains a collection of annuli, which we denote by $\partial_vX$. Denote the closure of $S\setminus\partial_vX$ by $\partial_hX$. Assign each component of $\partial_hX$ a plus or minus sign and use $\partial_h^\pm X$ to denote the union of the components with $\pm$-sign. Assign each component of $\partial_vX$ a $\pm$-sign or no sign. Denote the union of the annuli in $\partial_vX$ with $\pm$-sign by $\partial_v^\pm X$, and denote the union of the annuli in $\partial_vX$ with no sign by $\partial_v^0X$. Moreover, we give each annulus $A$ in $\partial_v^\pm X$ a numerical order, denoted by $o(A)$. Let $\partial^\pm X=\partial_h^\pm X\cup\partial_v^\pm X$. We say that these subsurface of $S$ with the signs and orders are **admissible** if
1. there is no annulus in $\partial_vX$ with both boundary circles in $\partial_h^-X$,
2. each annulus in $\partial_v^0X$ has one boundary circle in $\partial_h^+X$ and the other boundary circle in $\partial_h^-X$,
3. all the annuli in $\partial_v^+X$ and $\partial_v^-X$ have boundary curves in $\partial_h^+X$,
4. $\partial^+X$ is connected.
5. For each annulus $A$ in $\partial_v^-X$, there is an arc $\alpha$ in $\partial^+X$ that connects the two boundary circles of $A$, such that $\alpha$ only intersects annuli in $\partial_v^+X$ with order smaller than $o(A)$.
\[Ladmissible\] Let $X$ be a relative compression body. Then $\partial X$ and the signs and orders in Definition \[Drelative compression body\] are admissible.
By Lemma \[Lconnected\], $\partial^+X$ is connected, so condition (4) of Definition \[Dadmissible\] is satisfied. For any annulus $A$ in $\partial_v^-X$, let $D$ be its dual disk and let $\alpha=\partial D\cap\partial^+X$. By our definition, $\alpha$ satisfies condition (5) of Definition \[Dadmissible\]. Other conditions of Definition \[Dadmissible\] follow directly from Definition \[Drelative compression body\].
In the later sections, for a certain surface $S$ with admissible decompositions as in Definition \[Dadmissible\], we will construct a relative compression body $X$ with $\partial X=S$ and compatible boundary structure as in Lemma \[Ladmissible\].
Marked handlebodies and type I blocks {#StypeI}
=====================================
In this section, we describe some building blocks of the construction. Each building block is a topological handlebody with a special structure.
Before we proceed, we first describe an operation on surfaces that we use many times throughout the paper. Let $S_1$ and $S_2$ be two surfaces in a 3-manifold and let $\alpha$ be an arc connecting $S_1$ to $S_2$. Let $H=D^2\times I$ be a 1-handle along $\alpha$ with $\alpha=\{x\}\times I$, and suppose $H\cap (S_1\cup S_2)=D^2\times\partial I$. Let $S'$ be the surface obtained by first removing the two disks $H\cap (S_1\cup S_2)$ from $S_1$ and $S_2$ and then connecting the resulting surfaces by the annulus $(\partial D^2)\times I$. If $S_1\ne S_2$, we say that $S'$ is obtained by connecting $S_1$ and $S_2$ via a tube along $\alpha$ (topologically $S'\cong S_1\# S_2$). If $S_1=S_2$ and $\alpha$ is a trivial arc (i.e. parallel to an arc in $S_1$), then we say that $S'$ is obtained by adding a trivial tube.
We call a genus-$g$ handlebody $X$ a **marked handlebody** if $\partial X$ contains $m$ ($m\le g$) disjoint essential annuli $A_1,\dots, A_m$ and $X$ contains $m$ disjoint compressing disks $\tau_1,\dots, \tau_m$ such that (1) $\tau_i\cap A_j=\emptyset$ if $i\ne j$, and (2) $\tau_i\cap A_i$ is a single essential arc of $A_i$ for each $i$. These annuli $A_i$’s are called marked annuli in $\partial X$.
By connecting two parallel copies of $\tau_i$ via a band around $A_i$, we can construct a separating compressing disk in $X$ which cuts off a solid torus $T_i$ with $A_i\subset \partial T_i$, see Figure \[Fmono\] for a picture. So we can give an equivalent but slightly different description of $X$: Start with a collection of solid tori $T_1,\dots, T_m$ and suppose $T_i=\tau_i\times S^1$, where $\tau_i$ is a bigon disk. Let $a_i$ and $a_i'$ be the two boundary edges of the bigon $\tau_i$, and let $A_i=a_i\times S^1$ and $A_i'=a_i'\times S^1$ be the pair of annuli in $\partial T_i=\partial\tau_i\times S^1$. Let $B$ be a 3-ball. We first connect each $T_i$ to $B$ using a 1-handle $H_i$ ($i=1,\dots,m$) and require that each 1-handle $H_i$ ($i=1,\dots,m$) is attached to the annulus $A_i'$. Then we add $p$ 1-handles ($p=g-m$) $H_1',\dots, H_p'$ to $B$. The resulting handlebody is $X$, and the set of annuli $A_1,\dots A_m$ are our marked annuli on $\partial X$. We call $B$ in this construction the [*central $3$-ball*]{} of $X$.
The boundary of $X$ has two parts: (1) the vertical boundary $\partial_vX=\bigcup_{i=1}^m A_i$ and (2) the horizontal boundary $\partial_hX=\overline{\partial X\setminus \partial_vX}$.
\[LBM1\] Let $X$ be a marked handlebody. Assign a plus sign to $\partial_hX$ and an arbitrary $\pm$-sign to each annulus in $\partial_vX$. Assign an arbitrary order to each annulus in $\partial_vX$. Then $X$ is a relative compression body.
Let $A_1,\dots,A_m$ be the components of $\partial_vX$. In our definition, $X$ contains a collection of compressing disks $\tau_1,\dots,\tau_m$ such that $\tau_i\cap A_j=\emptyset$ if $i\ne j$, and $\tau_i\cap A_i$ is an essential arc of $A_i$. If $A_i$ has a minus sign, then $\tau_i$ is a dual disk disjoint from all other annuli in $\partial_vX$. In particular, $\tau_i$ is a dual disk for $A_i$ no matter what orders these annuli have. Hence $X$ is a relative compression body.
**Cross-section disks, suspension surfaces, and standard surfaces**:
Let $X$ be a marked handlebody with $T_i=\tau_i\times S^1$, $A_i$, $A_i'$, $B$ and $H_i$ ($i=1,\dots, m$) as above. A [**cross-section disk**]{} is a properly embedded disk in $X$ that cuts through each $H_i$ and each $T_i$, as shown in Figure \[Ftype1\](a). More precisely, a cross-section disk can be described as follows: we view the central 3-ball $B$ in the construction as a product $D\times I$, where $D$ is a disk, and view each 1-handle $H_i$ as a product $\Delta_i\times I$ with $\Delta_i\times\{0\}\subset A_i'\subset\partial T_i$ and $\Delta_i\times\{1\}\subset (\partial D)\times I\subset\partial B$. First take a disk $E_0=D\times\{t\}\subset D\times I=B$ and suppose $\partial E_0$ intersects each $\Delta_i\times\{1\}$ in a single arc $\delta_i\times\{1\}$, where $\delta_i$ is an arc in $\Delta_i$. Let $E_1$ be the union of $E_0$ and all the quadrilateral disks $\delta_i\times I$ in $H_i$ ($i=1,\dots,m$). In each solid torus $T_i=\tau_i\times S^1$, we take a meridional disk $\tau_i\times\{x\}$ and we suppose the intersection of the meridional disk $\tau_i\times\{x\}$ with the 1-handle $H_i$ is the arc $\delta_i\times\{0\}$. Let $E$ be the union of $E_1$ and all the disks $\tau_i\times\{x\}$ ($i=1,\dots, m$). So $E$ is a disk properly embedded in $X$ and we call $E$ a cross-section disk of $X$, see the shaded region in Figure \[Ftype1\](a) for a picture.
A cross-section disk $E$ has 3 parts: (1) $E_0=D\times\{t\}$ in the central 3-ball $B$, (2) the rectangles $\delta_i\times I$ in the 1-handles $H_i$ and (3) the bigons $\tau_i\times\{x\}$, $i=1,\dots,m$. Note that by choosing $E_0=D\times\{t\}$ using different $t\in I$, we can construct a collection of parallel cross-section disks.
[type1.eps]{} (20,-5)[(a)]{} (78,-5)[(b)]{} (27,37)[$B$]{} (5,22)[$H_i$]{} (19,8)[$T_i$]{} (77,40)[1-handle $H_j'$]{} (63,22)[$G'$]{} (80,22)[$\Gamma_i$]{}
Next, we use the cross-section disk $E$ to describe a type of properly embedded planar surfaces in $X$ which we call suspension surfaces. Let $\Gamma_1,\dots,\Gamma_k$ be a collection of $\partial$-parallel annuli in $X$ with each $\partial\Gamma_i\subset \partial_vX$. Let $\widehat{\Gamma}_i$ be the solid torus bounded by $\Gamma_i$ and the subannulus of $\partial_vX$ between the two curves of $\partial\Gamma_i$. We say that these annuli $\Gamma_1,\dots,\Gamma_k$ are [*non-nested*]{} if these solid tori $\widehat{\Gamma}_1, \dots, \widehat{\Gamma}_k$ are non-nested/disjoint. Suppose these $\partial$-parallel annuli $\Gamma_1,\dots,\Gamma_k$ are non-nested. Let $E$ be a cross-section disk described above. We may assume $E\cap \widehat{\Gamma}_i$ is a meridional disk of the solid torus $\widehat{\Gamma}_i$. Let $O$ be a point in the subdisk $E_0=D\times\{t\}$ of the cross-section disk $E$ and let $S$ be a 2-sphere around $O$. A [**suspension surface**]{} (over $\Gamma_1,\dots,\Gamma_k$ and based at the cross-section disk $E$) is the surface obtained by connecting each annulus $\Gamma_i$ to $S$ via a small tube along an arc in the cross-section disk. We will call $S$ the [**central sphere**]{} of the suspension surface. Alternatively, we can first connect each disk $E\cap\widehat{\Gamma}_i$ to $O$ using an arc in the cross-section disk $E$ and let $G$ be the union of these arcs. Let $N_\Gamma$ be a small neighborhood of $G\cup(\bigcup_{i=1}^k\widehat{\Gamma}_i)$. So $N_\Gamma$ is a marked handlebody whose marked annuli are the subannuli of $\partial_vX$ bounded by $\partial\Gamma_1,\dots, \partial\Gamma_k$. The frontier surface of $N_\Gamma$ in $X$ is a suspension surface over $\Gamma_1,\dots,\Gamma_k$. In particular, a suspension surface is a planar surface properly embedded in $X$ with boundary in $\partial_vX$. Moreover, we call $N_\Gamma$ the marked handlebody bounded by the suspension surface. Note that topologically $X\setminus N_\Gamma$ is also a handlebody.
Recall that the marked handlebody $X$ in the definition has $p$ extra 1-handles $H_1',\dots, H_p'$ attached to the central 3-ball $B$, which makes $\partial_hX$ a genus-$p$ surface. Let $F$ be a suspension surface constructed above. By the construction, $F$ is a planar surface. Next, we add some tubes to $F$ and change $F$ into a nonplanar surface. Each tube is either a trivial tube or a tube going through some 1-handle $H_j'$ and we require each 1-handle $H_j'$ contains at most one tube. First, view the genus-$p$ handlebody $(\bigcup_{i=1}^pH_i')\cup B$ as a product $P\times I$, where $P$ is the planar surface obtained by adding $p$ 2-dimensional 1-handles to the disk $D$. Then extend the cross-section disk $E$ to a planar surface $E'$ by extending $E_0=D\times\{t\}$ to $P\times\{t\}$. Now extend the graph $G\subset E$ (in the construction above) to a graph $G'\subset E'$ by first adding a collection of trivial loops to the point $O$ in $E_0=D\times\{t\}$ and then adding $q$ loops ($q\le p$) in $P\times\{t\}$ going through the 1-handles $H_1',\dots, H_p'$, such that each $H_j'$ contains at most one arc, see Figure \[Ftype1\](b) for a picture. Let $N_\Gamma'$ be a small neighborhood of $G'\cup(\bigcup_{i=1}^k \widehat{\Gamma}_i)$. Clearly, $N_\Gamma'$ is a marked handlebody. We call the frontier surface of $N_\Gamma'$ in $X$ a **standard surface** in $X$ and call $N_\Gamma'$ the marked handlebody bounded by the standard surface. Note that $X\setminus N_\Gamma'$ is also a topological handlebody.
**Type I blocks**:
Let $X$ be a marked handlebody as above. Let $F_1,\dots, F_n$ be a collection of mutually disjoint standard surfaces and let $N_i$ the marked handlebody bounded by $F_i$. Let $W$ be the closure of $X\setminus\cup_{i=1}^n N_i$ and we call $W$ a [**type I block**]{}. We call $(\cup_{i=1}^n F_i)\cup\partial_hX$ the horizontal boundary of $W$ and denote it by $\partial_h W$. The vertical boundary $\partial_v W$ is the closure of $\partial X\setminus\partial_h W$. So $\partial_v W$ is a collection of subannuli of $\partial_vX$.
**A basic construction:**
Let $X$ be a marked handlebody and let $D_1,\dots, D_k$ be a collection of parallel cross-section disks in $X$. Let $\Gamma_1,\dots,\Gamma_q$ be a collection of disjoint non-nested $\partial$-parallel annuli in $X$ with boundary circles in $\partial_vX$. We divide the annuli $\Gamma_1,\dots,\Gamma_q$ into $k$ disjoint sets of annuli $\widetilde{\Gamma}_1,\dots, \widetilde{\Gamma}_k$. Let $F_i$ be the suspension surface over the set of annuli $\widetilde{\Gamma}_i$ and based at the cross-section disk $D_i$ ($i=1,\dots, k$). By taking disjoint cross-sections disks, we may assume that $F_1,\dots, F_k$ are disjoint. $F_1,\dots F_k$ divide $X$ into $k+1$ submanifolds $W$, $X_1,\dots X_k$, where each $X_i$ is the marked handlebody bounded by $F_i$ and $W$ is the type I block between $\partial_hX$ and these $F_i$’s. The boundary of $W$ has two parts: (1) the horizontal boundary $\partial_hW=\partial_hX\cup F_1\cup\cdots\cup F_k$, and (2) the vertical boundary $\partial_vW$ consisting of subannuli of $\partial_vX$.
Consider the type I block $W$. Assign each $F_i$ a plus sign and assign $\partial_hX$ either a plus or a minus sign. We give each annulus in $\partial_vW$ a numerical order and either a $\pm$-sign or no sign. We define $\partial_h^\pm W$ and $\partial_v^\pm W$ to be the union of components of $\partial_hW$ and $\partial_vW$ respectively with $\pm$-sign, and let $\partial_v^0W$ be the union of annuli in $\partial_vW$ with no sign. We require that these surfaces with these signs and orders are admissible (see Definition \[Dadmissible\]).
Given any set of annuli in $\partial_v^+W$, we can first arrange the orders of the annuli in this set into a non-decreasing list and then consider the lexicographic order of this list. This gives an order on all subsets of annuli in $\partial_v^+W$. By condition (4) of Definition \[Dadmissible\], $\partial^+W$ is connected. So there is a set of annuli in $\partial_v^+W$ connecting all the components of $\partial_h^+W$ together. We call a set of annuli in $\partial_v^+W$ a [**minimal set of annuli**]{} connecting $\partial_h^+W$ if
1. the union of $\partial_h^+W$ and the annuli in this set is connected and
2. this set of annuli has minimal order among all such sets of annuli.
One may find it helpful to view the components of $\partial_h^+W$ as vertices and view the annuli in $\partial_v^+ W$ as edges connecting these vertices. So a minimal set of annuli is necessarily a set of edges that connect these vertices into a tree.
We have two situations depending on the sign of $\partial_h X$:
If $\partial_h X$ has a minus sign, then $\partial_h^+W=F_1\cup\cdots\cup F_k$. As $\partial^+ W$ is connected, there is a minimal set of $k-1$ annuli $\{x_1,\dots, x_{k-1}\}$ in $\partial_v^+W$ connecting $F_1,\dots, F_k$ together. We may name the annuli $x_i$’s and the surfaces $F_1,\dots, F_k$ (or arrange the subscripts) so that
1. $x_1$ has the smallest order among all the annuli in $\partial_v^+W$ that connect two distinct surfaces in $\{F_1,\dots,F_k\}$, and assume $F_1$ is attached to $x_1$.
2. $x_i$ connects $F_1\cup\cdots\cup F_i$ to $F_{i+1}$ for all $i=1,\dots, k-1$
3. for each $i=1,\dots, k-1$, $x_i$ has the smallest order among the annuli in the set $\{x_1,\dots, x_{k-1}\}$ that connect $F_1\cup\cdots\cup F_i$ to some $F_j$ with $j>i$.
If $\partial_h X$ has a plus sign, then $\partial_h^+W=\partial_hX\cup F_1\cup\cdots\cup F_k$. So there is a minimal set of $k$ annuli $\{x_1,\dots, x_{k}\}$ in $\partial_v^+W$ connecting $\partial_hX, F_1,\dots, F_k$ together. Similar to the case above, we may name the annuli $x_i$’s and the surfaces $F_1,\dots, F_k$ (or arrange the subscripts) so that
(i) $x_1$ has the smallest order among all the annuli in $\partial_v^+W$ that connect two distinct surfaces in $\{\partial_hX, F_1,\dots,F_k\}$, and assume $F_1$ is attached to $x_1$.
(ii) Let $\widehat{G}_i$ be the subset of the surfaces in $\{\partial_h X, F_1,\dots, F_k\}$ attached to the annuli $x_1,\dots, x_{i}$. The union of $x_1,\dots, x_{i}$ and the surfaces in $\widehat{G}_i$ is connected for each $i$, and the indices/subscripts are chosen so that, for any $F_p\in \widehat{G}_i$ and $F_q\notin \widehat{G}_i$, $p<q$.
(iii) for each $i=2,\dots, k$, $x_i$ has the smallest order among all the annuli in $\{x_1,\dots, x_{k}\}$ that connect a surface in $\widehat{G}_{i-1}$ to a surface in $\{\partial_hX, F_1,\dots,F_k\}\setminus\widehat{G}_{i-1}$.
In both cases, we say that these surfaces and cross-section disks are **well-positioned** if these $F_i$’s and $x_i$’s satisfy the respective conditions above and the cross-section disks $D_i$ and $D_{i+1}$ are adjacent for each $i$. Note that one can construct a suspension surface $F_i$ using an arbitrary cross-section disk, and different choices of cross-section disks do not affect whether or not the surfaces in $\partial W$ (with the fixed signs and orders of the $F_i$’s and $x_i$’s) are admissible, see Definition \[Dadmissible\].
\[LW2\] Let $X$, $W$, $F_1,\dots, F_k$ be as above. In particular, each $F_i$ is a suspension surface in $X$ with a plus sign. Suppose $\partial_hX$ is a planar surface and has a minus sign. Suppose the signs and orders of the surfaces in $\partial W$ are admissible. Let $\{x_1,\dots, x_{k-1}\}$ be a minimal set of annuli connecting $F_1,\dots, F_k$ and suppose these surfaces and the cross-section disks are well-positioned. Then $W$ is a relative compression body with respect to these signs and orders.
Since $\partial_hX$ has a minus sign, $\partial_h^+W=\bigcup_{i=1}^k F_i$. Let $G_i$ be the union of $x_1,\dots, x_i$ and $F_1,\dots, F_{i+1}$. By condition (2) on the $F_i$’s and $x_i$’s above, $G_i$ is a connected surface for each $i$.
Let $A$ be any annulus in $\partial_v^-W$. We first construct a dual disk for $A$. If both components of $\partial A$ lie in the same surface $F_j$, then $A$ has a dual disk $\Delta_A$, see the shaded disk in Figure \[Fdual\](a), and $\partial \Delta_A$ does not intersect any annulus in $\partial_v^+W$.
[dual.eps]{} (10,-3)[(a)]{} (46,-3)[(b)]{} (79,-3)[(c)]{} (18,11)[$A$]{} (52,1.4)[$A$]{} (51, 5)[$\Delta_A$]{} (43,2.2)[$x_q$]{} (82.5, 17)[$\Delta_i$]{} (73, 14.5)[$\alpha_i$]{} (73.8, 22)[$\alpha_j$]{} (64, 26)[$\partial_h X_0$]{}
Suppose the two components of $\partial A$ lie in two different surfaces $F_s$ and $F_t$ with $s<t$. By the construction of $G_i$, $\partial A\subset \partial G_{t-1}$. We have two cases to discuss:
[*Case (1)*]{}. The order $o(A)$ is larger than the orders of all the annuli $x_1,\dots, x_{t-1}$ in $G_{t-1}$
Recall that each $F_i$ is constructed by connecting a collection of $\partial$-parallel annuli to its central 2-sphere along arcs in the cross-section disk $D_i$. Denote the central 2-sphere of $F_i$ by $S_i$.
Next we perform some isotopies on $W$. Start with $F_1$, $F_2$, and $x_1$. Recall that $x_1$ connects $F_1$ to $F_2$. Let $\Gamma_1$ and $\Gamma_2$ be the two $\partial$-parallel annuli in the construction of $F_1$ and $F_{2}$ respectively that are attached to the annulus $x_1$. So $\Gamma'=\Gamma_1\cup x_1\cup\Gamma_2$ is an annulus. Let $\mathfrak{t}_1$ and $\mathfrak{t}_2$ be the two tubes in $F_1$ and $F_{1}$ that connect $\Gamma_1$ and $\Gamma_2$ to the central 2-spheres $S_1$ and $S_{2}$ respectively. Next, we show that $G_1=F_1\cup x_1\cup F_{2}$ can be viewed as a suspension surface. To see this, the first step is to push $\Gamma'$ into a $\partial$-parallel annulus in $X$, see the change from Figure \[Fslide\](a) to (b), where the shaded region denotes $x_1$. Now view $x_1$ as a subsurface in the interior of $\Gamma'$. Then perform a handle/tube slide on $\mathfrak{t}_1$, sliding $\mathfrak{t}_1$ across $x_1$ and then passing over $\mathfrak{t}_2$, which isotopes $\mathfrak{t}_1$ into a tube connecting the two central 2-spheres $S_1$ and $S_{2}$, see the isotopy from Figure \[Fslide\](b) to (c). Now $\mathfrak{t}_1$ merges $S_1$ and $S_{2}$ into a single 2-sphere. Since the cross-section disks $D_1$ and $D_{2}$ are adjacent, we can then push all the tubes in $F_1$ into a neighborhood of the cross-section disk $D_{2}$. This operation changes $G_1=F_1\cup x_1\cup F_{2}$ into a single suspension surface which we still denote by $G_1$. From the viewpoint of $W$, this operation is an isotopy and the annulus $x_1$ is isotoped into a subsurface of the new suspension surface $G_1$. Since the cross-section disk $D_1$ and $D_2$ are adjacent, this isotopy does not affect other $F_i$’s.
[slide.eps]{} (12,-3)[(a)]{} (48,-3)[(b)]{} (83,-3)[(c)]{} (2.5,15)[$\mathfrak{t}_1$]{} (83.5,24)[$\mathfrak{t}_1$]{} (56,14.7)[$\mathfrak{t}_2$]{} (89.5,15)[$\mathfrak{t}_2$]{}
Note that the annulus $x_{2}$ connects $G_1$ to $F_{3}$. Thus the new set of suspension surfaces $G_1, F_{3},\dots, F_t$ are connected by $x_{2},\dots, x_{t-1}$. By repeating the operation above, we can isotope the surface $G_{t-1}$ into a single suspension surface, and the annuli $x_1,\dots, x_{t-1}$ are viewed as subsurfaces of this suspension surface. Since the two curves in $\partial A$ are both boundary curves of $G_{t-1}$, there is a disk $\Delta_A$, as shown in Figure \[Fdual\](a), which intersects $A$ in a single essential arc. As the annuli $x_1,\dots, x_{t-1}$ are now viewed as subsurfaces of the suspension surface $G_{t-1}$, $\partial\Delta_A$ may intersect $x_1,\dots, x_{t-1}$. We may reverse the isotopy and isotope $G_{t-1}$ back to its original position, and this isotopy changes $\Delta_A$ into a disk that possibly intersects the annuli $x_1,\dots, x_{t-1}$ in their original positions. Since the order $o(A)$ is larger than the orders $o(x_1),\dots, o(x_{t-1})$ in Case (1), $\Delta_A$ is a dual disk for $A$.
[*Case (2)*]{}. The order $o(A)\le o(x_j)$ for some $1\le j\le t-1$.
Without loss of generality, suppose $j$ is the largest index such that $o(A)\le o(x_j)$ and $1\le j\le t-1$. So $o(A)$ is larger than the orders of $x_{j+1},\dots, x_{t-1}$.
Consider the surface $G_{k-1}$ which is the union of all the suspension surfaces $F_1,\dots, F_k$ and the annuli $x_1,\dots, x_{k-1}$. $G_{k-1}$ is a connected surface. Now remove the annulus $x_j$ from $G_{k-1}$. Since $\{x_1,\dots, x_{k-1}\}$ is a minimal set of annuli connecting $F_1,\dots, F_k$, $G_{k-1}\setminus x_j$ has two components, which we denote by $H_1$ and $H_2$. Without loss of generality, suppose $F_t\subset H_2$. Recall that $\partial A\subset \partial F_s\cup \partial F_t$ and $s<t$. We have two subcases:
The first subcase is that $F_s\subset H_1$. Since the signs and orders of the surfaces in $\partial W$ are admissible, by condition (5) in Definition \[Dadmissible\], there is an arc $\alpha\subset\partial^+W$ connecting the two components of $\partial A$ such that $\alpha$ only intersects annuli in $\partial_v^+W$ with order smaller than $o(A)$. Since $o(A)\le o(x_j)$, $\alpha\cap x_j=\emptyset$. Since $F_s\subset H_1$ and $F_t\subset H_2$, one component of $\partial A$ lies in $H_1$ and the other component of $\partial A$ lies in $H_2$. So $\alpha$ is an arc in $\partial^+W$ connecting $H_1$ to $H_2$. This means that one can find an annulus $x_j'$ in $\partial_v^+W$ such that (1) $\alpha$ intersects $x_j'$ and (2) $x_j'$ connects $H_1$ to $H_2$. Since $\alpha$ intersects $x_j'$, by condition (5) in Definition \[Dadmissible\], we have $o(x_j')< o(A)\le o(x_j)$. By replacing $x_j$ with the annulus $x_j'$, we obtain a new set of annuli with smaller order and connecting the surfaces $F_1,\dots, F_k$, contradicting the hypothesis that $\{x_1,\dots, x_{k-1}\}$ is a minimal set of such annuli.
The second subcase is that $F_s\subset H_2$. Consider the surface $G_{j-1}$. Recall that $G_{j-1}$ is connected and the annulus $x_j$ connects $F_{j+1}$ to $G_{j-1}$. If there is another annulus $x_p$ ($j<p\le t-1$) that connects $G_{j-1}$ to $F_{p+1}$, then by condition (3) on the $F_i$’s and $x_i$’s before the lemma, we have $o(x_j)\le o(x_p)$, and this contradicts the assumption at the beginning of Case (2) that $j$ is the largest index such that $o(A)\le o(x_j)$ and $1\le j\le t-1$. Thus none of $x_{j+1},\dots, x_{t-1}$ are attached to $G_{j-1}$. This implies that the annuli $x_{j+1},\dots, x_{t-1}$ connect $F_{j+1},\dots, F_t$ together into a connected surface, which we denote by $H'$. In particular $F_t\subset H'$. Since $F_t\subset H_2$ and since $H'$ is connected, we have $H'\subset H_2$. Since $G_{j-1}$ is connected and $F_s\subset H_2$, we must have $G_{j-1}\subset H_1$ and $F_s\not\subset G_{j-1}$. This implies that $F_s\subset H'$ and hence $\partial A\subset\partial H'$.
Now we apply the argument in Case (1) on $H'$. By performing tube slides, we can isotope $H'$ into a single suspension surface which we still denote by $H'$. The annuli $x_{j+1},\dots, x_{t-1}$ are now subsurfaces of $H'$. Moreover, since the cross-section disks $D_{j+1},\dots, D_t$ are adjacent to one another, this isotopy does not affect other surface $F_i$’s. As $\partial A\subset\partial H'$, there is a disk $\Delta_A$, as shown in Figure \[Fdual\](a), which intersects $A$ in a single essential arc. Note that $x_{j+1},\dots, x_{t-1}$ are subsurfaces of $H'$, so $\partial\Delta_A$ may intersect $x_{j+1},\dots, x_{t-1}$. By our assumption on $j$ at the beginning of Case (2), the order $o(A)$ is larger than the orders of $x_{j+1},\dots, x_{t-1}$. This means that, after isotope $H'$ back to its original position, $\Delta_A$ is a dual disk for $A$.
Therefore, in both Case (1) and Case (2), there is a dual disk for any annulus $A$ in $\partial_v^-W$. Moreover, since the surfaces in $\partial W$ are admissible, no annulus in $\partial_vW$ has both boundary curves in $\partial_h^-W=\partial_hX$. So every annulus of $\partial_vX$ must contain boundary curves of some $F_i$. Since $G_{k-1}$ can be isotoped into a single suspension surface and since $\partial_hX$ is planar, a maximal compression on $\partial^+W$ in $W$ yields a collar neighborhood of $\partial^-W=\partial_hX\cup\partial_v^-W$. This means that $W$ is a relative compression body.
\[Rgenus\] Given any relative compression body $W$, by Remark \[Rproperty\](3), if we maximally compress $\partial^+W$, the resulting manifold is a product neighborhood of $\partial^-W$. Thus we always have $g(\partial^+W)\ge g(\partial^-W)$. This is a major reason that we require $\partial_hX$ to be a planar surface in Lemma \[LW2\].
\[LW1\] Let $X$, $W$, $F_1,\dots, F_k$ be as above. In particular, each $F_i$ is a suspension surface in $X$ with a plus sign. Suppose $\partial_hX$ has a plus sign. Suppose the signs and orders of the surfaces in $\partial W$ are admissible. Let $\{x_1,\dots, x_{k}\}$ be a minimal set of annuli connecting $\partial_hX, F_1,\dots, F_k$ as above and suppose these surfaces and the cross-section disks are well-positioned. Then $W$ is a relative compression body with respect to these signs and orders.
The proof is similar to the proof of Lemma \[LW2\] except that $\partial_hX$ is a component of $\partial_h^+W$ in this lemma.
Let $A$ be any annulus in $\partial_v^-W$. We first construct a dual disk for $A$. If both components of $\partial A$ lie in the same surface $F_i$, then $A$ has a dual disk $\Delta_A$ as shown in Figure \[Fdual\](a) and $\partial \Delta_A$ does not intersect any annulus in $\partial_v^+W$. If both curves of $\partial A$ are boundary curves of $\partial_hX$, then $A$ must be a component of $\partial_vX$ and there is a dual disk $\Delta_A$, as in the proof of Lemma \[LBM1\], disjoint from all the $F_i$’s. Now suppose that the two curves of $\partial A$ lie in different surfaces of $\{\partial_h X, F_1,\dots, F_k\}$.
Let $\widehat{G}_i$ be the subset of the surfaces in $\{\partial_h X, F_1,\dots, F_k\}$ attached to $x_1, \dots, x_{i}$. Let $G_i$ be the union of $x_1,\dots, x_i$ and the surfaces in $\widehat{G}_i$. By condition (ii) on the set of annuli $\{x_1,\dots, x_k\}$ before Lemma \[LW2\], each $G_i$ is a connected surface.
Suppose $\partial A\subset\partial G_t$ and suppose $t$ is the smallest such index (i.e. $\partial A\not\subset G_{t-1}$). Similar to Lemma \[LW2\], we have two cases.
[*Case (1)*]{}. The order $o(A)$ is larger than the order of any annulus $x_i$ that lies in $G_{t}$, i.e. $o(A)>o(x_i)$ for any $1\le i\le t$.
If $\partial_hX$ is not in $\widehat{G}_t$, then the proof is the same as Case (1) of Lemma \[LW2\]. By isotoping $G_t$ into a single suspension surface, we have a cross-section disk $\Delta_A$ for $A$, as shown in Figure \[Fdual\](a). So, it remains to consider the case that $\partial_hX\subset G_t$, i.e. $G_t$ is the union of $x_1,\dots, x_t$, $\partial_hX$, and $F_1,\dots, F_t$.
[*Subcase (1a)*]{}. Exactly one annulus in $\{x_1,\dots, x_t\}$ is attached to $\partial_hX$.
Suppose $x_q$ ($1\le q\le t$) is the annulus attached to $\partial_hX$. Consider the subset of $t-1$ annuli $\{x_1,\dots, x_t\}\setminus x_q$. In this subcase, $F_1,\dots, F_t$ are connected by this subset of $t-1$ annuli into a connected surface $F'$. Moreover, we can perform the tube slides as in the proof of Lemma \[LW2\], which isotope $F'$ into a suspension surface. So, in this subcase, we may view $G_t$ as the union of $F'$, $\partial_hX$, and the annulus $x_q$ between them.
If both curves of $\partial A$ are boundary curves of $F'$, then as in Case (1) of Lemma \[LW2\], $A$ has a dual disk $\Delta_A$ as shown in Figure \[Fdual\](a). If one curve of $\partial A$ is a boundary curve of $\partial_hX$ and the other curve of $\partial A$ is a curve in $\partial F'$, then as illustrated in the schematic picture in Figure \[Fdual\](b), $A$ has a dual disk $\Delta_A$ that intersects the annulus $x_q$. Note that we can isotope $F'$ back to the original position and this isotopy carries $\Delta_A$ into a disk for $A$ that only intersects the annuli $x_1,\dots x_t$ of $\partial_v^+W$. Since $o(A)$ is larger than the order of any annulus in $\{x_1,\dots, x_t\}$, $\Delta_A$ is a dual disk for $A$.
[*Subcase (1b)*]{}. More than one annulus in $\{x_1,\dots, x_t\}$ is attached to $\partial_hX$.
Similar to the proof of Lemma \[LW2\], we try to isotope the $F_i$’s in $G_t$ into a single suspension surface. As before, for any two adjacent suspension surfaces connected by an annulus $x_i$ ($1\le i\le t$), we perform a tube slide and isotope them into a single suspension surface which contains $x_i$ as a subsurface. Since more than one annulus in $\{x_1,\dots, x_t\}$ is attached to $\partial_hX$, after some tube slides and isotopies, we arrive at a situation that (1) $x_p$ and $x_q$ connect two suspension surfaces $F'$ and $F''$ to $\partial_h X$ and (2) the cross-section disks of $F'$ and $F''$ are adjacent. Denote the cross-section disks of $F'$ and $F''$ by $D'$ and $D''$ respectively.
Let $\Gamma_1$ and $\Gamma_2$ be the $\partial$-parallel annuli (in the construction of the suspension surface $F'$ and $F''$) that are attached to $x_p$ and $x_q$ respectively. Let $\mathfrak{t}_1$ and $\mathfrak{t}_2$ be the tubes connecting $\Gamma_1$ and $\Gamma_2$ to the central 2-spheres of $F'$ and $F''$ respectively. Let $H_1=\partial_hX\cup x_p\cup \Gamma_1$. As illustrated in the schematic pictures Figure \[Fxslide1\](a, b), we perform an isotopy on $X$ by first pushing $x_p$ into the interior of $X$. Then, as illustrated in Figure \[Fxslide1\](c), we slide $\mathfrak{t}_1$ across $x_p$ into a tube (in a neighborhood of the cross-section disk $D'$) connecting $\partial_hX$ to the central sphere of $F'$.
[xslide1.eps]{} (14,-3)[(a)]{} (50,-3)[(b)]{} (83,-3)[(c)]{} (4,1.5)[$x_p$]{} (12,3)[$\Gamma_1$]{} (12.7,10)[$\mathfrak{t}_1$]{} (47,10)[$\mathfrak{t}_1$]{} (81.5,21)[$\mathfrak{t}_1$]{} (3,14)[$\partial_hX$]{}
Next we show that we can slide the tube $\mathfrak{t}_1$ across $x_q$ and passing over $\mathfrak{t}_2$ into a tube connecting the two central spheres of $F'$ and $F''$. To see this, note that we can similarly slide the tube $\mathfrak{t}_2$ across $x_q$ into a tube connecting $\partial_hX$ to the central sphere of $F''$. This operation can be viewed as an isotopy of $W$. As illustrated in Figure \[Fxslide2\](a, b), we can then slide $\mathfrak{t}_1$ over $\mathfrak{t}_2$ and into a tube connecting the two central spheres for $F'$ and $F''$. Since the cross-section disks for $F'$ and $F''$ are adjacent, this tube slide does not affect other suspension surfaces. Since these operations are all isotopies, this implies that, without isotoping the tube $\mathfrak{t}_2$ in the first step, we can slide $\mathfrak{t}_1$ across $x_q$ and passing over $\mathfrak{t}_2$ into a tube connecting the two central spheres of $F'$ and $F''$.
Now we merge the central 2-spheres of $F'$ and $F''$ together along the tube $\mathfrak{t}_1$. As in the proof of Lemma \[LW2\], we can merge $F'$ and $F''$ into a single suspension surface. Thus, after finite many such isotopes, we can merge $F_1,\dots, F_t$ into a single suspension surface $\widehat{F}$, and $\widehat{F}$ is connected to $\partial_hX$ by an annulus $x_q$. Now the configuration is the same as Subcase (1a), and we can construct a dual disk for $A$ as in Subcase (1a).
[xslide2.eps]{} (22,-3)[(a)]{} (72,-3)[(b)]{} (12.5,15)[$\mathfrak{t}_1$]{} (72,9)[$\mathfrak{t}_1$]{} (31.6,15)[$\mathfrak{t}_2$]{} (81.2,15)[$\mathfrak{t}_2$]{} (3,15)[$\partial_hX$]{} (60,15)[$\partial_hX$]{}
[*Case (2)*]{}. The order $o(A)\le o(x_j)$ for some annulus $x_j$ that lie in $G_{t}$, i.e. $1\le j\le t$.
Without loss of generality, suppose $j$ is the largest index such that $o(A)\le o(x_j)$ and $1\le j\le t$. So $o(A)$ is larger than the orders of $x_{j+1},\dots, x_{t}$.
Similar to Case (2) of Lemma \[LW2\], consider $G_k$ and remove the annulus $x_j$ from $G_{k}$. Since the $\{x_1,\dots, x_{k}\}$ is a minimal set of annuli connecting surfaces in $\partial_h^+W$, $G_{k}\setminus x_j$ has two components, denoted by $H_1$ and $H_2$.
The first subcase is that the one component of $\partial A$ lies in $H_1$ and the other component of $\partial A$ lies in $H_2$. This subcase is the same as the first subcase of Case (2) in Lemma \[LW2\]: Since the signs and orders of the surfaces in $\partial W$ are admissible, there is an arc $\alpha\subset\partial^+W$ connecting the two components of $\partial A$. By replacing $x_j$ with an annulus that $\alpha$ intersects, as in Case (2) in Lemma \[LW2\], we can obtain a new set of annuli with smaller order and connecting all the surfaces of $\partial_h^+W$. This contradicts the hypothesis that $\{x_1,\dots, x_{k}\}$ is a minimal set of such annuli.
The second subcase is that both components of $\partial A$ lie in the same surface, say $H_2$. This subcase is similar to the second subcase of Case (2) in Lemma \[LW2\], except that we have to take $\partial_hX$ into consideration.
Consider the surface $G_{j-1}$. By the assumption that $\partial A\not\subset G_{t-1}$ before Case (1), the two boundary curves of $A$ cannot both be in $\partial G_{j-1}$. If there is another annulus $x_p$ ($j<p\le t$) that connect $G_{j-1}$ to a surface in $\widehat{G}_t\setminus\widehat{G}_{j-1}$, then by condition (iii) on $x_1,\dots, x_k$ before Lemma \[LW2\], we have $o(x_j)\le o(x_p)$, and this contradicts the assumption above that $j$ is the largest index such that $o(A)\le o(x_j)$ and $1\le j\le t$. Thus none of $x_{j+1},\dots, x_{t}$ are attached to $G_{j-1}$. Let $\widehat{G}'$ be the collection of surfaces in $\widehat{G}_t\setminus\widehat{G}_{j-1}$, and let $H'$ be the union of $x_{j+1},\dots, x_{t}$ and all the surfaces in $\widehat{G}'$. Similar to the second subcase of Case (2) in Lemma \[LW2\], since none of $x_{j+1},\dots, x_{t}$ are attached to $G_{j-1}$, $H'$ must be a connected subsurface of $G_t$. Similarly, this implies that $H'\subset H_2$ and $\partial A\subset\partial H'$.
Now we apply the argument in Case (1) on $H'$. By performing tube slides, if $\partial_hX\not\subset H'$, we can isotope $H'$ into a single suspension surface, and if $\partial_hX\subset H'$, we can isotope $H'$ into the form of $\partial_hX\cup x_q\cup H''$, where $H''$ is a single suspension surface. As in Case (1) above, this means that there is a disk $\Delta_A$ which intersects $A$ in a single essential arc, as illustrated in Figure \[Fdual\](a or b). Note that $\partial\Delta_A$ may intersect $x_{j+1},\dots, x_{t}$, but by our assumption on $j$, the order $o(A)$ is larger than the orders of $x_{j+1},\dots, x_{t}$. This means that, after isotope $H'$ back to its original position, $\Delta_A$ is a dual disk for $A$.
Therefore, in both Case (1) and Case (2), there is a dual disk for any annulus $A$ in $\partial_v^-W$. Moreover, since $\partial_h X$ and $F_1,\dots, F_k$ all have plus signs, a maximal compression on $\partial^+W$ in $W$ yields a collar neighborhood of $\partial_v^-W$. Hence $W$ is a relative compression body.
\[Dsimilar\] Let $X_1$ and $Y_1$ be two relative compression bodies. We say that $X_1$ and $Y_1$ are [**$\partial$-similar**]{} if there is a homeomorphism $f\colon \partial X_1\to\partial Y_1$ such that
1. $f$ maps $\partial_v^0X_1$, $\partial_v^\pm X_1$ and $\partial_h^\pm X_1$ homeomorphically to $\partial_v^0Y_1$, $\partial_v^\pm Y_1$ and $\partial_h^\pm Y_1$ respectively, and
2. for each annulus $A$ in $\partial_v^\pm X_1$, $o(A)=o(f(A))$.
Note that the map $f$ is on the boundary only and it may not extend to a map from ${\mathrm{int}}(X_1)$ to ${\mathrm{int}}(Y_1)$.
Let $X$ and $Y$ be two stacks of relative compression bodies. Suppose $F_1,\dots, F_n$ are the horizontal surfaces in $X$ that divide $X$ into relative compression bodies $X_1,\dots, X_m$, and suppose $S_1,\dots, S_n$ are the horizontal surfaces in $Y$ that divide $Y$ into relative compression bodies $Y_1,\dots, Y_m$. Suppose $F_i\cong S_i$ for all $i$ and let $h_i\colon F_i\to S_i$ be the homeomorphism. Moreover, suppose these homeomorphisms $h_i$’s extend to homeomorphisms $f_j\colon \partial X_j\to \partial Y_j$ ($j=1,\dots, m$) which satisfy the two conditions above. In particular, $X_j$ and $Y_j$ are $\partial$-similar for all $j$. Then we say that the two stacks $X$ and $Y$ are [**similar**]{}.
Next, we use Lemma \[LW2\] and Lemma \[LW1\] to prove the main result on stacks of relative compression bodies.
\[LtypeI\] Let $X$ be a marked handlebody. Let $Y$ be a stack of relative compression bodies. Suppose that there is a homeomorphism $f\colon (\partial Y, \partial_vY)\to (\partial X,\partial_vX)$ and suppose every horizontal surface in the stack $Y$ is separating. Then there is a collection of properly embedded surfaces in $X$ dividing $X$ into a stack that is similar to the stack $Y$.
We prove the lemma using induction on the number of relative compression bodies in the stack $Y$. As $X$ is a marked handlebody, $\partial_hX$ is connected. Since $f\colon (\partial Y, \partial_vY)\to (\partial X,\partial_vX)$ is a homeomorphism, $\partial_hY$ is also connected. If the stack $Y$ has only one relative compression body, i.e. $Y$ itself, since $\partial_hY$ is connected and since $\partial_h^+ Y\ne\emptyset$ for any relative compression body $Y$, $\partial_hY$ must have a plus sign. Thus we can assign $\partial_hX$ a plus sign and assign each component of $\partial_vX=f(\partial_vY)$ a sign and an order according to the sign and order of the corresponding component of $\partial_vY$. By Lemma \[LBM1\], $X$ is $\partial$-similar to $Y$.
Suppose the lemma is true if the number of relative compression bodies in the stack $Y$ is at most $n$. Now suppose $Y$ contains $n+1$ relative compression bodies.
Let $W_Y$ be the relative compression body in the stack $Y$ that contains $\partial_hY$. Thus $\partial_hY$ is a component of $\partial_hW_Y$. Let $S_1,\dots,S_k$ be the components of $\partial_h W_Y\setminus\partial_hY$. So each $S_i$ is a properly embedded separating surface in $Y$. Let $y_1,\dots, y_p$ be the annuli in $\partial_vW_Y$. The complement $\overline{\partial_v Y-\cup_{i=1}^p y_i}$ is a collection of annuli, which we denote by $y_1',\dots, y_q'$. Each $y_i'$ ($i=1,\dots, q$) lies outside $W_Y$.
\[Claim1\] For each $i$, both boundary curves of $y_i'$ belong to in the same surface $S_j$ for some $j$.
To see this, we first take an essential arc $\alpha$ of the annulus $y_i'$. So $\alpha$ lies outside $W_Y$. Since $\partial\alpha\subset\partial W_Y$, we can connect the two endpoints of $\alpha$ using an arc properly embedded in $W_Y$. Now we have a closed curve that intersects each component of $\partial y_i'$ in a single point. Suppose the two curves of $\partial y_i'$ belong to different components of $\partial_hW_Y$, say $S_s$ and $S_t$, then this closed curve intersects $S_t$ in a single point, which means that $S_t$ is non-separating, contradicting our hypothesis.
Let $x_i=f(y_i)$ and $x_i'=f(y_i')$ be the corresponding annuli in $\partial_v X$. Let $\Gamma_i$ be a properly embedded $\partial$-parallel annulus in $X$ with $\partial\Gamma_i=\partial x_i'$ ($i=1,\dots, q$). By Claim \[Claim1\], we can divide these $\Gamma_i$’s into $k$ sets of annuli $\widetilde{\Gamma}_1,\dots, \widetilde{\Gamma}_k$ such that $\Gamma_a\in\widetilde{\Gamma}_j$ if and only if the curves $f^{-1}(\partial\Gamma_a)$ belong to the same surface $S_j$. Let $D_1,\dots,D_k$ be a sequence of parallel cross section disks and let $F_i$ be the suspension surface over the set of annuli $\widetilde{\Gamma}_i$ and based at the cross-section disk $D_i$. So a curve $\gamma$ is a boundary component of $F_i$ if and only if $f^{-1}(\gamma)$ is a boundary component of $S_i$. Note that each suspension surface $F_i$ is planar, so $F_i$ may not be homeomorphic to $S_i$. Nonetheless, $f(\partial S_i)=\partial F_i$.
Let $W_X$ be the submanifold of $X$ between $\partial_hX$ and these surfaces $F_1,\dots, F_k$. So $W_X$ is a type I block and $\partial_vW_X=\bigcup_{i=1}^p x_i$. We give a $\pm$-sign to $\partial_hX$ and each $F_i$ according to the signs of $\partial_hY$ and $S_i$ respectively. Moreover, we give a sign and an order to each annulus $x_i$ according to the sign and order of $y_i$ ($i=1,\dots, p$). Since $W_Y$ is a relative compression body, this one-to-one correspondence between surfaces in $\{\partial_hW_Y, \partial_vW_Y\}$ and $\{\partial_hW_X, \partial_vW_X\}$ implies that the surfaces $\partial_hW_X$, $\partial_vW_X$ with the signs and orders are admissible, see Definition \[Dadmissible\].
We have 4 cases to discuss depending on the signs of $S_1,\dots, S_k$, and $\partial_hY$.
*Case 1*. $S_1,\dots, S_k$ and $\partial_hY$ all have plus signs.
In this case, $\partial_h^+W_X=\partial_hX\cup F_1\cup\cdots\cup F_k$. Let $\{x_1,\dots,x_k\}$ be a minimal set of annuli connecting $\partial_hX$ and $F_1,\dots, F_k$ together, and we may choose these surfaces and cross-section disks $D_1,\dots, D_k$ to be well-positioned. By Lemma \[LW1\], $W_X$ is a relative compression body.
Let $F_i'$ be the standard surface obtained by adding $g(S_i)$ trivial tubes to $F_i$ ($i=1,\dots, k$). So $F_i'\cong S_i$ for each $i$. Let $W_X'$ be the submanifold of $W_X$ between $\partial_hX$ and $F_1'\dots, F_k'$. Since $F_1,\dots, F_k$ and $\partial_hX$ all have plus signs and since $F_i'\cong S_i$, $W_X'$ is a relative compression body $\partial$-similar to $W_Y$.
By our construction, each standard surface $F_i'$ bounds a marked handlebody $X_i$ in $X$. As each $S_i$ ($i=1,\dots, k$) is separating, $S_i$ cuts off a smaller stack $Y_i$ from $Y$. By the induction hypotheses, there is a collection of surfaces in each $X_i$ that divides $X_i$ into a stack that is similar to the stack $Y_i$. As $W_X'$ is $\partial$-similar to $W_Y$, all these surfaces together divide $X$ into a stack that is similar to $Y$.
*Case 2*. $S_1,\dots, S_k$ all have plus signs, but $\partial_hY$ has a minus sign.
In this case, $\partial_h^+W_X=F_1\cup\cdots\cup F_k$. Let $x_1,\dots,x_{k-1}$ be a minimal set of annuli connecting the $F_i$’s, and we may choose these surfaces and cross-section disks $D_1,\dots, D_k$ to be well-positioned. Note that if $\partial_hY$ is planar, then $\partial_hX$ is planar, and the lemma follows from Lemma \[LW2\] and the argument in Case 1. So we suppose $g(\partial_hX)=g(\partial_hY)\ge 1$.
We may view $X$ as the manifold obtained by adding $g(\partial_hX)$ 1-handles to a “smaller" marked handlebody $X_0$, where $P_0=\partial_hX_0$ is a planar surface and $\partial_v X_0=\partial_vX$. We may view $F_1,\dots, F_{k}$ as suspension surfaces in $X_0$. let $W_0$ be the submanifold of $X_0$ between $P_0$ and $F_1,\dots, F_k$. Assign $P_0$ a minus sign. Since $P_0$ is a planar surface, by Lemma \[LW2\], $W_0$ is a relative compression body.
*Subcase 2a*. $\sum_{i=1}^k g(S_i)\ge g(\partial_hY)$
Since $\partial_hX\cong\partial_hY$, $\sum_{i=1}^k g(S_i)\ge g(\partial_hX)$. Consider the $g(\partial_hX)$ 1-handles attached to $X_0$. Next, we add $g(S_i)$ tubes to each suspension surface $F_i$ ($i=1,\dots, k$), changing $F_i$ into a standard surface of genus $g(S_i)$. Since $\sum_{i=1}^k g(S_i)\ge g(\partial_hX)$, we can arrange $g(\partial_hX)$ of the total $\sum_{i=1}^k g(S_i)$ tubes going through the $g(\partial_hX)$ 1-handles attached $X_0$, exactly one tube for each 1-handle, and set the remaining $\sum_{i=1}^k g(S_i)-g(\partial_hX)$ tubes as trivial tubes. Denote the resulting standard surfaces by $F_1',\dots, F_k'$. So $g(F_i')=g(S_i)$ and by the construction of $F_i$, $F_i'\cong S_i$ for all $i$.
Let $W_X'$ be the submanifold of $X$ between $\partial_hX$ and $F_1',\dots, F_k'$. So $\partial_h^+W_X'=\bigcup_{i=1}^k F_i'$ and $\partial_h^-W_X'=\partial_hX$. Since $W_0$ is a relative compression body, a maximal compression on $\partial^+W_0$ yields a product neighborhood of $\partial^-W_0$. Since $P_0=\partial_h^-W_0$ and since each 1-handle of $X$ contains exactly one tube of $F_1',\dots, F_k'$, this implies that a maximal compression on $\partial^+W_X'$ yields a product neighborhood of $\partial^-W_X'$. Moreover, since $W_0$ is a relative compression body, each annulus of $\partial_v^- W_0$ has a dual disk in $W_0$. As $\partial_v^- W_0=\partial_v^- W_X'$, each annulus of $\partial_v^- W_X'$ has an induced dual disk in $W_X'$. Hence, $W_X'$ is a relative compression body $\partial$-similar to $W_Y$. Now the lemma follows from the induction as in Case 1.
*Subcase 2b*. $\sum_{i=1}^k g(S_i)< g(\partial_hY)$
As $\partial_hY$ has a minus sign, we have $\partial_h^-W_Y=\partial_hY$ and $\partial_h^+W_Y=\cup_{i=1}^kS_i$. As in Remark \[Rgenus\], we have $g(\partial^- W_Y)\le g(\partial^+ W_Y)$, which implies that $g(\partial_hY)=g(\partial_h^-W_Y)\le g(\partial^+W_Y)$. Hence the hypothesis $\sum_{i=1}^k g(S_i)< g(\partial_hY)$ implies that $\sum_{i=1}^k g(S_i)< g(\partial^+W_Y)$. Since $\partial_h^+W_Y=\cup_{i=1}^kS_i$ and $\partial^+W_Y=\partial_h^+W_Y\cup \partial_v^+W_Y$, this inequality is possible because of the annuli in $\partial_v^+W_Y$.
Recall that $\{x_1,\dots,x_{k-1}\}$ is a minimal set of annuli connecting the $F_i$’s and $x_i=f(y_i)$. So $y_1,\dots, y_{k-1}$ are $k-1$ annuli in $\partial_v^+W_Y$ connecting $S_1,\dots, S_k$ together. Thus each additional annulus in $\partial_v^+W_Y\setminus\{y_1,\dots, y_{k-1}\}$ contributes an extra genus for $\partial^+W_Y=\partial_h^+W_Y\cup \partial_v^+W_Y$, so we have $g(\partial^+W_Y)=\sum_{i=1}^k g(S_i)+ |\partial_v^+W_Y|-(k-1)$. Let $g= |\partial_v^+W_Y|-(k-1)=g(\partial^+W_Y)-\sum_{i=1}^k g(S_i)$. So $g(\partial^+W_Y)=\sum_{i=1}^k g(S_i)+ g$. Without loss of generality, suppose $y_k,\dots, y_{k+g-1}$ are the $g$ additional annuli in $\partial_v^+W_Y$. Hence $x_k,\dots, x_{k+g-1}$ are $g$ annuli in $\partial_v^+W_X$.
Next, we consider the marked handlebody $X_0$ and the relative compression body $W_0$ constructed at the beginning of Case 2.
Let $\widehat{F}$ be the union of $x_1,\dots, x_{k-1}$ and the suspension surfaces $F_1,\dots, F_k$. So $\widehat{F}$ is a connected planar surface. Similar to the proof of Lemma \[LW2\], we can perform tube slides and isotope $\widehat{F}$ into a suspension surface in $X_0$. Next, we view $\widehat{F}$ as a suspension surface and view $F_1,\dots, F_k$ as subsurfaces of $\widehat{F}$. As in the proof of Lemma \[LW2\], all the dual disks $\Delta_i$ in this configuration are as shown in Figure \[Fdual\](a). Since $x_k,\dots, x_{k+g-1}$ are $g$ annuli in $\partial_v^+W_0$, the corresponding disks $ \Delta_k,\dots, \Delta_{k+g-1}$, as in Figure \[Fdual\](a), are $g$ compressing disks for $\partial^+W_0$.
Now we add $g(S_i)$ trivial tubes to each $F_i$ ($i=1,\dots, k$) and denote the resulting surface by $F_i'$. So $F_i'\cong S_i$. As each $F_i$ is a subsurface of the suspension surface $\widehat{F}$, these trivial tubes can also be viewed as trivial tubes for $\widehat{F}$. Let $W_0'$ be the submanifold of $W_0$ bounded by $P_0=\partial_hX_0$, $F_1',\dots, F_k'$ and $\partial_vW_0$. Let $\mathfrak{m}= \sum_{i=1}^k g(S_i)$. So $W_0'$ is obtained from $W_0$ by drilling out $\mathfrak{m}$ trivial tunnels. Since each $F_i$ has a plus sign, $W_0'$ is also a relative compression body. Moreover, each trivial tube determines a compressing disk for $F_i'$ in $W_0'$, see the top shaded disk in Figure \[Fdual\](c). Let $D_1',\dots, D_\mathfrak{m}'$ be the compressing disks in $W_0'$ corresponding to these $\mathfrak{m}$ trivial tubes.
Let $\omega=g(\partial_hY)-\mathfrak{m}$ ($\mathfrak{m}= \sum_{i=1}^k g(S_i)$). By the hypothesis $\sum_{i=1}^k g(S_i)< g(\partial_hY)$ in this subcase, we have $\omega> 0$. Since $g=g(\partial^+W_Y)-\mathfrak{m}$ and since $g(\partial_hY)=g(\partial_h^-W_Y)\le g(\partial^+W_Y)$, we have $g\ge \omega$. Note that $g(\partial_hY)=\omega+\mathfrak{m}$. Recall that each annulus $x_i$ ($i=k,\dots, g+k-1$) is associated with a disk $\Delta_i$, as illustrated in Figure \[Fdual\](c), and since $x_k,\dots, x_{k+g-1}$ are $g$ annuli in $\partial_v^+W_X$, each $\Delta_i$ is a compressing disk for $\partial^+W_0'$. As $g\ge\omega$, we consider the first $\omega$ disks $\Delta_k,\dots,\Delta_{\omega+k-1}$.
Consider the $\omega+\mathfrak{m}$ compressing disks $\Delta_k,\dots,\Delta_{\omega+k-1}$ and $D_1',\dots, D_\mathfrak{m}'$ for $\partial^+W_0'$ in $W_0'$. Recall that $g(\partial_hY)=\omega+\mathfrak{m}$. Since $P_0=\partial_hX_0=\partial_h^-W_0'$, by Remark \[Rproperty\](4), there are (core) arcs $\alpha_1,\dots,\alpha_{\omega+\mathfrak{m}}$ dual to the $\omega+\mathfrak{m}$ disks $\Delta_k,\dots,\Delta_{\omega+k-1}$ and $D_1',\dots, D_\mathfrak{m}'$ with $\partial\alpha_i\subset\partial_hX_0=P_0$, see Figure \[Fdual\](c), such that after drilling out tunnels along these arcs $\alpha_i$’s, $W_X=W_0'\setminus\bigcup_{i=1}^{\omega+\mathfrak{m}}N(\alpha_i)$ is still a relative compression body.
Note that there are $\omega$ disjoint circles $\gamma_1,\dots,\gamma_\omega$ in the central sphere of $\widehat{F}$ such that each $\gamma_j$ intersects the disk $\Delta_{j+k-1}$ exactly once and disjoint from other disks. Similarly, each of the meridional circles of the trivial tubes meets the corresponding disk $D_i'$ exactly once, see Figure \[Fdual\](c). So these arcs $\alpha_i$’s basically go around these circles. Moreover, if we ignore $\widehat{F}$, then these $\alpha_i$’s are arcs in a large 3-ball in $X_0$ that contain all the cross-section disks of $F_1,\dots, F_k$. Thus, $\alpha_1,\dots,\alpha_{\omega+\mathfrak{m}}$ are trivial arcs in the marked handlebody $X_0$. Hence $X=X_0\setminus \bigcup_{i=1}^{\omega+\mathfrak{m}}N(\alpha_i)$ is a marked handlebody with $\partial_vX=\partial_vX_0$ and $g(\partial_hX)=\omega+\mathfrak{m}=g(\partial_hY)$. This implies that $W_X$ is a submanifold of the marked handlebody $X$ and $W_X$ is $\partial$-similar to $W_Y$. Note that we may isotope $\widehat{F}$ back into the original positions of $F_1,\dots, F_i$ and $x_1,\dots, x_{k-1}$ and the isotopy carries these disks and arcs $\alpha_i$’s to a different position. By our construction, after this isotopy, each $F_i'$, together with subannuli of $\partial_vX$, bounds a marked handlebody in $X$. Now by the induction as in Case 1, we can build a collection of surfaces that divide $X$ into a stack of relative compression bodies that is similar to the stack $Y$.
\[RfixH\] In the operation above, we fix $X_0$ and obtain the marked handlebody $X$ by drilling out $g(\partial_hX)$ trivial tunnels from $X_0$. Instead of drilling tunnels, one can also carry out an equivalent (dual) operation by fixing $X$ while dragging some tubes of $F_i'$ through the $g(\partial_hX)$ 1-handles of $X$, in other words, re-embedding some tube of $F_i'$ into tubes that go through the 1-handles of $X$. Note that, since the tubes of the suspension surface $F_i$ may be re-embedded into tubes through the 1-handles of $X$, the final surface $F_i'$ may not be a standard surface in $X$. Nonetheless, each $F_i'$, together with subannuli of $\partial_vX$, still bounds a marked handlebody in $X$, and we can carry out the induction as in Case 1.
*Case 3*. $S_1,\dots, S_k$ all have minus signs.
By the definition of relative compression body, $\partial_h^+W_Y\ne\emptyset$. Hence $\partial_hY$ must have a plus sign and $\partial_h^+ W_Y=\partial_hY$.
Let $y_1',\dots, y_q'$ be the collection of subannuli in $\partial_vY$ defined before Claim \[Claim1\]. We first show that each component of $\partial_vY$ contains at most one $y_i'$. If two such annuli, say $y_i'$ and $y_j'$, lie in the same component of $\partial_vY$, then there is a component $y_t$ of $\partial_v W_Y$ between $y_i'$ and $y_j'$ with both boundary circles in $\partial y_i'\cup\partial y_j'$. This implies that $y_t$ has both boundary circles in $\bigcup_{i=1}^k \partial S_i$. Since $S_1,\dots, S_k$ all have negative signs, $y_t$ is an annulus in $\partial_vW_Y$ with both boundary curves in $\partial_h^-W_Y$, contradicting the hypothesis that $W_Y$ is a relative compression body (by Definition \[Drelative compression body\], there is no such annulus).
Let $x_i'=f(y_i')$ ($i=1,\dots, q$). By the conclusion above, each component of $\partial_vX$ contains at most one $x_i'$. Consider the $\partial$-parallel annulus $\Gamma_i$ with $\partial\Gamma_i=\partial x_i'$ defined after Claim \[Claim1\]. Let $W_X'$ be the submanifold of $X$ obtained by deleting the solid tori bounded by $\Gamma_i\cup x_i'$ ($i=1,\dots, q$). So $\partial W_X'$ consists of $\partial_hX$, $\Gamma_1,\dots,\Gamma_q$ and the annuli $x_1,\dots, x_p$, where $x_i=f(y_i)$. We assign a sign and an order to each $x_i$ according to the sign and order of $y_i$.
As $X$ is a marked handlebody, if we maximally compress $\partial_hX$ in $X$, the resulting manifold is a collar neighborhood of $\partial_vX$. Since each component of $\partial_vX$ contains at most one $x_i'$, if we maximally compress $\partial_hX$ in $W_X'$, we obtain a collection of solid tori, each of which is either a product neighborhood of some $\Gamma_i$ or a collar neighborhood of a component of $\partial_vX$ that does not contain any $x_i'$. Assign a plus sign to $\partial_hX$ and a minus sign to each annulus $\Gamma_i$. Similar to the proof of Lemma \[LBM1\], $W_X'$ is a relative compression body with $\partial_h^-W_X'=\bigcup_{i=1}^q\Gamma_i$, $\partial_h^+W_X'=\partial_hX$ and $\partial_v W_X'=\bigcup_{i=1}^p x_i$.
By Remark \[Rproperty\](4), there is a graph $Z$ in $W_X'$ connecting all the annuli $\Gamma_i$’s together such that $W_X'\setminus N(Z)$ is a product neighborhood of $\partial^+W_X'$. Moreover, by Remark \[Rproperty\](4), for any nontrivial subgraph $Z'$ of $Z$, $W_X'\setminus N(Z')$ is also a relative compression body. The graph $Z$ can be constructed to have $k$ disjoint subgraphs $Z_1,\dots Z_k$, such that
1. each $Z_j$ connects a subset $\widetilde{\Gamma}_j$ of the annuli $\Gamma_i$’s,
2. $\Gamma_a\in\widetilde{\Gamma}_j$ if and only if $\partial\Gamma_a\subset f(\partial S_j)$, $j=1,\dots, k$,
3. let $F_j'$ be the frontier surface of $\widetilde{\Gamma}_j\cup N(Z_j)$ in $W_X'$, then $g(S_j)=g(F_j')$.
Let $W_X=W_X'\setminus \cup_{j=1}^kN(Z_j)$. We set $\partial_h^+W_X=\partial_hX$, $\partial_vW_X=\partial_vW_X'$, and $\partial_h^-W_X=\bigcup_{i=1}^k F_i'$. Conditions (2) and (3) above imply that each $F_j'$ is homeomorphic to $S_j$. So $W_X$ is a relative compression body $\partial$-similar to $W_Y$. Moreover, each $F_i'$ (together with some subannuli of $\partial_vX$) bounds a marked handlebody in $X$. Thus we can proceed with the induction as in Case 1 and this proves Case 3.
Case 4. $S_1,\dots, S_k$ do not have the same sign.
The proof for Case 4 is a mix of the proofs for Cases 1, 2, and 3.
Let $y_i$ $S_i$ and $\Gamma_i$ be as above. Without loss of generality, we may suppose $S_i$ has a plus sign if $i\le t$ and has a minus sign if $t+1\le i\le k$. We divides these $\partial$-parallel annuli $\Gamma_i$’s into $k$ sets of annuli $\widetilde{\Gamma}_1,\dots, \widetilde{\Gamma}_k$ such that $\Gamma_a\in\widetilde{\Gamma}_j$ if and only if $\partial\Gamma_a\subset f(\partial S_j)$ ($j=1,\dots,k$).
Now we temporarily ignore $\widetilde{\Gamma}_{t+1},\dots,\widetilde{\Gamma}_k$ and proceed as in Case 1 and Case 2: We first construct suspension surfaces $F_1,\dots, F_t$ over the sets of annuli $\widetilde{\Gamma}_1,\dots,\widetilde{\Gamma}_t$ respectively. Let $V$ be the submanifold of $X$ bounded by $\partial_hX$, $F_1,\dots, F_t$, and the collection of subannuli of $\partial_vX$ between these surfaces. So $\partial_hV=\partial_hX\cup(\bigcup_{i=1}^t F_i)$ and $\partial_v V$ consists of subannuli of $\partial_vX$. Denote the annuli in $\partial_vV$ by $x_1,\dots, x_m$, $w_1,\dots,w_s$, where no $x_i$ ($i=1,\dots,m$) contains boundary curves of any annulus in $\{\widetilde{\Gamma}_{t+1},\dots,\widetilde{\Gamma}_k\}$, and each $w_j$ ($j=1,\dots,s$) contains boundary curves of at least one annulus in $\{\widetilde{\Gamma}_{t+1},\dots,\widetilde{\Gamma}_k\}$. Note that since the annuli $\Gamma_i$’s are non-nested, each annulus in $\{\widetilde{\Gamma}_{t+1},\dots,\widetilde{\Gamma}_k\}$ must have both boundary curves in the same $w_j$ for some $j=1,\dots, s$.
By our construction, $\partial F_i=f(\partial S_i)$ ($i=1,\dots,t$). We assign $F_1,\dots, F_t$ plus signs and assign $\partial_hX$ a $\pm$-sign according to the sign of $\partial_hY$. Note that each annulus $x_i$ ($i=1,\dots, m$) corresponds to a component $y_i$ of $\partial_vW_Y$ ($y_i=f(x_i)$). Assign a sign and an order to each $x_i$ according to the sign and order of $y_i$.
\[Claim2\] Each $w_j$ contains the boundary curves of exactly one annulus in $\{\widetilde{\Gamma}_{t+1},\dots,\widetilde{\Gamma}_k\}$, and both curves of $\partial w_j$ are boundary curves of $\partial_h^+ V$.
This proof is similar to the argument in Case 3. If two annuli $\Gamma_a$ and $\Gamma_b$ in $\{\widetilde{\Gamma}_{t+1},\dots,\widetilde{\Gamma}_k\}$ have boundary curves in the same $w_j$, then $w_j$ must contain a subannulus $x_c$ between $\Gamma_a$ and $\Gamma_b$ such that $\partial x_c\subset\bigcup_{i=t+1}^k\widetilde{\Gamma}_{i}$. Let $y_c = f^{-1}(x_c)$ be the corresponding annulus in $\partial_v W_Y$. Since $S_{t+1},\dots, S_k$ have minus signs, this means that $\partial y_c$ has both curves in $\partial_h^-W_Y$. However, this is impossible because $W_Y$ is a relative compression body and, by Definition \[Drelative compression body\], no annulus in $\partial_vW_Y$ has both boundary curves in $\partial_h^-W_Y$.
Similarly, if a curve of $\partial w_j$ is a boundary curve of $\partial_h^- V$ (this occurs only if $\partial_h X$ has a minus sign, since $F_1,\dots, F_t$ all have plus signs), then since $w_j$ contains the boundary curves of an annulus in $\{\widetilde{\Gamma}_{t+1},\dots,\widetilde{\Gamma}_k\}$, a subannulus $x_d$ of $w_j$ has one boundary curve in some $\widetilde{\Gamma}_{p}$ ($t+1\le p\le k$) and the other boundary curve in $\partial_h^- V$. Thus $y_d = f^{-1}(x_d)$ is a component of $\partial_v W_Y$ with both boundary curves in $\partial_h^-W_Y$, which is impossible.
Now we assign each annulus $w_i$ a minus sign and an order larger than the order of any annulus in $\partial_v^+W_Y$, and we view each $w_i$ as an annulus in $\partial_v^-V$. Since $\partial^+ W_Y$ is connected (see Lemma \[Lconnected\]), the construction of $F_i$ implies that $\partial^+V=\partial_h^+V\cup\partial_v^+V$ is connected. Thus the assumptions on $W_Y$ and the $S_i$’s imply that these signs and orders on $\partial V$ are admissible, see Definition \[Dadmissible\]. As before, we can find a minimal set of annuli connecting the components of $\partial_h^+V$, and we can assume the $F_i$’s and their cross-section disks are well-positioned. As in Case 1 and Case 2, we can construct a surface $F_i'$ by modifying $F_i$ and adding $g(S_i)$ tubes to $F_i$ ($i=1,\dots,t$), such that the submanifold $V'$ between $\partial_hX$ and $\cup_{i=1}^t F_i'$ is a relative compression body with each $w_i$ an annulus in $\partial_v^-V'$.
By Claim \[Claim2\], for each annulus $w_i$, there is exactly one annulus in $\{\widetilde{\Gamma}_{t+1},\dots,\widetilde{\Gamma}_k\}$ having boundary in $w_i$. Similar to Case 3, let $V''$ be the submanifold of $V'$ obtained by deleting the solid tori bounded by the $\partial$-parallel annuli in $\{\widetilde{\Gamma}_{t+1},\dots,\widetilde{\Gamma}_k\}$ and the corresponding subannuli of $w_1,\dots,w_s$. Now view the annuli in $\{\widetilde{\Gamma}_{t+1},\dots,\widetilde{\Gamma}_k\}$ as components of $\partial_h^-V''$. Similar to Case 3, since each $w_i$ is a component of $\partial_v^- V'$ and since $V'$ is a relative compressing body, if we maximally compress $\partial^+V'$ in $V''$, the resulting manifold is a product neighborhood of $\partial^-V''$. Therefore, $V''$ is a relative compression body (with orders and signs the same as corresponding components of $\partial_h^\pm V'$ and $\partial_v^\pm V'$). Note that, if $\partial_hX$ has a plus sign, then $\partial_h^- V''=\bigcup_{i=t+1}^k\widetilde{\Gamma}_{i}$, and if $\partial_hX$ has a minus sign, then $\partial_h^- V''=(\bigcup_{i=t+1}^k\widetilde{\Gamma}_{i})\cup\partial_hX$.
By Remark \[Rproperty\](4), there is a graph $Z$ in $V''$ connecting all the components of $\partial^- V''$, such that $V''\setminus N(Z)$ is a product neighborhood of $\partial^+V''$. Similar to Case 3, there are graphs $Z_{t+1},\dots, Z_k$ in $V''$ such that
1. each $Z_j$ connects all the annuli in $\widetilde{\Gamma}_j$, and
2. let $F_j'$ be the frontier surface of $\widetilde{\Gamma}_j\cup N(Z_j)$ in $V''$ ($j=t+1,\dots, k$), then $g(S_j)=g(F_j')$.
Let $W_X$ be the manifold obtained by deleting a small neighborhood of $\cup_{i=t+1}^kZ_i$ from $V''$. Note that after deleting $N(Z_j)$, the annuli in $\widetilde{\Gamma}_j$ merge into a connected surface $F_j'$ which is homeomorphic to $S_j$ ($j=t+1,\dots, k$). Since $F_i'\cong S_i$ for $i=1,\dots, t$, this means that $W_X$ is a relative compression body $\partial$-similar to $W_Y$. Thus we can proceed with the induction and this proves Case 4.
Separating surfaces {#Ssep}
===================
Let $M=\mathcal{W}\cup_\mathcal{T} \mathcal{V}$ and $N=\widehat{\mathcal{T}}\cup_\mathcal{T} \mathcal{V}$ be as in Theorem \[Ttorus\], where $H^1(\mathcal{W})=\mathbb{Z}$ and $\widehat{\mathcal{T}}$ is a solid torus. We view $\mathcal{V}$ as a submanifold of both $M$ and $N$. Let $\Sigma$ be the collection of incompressible and strongly irreducible surfaces in an untelescoping of a minimal genus Heegaard splitting of $M$. By Lemma \[Ldivide\] and Remark \[Rtorus\], we may assume that $\Sigma$ is in good position with respect to $\mathcal{T}$.
First, we would like to point out that we only need to consider the case that $\Sigma$ has no component entirely in $\mathcal{W}$. If $\Sigma $ has a strongly irreducible surface entirely in $\mathcal{W}$, then since $\mathcal{T}$ is incompressible, by maximally compressing this component on either side, we obtain an incompressible component of $\Sigma$ entirely in $\mathcal{W}$. For homology reason, each closed surface in $\mathcal{W}$ is separating. Thus there is a collection of incompressible components $\mathcal{F}_1,\dots, \mathcal{F}_n$ of $\Sigma$ entirely in $\mathcal{W}$, such that (1) each $\mathcal{F}_i$ bounds a submanifold $Z_i$ of $\mathcal{W}$, (2) these $Z_i$’s are non-nested, and (3) $\Sigma$ has no component entirely in $\mathcal{W}\setminus\bigcup_{i=1}^n Z_i$. Let $\Sigma'$ be the components of $\Sigma$ lying outside $\bigcup_{i=1}^n Z_i$. So no component of $\Sigma'$ is entirely in $\mathcal{W}$. Since $\mathcal{F}_i$ is an incompressible component in the untelescoping, $\Sigma\cap {\mathrm{int}}(Z_i)$ gives a generalized Heegaard splitting of $Z_i$. Let $\mathcal{A}_i$ be the compression body in the untelescoping that lies outside $Z_i$ and with $\mathcal{F}_i\subset\partial_-\mathcal{A}_i$. Let $\mathcal{P}_i=\partial_+\mathcal{A}_i$. So $\mathcal{P}_i$ is a component of $\Sigma'$. By rearranging handles in $Z_i$ (as a converse of the untelescoping), we first convert $\Sigma\cap {\mathrm{int}}(Z_i)$ into a Heegaard surface of $Z_i$. Then we amalgamate this Heegaard surface of $Z_i$ with $\mathcal{P}_i$ along $\mathcal{F}_i$, see [@Sch]. As in [@Sch], also see [@La2; @L4; @L14], this amalgamation operation is basically using a tube across $\mathcal{F}_i$ to connect $\mathcal{P}_i$ with a closed surface in $Z_i$. In particular, it does not affect how these surfaces intersect $\mathcal{T}$. We perform this amalgamation operation for each $Z_i$ and let $\Sigma''$ be the resulting collection of surfaces. $\Sigma''$ decomposes $M$ into a generalized Heegaard splitting with the same genus (it can be viewed as rearranging the same set of handles). Moreover, no component of $\Sigma''$ is entirely in $\mathcal{W}$. This amalgamation operation is away from the torus $\mathcal{T}$, so the intersection of $\mathcal{T}$ with each compression body is still a collection of incompressible annuli. Hence it follows from the proof of Lemma \[Ldivide\] that $\Sigma''$ is still in good position with respect to $\mathcal{T}$. We can continue the proof using $\Sigma''$ instead of $\Sigma$ and the proof is the same.
By the argument above, we may suppose that $\Sigma$ has no component entirely in $\mathcal{W}$.
For homology reasons, there is a nontrivial simple closed curve in $\mathcal{T}$ that is null-homologous in $\mathcal{W}$ (this curve bounds a non-separating Seifert surface in $\mathcal{W}$). Choose a framing for $H_1(\mathcal{T})$ by setting the slope of this curve to be $1/0$ (or $\infty$). We also call this slope a meridional slope. Note that $N$ is obtained from $\mathcal{V}$ by a Dehn filling along the $\infty$-slope.
In this section, we prove Theorem \[Ttorus\] in the case that every component of $\Sigma\cap\mathcal{W}$ is separating in $\mathcal{W}$. Suppose every component of $\Sigma\cap\mathcal{W}$ is separating in $\mathcal{W}$. We have 3 cases to discuss.
**Case (a)**. A component of $\Sigma\cap \mathcal{W}$ has genus at least one.
Let $\Sigma_0$ be a component of $\Sigma\cap \mathcal{W}$ with $g(\Sigma_0)\ge 1$. Since every component of $\Sigma\cap\mathcal{W}$ is separating in $\mathcal{W}$, $\partial\Sigma_0$ has an even number of components. Let $c_1,\dots, c_{2k}$ be the boundary curves of $\Sigma_0$. The curves $c_1,\dots, c_{2k}$ divide $\mathcal{T}$ into $2k$ annuli $A_1,\dots, A_{2k}$. Suppose $A_i$ is adjacent to $A_{i+1}$ for each $i$. As $\Sigma_0$ is separating in $\mathcal{W}$, $A_{2i}$ and $A_{2i-1}$ lie on different sides of $\Sigma_0$ for each $i$.
Now consider $N= \widehat{\mathcal{T}} \cup_\mathcal{T} \mathcal{V}$ and view $\mathcal{T}$ as the boundary of the solid torus $\widehat{\mathcal{T}}$.
We first construct a surface $\Sigma_0'$ properly embedded in $\widehat{\mathcal{T}}$ such that
1. $\partial\Sigma_0'=c_1\cup\cdots\cup c_{2k}$,
2. $g(\Sigma_0')=g(\Sigma_0)$ and
3. $\Sigma_0'$ divides the solid torus $\widehat{\mathcal{T}}$ into two marked handlebodies.
The construction is fairly straightforward. Let $\Gamma_i$ be a $\partial$-parallel annulus in the solid torus $\widehat{\mathcal{T}}$ with $\partial\Gamma_i=\partial A_{2i}$ ($i=1,\dots, k$). Let $N_i$ be the solid torus bounded by $\Gamma_i\cup A_{2i}$. Let $c$ be a core curve of the solid torus $\widehat{\mathcal{T}}$ and suppose $c$ lies outside each $N_i$. So $\widehat{\mathcal{T}}\setminus N(c)\cong T^2\times I$. Let $J=\alpha\times I$ be a vertical annulus in $\widehat{\mathcal{T}}\setminus N(c)\cong T^2\times I$ and we choose the slope of $\alpha$ so that $J$ meets each curve $c_i$ in a single point. Hence $J\cap N_i$ is a bigon meridional disk of $N_i$. Let $h_1,\dots h_k$ be a collection of 1-handles connecting $\overline{N(c)}$ to $N_1,\dots, N_k$ respectively along vertical arcs in the vertical annulus $J$. Let $X_0=\overline{N(c)}\cup (\bigcup_{i=1}^k (N_i\cup h_i)) $. So $X_0$ is a marked handlebody with $\partial_v X_0=\cup_{i=1}^k A_{2i}$ and $\partial_hX_0$ a genus-one surface properly embedded in $\widehat{\mathcal{T}}$. Since these 1-handles $h_i$ are unknotted, the closure of $\widehat{\mathcal{T}}\setminus X_0$ is a marked handlebody whose vertical boundary is $\cup_{i=1}^k A_{2i-1}$.
Now we add $g$ trivial 1-handles to $X_0$, where $g=g(\Sigma_0)-1$, and denote the resulting manifold by $X_0'$. Hence $X_0'$ is a marked handlebody with $\partial_h X_0'\cong\Sigma_0$. Let $X^c$ be the closure of the $\widehat{\mathcal{T}}\setminus X_0'$. Since these $g$ 1-handles are trivial 1-handles, $X^c$ is a marked handlebody with $\partial_v X^c=\cup_{i=1}^k A_{2i-1}$. Set $\Sigma_0'=\partial_hX_0'=\partial_hX^c$, so we have $\Sigma_0'\cong\Sigma_0$
$\Sigma_0'$ divides $\widehat{\mathcal{T}}$ into two marked handlebodies and $\Sigma_0$ divides $\mathcal{W}$ into two stacks of relative compression bodies. By applying Lemma \[LtypeI\] to each of the two stacks, we can construct a stack of relative compression bodies in $\widehat{\mathcal{T}}$ that is similar to the stack $\mathcal{W}$. By Lemma \[Lmerge\], we can connect the horizontal surfaces in $\widehat{\mathcal{T}}$ to $\Sigma\cap\mathcal{V}$ and obtain a generalized Heegaard splitting for $N$ with the same genus. Thus $g(N)\le g(W)$ and Theorem \[Ttorus\] holds in Case (a).
**Case (b)**. The curves in $\Sigma\cap \mathcal{T}$ have an integer slope.
This case is similar to Case (a). Let $\Sigma_0$ be a component of $\Sigma\cap \mathcal{W}$. Since $\Sigma_0$ is separating in $\mathcal{W}$, $\partial\Sigma_0$ has an even number of components. Let $c_1,\dots, c_{2k}$ be the boundary curves of $\Sigma_0$. Let $A_i$ ($i=1,\dots, 2k$), $\Gamma_j$ and $N_j$ ($j=1,\dots, k$) be as in Case (a).
Consider $N= \widehat{\mathcal{T}} \cup_\mathcal{T} \mathcal{V}$ and view $\mathcal{T}$ as the boundary of the solid torus $\widehat{\mathcal{T}}$. Let $\Delta$ be a compressing disk in the solid torus $\widehat{\mathcal{T}}$. Since the curves in $\Sigma\cap \mathcal{T}$ have an integer slope, $\partial\Delta$ intersects each curve $c_i$ in a single point. Hence $\Delta\cap N_i$ is a bigon compressing disk of $N_i$. Let $O$ be a point in ${\mathrm{int}}(\Delta)$ and outside each $N_i$. Let $h_1,\dots, h_k$ be a collection of 1-handles connecting a 3-ball $N(O)$ to $N_1,\dots, N_k$ respectively along arcs in $\Delta$. Let $X_0$ be the union of $N(O)$, the solid tori $N_i$’s, and the 1-handles $h_i$’s. So $X_0$ is a marked handlebody with $\partial_v X_0=\cup_{i=1}^k A_{2i}$. Since $\Delta$ is a compressing disk in the solid torus, the closure of $\widehat{\mathcal{T}}\setminus X_0$ is a marked handlebody whose vertical boundary is $\cup_{i=1}^k A_{2i-1}$. Then we add $g$ trivial 1-handles to $X_0$, where $g=g(\Sigma_0)$, and denote the resulting manifold by $X_0'$. Similar to Case (a), $X_0'$ and its complement are both marked handlebodies. Let $\Sigma_0'=\partial_h X_0'$. Thus $\Sigma_0'\cong \Sigma_0$. Now the proof is the same as Case (a).
**Case (c)**. Every component of $\Sigma\cap \mathcal{W}$ is a planar surface, but there is a relative compression body $Y$ in the stack $\mathcal{W}$ with $g(\partial^+Y)\ge 1$.
Recall that if one maximally compresses $\partial^+Y$ in $Y$, the resulting manifold is a product neighborhood of $\partial^-Y$. As in Remark \[Rproperty\](4), this means that there is a graph $G$ connecting all the components of $\partial^-Y$ such that $Y\setminus N(G\cup \partial^-Y)\cong (\partial^+Y)\times I$. Denote $Y'=Y\setminus N(G\cup \partial^-Y)$ and let $\Sigma_0$ be the frontier surface of $\overline{N(G\cup\partial^-Y)}$ in $Y$. So $\Sigma_0$ is a properly embedded surface in $Y$ and $\Sigma_0\subset\partial Y'$. We may view $Y'$ as a (trivial) relative compression body with $\partial^+Y'=\partial^+Y$ and $\partial_h^-Y'=\Sigma_0$. As $Y'\cong (\partial^+Y)\times I$, we have $\Sigma_0\cong\partial^+Y$. By Lemma \[Lconnected\], $\partial^+Y$ is connected. Hence $\Sigma_0$ is connected. Since $g(\partial^+Y)\ge 1$, $g(\Sigma_0)\ge 1$.
$\Sigma_0$ divides $Y$ into two submanifolds $Y'$ and $Y''$, where $Y''$ is a neighborhood of $G\cup\partial^-Y$. If we maximally compress $\Sigma_0$ in $Y''$, the resulting manifold is a product neighborhood of $\partial^-Y$. Hence we may view $Y''$ as a relative compression body with $\partial_h^+Y''=\Sigma_0$ and $\partial^-Y''=\partial^-Y$. Although $Y'$ and $Y''$ are both relative compression bodies, $Y=Y'\cup_{\Sigma_0}Y''$ is not a stack by definition because $\Sigma_0$ has a minus sign in $Y'$ but has a plus sign in $Y''$. Nonetheless, the closure of each component of $\mathcal{W}\setminus \Sigma_0$ is a stack.
As $g(\Sigma_0)\ge 1$, we can repeat the construction in Case (a) and construct a surface $\Sigma_0'$ in the solid torus $\widehat{\mathcal{T}}$, such that $\Sigma_0'\cong\Sigma_0$, $\partial\Sigma_0'=\partial\Sigma_0$ in $\mathcal{T}$, and $\Sigma_0'$ divides $\widehat{\mathcal{T}}$ into two marked handlebodies. Now the proof is the same as Case (a): in each of the two marked handlebodies, we apply Lemma \[LtypeI\] and construct a collection of surfaces, dividing the marked handlebody into a stack that is similar to the corresponding stack of $\overline{\mathcal{W}\setminus \Sigma_0}$. After removing $\Sigma_0'$, we obtain a stack $\widehat{\mathcal{T}}$ that is similar to the stack $\mathcal{W}$. Now the proof is the same as Case (a) and this finishes Case (c).
Suppose every component of $\Sigma\cap \mathcal{W}$ is a planar surface and, for each relative compression body $Y$ in the stack $\mathcal{W}$, $\partial^+Y$ is a planar surface. Then the slope of the curves in $\Sigma\cap \mathcal{T}$ must be an integer.
$\Sigma\cap \mathcal{W}$ divides $\mathcal{W}$ into a stack of relative compression bodies. Let $Y$ be any relative compression body in this stack. By Remark \[Rproperty\](5), the manifold $\widehat{Y}$ obtained by adding a 2-handle along each annulus in $\partial_v^-Y$ is also a relative compression body. Next, add 2-handles to $\widehat{Y}$ along every annulus in $\partial_v^0Y$. By Remark \[Rproperty\](6), the resulting manifold, denoted by $\widehat{Y}'$, is also a relative compression body. Since each component of $\Sigma\cap \mathcal{W}$ is a planar surface, $\partial^-\widehat{Y}'$ is a collection of 2-spheres. By the hypothesis of the claim, $\partial^+Y$ is a planar surface, which implies that $\partial^+\widehat{Y}'$ is a 2-sphere. Hence $\widehat{Y}'$ is a punctured 3-ball (i.e. a 3-ball with a collection of 3-balls removed from its interior). Let $\widehat{Y}''$ be the manifold obtained from $\widehat{Y}'$ by adding a 2-handle along each annulus in $\partial_v^+Y$. Since $\widehat{Y}'$ is a punctured 3-ball, so is $\widehat{Y}''$.
Let $\mathcal{W}'=\mathcal{W}\cup (D^2\times S^1)$ be the manifold obtained by a Dehn filling on $\mathcal{W}$ with each curve of $\Sigma\cap \mathcal{T}$ bounding a disk in the solid torus $D^2\times S^1$. The argument above implies that each relative compression body in the stack $\mathcal{W}$ extends to a punctured 3-ball in $\mathcal{W}'$. Hence $\mathcal{W}'$ is either $S^3$ or a connected sum of $S^2\times S^1$.
Now consider the slope of the curves in $\Sigma\cap \mathcal{T}$. If the slope is neither an integer nor the $\infty$-slope, then $H_1(\mathcal{W}')$ is a nontrivial finite cyclic group, contradicting our conclusion that $\mathcal{W}'$ is either $S^3$ or a connected sum of $S^2\times S^1$. If the slope is the $\infty$-slope, then $H_1(\mathcal{W}')=\mathbb{Z}$ and $\mathcal{W}'$ must be $S^2\times S^1$. Since each relative compression body in the stack $\mathcal{W}$ extends to a punctured 3-ball in $\mathcal{W}'$, $\Sigma\cap\mathcal{W}$ extends into a collection of 2-spheres that divide $\mathcal{W}'$ into a collection of punctured 3-balls. This implies that there must be a component of $\Sigma\cap\mathcal{W}$ that extends to a non-separating 2-sphere in $\mathcal{W}'=S^2\times S^1$. This means that a component of $\Sigma\cap\mathcal{W}$ has an odd number of boundary curves and is non-separating in $\mathcal{W}$, a contradiction to the hypothesis of this section.
Theorem \[Ttorus\] in the case that every surface in $\Sigma\cap\mathcal{W}$ is separating now follows from the claim above and the discussions in the 3 cases: By Cases (a), we may assume every component of $\Sigma\cap\mathcal{W}$ is a planar surface. By Case (c), we may assume that for each relative compression body $Y$ in the stack $\mathcal{W}$, $\partial^+Y$ is a planar surface. So it follows from the Claim that the slope of the curves in $\Sigma\cap\mathcal{T}$ is an integer. Now Theorem \[Ttorus\] follows from Case (b).
Next, we consider the case that a component of $\Sigma\cap\mathcal{W}$ is separating.
Type II blocks {#StypeII}
==============
In this section, we describe the second type of building block in the construction.
We call a genus-$g$ handlebody $X$ an **extended marked handlebody** if $\partial X$ contains $m+1$ ($m\le g$) disjoint annuli $A_0, A_1,\dots, A_m$, such that
1. $A_0$ is trivial (i.e. $\partial A_0$ is trivial in $\partial X$) and each $A_i$ ($i\ge 1$) is an essential annulus in $\partial X$
2. $X$ contains $m$ disjoint meridional disks $\tau_1,\dots, \tau_m$ such that $\tau_i\cap A_j=\emptyset$ if $i\ne j$, and $\tau_i\cap A_i$ is a single essential arc of $A_i$ for each $i\ge 1$.
The only difference between a marked handlebody and an extended marked handlebody is that one marked annulus is trivial in an extended marked handlebody.
Similar to the description of a marked handlebody in section \[StypeI\], we can describe $X$ as follows: Start with a marked handlebody $X_1$ constructed by first connecting a central 3-ball $B$ to a collection of solid tori $T_1,\dots, T_m$ using 1-handles, and then adding $g-m$ 1-handles to $B$ (here we use the same notation as in section \[StypeI\]). Let $A_1, \dots, A_m$ be the annuli in $\partial T_1,\dots, \partial T_m$ with $\partial_vX_1=\bigcup_{i=1}^m A_i$. Next, connect a 3-ball $B'=D^2\times I$ to $B$ using a 1-handle $H_0$ and require $H_0$ is attached to $D^2\times\{1\}\subset\partial B'$. Denote the resulting manifold by $X$ and denote $(\partial D^2)\times I$ by $A_0$. The set of annuli $A_0, A_1,\dots A_m$ are our marked annuli on $\partial X$.
The boundary of $X$ has two parts: (1) the vertical boundary $\partial_vX=\bigcup_{i=0}^m A_i$ and (2) the horizontal boundary $\partial_hX=\overline{\partial X\setminus\partial_vX}$. So $\partial_hX$ has two components, one of which is the disk $D^2\times\{0\}\subset\partial B'$.
Next, we describe the cross-section disks and suspension surfaces for $X$. We view $X$ as the manifold obtained by connecting a marked handlebody $X_1$ to a 3-ball $B'=D^2\times I$ via a 1-handle $H_0$ as above. A cross-section disk for $X$ is simply an extension of a cross-section disk for $X_1$ into the 1-handle $H_0$ and the 3-ball $B'$. More precisely, start with a cross-section disk $E'$ for the marked handlebody $X_1$. Let $H_0=\Delta_0\times I$ be the 1-handle connecting $X_1$ to $B'=D^2\times I$ with $\Delta_0\times\{0\}\subset D^2\times\{1\}\subset\partial B'$. We may suppose that there is a disk $E_H$ properly embedded in the 1-handle $H_0=\Delta_0\times I$ in the form of $\delta_0\times I$ such that $E_H\cap E'$ is the arc $\delta_0\times \{1\}$ in $\partial H_0$. Moreover, we may suppose that there is a disk $E_D$ properly embedded in $B'=D^2\times I$ in the form of $\delta\times I$, such that $E_D\cap E_H$ is the arc $\delta_0\times \{0\}$ in $\partial H_0$. Let $E=E'\cup E_H\cup E_D$. We call $E$ a [**cross-section disk**]{} for $X$.
A [**suspension surface**]{} in $X$ is constructed in the same way as in section \[StypeI\]: Start with a collection of non-nested $\partial$-parallel annuli, each of which is parallel to a subannulus of some $A_i$ ($i=0, 1,\dots, m$). Then connect these $\partial$-parallel annuli to a central 2-sphere along arcs in a cross-section disk.
Denote the two components of $\partial_hX$ by $F_0$ and $F_1$, where $F_0=D^2\times\{0\}\subset\partial B'$ is the disk component. Let $S$ be a suspension surface in $X$ over a set of $\partial$-parallel annuli. Let $\hat{S}$ be the surface obtained by connecting $S$ and the disk $F_0$ by a tube along an arc in the cross-section disk, see Figure \[Ftype2\](a) for a 1-dimensional schematic picture. We call $\hat{S}$ an [**extended suspension surface**]{}. Let $\hat{S}'$ be the surface obtained by adding some (possibly none) trivial tubes to $\hat{S}$, and let $X_0$ be the submanifold of $X$ bounded by $F_1$, $\hat{S}'$, and the subannuli of $\partial_vX$ between them. We call $X_0$ a [**type II block**]{}. Similarly, define the horizontal and vertical boundaries of $X_0$ as $\partial_hX_0=F_1\cup \hat{S}'$ and $\partial_vX_0=\overline{\partial X_0\setminus\partial_hX_0}$ respectively.
[**Conversion between type I and type II blocks:**]{}
Let $T_0, T_1,\dots, T_m$ be a collection of solid tori and let $\mathfrak{X}$ be a marked handlebody constructed by connecting these $T_i$’s to a central 3-ball $B$ using 1-handles, as in section \[StypeI\]. Suppose $\partial_v\mathfrak{X}$ consists of annuli $A_0,\dots, A_m$, where each $A_i$ is an annulus in $\partial T_i$. Let $\Gamma_0,\dots,\Gamma_n$ be a collection of non-nested $\partial$-parallel annuli in $\mathfrak{X}$ with $\partial\Gamma_i\subset\partial_v\mathfrak{X}$, and let $S$ be a suspension surface over these $\Gamma_i$’s. Let $\mathfrak{X}_0$ be the submanifold of $\mathfrak{X}$ bounded by $S$, $\partial_h\mathfrak{X}$, and the subannuli of $\partial_v\mathfrak{X}$ between them. So $\mathfrak{X}_0$ is a type I block by definition.
Suppose $\partial\Gamma_0\subset A_0$, and suppose a component of $\partial A_0$ and a component of $\partial\Gamma_0$ bounds a subannulus $A_0'\subset A_0$ such that ${\mathrm{int}}(A_0')$ does not contain boundary curves of any $\Gamma_i$. In particular, $A_0'$ is a component of $\partial_v \mathfrak{X}_0$. Now add a 2-handle to $\mathfrak{X}_0$ along the annulus $A_0'$ and denote the resulting manifold by $X_0$. We claim that $X_0$ is a type II block. To see this, start with the solid torus $T_0$ in the construction of $\mathfrak{X}$ and view $\Gamma_0$ as a $\partial$-parallel annulus in $T_0$. Let $\widehat{\Gamma}_0$ be the solid torus bounded by $\Gamma_0$ and the subannulus of $A_0$ between $\partial\Gamma_0$. We may view the closure of $T_0\setminus\widehat{\Gamma}_0$ as a product $A\times I$, where $A$ is an annulus, $\Gamma_0=A\times\{0\}$, and the annulus $A_0'$ described above is a component of $(\partial A)\times I$. So adding a 2-handle along $A_0'$ changes $A\times I$ to a 3-ball $B'=D^2\times I$, and $\Gamma_0$ extends to the disk $D^2\times\{0\}$. Moreover, the set of 1-handles connect $B'$, $T_1,\dots, T_m$ and $B$ into an extended marked handlebody. Further, the suspension surface $S$ in $\mathfrak{X}$ becomes an extended suspension surface in this extended marked handlebody. Thus $X_0$ is a type II block.
Conversely, given a type II block $X_0$ as above, let $\alpha$ be an arc of the form $\{x\}\times I\subset D^2\times I=B'$ and suppose $\alpha$ is disjoint from the cross-section disk of $\hat{S}$. Then $\mathfrak{X}_0=X_0\setminus N(\alpha)$ is a type I block (one can view $N(\alpha)$ as the 2-handle above and view $\alpha$ to be a co-core of the 2-handle).
\[Rconversion\] Let $X_0$ be a type II block and let $\alpha$ be the arc of the form $\{x\}\times I\subset D^2\times I=B'$ as above. Since $\alpha$ is unknotted, $X_0\setminus N(\alpha)$ is a topological handlebody. We may view the closure of $N(\alpha)$ as a cylinder $\Delta_\alpha\times I$, where $\Delta_\alpha$ is a disk and $\Delta_\alpha\times \partial I\subset\partial_hX_0$. In the description above, we view $\mathfrak{X}_0=X_0\setminus N(\alpha)$ as a type I block and view $A_0'=\partial\Delta_\alpha\times I$ as a component of $\partial_v\mathfrak{X}_0$. Moreover, $\partial_h \mathfrak{X}_0=\partial_h X_0\setminus (\Delta_\alpha\times \partial I)$, and $\partial_h \mathfrak{X}_0$ has two components which we denote by $S_1$ and $S_2$. Note that we may view $\mathfrak{X}_0$ and its boundary surface in a slightly different way: If we include the annulus $A_0'$ as part of $\partial_h \mathfrak{X}_0$ instead, in other words, if we define $\partial_h\mathfrak{X}_0=S_1\cup A_0'\cup S_2$, then it follows from the proof of Lemma \[LW1\] that $\mathfrak{X}_0$ is a marked handlebody with $\partial_h\mathfrak{X}_0=S_1\cup A_0'\cup S_2$.
[type2.eps]{} (20,-5)[(c)]{} (67,-5)[(d)]{} (20,34)[(a)]{} (67,34)[(b)]{} (23,60)[$\hat{S}$]{} (34,51)[$F_1$]{} (0,21)[$D^-$]{} (26,16)[$D^+$]{} (12,11)[$P_{k-1}$]{} (60,55)[$P_{k-1}^-$]{} (84,55)[$P_{k-1}^+$]{} (86.5,35)[$\Gamma_{k-2}$]{} (48,21)[$P_{k-2}^-$]{} (74,20)[$P_{k-2}^+$]{} (58,17.5)[$\mathfrak{X}_{k-2}$]{}
Non-separating surfaces {#Snonsep}
=======================
Let $M=\mathcal{W}\cup_\mathcal{T} \mathcal{V}$ and $N=\widehat{\mathcal{T}}\cup_\mathcal{T} \mathcal{V}$ be as in Theorem \[Ttorus\]. Let $\Sigma$ be the collection of surfaces in an untelescoping of a minimal genus Heegaard splitting of $M$. In this section, we prove Theorem \[Ttorus\] in the case that $\Sigma\cap\mathcal{W}$ contains a component that is non-separating in $\mathcal{W}$. This, combined with section \[Ssep\], gives a full proof of Theorem \[Ttorus\].
As in section \[Ssep\], we view $\mathcal{T}$ as a torus in both $M$ and $N$, and we may assume that $\Sigma$ has no component that is entirely in $\mathcal{W}$. Suppose at least one component of $\Sigma\cap \mathcal{W}$ is non-separating in $\mathcal{W}$. By our choice of framing, the curves $\Sigma\cap \mathcal{T}$ in $\mathcal{T}$ must have the $\infty$-slope.
Let $S$ be a non-separating component of $\Sigma\cap \mathcal{W}$. Denote the set of boundary curves of $S$ by $C_0$. We fix a normal direction for $S$ which induces a normal direction for each circle of $C_0$ in $\mathcal{T}$. By the homology assumption on $\mathcal{W}$, $C_0$ (with the induced orientation) represents a primitive element in $H_1(\mathcal{T})$. In particular, $C_0$ contains an odd number of curves. Curves in $C_0$ divide the torus $\mathcal{T}$ into an odd number of annuli and denote this set of annuli by $\mathcal{A}_0$. We call an annulus in $\mathcal{A}_0$ an inner annulus if the normal directions of both of its boundary curves point out of this annulus. The homology conclusion on $C_0$ implies that, if $|C_0|>1$, then there must be at least one inner annulus in $\mathcal{A}_0$. Let $\mathcal{A}_0'$ be the set of all inner annuli in $\mathcal{A}_0$.
Next, we remove the boundary curves of the annuli in $\mathcal{A}_0'$ from $C_0$ and let $C_1$ denote the remaining set of curves. Clearly $C_0$ and $C_1$ represent the same element in $H_1(\mathcal{T})$. Now we apply the same procedure on $C_1$: $C_1$ divides $\mathcal{T}$ into a collection of annuli, which we denote by $\mathcal{A}_1$, and we define inner annulus for $\mathcal{A}_1$ similarly. If $|C_1|>1$, there must be at least one inner annulus in $\mathcal{A}_1$. Let $\mathcal{A}_1'$ be the set of inner annuli in $\mathcal{A}_1$. Note that, by our construction, each inner annulus in $\mathcal{A}_1'$ must contain at least one inner annulus of $\mathcal{A}_0'$ as a subannulus. Similarly, let $C_2$ be the subset of $C_1$ obtained by deleting all the boundary curves of the annuli in $\mathcal{A}_1'$. This procedure produces a sequence of sets of curves $C_0\supset C_1\supset\cdots\supset C_k$ and eventually $C_k$ is a single curve. Moreover, each annulus in $\mathcal{A}_i'$ must contain at least one annulus of $\mathcal{A}_{i-1}'$ as a subannulus.
Next, we build a planar surface $P$ in $\widehat{\mathcal{T}}$, such that $\partial P$ is the same set of curves as $\partial S$ in $\mathcal{T}$. Let $D$ be a meridional disk of $\widehat{\mathcal{T}}$ bounded by $C_k$ (recall that $C_k$ is a single curve). For each inner annulus $A$ in $\mathcal{A}_i'$, build a $\partial$-parallel annulus $\Gamma$ in $\widehat{\mathcal{T}}$ parallel to $A$ and with $\partial\Gamma=\partial A$. Let $\Gamma_i$ be the set of such $\partial$-parallel annuli parallel to the set of inner annuli in $\mathcal{A}_i'$. We may suppose $D$ and all the annuli in the $\Gamma_i$’s are pairwise disjoint. Assign normal directions to $D$ and these $\partial$-parallel annuli according to the normal directions of their boundary curves. So the union of $D$ and these $\partial$-parallel annuli is a disconnected surface whose boundary is the same as $\partial S$.
Now we use tubes to connect $D$ and the annuli in the $\Gamma_i$’s together in a way that preserves the induced normal directions. The meridional disk $D$ cuts the solid torus $\widehat{\mathcal{T}}$ into a product $D\times I$. Denote $D\times\{0\}$ by $D^-$ and $D\times\{1\}$ by $D^+$, and view $D^\pm$ as the $\pm$-side of $D$. Suppose the induced normal direction of $D^+$ points out of $D\times I$ and the direction of $D^-$ points into $D\times I$.
As illustrated in the 1-dimensional schematic picture in Figure \[Ftype2\](c), we first connect each annulus in $\Gamma_{k-1}$ to $D^-$ using a trivial arc. Then we add tubes along these arcs to connect $D^-$ and the annuli in $\Gamma_{k-1}$. Denote the resulting planar surface by $P_{k-1}$. By the definition of inner annulus, the induced normal directions for $\Gamma_{k-1}$ and $D^-$ are compatible in $P_{k-1}$. We may view $P_{k-1}$ as a non-separating surface in $\widehat{\mathcal{T}}$.
As illustrated in Figure \[Ftype2\](b), if we cut $\widehat{\mathcal{T}}$ open along $P_{k-1}$, the resulting manifold, denoted by $\mathfrak{X}_{k-1}$, is a type II block. Denote the two components of $\partial_h\mathfrak{X}_{k-1}$ by $P_{k-1}^-$ and $P_{k-1}^+$, and suppose the induced normal direction at $P_{k-1}^-$ points into $\mathfrak{X}_{k-1}$, see Figure \[Ftype2\](b). Moreover, $\Gamma_{k-2}$ can be viewed as a collection of $\partial$-parallel annuli in $\mathfrak{X}_{k-1}$.
Similarly, we connect $P_{k-1}^-$ to the annuli in $\Gamma_{k-2}$ using tubes along unknotted arcs, and denote the resulting planar surface by $P_{k-2}$. We may view $P_{k-2}$ as a non-separating surface in $\widehat{\mathcal{T}}$ and let $\mathfrak{X}_{k-2}$ be the manifold obtained by cutting $\widehat{\mathcal{T}}$ open along $P_{k-2}$. As illustrated in the schematic picture in Figure \[Ftype2\](d), $\mathfrak{X}_{k-2}$ is also a type II block with the two components of $\partial_h\mathfrak{X}_{k-2}$ being the two sides of $P_{k-2}$. Denote the two sides of $P_{k-2}$ by $P_{k-2}^\pm$.
We continue this procedure: Let $P_{j-1}$ be the planar surface obtained by connecting the annuli in $\Gamma_{j-1}$ to $P_j^-$ using tubes along unknotted arcs for each $j$. Denote the final planar surface $P_0$ by $P$. So $P$ is a non-separating planar surface with $\partial P=C_0=\partial S$ and with compatible normal direction. Moreover, the manifold obtained by cutting $\widehat{\mathcal{T}}$ open along $P$ is a type II block.
Suppose $\Sigma\cap \mathcal{W}$ contains another non-separating surface $S'$. Next we build a non-separating planar surface $P'$ in the solid torus $\widehat{\mathcal{T}}$ with $\partial P'=\partial S'$ in $\mathcal{T}$ and such that $P'$ and $P$ divide $\widehat{\mathcal{T}}$ into a pair of type II blocks.
Assign a normal direction to $S'$ compatible with the normal direction of $S$, i.e. $S$ and $S'$ (with this orientation) represent the same element in $H_2(\mathcal{W},\mathcal{T})$. The normal direction of $S'$ induces an orientation for each boundary curve. We say two curves in $\partial S$ and $\partial S'$ have the same orientation if they represent the same element in $H_1(\mathcal{T})$, otherwise, they have opposite orientations.
Let $Y_P$ be the type II block obtained by cutting the solid torus $\widehat{\mathcal{T}}$ open along $P$. We may view $\partial S'$ as a set of curves in $\partial_vY_P$.
\[Claimcurves\] Let $A$ be a component of $\partial_vY_P$, and let $\gamma'$ and $\gamma''$ be the two boundary curves of $A$. Let $\gamma_1,\dots,\gamma_n$ be the components of $\partial S'$ that lie in $A$. Suppose $\gamma_i$ is adjacent to $\gamma_{i+1}$ for any $i=1,\dots, n-1$, and suppose $\gamma'$ (resp. $\gamma''$) is adjacent to $\gamma_1$ (resp. $\gamma_n$) in $A$. Then
1. if $\gamma'$ and $\gamma''$ have the same orientation, then $A$ contains at least one curve of $\partial S'$,
2. any two adjacent curves $\gamma_i$ and $\gamma_{i+1}$ have opposite orientations,
3. $\gamma'$ and $\gamma_1$ have the same orientation, and
4. $\gamma''$ and $\gamma_n$ have the same orientation
Suppose part (1) is false, i.e. $\gamma'$ and $\gamma''$ have the same orientation and $A\cap\partial S'=\emptyset$. Let $\tau$ be an essential arc in $A$ connecting $\gamma'$ to $\gamma''$. Since $\gamma'$ and $\gamma''$ have the same orientation, $\tau$ is an arc connecting the plus side of $S$ to its minus side and $\tau\cap S'=\emptyset$. However, since $S$ and $S'$ represent the same element in $H_2(\mathcal{W}, \mathcal{T})\cong\mathbb{Z}$, any arc connecting the plus side to the minus side of $S$ must intersect $S'$. We have a contradiction.
Similarly, if two adjacent curves $\gamma_i$ and $\gamma_{i+1}$ have the same orientation, then an arc in $A$ connecting $\gamma_i$ to $\gamma_{i+1}$ is an arc connecting the plus side of $S'$ to its negative side without intersecting $S$. Again, this is impossible since $H_2(\mathcal{W}, \mathcal{T})\cong\mathbb{Z}$. Hence part (2) of the claim holds.
Now consider $\gamma'$ and $\gamma_1$. Since $\gamma'$ and $\gamma_1$ are adjacent in $A$, there is an arc $\tau$ in $A$ connecting $\gamma'$ to $\gamma_1$ and disjoint from all other curves. Without loss of generality, suppose $\tau$ is on the plus side of $S$. If $\gamma'$ and $\gamma_1$ have opposite orientations, then $\tau$ connects the plus side of $S$ to the plus side of $S'$. This is also impossible since $S$ and $S'$ have compatible orientations and represent the same element in $H_2(\mathcal{W},\mathcal{T})$, which means that the plus side of $S$ faces the minus side of $S'$. Thus part (3) of the claim holds. Part (4) is similar to part (3).
Next, we construct a non-separating planar surface $P'$ in the solid torus $\widehat{\mathcal{T}}$, such that $\partial P'=\partial S'$, and $P$ and $P'$ divide $\widehat{\mathcal{T}}$ into two type II blocks.
Before we proceed, we describe the type II block $Y_P$ (obtained by cutting the solid torus $\widehat{\mathcal{T}}$ open along $P$) as in the definition of type II block in section \[StypeII\]: First let $X_P$ be an extended marked handlebody obtained by connecting $D^2\times I$ and a collection of solid tori $T_1,\dots, T_m$ to a 3-ball $B$ using 1-handles. Set $A_0=(\partial D^2)\times I$, $A_i\subset\partial T_i$ and $\partial_vX_P=\bigcup_{i=0}^mA_i$ as in section \[StypeII\]. The two components of $\partial_hX_P$ consist of the disk $D^2\times\{0\}$ and a surface that we denote by $P^+$. Suppose $P^+\cong P$ ($P^+$ denotes the plus side of $P$). Let $P^-$ be an extended suspension surface obtained by tubing together $D^2\times\{0\}$ and a collection of non-nested $\partial$-parallel annuli $\Gamma_1,\dots,\Gamma_s$. $Y_P$ can be defined to be the submanifold of $X_P$ bounded by $P^-$, $P^+$, and a collection of subannuli in $\partial_vX_P$. We view $P^\pm$ as the two sides of $P$. Suppose the induced normal direction for $P^+$ points out of $Y_P$ and the direction for $P^-$ points into $Y_P$.
Consider the curves in $\partial S'$. Since $Y_P\subset X_P$, we may view $\partial_vY_P$ as a collection of subannuli of $\partial_vX_P$ and $\partial S'\subset\partial_vY_P\subset \partial_vX_P$. In particular, we also view $\partial S'$ as curves in $\partial_vX_P=\bigcup_{i=0}^mA_i$.
\[ClaimS’\] Let $X_P$ be the extended marked handlebody as above and $\partial S'\subset\partial_vX_P$. Then
1. any two adjacent curves of $\partial S'$ in each $A_i$ ($i=0,\dots, m$) have opposite orientations
2. $A_0\cap \partial S'$ contains an odd number of curves, and
3. $A_i\cap \partial S'$ contains an even number of curves for each $i\ge 1$.
We first prove part (1). By Claim \[Claimcurves\](2), any two adjacent curves of $\partial S'$ in each component of $\partial_vY_P$ have opposite signs. So either part (1) holds or there are two curves $\gamma_1$, $\gamma_2$ of $\partial S'$ that are adjacent in $A_i$ ($i=0,\dots, m$) but lie in different components of $\partial_vY_P$. Let $A_\gamma$ be the subannulus of $A_i\subset \partial_vX_P$ bounded by $\gamma_1$ and $\gamma_2$. Since $\partial S'\subset\partial_vY_P$ and since $\gamma_1$ and $\gamma_2$ lie in different components of $\partial_vY_P$, $A_\gamma$ contains two subannuli of $\partial_vY_P$ next to $\partial A_\gamma$ and possibly some whole components of $\partial_vY_P$. In particular, $A_\gamma$ contains an even number of curves of $\partial S$.
If $A_\gamma$ contains a whole component $E$ of $\partial_vY_P$, then since $\gamma_1$ and $\gamma_2$ are adjacent in $A_i$, $E$ does not contain any curve of $\partial S'$, and by Claim \[Claimcurves\](1), the two boundary curves of $E$ must have opposite orientations. Recall that we view $P^-$ as an extended suspension surface obtained by tubing together $\Gamma_1,\dots,\Gamma_s$ and $D^2\times\{0\}$. So the two boundary curves of each $\partial$-parallel annulus $\Gamma_i$ have opposite orientations. Any two adjacent curves of $\partial S$ in $A_\gamma$ are either boundary curves of a component $E$ of $\partial_vY_P$ or the boundary curves of $\Gamma_i$ for some $i=1,\dots, s$. This implies that any two adjacent curves of $\partial S$ in $A_\gamma$ have opposite orientations. Since $A_\gamma$ contains an even number of curves of $\partial S$, it follows from Claim \[Claimcurves\](3, 4) that $\gamma_1$ and $\gamma_2$ must have opposite orientations. Hence part (1) of the claim holds.
An essential arc of the annulus $A_0$ is an arc connecting $P^-$ to $P^+$, which gives rise to an arc in $\mathcal{W}$ connecting the plus side of $S$ to its minus side. Since $S$ and $S'$ represent the same element in homology, this arc must intersect $S'$ an odd number of times, and this implies that $A_0\cap\partial S'$ contains an odd number of curves. Hence part (2) of the claim holds. Similarly, an essential arc of $A_i$ ($i\ge 1$) is an arc connecting $P^+$ to $P^+$, so part (3) holds for the same homological reason.
Next we temporarily ignore $P^-$ and construct a surface $P'$ in $X_P$ with $\partial P'=\partial S'$.
Let $\delta_0$ be the component of $A_0\cap \partial S'$ that is adjacent to the curve $\partial(D^2\times\{0\})$ in $A_0$, and let $\hat{\delta}_0$ be the disk in $B'=D^2\times I$ in the form of $D^2\times\{x\}$ bounded by $\delta_0$, see the 1-dimensional schematic picture in Figure \[Fnonsep\](a). By Claim \[ClaimS’\](1), any two adjacent curves of $A_i\cap\partial S'$ have opposite orientations. So we can use a collection of non-nested $\partial$-parallel annuli $\Gamma_1',\dots,\Gamma_t'$ to connect the remaining curves $\partial S'\setminus\delta_0$ in pairs, see Figure \[Fnonsep\](a). Then we use tubes to connect the disk $\hat{\delta}_0$ and these $\partial$-parallel annuli $\Gamma_1',\dots,\Gamma_t'$ together similar to the construction of the extended suspension surface in section \[StypeII\]. Denote the resulting surface by $P'$, so $\partial P'=\partial S'$. See the red curves in Figure \[Fnonsep\](b) for a schematic picture of $P'$. Moreover, $P'$ has a normal direction compatible with the orientations of the curves in $\partial S'$.
[nonsep.eps]{} (22,-5)[(a)]{} (74,-5)[(b)]{} (3.5, 25)[$\hat{\delta}_0$]{} (34, 2)[$\Gamma_i'$]{} (73, 19)[$P'$]{} (36.5, 29)[$D^2\times\{0\}$]{} (63, 2)[$X_0$]{} (75, 29.5)[$X_1$]{}
As illustrated in Figure \[Fnonsep\](b), $P'$ divides $X_P$ into two submanifolds $X_0$ and $X_1$ with $D^2\times\{0\}\subset \partial X_0$ and $P^+\subset\partial X_1$. By the construction of $P'$, we may view $X_0$ as an extended marked handlebody and view $X_1$ as a type II block. Since $\partial S$ and $\partial S'$ represents the same element in $H_1(\mathcal{T})$, the normal direction of $P'$ is compatible with the direction of $P^+$. So the direction of $P'$ points into $X_1$ and out of $X_0$.
Now we consider $P^-$. Recall that the extended suspension surface $P^-$ was constructed by connecting $D^2\times\{0\}$ and a collection of $\partial$-parallel annuli $\Gamma_1,\dots,\Gamma_s$ using tubes. Let $\widehat{\Gamma}_i$ be the solid torus bounded by $\Gamma_i$ and the subannulus of $\partial_vX_P$ between $\partial\Gamma_i$ . The type II block $Y_P$ and each $\widehat{\Gamma}_i$ lie on different sides of $P^-$ in $X_P$. Since $\partial P'$ lies in $\partial_vY_P$, this implies that no curve of $\partial P'=\partial S'$ lies inside the solid torus $\widehat{\Gamma}_i$. Thus, each solid torus $\widehat{\Gamma}_i$ ($i=1,\dots, s$) lies entirely in either $X_0$ or $X_1$ (after possibly shrinking $\widehat{\Gamma}_i$ if necessary).
\[Claimannuli\] $\Gamma_1,\dots,\Gamma_s$ all lie in $X_0$ (after possibly an isotopy relative to $\partial\Gamma_i$).
Suppose the claim is false. Then, since each solid torus $\widehat{\Gamma}_i$ ($i=1,\dots, s$) lies entirely in either $X_0$ or $X_1$, there must be some annulus $\Gamma_i$ with $\widehat{\Gamma}_i\subset X_1$. We may choose such a $\Gamma_i$ to be adjacent to some curve $\gamma$ of $\partial P'=\partial S'$. In other words, choose $\Gamma_i$ so that there is an annulus $C$ between $\Gamma_i$ and a curve $\gamma$ of $\partial P'$ with the following properties:
1. the two components of $\partial C$ are $\gamma$ ($\gamma\subset\partial P'$) and a component of $\partial\Gamma_i$,
2. $C\subset X_1$ and ${\mathrm{int}}(C)$ does not contain a boundary curve of any $\Gamma_j$ ($j=1,\dots, s$).
By the construction of $P^-$, the normal direction of $\Gamma_i$ points out of the solid torus $\widehat{\Gamma}_i$. We have concluded earlier that the induced direction of $P'$ points into $X_1$. This implies that the normal directions at the two boundary curves of $C$ must both point into $C$. However, this means that the two curves of $\partial C$ have opposite orientations, contradicting Claim \[Claimcurves\](3, 4).
Since $P^-$ is an extended suspension surface obtained by tubing together the $\partial$-parallel annuli $\Gamma_1,\dots,\Gamma_s$ and the disk $D^2\times\{0\}$, Claim \[Claimannuli\] implies that $P^-$ can be constructed as an extended suspension surface in the extended marked handlebody $X_0$. Hence the submanifold of $X_0$ between $P^-$ and $P'$ is a type II block. In other words, $P'$ is a surface in $Y_P$ with $\partial P'=\partial S'$ and $P'$ divides $Y_P$ into two type II blocks. By gluing $P^-$ to $P^+$, we get back the solid torus $\widehat{\mathcal{T}}$. Hence, $P$ and $P'$ divide $\widehat{\mathcal{T}}$ into a pair of type II blocks.
Let $S_1,\dots, S_n$ be the components of $\Sigma\cap \mathcal{W}$ that are non-separating in $\mathcal{W}$. As before, we may assign each $S_i$ a compatible normal direction so that they represent the same element in $H_2(\mathcal{W},\mathcal{T})$. By repeating the arguments above, we get a collection of non-separating planar surfaces $P_1,\dots, P_n$ in the solid torus $\widehat{\mathcal{T}}$ such that $\partial P_i=\partial S_i$ and $P_1,\dots, P_n$ divide $\widehat{\mathcal{T}}$ into a collection of type II blocks.
Assign each $P_i$ a normal direction compatible with the induced normal direction of $\partial P_i=\partial S_i$. We refer the two sides of each $P_i$ as the left and right sides with the normal direction of $P_i$ pointing from the left to the right side.
[**Case (1)**]{}. $n=1$, i.e. $\Sigma\cap\mathcal{W}$ contains only one non-separating surface $S_1$.
Let $\widetilde{Y}$ be the manifold obtained by cutting $\mathcal{W}$ open along $S_1$, and let $\widetilde{X}_0$ be the type II block obtained by cutting $\widehat{\mathcal{T}}$ open along $P_1$. Let $P^+$ and $P^-$ be the two surfaces in $\partial_h\widetilde{X}_0$ representing the two sides of $P_1$, and suppose the induced normal direction of $P^-$ points into $\widetilde{X}_0$. Let $P_1'$ be the surface obtained from $P_1$ by adding $g_1$ trivial tubes ($g_1=g(S_1)$) on the right side of $P_1$, and let $\widetilde{X}$ be the manifold obtained by cutting $\widehat{\mathcal{T}}$ open along $P_1'$. So $\widetilde{X}$ can be obtained from $\widetilde{X}_0$ by first adding $g_1$ 1-handles at $P^+$ and then drilling out $g_1$ trivial tunnels at $P^-$. $\widetilde{X}$ is also a type II block. Let $P_l$ and $P_r$ be the two components of $\partial_h\widetilde{X}$ representing the two sides of $P_1'$. Similarly, denote the two surfaces in $\widetilde{Y}$ representing the two sides of $S_1$ by $S_l$ and $S_r$ respectively.
The surfaces of $\Sigma\cap\mathcal{W}$ that lie in ${\mathrm{int}}(\widetilde{Y})$ divide $\widetilde{Y}$ into a stack of relative compression bodies. Since $S_1$ is the only non-separating component of $\Sigma\cap\mathcal{W}$, there is a relative compression body $Y$ in the stack $\widetilde{Y}$ that contains both $S_l$ and $S_{r}$. So $\partial_hY$ consists of $S_l$, $S_{r}$, and a collection of separating surfaces $Q_1,\dots,Q_k$ in $\mathcal{W}$.
Next, we construct a collection of surfaces $Q_1',\dots, Q_k'$ in $\widetilde{X}$ such that (1) $\partial Q_j'=\partial Q_j$ in $\mathcal{T}$, (2) $Q_j'\cong Q_j$ for all $j$, and (3) the submanifold $X$ bounded by $P_l$, $P_{r}$, $Q_1',\dots, Q_k'$ (and subannuli of $\partial_v \widetilde{X}$) is a relative compression body $\partial$-similar to $Y$.
The argument is similar to section \[StypeI\], and the only difference here is that $\widetilde{X}$ is a type II block. Instead of repeating the argument in section \[StypeI\], we convert $\widetilde{X}$ into a type I block or a marked handlebody and then apply Lemma \[LtypeI\].
In section \[StypeII\], we described a way of converting a type II block into a type I block by drilling out a tunnel $N(\alpha)$, where $\alpha$ is of the form $\{x\}\times I\subset D^2\times I=B'$ in the construction of an extended marked handlebody. We view $N(\alpha)=\Delta_\alpha\times I$, where $\Delta_\alpha$ is a disk, and suppose $\Delta_\alpha\times\{0\}\subset P_l$ and $\Delta_\alpha\times\{1\}\subset P_{r}$. Let $P_l'=P_l\setminus(\Delta_\alpha\times\{0\})$, $P_{r}'=P_{r}\setminus (\Delta_\alpha\times\{1\})$, and $\mathcal{A}_\alpha=(\partial\Delta_\alpha)\times I$. Let $\widetilde{X}'=\widetilde{X}\setminus N(\alpha)$.
Next, we construct a relative compression body $X$ that is $\partial$-similar to $Y$.
[*Subcase (1a)*]{}. $S_1$ has a minus sign.
In this case, $P_l$ and $P_r$ have minus signs. As in Remark \[Rconversion\], instead of viewing $\widetilde{X}'=\widetilde{X}\setminus N(\alpha)$ as a type I block, we may view $\widetilde{X}'$ as a marked handlebody with $\partial_h \widetilde{X}'=P_l'\cup P_{r}'\cup \mathcal{A}_\alpha$.
Now consider the relative compression body $Y$. In this subcase, both $S_l$ and $S_{r}$ are components of $\partial_h^-Y$. By Remark \[Rproperty\](4), there is a core arc $\beta$ in $Y$ connecting $S_l$ to $S_{r}$, such that (1) $Y\setminus N(\beta)$ is a relative compression body and (2) $S_l$ and $S_{r}$ are tubed together and become a component of $\partial_h^-(Y\setminus N(\beta))$.
By Lemma \[LtypeI\], we can build a collection of surfaces $Q_1',\dots, Q_k'$ in the marked handlebody $\widetilde{X}'=\widetilde{X}\setminus N(\alpha)$ such that (1) $\partial Q_j'=\partial Q_j$ in $\mathcal{T}$ and $Q_j'\cong Q_j$ for all $j$ and (2) the type I block, denoted by $X'$, between $\partial_h \widetilde{X}'$ and $Q_1',\dots, Q_k'$ is a relative compression body $\partial$-similar to $Y\setminus N(\beta)$. Let $X$ be the manifold obtained by adding a 2-handle to $X'$ along the annulus $\mathcal{A}_\alpha$ (i.e. filling the tunnel $N(\alpha)$). Since the 2-handle is added to $\partial_h^- X'$, by Remark \[Rproperty\](5), $X$ is a relative compression body $\partial$-similar to $Y$.
[*Subcase (1b)*]{}. $S_1$ has a plus sign.
In this case, both $S_l$ and $S_r$ have plus signs. Let $\gamma$ be a $\partial$-parallel arc in $Y$ connecting $S_l$ to $S_r$ and parallel to an arc in $\partial^+Y$. Let $N(\gamma)$ be a small neighborhood of $\gamma$ in $Y$. We view $N(\gamma)=\Delta_\gamma\times I$ with $\Delta_\gamma\times\partial I\subset S_l\cup S_{r}$. Let $\mathcal{A}_\gamma=(\partial\Delta_\gamma)\times I$. Let $S_l'=S_l\setminus (\Delta_\gamma\times\partial I)$ and $S_r'=S_r\setminus (\Delta_\gamma\times\partial I)$. As in Remark \[Rproperty\](7), after setting $\mathcal{A}_\gamma$ as a component of $\partial_v^- (Y\setminus N(\gamma))$ with a large order, $Y\setminus N(\gamma)$ is a relative compression body, where $S_l'$ and $S_r'$ are two components of $\partial_h^+(Y\setminus N(\gamma))$.
Now consider $\widetilde{X}'=\widetilde{X}\setminus N(\alpha)$ defined before Subcase (1a). The difference from Subcase (1a) is that, in this subcase, we view $\widetilde{X}'=\widetilde{X}\setminus N(\alpha)$ as a type I block (see section \[StypeII\]) and view $\mathcal{A}_\alpha$ as a component of $\partial_v^-\widetilde{X}'$ with induced order from $\mathcal{A}_\gamma\subset\partial_v^- (Y\setminus N(\gamma))$. Moreover, in this subcase, $\partial_h\widetilde{X}'=P_l'\cup P_r'$.
Since $\widetilde{X}'$ is a type I block, we may view $\widetilde{X}'$ as a submanifold of a marked handlebody as in the definition of type I block in section \[StypeI\]: First take a marked handlebody $Z$ with $\partial_hZ=P_{r}'$, then view $P_l'$ as a standard surface in $Z$ such that the submanifold of $Z$ between $P_r'$ and $P_{l}'$ is the type I block $\widetilde{X}'$.
Now we apply Lemma \[LtypeI\] to the marked handlebody $Z$. By the construction in Lemma \[LtypeI\], we can build a collection of standard surfaces $Q_1',\dots, Q_k'$ in $Z$, such that the cross-section disks for $P_l'$, $Q_1',\dots, Q_k'$ are well-positioned in $Z$, $\partial Q_i'=\partial Q_i$, and $Q_i'\cong Q_i$. Moreover, the type I block, denoted by $X'$, bounded by $P_{r}'$, $P_l'$, $Q_1',\dots, Q_k'$ (and subannuli of $\partial_v\widetilde{X}'$) is a relative compression body that is $\partial$-similar to $Y-N(\gamma)$. Let $X$ be the manifold obtained by adding a 2-handle to $X'$ along the annulus $\mathcal{A}_\alpha$ (i.e. filling the tunnel $N(\alpha)$). Since $\mathcal{A}_\alpha$ is an annulus in $\partial_v^- X'$, by Remark \[Rproperty\](5), $X$ is a relative compression body $\partial$-similar to $Y$.
In both subcases, each $Q_i'$ in $\widetilde{X}$ cuts off a marked handlebody from $\widetilde{X}$. By Lemma \[LtypeI\], we can construct a collection of surfaces in each of these marked handlebodies, such that these surfaces, together with $Q_1',\dots, Q_k'$, divide $\widetilde{X}$ into a stack that is similar to the stack $\widetilde{Y}$. After gluing $P_l$ to $P_r$, we have a collection of surfaces that divide $\widehat{\mathcal{T}}$ into a stack of relative compression bodies similar to the stack $\mathcal{W}$.
[**Case (2)**]{}. $n>1$.
Consider the non-separating surfaces $S_1,\dots, S_n$ in $\Sigma\cap\mathcal{W}$ and suppose each $S_i$ is adjacent to $S_{i+1}$. We first show that there must be two adjacent non-separating surfaces $S_i$ and $S_{i+1}$ (set $S_{n+1}=S_1$) with the same sign. The reason for this comes from the untelescoping construction. Recall in section \[SGHS\], $\Sigma$ divides $M$ into a collection of compression bodies $\mathcal{A}_0,\dots,\mathcal{A}_m$ and $\mathcal{B}_0,\dots,\mathcal{B}_m$, where each $\mathcal{A}_i$ is obtained by adding 1-handles either to 0-handles or to $\mathcal{B}_{i-1}$ along $\mathcal{F}_i$, and each $\mathcal{B}_i$ is obtained by adding 2- and 3-handles to $\mathcal{A}_i$ along $\mathcal{P}_i$ ($i=0,\dots , m$), see Figure \[Funtel\]. We can set a direction of “adding handles" on these surfaces as follows: the direction for each $\mathcal{F}_i$ points from $\mathcal{B}_{i-1}$ to $\mathcal{A}_i$, and the direction for each $\mathcal{P}_i$ points from $\mathcal{A}_i$ to $\mathcal{B}_i$. This is the direction of the handle addition, where one starts from the handlebody $\mathcal{A}_0$ and ends at the handlebody $\mathcal{B}_m$ by adding handles in this direction, see Figure \[Funtel\]. Note that Figure \[Funtel\] is the simplest diagram describing generalized Heegaard splittings, see [@SSS] for more complicated diagrams. Therefore, if any two adjacent non-separating surfaces $S_i$ and $S_{i+1}$ (set $S_{n+1}=S_1$) have different signs, then the surfaces $\mathcal{F}_i$’s and $\mathcal{P}_i$’s appear alternately in $\mathcal{W}$, which implies that the handle-addition direction gives an oriented cycle. This is impossible because this is a direction for adding handles.
Without loss of generality, suppose $S_1$ and $S_2$ have the same sign.
The non-separating surfaces $S_1,\dots, S_n$ divide $\mathcal{W}$ into submanifolds $\widetilde{Y}_1,\dots, \widetilde{Y}_n$, where $\widetilde{Y}_i$ is the submanifold between $S_i$ and $S_{i+1}$. Similarly, $P_1,\dots, P_n$ divide $\widehat{\mathcal{T}}$ into a collection of type II blocks $\widetilde{X}_1,\dots, \widetilde{X}_n$, where $\widetilde{X}_i$ is the type II block between $P_i$ and $P_{i+1}$.
First, add $g_1$ trivial tubes to $P_1$ on its right side (i.e. in $\widetilde{X}_1$), where $g_1=g(S_1)$. Denote the resulting surface by $P_1'$. So $P_1'\cong S_1$. Now replace $P_1$ by $P_1'$ and consider the collection of non-separating surfaces $P_1', P_2,\dots, P_n$ which divide $\widehat{\mathcal{T}}$ into submanifolds $\widetilde{X}_1', \widetilde{X}_2, \dots, \widetilde{X}_{n-1}, \widetilde{X}_n'$, where $\widetilde{X}_n'$ is obtained by adding $g_1$ 1-handles to $\widetilde{X}_n$, and $\widetilde{X}_1'$ is obtained from $\widetilde{X}_1$ by drilling out $g_1$ trivial tunnels.
Next we add $g_n$ tubes to $P_n$ on its right side ($g_n=g(S_n)$) and modify $\widetilde{X}_n'$. We divide the discussion into two subcases.
[*Subcase (2a)*]{}. $S_n$ and $S_1$ have the same sign.
In this subcase, we simply add $g_n$ trivial tubes to $P_{n}$ on its right side and denote the resulting surface by $P_{n}'$. So $P_n'\cong S_n$. Now we replace $P_{n}$ by $P_{n}'$ and consider the new collection of surfaces $P_1', P_2,\dots, P_{n-1}, P_n'$ which divide $\widehat{\mathcal{T}}$ into submanifolds $\widetilde{X}_1', \widetilde{X}_2, \dots, \widetilde{X}_{n-1}', \widetilde{X}_n''$, where $\widetilde{X}_n''$ is the submanifold between $P_n'$ and $P_1'$. So $\widetilde{X}_n''$ is obtained from $\widetilde{X}_n'$ by drilling out $g_n$ trivial tunnels, and $\widetilde{X}_{n-1}'$ is obtained by adding $g_n$ 1-handles to $\widetilde{X}_{n-1}$.
Since $P_n'$ and $P_1'$ have the same sign in this subcase, $\widetilde{X}_n''$ has a similar structure as the manifold $\widetilde{X}$ in Case (1). So we can apply the argument on $\widetilde{X}$ in Case (1) to $\widetilde{X}_n''$ and construct a sequence of separating surfaces in $\widetilde{X}_n''$, such that these surfaces divide $\widetilde{X}_n''$ into a stack of relative compression bodies that is similar to the stack $\widetilde{Y}_n$.
[*Subcase (2b)*]{}. $S_n$ and $S_1$ have opposite signs.
Let $Y$ be the relative compression body in the stack $\widetilde{Y}_n$ that contains both $S_n$ and $S_1$. Since $S_n$ and $S_1$ have different signs, by Remark \[Rproperty\](6), there is a vertical arc $\beta$ in $Y$ that connects $S_n$ to $S_{1}$, such that $Y-N(\beta)$ is a relative compression body. More precisely, let $N(\beta)$ be a small neighborhood of $\beta$ in $Y$ and we view $N(\beta)=\Delta_\beta\times I$ with $\Delta_\beta\times\partial I\subset S_n\cup S_{1}$. Let $\mathcal{A}_\beta=(\partial\Delta_\beta)\times I$. Let $S_n'=S_n\setminus (\Delta_\beta\times\partial I)$ and $S_1'=S_1\setminus (\Delta_\beta\times\partial I)$. As in Remark \[Rproperty\](6), we can choose an arc $\beta$ such that $Y\setminus N(\beta)$ is a relative compression body with $S_n'$ and $S_1'$ being two components of $\partial_h^\pm (Y\setminus N(\beta))$ and $\mathcal{A}_\beta$ a component of $\partial_v^0 (Y\setminus N(\beta))$.
Now we convert the type II block $\widetilde{X}_n'$ into a type I block, as in section \[StypeII\], by drilling out a tunnel $N(\alpha)$, where $\alpha$ is of the form $\{x\}\times I\subset D^2\times I=B'$ and $B'$ is the 3-ball in the construction of an extended marked handlebody. As in Case (1), we view $N(\alpha)=\Delta_\alpha\times I$, where $\Delta_\alpha$ is a disk, $\Delta_\alpha\times\{0\}\subset P_n$ and $\Delta_\alpha\times\{1\}\subset P_{1}'$. Let $P_n^c=P_n\setminus(\Delta_\alpha\times\{0\})$ and $P_{1}^c=P_{1}'\setminus (\Delta_\alpha\times\{1\})$. Let $\widetilde{X}_n^c=\widetilde{X}_n'\setminus N(\alpha)$. Let $\mathcal{A}_\alpha=(\partial\Delta_\alpha)\times I$. So $\widetilde{X}_n^c$ is a type I block with $\partial_h \widetilde{X}_n^c=P_n^c\cup P_{1}^c$ and $\partial_v \widetilde{X}_n^c=\partial_v\widetilde{X}_n'\cup\mathcal{A}_\alpha$.
Similar to Subcase (1b), we may view the type I block $\widetilde{X}_n^c$ as a submanifold of a marked handlebody in section \[StypeI\] as follows: First take a marked handlebody $Z$ with $\partial_hZ=P_{1}^c$, then view $P_n^c$ as a suspension surface in $Z$ such that the submanifold of $Z$ between $P_n^c$ and $P_{1}^c$ is the type I block $\widetilde{X}_n^c$.
Now we apply Lemma \[LtypeI\] to the marked handlebody $Z$. By the construction in Lemma \[LtypeI\], we can build a collection of surfaces $Q_1',\dots, Q_k'$ in $Z$, such that the cross-section disks for $P_n^c$ and $Q_1',\dots, Q_k'$ are well-positioned. We also add $g_n$ tubes to $P_n^c$ ($g_n=g(S_n)$) and denote the resulting surface by $P_n^d$. Moreover, as in Lemma \[LtypeI\], we can construct these surfaces and add these $g_n$ tubes so that the submanifold $Z^c$ between $P_1^c$, $P_n^d$, $Q_1',\dots, Q_k'$ is a relative compression body $\partial$-similar to the relative compression body $Y-N(\beta)$.
Note that, in the construction of Lemma \[LtypeI\] and as explained in Remark \[RfixH\], some tubes of $P_n^d$ may be dragged through the 1-handles of $Z$, in other words, some tube of $P_n^d$ may be re-embedded into tubes that go through the $g_1$ tubes of $P_1'$. This is the main difference between Subcase (2a) and Subcase (2b).
Let $X_n$ be the relative compression body obtained by adding a 2-handle to $Z^c$ along the annulus $\mathcal{A}_\alpha$ (i.e. filling the tunnel $N(\alpha)$). By Remark \[Rproperty\](6), $X_n$ a relative compression body $\partial$-similar to $Y$.
Adding the 2-handle also extends the surface $P_n^d$ to a surface $P_n'$ with $P_n'\cong S_n$. Denote the resulting manifold between $P_n'$ and $P_1'$ by $\widetilde{X}_n''$. Similarly, we view $Q_1',\dots, Q_k'$ as separating surfaces in $\widetilde{X}_n''$. Since each $Q_i'$ cuts off a marked handlebody from $\widetilde{X}_n''$, as in Case (1), we can construct surfaces in $\widetilde{X}_n''$ which divide $\widetilde{X}_n''$ into a stack of relative compression bodies that is similar to the stack $\widetilde{Y}_n$.
So, in both subcases, we can change $P_n$ to $P_n'$ and constructed surfaces that divide $\widetilde{X}_n''$ into a stack of relative compression bodies that is similar to the stack $\widetilde{Y}_n$.
The operation of changing $P_n$ to $P_n'$ also affect the type II block $\widetilde{X}_{n-1}$. For $\widetilde{X}_{n-1}$, the effect of adding tubes to $P_n$ on its right side is simply adding 1-handles to $\widetilde{X}_{n-1}$, and the topological type of the resulting manifold $\widetilde{X}_{n-1}'$ does not depend on whether or not the tubes that we added to $P_n$ are trivial tubes. Moreover, the possible tube re-embedding explained above and in Remark \[RfixH\] does not change the topological type of $\widetilde{X}_{n-1}'$ either. Thus $\widetilde{X}_{n-1}'$ can be viewed as the manifold obtained by adding $g_n$ 1-handles to $\widetilde{X}_{n-1}$ at $P_n$.
Consider the new collection of surfaces $P_1', P_2,\dots, P_{n-1}, P_n'$ which divide $\widehat{\mathcal{T}}$ into submanifolds $\widetilde{X}_1', \widetilde{X}_2,\dots, \widetilde{X}_{n-2},\widetilde{X}_{n-1}', \widetilde{X}_n''$. Next, we inductively carry out this construction on $P_{n-1},\dots, P_2$ and successively modify the planar surfaces $P_{n-1},\dots, P_2$ into $P_{n-1}',\dots, P_2'$, with $P_i'\cong S_i$ for each $i$. $P_1',\dots, P_n'$ divide $\widehat{\mathcal{T}}$ into submanifolds $\widetilde{X}_{1}'',\dots, \widetilde{X}_n''$. Similarly, we construct separating surfaces in $\widetilde{X}_{n-1}'',\dots, \widetilde{X}_2''$ which divide each $\widetilde{X}_{i}''$ into a stack of relative compression bodies similar to the stack $\widetilde{Y}_{i}$ ($i=2,\dots, n-1$).
The last step is to consider the type II block $\widetilde{X}_{1}''$ between $P_1'$ and $P_2'$. Recall that $P_1'$ is obtained by adding trivial tubes to $P_1$, so the topological structure of $\widetilde{X}_{1}''$ is similar to that of $\widetilde{X}$ in Case (1). Since both $S_1$ and $S_2$ have the same sign, we can apply the construction on $\widetilde{X}$ in Case (1) to $\widetilde{X}_{1}''$ which makes $\widetilde{X}_{1}''$ a stack of relative compression bodies that is similar to the stack $\widetilde{Y}_{1}$. By gluing these stacks together, we obtain a stack in $\widehat{\mathcal{T}}$ similar to the stack $\mathcal{W}$.
In both Case (1) and Case (2) above, by gluing the stack of relative compression bodies in $\widehat{\mathcal{T}}$ to the stack of relative compression bodies in $\mathcal{V}$, as in section \[Ssep\], we obtain a generalized Heegaard splitting for $N=\widehat{\mathcal{T}}\cup_\mathcal{T} \mathcal{V}$ with genus equal to $g(M)$. This means that $g(N)\le g(M)$ and Theorem \[Ttorus\] holds.
[99]{}
David Bachman, Saul Schleimer, Eric Sedgwick, *Sweepouts of amalgamated 3-manifolds*. Algebraic & Geometric Topology **6** (2006) 171–194
Michel Boileau and Shicheng Wang, *Degree one maps between small 3-manifolds and Heegaard genus*. Algebraic & Geometric Topology **5** (2005) 1433–1450
Andrew Casson and Cameron Gordon, *Reducing Heegaard splittings*. Topology and its Applications, **27** 275–283 (1987).
Allan Edmonds, *Deformation of maps to branched covering in dimension 2*. Ann. of Math. **110** (1979), 113–125
Wolfgang Haken, *Some results on surfaces in 3–manifolds*. Studies in Modern Topology, Math. Assoc. Amer., distributed by Pretice-Hall, (1968) 34–98.
Wolfgang Haken, *On homotopy 3-spheres*, Illinois J. Math. **10** (1966) 159–178
C. Hayat-Legrand, S. Wang and H. Zieschang, *Any 3-manifold 1-dominates only finitely many 3-manifolds supporting $S^3$ geometry*, Proc. Amer. Math. Soc. **130** (2002) 3117–3123
John Hempel, *3-manifolds as viewed from the curve complex*. Topology **40** (2001), 631–657.
Marc Lackenby, *The Heegaard genus of amalgamated 3-manifolds*. Geom. Dedicata **109** (2004), 139–145.
Tao Li, *Heegaard surfaces and measured laminations, I: the Waldhausen conjecture*, Invent. Math. **167** (2007), 135-177
Tao Li, *Heegaard surfaces and the distance of amalgamation*. Geometry & Topology, **14** (2010), 1871–1919.
Tao Li *On the Heegaard splittings of amalgamated 3-manifolds*. Geometry & Topology Monographs , 12 (2007) 157–190
Tao Li and Ruifeng Qiu *On the degeneration of tunnel numbers under connected sum*. Transactions of the AMS **368** (2016), 2793–2807
Yongwu Rong, *Maps between Seifert fibered spaces of infinite $\pi_1$*, Pacific J. Math. **160** (1993), 143–154
Yongwu Rong; Shicheng Wang, *The preimages of submanifolds*, Math. Proc. Cambridge Philos. Soc. **112** (1992) 271–279
Toshio Saito, Martin Scharlemann and Jennifer Schultens, *Lecture notes on generalized Heegaard splittings*, arXiv:math/0504167, World Scientific Publishing Company, (2016)
Martin Scharlemann and Abigail Thompson, *Thin position for 3-manifold*, Comptemp. Math., **164** (1992), 231–238
Trent Schirmer, *A lower bound on tunnel number degeneration*. Algebraic & Geometric Topology **16** (2016) 1279–1308
Jennifer Schultens, *The classification of Heegaard splittings for (compact orientable surface)$\times S^1$*, Proc. London Math. Soc. 67 (1993) 425–448.
Teruhiko Soma, *Non-zero degree maps onto hyperbolic 3-manifolds*, J. Diff. Geom. **49** (1998) 517–546
Friedhelm Waldhausen, *On mappings of handlebodies and of Heegaard splittings*, *Topology of Manifolds* (Proc. Inst. Univ. of Georgia, Athens, Ga. 1969), Markham, Chicago, Ill. (1970) 205–211
Shicheng Wang; Qing Zhou, *Any 3-manifold 1-dominates at most finitely many geometric 3-manifolds*, Math. Ann. **332** (2002) 525–535
[^1]: Partially supported by an NSF grant
| {
"pile_set_name": "ArXiv"
} | ArXiv |