Entanglement dualities in supersymmetry

1, 2, 3, 4, Robert H. Jonsson, ∗ Lucas Hackl, † and Krishanu Roychowdhury ‡ 1Max-Planck-Institut fur¨ Quantenoptik, Hans-Kopfermann-Str. 1, 85748 Garching, Germany 2School of Mathematics and Statistics & School of Physics, The University of Melbourne, Parkville, VIC 3010, Australia 3QMATH, Department of Mathematical Sciences, University of Copenhagen, Universitetsparken 5, 2100 Copenhagen, Denmark 4Department of Physics, Stockholm University, SE-106 91 Stockholm, Sweden We derive a general relation between the bosonic and fermionic entanglement in the ground states of su- persymmetric quadratic Hamiltonians. For this, we construct canonical identifications between bosonic and fermionic subsystems. Our derivation relies on a unified framework to describe both, bosonic and fermionic Gaussian states in terms of so-called linear complex structures J. The resulting dualities apply to the full en- tanglement spectrum between the bosonic and the fermionic systems, such that the von Neumann entropy and arbitrary Renyi entropies can be related. We illustrate our findings in one and two-dimensional systems, includ- ing the paradigmatic Kitaev honeycomb model. While typically SUSY preserves features like area law scaling of the entanglement entropies on either side, we find a peculiar phenomenon, namely, an amplified scaling of the entanglement entropy (“super area law”) in bosonic subsystems when the dual fermionic subsystems develop almost maximally entangled modes.

CONTENTS and nonrelativistic settings [1–5]. In a nutshell, SUSY posits a fundamental equivalence between the two classes of ele- I. Introduction 1 mentary particles with distinct statistics. Mathematically, it maps the fermionic degrees of freedom to the bosonic ones II. Gaussian states and supersymmetry 2 and vice-versa. From this perspective, they are equivalent and A. Bosonic and fermionic Gaussian states 2 dubbed superpartners of each other. B. Supercharge operator and supersymmetric While normally SUSY is conceived as a symmetry in quan- Gaussian states 4 tum field theories, it as well applies to much simpler models C. Supersymmetric identification maps 5 of quantum mechanics such as harmonic oscillators or the hy- D. Application: supersymmetric Kitaev chain 6 drogen atom [6–9]. The SUSY Hamiltonian Hˆ can be con- structed from a generating operator Qˆ (also called the super- III. Entanglement duality 8 charge operator) which, for the harmonic oscillator problem, A. Reduced Gaussian states and entanglement 8 √ takes a remarkably simple form Qˆ = ωˆb†cˆ where ˆb (cˆ) de- B. Supersymmetric ground states and identification notes the bosonic (fermionic) annihilation operator. The cor- maps 9 responding SUSY Hamiltonian C. Dual supersymmetric subsystems 10

D. Duality for Gaussian states and their Hˆ = {Qˆ, Qˆ†} = ω(ˆb†ˆb +c ˆ†cˆ) ≡ Hˆb + Hˆf (1) entanglement 10 E. Application: two-mode system 12 then decomposes into two simple quadratic Hamiltonians: one F. Application: supersymmetric Kitaev honeycomb for a bosonic oscillator (Hˆb) and the other for a fermionic one model 14 (Hˆf ). When it comes to dealing with real bosons or fermions a hermitian form of the generating operator Qˆ = Qˆ† (and IV. Discussion 16 accordingly, Hˆ = Qˆ2) is useful, as also is the case for the present work. Acknowledgments 18 Such a simple setting is readily amenable to accommodate multiple bosonic and fermionic modes, or in other words, sys- References 18 tems of free (noninteracting) bosons and fermions (in the con- tinuum or on a lattice) as long as the generating operator Qˆ I. INTRODUCTION involves the bosonic and fermionic operators to linear order [10, 11], as shown in the previous harmonic oscillator exam- arXiv:2103.09657v1 [quant-ph] 17 Mar 2021 ple and also will be demonstrated later. The resulting part- As a long-established concept in quantum physics, super- ˆ ˆ symmetry (SUSY) finds applications in a wide range of fields ner Hamiltonians (referred to as Hb and Hf for bosons and from particle physics to condensed matter in both relativistic fermions, respectively) are isospectral in their one-particle ex- citations except for zero modes. Inclusion of zero modes in SUSY has, in addition, a topological aspect (referred to as “Witten index” [12] and interpreted in several other contexts ∗ [email protected] e.g. see [13]) and has been studied to a great extent, however, † [email protected] that discussion is not relevant to this work. ‡ [email protected] Ground states of a quadratic Hamiltonian (bosonic or 2 fermionic) garner special attention as they provide a fertile However, we will show that even when well localized subre- ground to trace several properties of the system, which they gions are identified of both lattices, the scaling of the entan- are part of, analytically. These states are also known as Gaus- glement entropy of the dual supersymmetric subsystem can be sian states [14–18]. The study of the von Neumann bipar- very different – on the bosonic side, it can drastically exceed tite entanglement entropy plays a central role in the quan- the area law exhibited by the original fermionic subsystem. tum foundations of statistical mechanics [19–35], in quantum In summary, this study extends the concept of SUSY information theory [36–45] and condensed matter dedicated beyond a spectral mapping between (supersymmetric) to classifying novel states of matter, particularly those with quadratic Hamiltonians to discuss the general identification topological quantum order [46–54]. While measuring entan- of fermionic and bosonic supersymmetric Gaussian systems, glement is numerically costly for a generic quantum state, it their subsystems, and entanglement spectra as implied by greatly simplifies for the Gaussian states [55, 56]. the supercharge operator. Exemplifying lattice models in The main result of this work is a duality between the eigen- 1D and 2D, we investigate the locality properties of these values of reduced density operators in the bosonic and the identification maps, and their consequences in the context fermionic system, i.e., the so called entanglement spectra. For of entanglement-area laws. In doing so, we employ the idea Gaussian states, these spectra are fully encoded in the eigen- of Kahler¨ structure which brings the bosonic and fermionic values ±iλ of the so-called restricted complex structure J, Gaussian states within a unified frame to work in. A further where λf ∈ [0, 1] for fermions and λb ∈ [1, ∞) for fermions. merit of this approach lies in treating the involved geometric In super-symmetric systems, the charge operator Qˆ provides structures independent of their matrix representation in a an identification between the bosonic and the fermionic sys- given basis, as discussed at length, e.g., in [62–64]. tem, so that picking a subsystem on the bosonic side automat- ically defines a related subsystem on the fermionic side and The article is structured as follows: In Sec. II, we review vice versa. Our key finding is that under identification, we the unified Kahler¨ structure formalism to describe bosonic 1 and fermionic Gaussian states and apply it to supersymmet- have λb = 1/λf and Jb = −Jf− , where we use b and f to refer to the bosonic and fermionic structure, respectively. ric quadratic Hamiltonians, where a charge operator induces Applying our results to examples, we also discuss conse- an identification map at the classical phase space level. In Sec. quences of the derived duality for the entanglement entropy III, we explore how the entanglement entropies in the bosonic in Gaussian states related by SUSY. Though not always, en- and fermionic systems are related and introduce a general the- tanglement entropy often turns out to be a sufficient measure orem on their entanglement spectra. In Sec. IV, we summarize (among others) of the entanglement information encoded in a our key findings complemented by lattice models as applica- quantum state [57–59]. In fact, in a number of strongly corre- tions and discuss future work. lated systems, this quantity serves as a smoking-gun to iden- tify topological quantum order in the ground states. Examples include Kitaev’s celebrated model of Majorana fermions on II. GAUSSIAN STATES AND SUPERSYMMETRY a honeycomb lattice [60]. In earlier works [11], the bosonic SUSY analog of this model has been realized and shown to In this section, we review the unified formalism that treats inherit the topological properties from its fermionic partner. both bosonic and fermionic Gaussian states on the same foot- We will also regard this model here, as one of our examples ing. For this, we present a hand-on introduction to the for- to illustrate the aspects of entanglement dualities considering malism of [63], which can be consulted for a more rigorous the SUSY-related Gaussian states. exposition. Other reviews of Gaussian states include [15]. Generally speaking, for non-critical ground states in d di- mensions (for both fermionic and bosonic systems), the entan- glement entropy of a subsystem A obeys the so-called “area A. Bosonic and fermionic Gaussian states law” (for a review, see [58, 61] and references therein)

d 1 We consider a bosonic or fermionic system with N degrees S(A) ∝ L − + ..., (2) of freedom described by a Hilbert space H. We can always meaning that, in the thermodynamic limit, the leading order find a basis of creation and annihilation operators which we ˆ ˆ contribution to the entanglement entropy of A with the rest denote as bi and bi† for bosons, and as cˆi and cˆi† for fermions, d 1 of the system scales with its surface area L − when L de- but we use aˆi and aˆi† in expressions valid for both bosons and notes the linear dimension of A. For critical states, however, fermions (see Tab. I). These operators satisfy the canonical the ellipses in (2) can contain sublinear corrections (e.g., log- commutation or anti-commutation relations arithmic corrections for free fermions), and for topologically ˆ ˆ [bi, bj†] = δij , (bosons) ordered states, a universal constant called “topological entan- (3) glement entropy”. {cˆi, cˆj†} = δij . (fermions) The identification provided by the supercharge Qˆ facilitates Out of these, we can construct a set of 2N Hermitian operators a natural connection between a subsystem in one lattice and a 1 i subsystem in the superpartner lattice. A priori, this identifica- ˆ† ˆ ˆ† ˆ qˆi = √ (bi + bi)p ˆi = √ (bi − bi) , (bosons) tion does not warrant a local subsystem in one system to get 2 2 (4) 1 i γˆ = (ˆc† +c ˆ ) , ηˆ = (ˆc† − cˆ ) , (fermions) mapped to a localized subsystem in its superpartner system. i √2 i i i √2 i i 3 which satisfy the commutation or anti-commutation relations Real basis Complex basis

[ˆqi, qˆj] = [ˆpi, pˆj] = 0 , [ˆqi, pˆj] = iδij, (bosons) (5) Bosons Quadrature operators CCR operators {γˆ , γˆ } = {ηˆ , ηˆ } = δ , {γˆ , ηˆ } = 0. (fermions) ˆ ˆ ˆ† i j i j ij i j ξb ≡ (ˆqj , pˆk) (bj , bk) For bosons, these operators are commonly called quadra- Fermions Majorana operators CAR operators ˆ † ture operators (generalized positions and momenta), while for ξf ≡ (ˆγj , ηˆk) (ˆcj , cˆk) fermions, they are called the Majorana operators. Unified Hermitian operators ξˆ Ladder operators (ˆa , aˆ† ) Up to normalization, there is a unique state |0i ∈ H, such j k that aˆi |0i = 0 ∀i, which is called the vacuum state with re- spect to our choice of operators. An orthonormal basis of H TABLE I. Overview of notations for operator bases. Listed are real can then be constructed by successively applying creation op- (self-adjoint) and complex operator bases for bosons and fermions, erators on |0i, as well as a unified notation used throughout this work. For an N- mode quantum system, indices are in the range j, k ∈ {1,...,N}. N ni The creation and annihilation operators, in a complex basis, satisfy Y (ˆai†) |n1, . . . , nN i = √ |0i , (6) canonical commutation/anti-commutation relations (CCR/CAR). n ! i=1 i ab a cb where ni ∈ N for bosons and ni = 0, 1 for fermions. • For bosons, G := −J cΩ is a metric, i.e., symmet- We can now collect the 2N operators to form the vector ric and positive definite. ab a cb ( • For fermions, Ω := J cG is a symplectic form, i.e., (ˆq ,..., qˆN , pˆ ,..., pˆN ) (bosons) ξˆa ≡ 1 1 , (7) anti-symmetric and non-degenerate. (ˆγ1,..., γˆN , ηˆ1,..., ηˆN ) (fermions) The matrix J is called a linear complex structure. where we have the index a = 1,..., 2N (later, we use Latin In (10) and the rest of this manuscript, we use Einstein’s indices exclusively for bosons and Greek indices for fermions, summation convention3, where a sum is implied over repeated but for now we use Latin indices for both). It is well known indices (index contraction). The position of the index indicates that, for both bosons and fermions, any operator O can be if it can be contracted with vectors va ∈ V in phase space or a described as a power series in ξˆ or as a limit of such a se- dual vectors wa ∈ V ∗ in dual phase space. Objects with two ries. For many physically relevant operators, this series will indices are often written as matrices, where matrix multipli- be finite and of low order. The canonical commutation or anti- cation is the same as contraction over adjected indices. This a ac b commutation relations in terms of ξˆ read may require a transpose, e.g., Ω J c needs to be written as ab ac b (ΩJ |) = Ω (J |)c to make the indices c adjacent. a b ab [ξˆ , ξˆ ] = iΩ , (bosons) The above relations introduce for every Gaussian state |Ji (8) ab ab {ξˆa, ξˆb} = Gab , (fermions) the object G for bosons and Ω for fermions, such that we have in both cases a so-called Kahler¨ structure: This is where Ωab is called the symplectic form and Gab is a metric. a triplet (G, Ω,J) such that With respect to our choice of basis in (7), they are represented Gab = −J a Ωcb ⇔ Ωab = J a Gcb , (11) by the matrices c c the equivalence following from J 2 = −1. Moreover, we have  0 1 1 0 Ω ≡ and G ≡ , (9) JΩJ | = Ω and JGJ | = G. −1 0 0 1 This definition of Gaussian states, unifying bosons and and will play an important role in later formulas. fermions, may appear surprising to readers more familiar with the definition of Gaussian states in terms of covariance ma- We define1 a Gaussian state |Ji ∈ H as the solution of2 trices or Bogoliubov transformations. However, as shown in 1 a a b [63], these definitions are fully equivalent, as we review in the (δ b + iJ b)ξˆ |Ji = 0 . (10) 2 following. Covariance matrix. The covariance matrix of a quantum 2 − As shown in [63], a solution of (10) exists only if J = 1 state |ψi with hψ|ξˆa|ψi = 0 is defined as4 and the following compatibility conditions are satisfied: ( hψ|ξˆaξˆb + ξˆbξˆa|ψi (bosons) Γab = , (12) −i hψ|ξˆaξˆb − ξˆbξˆa|ψi (fermions) 1 Here, we restrict to Gaussian states with hJ|ξˆa|Ji = 0, i.e., the 1-point correlation function vanishes. However, the formalism extends to also in- clude displacements za = hJ|ξˆa|Ji for bosons, as explained in [63]. 2 Note that (10) only fixes |Ji up to a complex phase. This does not cause 3 All our equations with indices are fully basis independent and compati- any problems when considering individual Gaussian states, where the com- ble with Penrose’s abstract index notation [65]. In fact, we can even use ˆa † † plex phase is unphysical. However, if considering superpositions of Gaus- complex bases, such as ξ ≡ (ˆa1,..., aˆN , aˆ1,..., aˆN ) (see [63]). sain states |Ji + |J˜i, we would need to parametrize explicitly how the 4 Some authors use a different normalization or sign. The extension to states respective complex phases are related. with hψ|ξˆa|ψi 6= 0 is also straight-forward and explained in [63]. 4 i.e., the covariance matrix is exactly the expression that is Example 1. The simplest bosonic Gaussian state is the not already fixed by the canonical commututation or anti- ground state of the harmonic oscillator with Hamiltonian commutation relations. Given a Gaussian state |Ji with as- ˆ 1 2 2 2 H = 2 (ˆp + ω qˆ ) that takes the form sociated Kahler¨ structures (G, Ω,J), it follows from (10) that we have the 2-point function 1 X∞ |Ji = (− tanh ρ )n |2ni (20) 1 cosh ρ 2 Cab := hJ|ξˆaξˆb|Ji = (Gab + iΩab) , (13) 2 n=0 2 2 with respect to the basis (6) and . Its covariance a 1 a a b ρ = log ω ˆ ∓ ˆ ab ab To prove this, we define ξ = 2 (δ b iJ b)ξ , which depend matrix Γ = G and complex structure as ˆ±a ˆa ˆa ˆa on J. With this, we find ξ = ξ+ + ξ , and we have ξ |Ji = − −     0 and hJ| ξˆa = 0, due to (10). This implies ω 0 0 ω + Γ = G ≡ 1 and J ≡ 1 (21) 0 ω − ω 0 ( ˆa ˆb ab a b hJ|[ξ , ξ+]|Ji (bosons) h |ˆ ˆ | i (14) a C2 = J ξ ξ+ J = ˆ−a ˆb with respect to the basis ξˆ ≡ (ˆq, pˆ). − hJ|{ξ , ξ+}|Ji (fermions) − The simplest fermionic Gaussian states are the basis states due to h |ˆb ˆa | i . Finally, the commutator or anti- |J+i = |0i and |J i = |1i, which are also the only Gaussian J ξ+ξ J = 0 − commutator above− can be evaluated using (8) to be states for a single degree of freedom. Their covariance matri- ces Γ = Ω and complex structures J happen to coincide ˆa ˆb 1 a cd b ± ±a ± [ξ , ξ+] = 4 (1 + iJ) ciΩ (1 − iJ) d , (bosons) in the basis ξˆ ≡ (ˆq, pˆ) as − (15) ˆa ˆb 1 a cd b {ξ , ξ+} = 4 (1 + iJ) cG (1 − iJ) d , (fermions)   − 0 ±1 ≡ ≡ (22) 1 Γ = Ω J ∓ . which in both cases combines to 2 (G + iΩ) via (11). ± ± ± 1 0 ab We can reverse this argument to use C2 (and thus the co- variance matrix Γab contained in it) of a general state |ψi, In summary, this section has reviewed how bosonic and fermionic Gaussian states can be efficiently described in a uni- with hψ|ξˆa|ψi = 0, to check if |ψi is a Gaussian state i.e., fied formalism using the triplet of Kahler¨ structures. |ψi = |Ji and find J. For this, we first compute Gab = (G, Ω,J) Physical properties, such as expectation values or entangle- h |ˆa ˆb| i ab h |ˆa ˆb| i 2 Re ψ ξ ξ ψ and Ω = 2 Im ψ ξ ξ ψ and then invert ment entropies can be directly computed from them. (11) to compute a ac 1 J b = Ω (G− )cb . (16) B. Supercharge operator and supersymmetric Gaussian states One can then show [63] that J 2 = −1 is necessary and suf- ficient for |ψi to be the Gaussian state |Ji, i.e., a solution of We will now consider a system that contains both, bosonic 2 (10). However, if J =6 −1, |ψi is not a Gaussian state. and fermionic degrees of freedom. We denote the bosonic Bogoliubov transformations. These transformations map operators by ξˆa and the fermionic ones by ξˆα, where we use Gaussian states into Gaussian states, hence, are also termed b f Latin letters for bosons and Greek letters for fermions. The Gaussian transformation. For a Gaussian state | i annihi- J commutation and anti-commutation relations then read lated by a set of annihilation operators aˆi0 i.e., aˆi0 |Ji = 0, the following transformation relates them to the original aˆi ˆa ˆb ab ˆα ˆβ αβ [ξb, ξb] = iΩb and {ξf , ξf } = Gf , (23) X aˆi0 = (αijaˆi + βijaˆj†) , (17) while the bosonic and the fermionic operators commute j ˆa ˆα [ξb, ξf ] = 0. where the matrix elements αij and βij characterize the trans- The SUSY transformation between the bosonic and the formation. Defining a Gaussian state |J0i as the state anni- fermionic degrees of freedom can be generated by a Hermi- hilated by all aˆi i.e., aˆi |J0i = 0, we can use (16) and (9) to tian supercharge operator [10] compute that J0 is represented by the matrix α a Qˆ = Rαaξˆ ξˆ , (24)  0 1 f b J0 ≡ , (18) −1 0 with a real-valued R. As mentioned already in (1), this super- from which we deduce the resulting Bogoliubov transformed charge defines a supersymmetric Hamiltonian 1 state |Ji with J = MJ0M − where the matrix M is [64] ˆ 1 ˆ ˆ 1 b ˆa ˆb i f ˆα ˆβ ˆ ˆ H = 2 {Q, Q} = 2 habξbξb + 2 hαβξf ξf ≡ Hb + Hf , (25) Re α + Re β Im β − Im α M = . (19) Im α + Im β Re α − Re β which splits into a bosonic part Hˆb and a fermionic part Hˆf . Their Hamiltonian forms are: The matrix M is a group element of the symplectic group Sp(2N, R) for bosons or the orthogonal group O(2N, R) for f ab | hαβ = RαaΩ Rbβ , (26) fermions, which induces the unitary representation of Gaus- b | αβ sian transformations on the Hilbert space [63]. hab = RaαG Rβb , (27) 5

f f b b which satisfy hαβ = −hβα and hab = hba. Note, that the full for which we find the Hamiltonian Hamiltonian’s ground state energy E = iR R GβαΩba = 0 αa βb Hˆ = Qˆ2 = 1 (ˆq2 +p ˆ2) + i (ˆγηˆ − ηˆγˆ) . (37) i tr(GRΩ|R|) = 0 vanishes, as the bosonic and the 2 2 fermionic contributions cancel each other. (equivalent forms of Qˆ and Hˆ in terms of complex bosonic The excitation spectrum of ˆ and ˆ can be derived by Hb Hf and fermionic operators are shown in the introduction). The diagonalizing the Lie generators and , defined via the Kb Kf associated Lie algebra generators are then given by relations5  0 1 [H,ˆ ξˆa] = (K )a ξˆb and [H,ˆ ξˆα] = (K )α ξˆβ . (28) K ≡ K ≡ , (38) b b b b f f β f b f −1 0 One can show [63] that these matrices are Lie algebra ele- and the associated ground state is |GSi = |0 i ⊗ |0 i. ments satisfying b f

| | KbΩ = −ΩK and Kf G = −GK , (29) b f C. Supersymmetric identification maps which implies Kb ∈ sp(2N, R) and Kf ∈ so(2N, R). Using the relations (3) allows us to compute them explicitly as We introduced supersymmetric Hamiltonians through the a ac ac αβ supercharge operator Qˆ as Hˆ = Hˆ + Hˆ , where Hˆ and Hˆ (K ) = 1 Ω (h + h ) = Ω R| G R , (30) b f b f b b 2 cb bc cα βb have identical one-particle spectrum. Both the bosonic and α 1 αγ αγ ab | (Kf ) β = 2 G (hγβ − hβγ ) = G RγaΩ Rbβ . (31) the fermionic part are described classically by phase spaces 2N 2N Vb ' R and Vf ' R (with the corresponding dual From this, it is evident that Kb and Kf are isospectral except spaces denoted by V and V ), respectively, such that V is for the degeneracy of potential zero eigenvalues. b∗ f∗ b equipped with the symplectic form Ωb, and Vf is equipped The ground state of Hˆ is given by the tensor product with a metric Gf . The respective other structure in each space, i.e., a metric G on Vb and a symplectic form Ω on Vf , is de- |GSi = |Jbi ⊗ |Jf i , (32) fined by the ground state |Ji = |Jbi ⊗ |Jf i of Hˆ . where the associated Jb and Jf are computed from the gener- In this section, we now use the supercharge Qˆ to construct ators as [63, 66] linear maps between the two phase spaces, L1 : Vb → Vf and L : V → V , that identify the spaces in such a way that the 1 and 1 (33) 2 f b Jb = Kb− Kb Jf = Kf− Kf . symplectic forms and metrics are mapped onto each other. These formulas may be surprising at first sight, but they can Under the above assumption that Raα is real, the super- ˆ ˆα ˆa be readily checked using a basis, where the individual normal charge operator Q = Rαaξf ξb induces the supersymmetric modes of Hˆ decouple. In this basis, we have identification maps T1 : Vb → Vf and T2 : Vf → Vb as α αβ a ab | X ωi (T ) a = G Rβa and (T ) α = Ω R . (39) Hˆ = (ˆnb +n ˆf ) , (34) 1 2 bα 2 i i i These are related to the Lie generators noting Kb = T2T1 and Kf = T1T2. Hence T2 maps the eigenvectors of Kf (in Vf,C, where nˆi =a ˆ†aˆi are the normal mode number operators and i the complexification on Vf ) to the eigenvectors of Kb with the ωi are the one-particle excitation energies. Note that due to b f same eigenvalue, and for T1 the analogous holds: hab being positive, all ωi are positive and we choose nˆi, such that excitations increase energy. If we go into the associated Kbvb = ±iλvvb ⇒ Kf T1vb = ±iλvT1vb ˆa b 1 b 2 b 2
f (40) basis ξ , where nˆi = 2 (ˆqi ) + (ˆpi ) and nˆi = iˆγiηˆi, the Kf wf = ±iλwwf ⇒ KbT2wf = ±iλwT2wf matrix representations of the generators are If only the spaces V and V are given, each equipped with   b f 0 ωi Kahler¨ structures (G, Ω,J), then there exists a large class Kb ≡ Kf ≡ ⊕i . (35) 6 −ωi 0 of potential identification maps , however, the choice of Rαb fixes this freedom. In this specific basis, Jb and Jf assume the standard form from (18), which then implies (33).

Example 2. The simplest supersymmetric Hamiltonian con- 6 Given an identification map T1 : Vb → Vf , we can define a new identi- sists of one bosonic and one fermionic degree of freedom. The 0 fication T1 = Uf T1Ub, where both Ub : Vb → Vb and Uf : Vf → Vf respective supercharge operator is given by need to preserve the respective Kahler¨ structures. This implies that Ub and Uf form a representation of the group U(N). In our case, we also would Qˆ =q ˆγˆ +p ˆηˆ , (36) like that T1 maps Kb onto Kf , which implies that the respective symme- try group will depend on the degeneracy of the one-particle spectrum. If Kb (and thus also Kf ) has m distinct eigenvalue pairs ±iλi with degen- Pm eracy di such that i=1 di = N, the resulting symmetry group will be U(d1)×· · ·×U(dm). Only if the Hamiltonian is fully degenerate with N 5 Alternatively, one can also exploit the Heisenberg equation of motion lead- eigenvalue pairs ±λ, this will lead to the maximal symmetry group U(N) d ˆa ˆ ˆa a ˆb ˆα 0 ing to dt ξb = i[H, ξb ] = i(Kb) bξb and similarly for ξf . of possible identification maps T1. 6

We can use the supersymmetric identification maps to con- struct normalized identification maps L1 : Vb → Vf and 7 1.5 L2 : Vf → Vb as

1 1/2 1 1/2 L1 = |K− | T1,L2 = |K− | T2, (41) f b 1 bulk mode continuum

(where the form of L1 was identified in [11]). These have the property that their products exactly reproduce the linear complex structures 0.5 edge mode L1L2 = Jf and L2L1 = Jb, (42) 0 of the ground state of Hˆ . 0.5 1 To see this, it is convenient to work in the eigenbases of the trivial phase topological phase ( k) generators Kb and Kf . Let v ± ∈ Vb,C denote a basis of ( k) eigenvectors of Kb with eigenvalues ±iλk. Then {T1v ± } FIG. 1. Spectrum of the Kitaev chain with open ends. The system is a basis of Vf,C diagonalizing Kf . In fact, with respect to is in a trivial phase for |t/µ| < 1/2, critical at t/µ = ±1/2, and these bases Kb and Kf are represented by the same matrix. 1 1/2 1 1/2 topological, otherwise, with edge modes appearing. Accordingly, also |Kf− | and |Kb− | are represented by the same matrices. From this follows, in particular, operator, the identification map L associates the bosonic op- | 1|1/2 | 1|1/2 (43) 1 L1 = Kf− T1 = T1 Kb− erator 1 1/2 1 1/2 L = |K− | T = T |K− | , (44) 2 b 2 2 f α ˆa L1(ˆr) = rα(L1) aξb (48) and hence, we have with it. In this sense, the identification maps always identify 1 1/2 1 1/2 corresponding pairs of eigenmodes of the SUSY Hamiltonian L1L2 = |K− | T1|K− | T2 f b (45) with each other: If we diagonalize the SUSY Hamiltonian as 1 1 = |K− |T T = |K− |K = J , f 1 2 f f f   ˆ2 X ˆ ˆ 1 1/2 1 1/2 Q = ωi bi†bi +c ˆi†cˆi (49) L2L1 = |Kb− | T2|Kf− | T1 (46) i 1 1 = |Kb− |T2T1 = |Kb− |Kb = Jb . then, assuming that all ωi are different, we always have In the following we use the identification maps to asso-   iφi,1 iφi,2 ciate both linear observables and quadratic forms between the L1 (ˆci) = e ˆbi,L2 ˆbi = e cˆi (50) two supersymmetric partner systems. For this, it is important to note that, since the identification maps and their inverses for all i = 1, ..., N, because of (41). And, due to (42) act on the phase spaces, i.e., they act on upper indices from the complex phases are such that eiφi,1 eiφi,2 = −i, since the left, their corresponding transposes act on the dual phase Jb(ˆbi) = −iˆbi and Jf (ˆci) = −iˆci as follows from (11) and (9) spaces, i.e., on lower indices, as (expressed in the complex bases).

−1 | −1 | Example 3. The supercharge operator Qˆ introduced in Ex- L1 (L1) (L1 ) L1 Vb −−→ Vf ,Vf −−−−→ Vb,Vb∗ −−−−→ Vf∗,Vf∗ −−→ Vb∗, ample 2 induces the rather simple identification maps repre- −1 | −1 | L2 (L2) (L2 ) L2 sented by the matrices Vf −−→ Vb,Vb −−−−→ Vf ,Vf∗ −−−−→ Vb∗,Vb∗ −−→ Vf∗ . 1 0  0 1 ˆa L1 ≡ ,L2 ≡ . (51) For example, let sˆ = saξb be a linear operator on the bosonic 0 1 −1 0 system, then Accordingly the Hermitian mode operators are identified as a ˆα L2(ˆs) = sa(L2) αξf (47) L (ˆγ) =q, ˆ L (ˆη) =p ˆ , 1 1 (52) is the linear fermionic operator associated to it by the iden- L2(ˆq) =η, ˆ L2(ˆp) = −γˆ . ˆα tification map L1. Analogously, if rˆ = rαξf is a fermionic D. Application: supersymmetric Kitaev chain

7 Note that f(K) for a diagonalizable matrix K = U −1DU, where D is a In this section, we choose the well known Kitaev chain [67] diagonal matrix containing the eigenvalues of K, is equivalent to applying of N sites with open boundary conditions as a concrete ap- f to its eigenvalues i.e. f(K) = U −1f(D)U. If f can be expanded as a plication for the formalism above, and investigate the physi- power series, we can define f(K) even for non-diagonalizable K. cal properties of the identification maps. In our construction, 7 the supersymmetric partner of the Kitaev chain resembles the nonlinear zero mode (soliton) that can reverse the location of Kane-Lubensky chain [13]. In particular, we are interested in the edge mode [68], however, that falls beyond the ambit of addressing the question: to what extent do the identification the present setting. The localization of the edges mode at the maps preserve the localization properties of operators, when boundaries of the chain is exponential, in the sense that when mapping them from one system to its SUSY partner? P writing the edge mode operator as cˆN0 = j αjcˆj + βjcˆj†, or The form of the fermionic Kitaev chain Hamiltonian which ˆ P ˆ ˆ 2 2 bN0 = j αjbj + βjbj†, the quantities |αj| and |βj| decay we study is obtained by considering a real pairing, and setting exponentially away from the concerned edge. its magnitude equal to the hopping (t) in the original model The appearance and localization of the edge modes have proposed in [67]: consequences for the properties of the identification maps. In N N 1 particular, they affect to what extent the identification maps µ X
X−
preserve the locality of the onsite observables in a system Hˆf = cˆ†cˆi − cˆicˆ† + t cˆ† cˆi − cˆ† cˆ† + H.c. , 2 i i i+1 i+1 i when mapping them onto its SUSY partner, as visualized in i=1 i=1 (53) Fig. 2. From above, we know that the identification maps ex- actly map corresponding eigenmodes of the partner Hamilto- where µ denotes the chemical potential. A supercharge which nians to each other, and that we can choose the relative phase generates this Hamiltonian as the fermionic part of Qˆ2 = factor such that ˆ ˆ Hf + Hb is given by ˆ ˆ L1(ˆci0 ) = bi0 ,L2(bi0 ) = −iˆci0 . (57) N N 1 √ X t X−   Thus, at the point (t = 0), where the individual chain sites Qˆ = µ cˆiˆb† + √ cˆiˆbi +c ˆiˆb† + H.c. i µ +1 i+1 can be chosen as eigenmodes of the partner Hamiltonians, i=1 i=1 N N 1 the identification maps exactly associate the fermionic and √ X 2t X− bosonic chain sites one-to-one, maintaining their ordering. = µ (ˆγiqˆi +η ˆipˆi) + √ γˆi qˆi (54) µ +1 This feature of locality of the identification maps is con- i=1 i=1 spicuous throughout the trivial phase, except the onsite local- Its bosonic part resembles the Kane-Lubensky (KL) chain, a ization at t = 0 now transforms to an exponential one (with a well-studied model in topological mechanics [13]: length scale falling with the spectral gap), as seen in Fig. 2a for a chain of N = 30 sites. In detail, in the trivial phase, the N N N 1 2 2 − identification maps associate single site operators cˆ and ˆb ˆ µ X 2 4t + µ X 2 µ 2 X i i Hb = pˆi + qˆi + qˆ1 + 2t qˆiqˆi+1 P ˆ ˆ 2 2 2µ 2 with operators L1(ˆck) = j αkjbj + βkjbj, such that |αkj| i=1 i=2 i=1 and | |2 decay exponentially in | − |. Likewise, in the triv- N βkj k j   h  2    µ ˆ ˆ ˆ ˆ X µ 2t ˆ ˆ ˆ ˆ ial phase, L2 maps onsite bosonic operators to exponentially = b b† + b†b + 1 + 2 b†bi + bib† 2 1 1 1 1 2 µ i i localized fermionic operators. i=2 In the topological phase, however, the identification maps 2  ˆ ˆ ˆ ˆ  t ˆ ˆ  develop non-local features as can be seen in Fig. 2c. Here a +t bi 1bi† + bi 1bi + bibi + H.c. . − − µ fermionic site operator cˆk (e.g., in the figure, k = 15 in a chain (55) of N = 30 sites) when mapped to the operator L1(ˆck) on the bosonic side, can acquire a significant component located at Denoting the energy eigenmodes of the system with primed the left edge of the bosonic chain, which is the edge where also operators, the SUSY Hamiltonian can be diagonalized as the bosonic edge mode is localized. If we shift the original

N fermionic site to further right, the edge contribution to L1(ˆck) 2 X   decays and the localization of the resulting observable gains Qˆ = Hˆ + Hˆ = ω ˆb †ˆb +c ˆ cˆ . (56) f b i 0i i0 i0 † i0 prominence. On the other hand, if we move the original site i=1 to the left, the edge contribution to L1(ˆck) starts dominating Figure 1 schematically shows the spectrum of the Kitaev over the bulk coming from (bosonic) sites in the neighborhood chain which is in a trivial phase for |t/µ| < 1/2, and in a of the k-th site. 1 topological phase, otherwise. The bulk gap closes at the criti- Instead of the map L1, we may as well employ L2− to map cal point t/µ = ±1/2 in the limit of large N. The trivial phase the fermionic site operators to their bosonic counterparts. The 1 is featureless; all eigenmodes together form a bulk mode con- observed behaviour is similar, however, for L2− (ˆck), in the tinuum. However, as the system enters the topological phase topological phase, the edge contribution at the left end of the for |t/µ| > 1/2, an edge mode gradually separates from the bosonic chain dominates as k → N, i.e., when the original continuum and stabilizes at zero energy (albeit with an ex- fermionic operator approaches the right end of the chain. ponentially small gap with N) as a telltale signature of the The converse association of bosonic onsite operators with topological phase. On the fermionic side, i.e., for the Kitaev the corresponding fermionic observables, via the identifica- 1 chain, the edge modes are localized at both ends of the chain. tion maps L2 or L1− behaves very similar: in the trivial phase, In contrast, on the bosonic side, i.e., for the KL chain, they are they are exponentially localized as above, and in the topolog- localized only at one end (here the left end) of the chain. For ical phase, they exhibit similar non-local features. However, ˆ 1 ˆ completeness, we mention that in the KL chain, there exists a here both L2(bk) and L1− (bk) develop a dominant edge con- 8

0 0 10 + 10 +* ++* * + +* + -2 + + -2 +* 10 + 10 +* * * + * * +* + + + * + * +* + * +* 10-4 + * * ++* * * 10-4 +* * * * ++ * ++* (ˆq , pˆ†) (ˆq , pˆ†) (ˆq , pˆ†) (ˆq , pˆ†) (ˆq , pˆ†) +* +* + * + * * +* * 1 1 2 2 3 3 4 4 5 5 +* +* +* + 10-6 ++ * ++* 10-6 +* ++ * * +* +* +* ++ * * +* * +* * * ++* * 5 10 15 20 25 30 5 10 15 20 25 30

100 + 100 + L1 10-2 10-2 10-4 10-4 10-6 10-6

5 10 15 20 25 30 (ˆγ1, ηˆ1†) (ˆγ2, ηˆ2†) (ˆγ3, ηˆ3†) (ˆγ4, ηˆ4†) (ˆγ5, ηˆ5†) 5 10 15 20 25 30

(a) Trivial phase, t = 0.35µ. (b) Schematic visualization of L1(ˆc) (c) Topological phase, t = µ.

FIG. 2. Locality of the identification map L1 and its dependence on the relative coupling t/µ. The plots shows, for a system of N = 30 P ˆ ˆ† modes, how L1 associates the onsite operator cˆ15 in the fermionic Kitaev chain (53) to the operator L1(ˆc15) = j αj bj +βj bj on the bosonic Kane-Lubensky chain (55). In the trivial phase, the identification map preserves locality to a very high degree, namely, with an exponential 2 2 decay of the coefficients |αj | and |βj | with the distance |k − j| (here k ≡ 15). In distinction, in the topological phase, the operator L1(ˆck) can be non-local with a strong contribution from the boundary sites.

tribution when the original bosonic operator ˆbk approaches the Note that the bosonic condition also implies that A is even left end of the chain. For L2(ˆbk), the edge contribution ap- dimensional, as any anti-symmetric odd-dimensional matrix 1 has a vanishing determinant. pears on the left edge of the fermionic chain, for L− (ˆbk) it 1 ˆa ˆa ˆa appears on the right edge. In practice, we choose a basis ξ = (ξA, ξB) that splits This example demonstrates that the identification between V = A ⊕ B into a direct sum, where B is the complementary the bosonic and the fermionic parts of a SUSY Gaussian state system to A defined as via the identification maps may or may not coincide with an ( a a 1 b b identification based intuitively on some underlying (lattice) v ∈ V v Ωab− u = 0 ∀ u ∈ A (bosons) B =  a a 1 b b , geometry of the SUSY Hamiltonians. Whereas we observe v ∈ V v Gab− u = 0 ∀ u ∈ A (fermions) agreement in the trivial phase of the SUSY Kitaev chain, in (58) the topological phase, the identification maps behave vastly which is called the symplectic complement for bosons and the differently and disengage from notions based on the geometric orthogonal complement for fermions.8 We have the two bases intuition. The duality relations of the next section will show ξˆA and ξˆB with NA +NB = N, such that the resulting matrix that whereas the geometrical appearance of modes can be dis- representations of Ωab and Gab take the forms torted by the identification maps, their entanglement proper- ties remain intimately related.  1  −1  Ω  Ωab ≡   ≡ A , (bosons)  1  ΩB III. ENTANGLEMENT DUALITY −1  1  In this section, we derive how subsystem decompositions 1  G  Gab ≡   ≡ A . (fermions) V = A ⊕ B behave under the supersymmetric identification  1  GB maps L1 and L2 which leads to a duality between the bosonic 1 and fermionic (mixed) Gaussian states. We can also use this (59) to relate the associated entanglement entropies. Note that this implies that the restrictions ΩA and ΩB, or GA and GB, respectively, reproduce the standard forms from (9), i.e., the subsystems are themselves a bosonic or fermionic sys- A. Reduced Gaussian states and entanglement tem consisting NA and NB degrees of freedom. When quantizing the subsystems A and B, we can construct N Given a classical phase space V ' R2 , a subspace A ⊂ Fock spaces HA and HB as described in section II A, such that V defines a physical subsystem if the following condition is the full Hilbert space is a tensor product H = HA ⊗ HB. In satisfied:

• Bosonic: The restriction of Ωab to the subspace A is non-degenerate, i.e., has non-zero determinant. 8 Here, we used the inverse matrices Ω−1 and G−1 which are bilinear forms on the phase space (rather than its dual). In [63, 64, 69], they are denoted Ω−1 ≡ ω G−1 ≡ g • Fermionic: The subspace A is even dimensional. by ab ab and ab ab. 9

general, a pure Gaussian state |Ji ∈ H will itself not be a computational resources for large systems and appropriate ap- tensor product state with respect to this decomposition, which proximations or truncation for infinite dimensional Hilbert means that the subsystems are entangled. spaces (in the case of bosons). However, if the state |ψi hap- It is well-known that the bipartite entanglement encoded in pens to be a Gaussian state |Ji, we can exploit the relation a general pure state |ψi can be characterized by the spectrum between the spectra of ρA and JA to find analytical formulas of the mixed state ρA = TrB |ψi hψ| that results from tracing in terms of the restriction JA to the subsystem A. These re- over HB. If |ψi is a pure Gaussian state |Ji, the reduced state strictions correspond exactly to the symplectic or orthogonal ρA is a mixed Gaussian state. It can be expressed in terms of decomposition V = A ⊕ B introduced at the beginning of the linear complex structure J as [63] this section. The formulas for the von Neumann entropies are given by [55, 56] ( ˆr ˆs Qˆ qrsξ ξ + c0 (bosons) ρ = e− with Qˆ = , (60)  PNA A ˆr ˆs sb(λi) (bosons) iqrsξ ξ + c0 (fermions) S(ρ ) = i=1 , (65) A PNA i=1 sf (λi) (fermions) 9 with qrs is a 2NA-by-2NA matrix given by [63] x+1
x+1
x 1
x 1
with sb(x) = 2 log 2 − −2 log −2 for ( 1 l bosons, and − 1+x 1+x
− 1 x 1 x
for −i(Ω− )rl arccoth (iJA) s (bosons) sf (x) = 2 log 2 −2 log −2 q = A fermions, which can be unified by the single trace formula rs 1 l +i(GA− )rl arctanh (iJA) s (fermions) [63, 70] (61) ( 1 l +(ΩA− )rl arccot (JA) s (bosons) "   2# = . 1 1 + iJA 1 + iJA 1 l S(ρ ) = Tr log . (66) −(G− )rl arctanh (JA) s (fermions) A A 2 2 2 Here JA is the restriction of J to the 2NA-by-2NA subblock Formula (65) can also be used to compute the Renyi entropy representing the action of J onto the subspace A ⊂ V , and 1 1 1 of order n if we replace sb and sf by the respective Renyi similarly ΩA− and GA− denote the restrictions of Ω− and 1 entropy functions [63]: G− . The coefficient c0 is given by (k) 1 λ+1 k λ 1 k
  2  − 1 1+JA rb (λ) = log ( 2 ) ( −2 ) , (67)  4 log det 4 (bosons) k − 1 c0 =  2  . (62) 1 1+JA (k) 1 1+λ k 1 λ k
− log det (fermions) r (λ) = − log ( ) + ( − ) . (68) 4 4 f k − 1 2 2

It can be shown [62, 63] that the eigenvalues of JA are purely It follows from the above, that a subsystem A (bosonic or imaginary and appear in NA conjugate pairs ±iλi, where λi ∈ fermionic) of a system in a pure Gaussian state, is not entan- [1, ∞) for bosons and λi ∈ [0, 1] for fermions. gled with the rest of the system, i.e., it is in a product state with These relations have the consequence that for Gaussian the rest of the system, if and only if λi = 1 for all eigenvalues 2 states, the rather complicated spectrum of ρA simplifies, so of JA. In that case, we have JA = −1A, and the subsystem is that it can be efficiently calculated from the much simpler in a pure Gaussian state on its own. In particular, this is equiv- spectrum of JA given by ±iλi. Specifically, the eigenvalues alent to J(A) = A, i.e., the full (unrestricted) linear complex of ρA are structure mapping A onto itself.   ni 2  QNA (tanh ri) (bosons)  i=1 cosh ri µ(n1, . . . , nN ) = , (63) B. Supersymmetric ground states and identification maps  ni 2  QNA (tan ri) (fermions)  i=1 sec ri Above in (42), we saw that L1 and L2 together encode 1 1 the linear complex structures of both the bosonic and the where ri = 2 cosh− (λi), ni ∈ N for bosons, and ri = 1 1 fermionic part of the ground state (32) of Hˆ . In the following, 2 cos− (λi), ni = 0, 1 for fermions. The entanglement entropy SA(|ψi) = SB(|ψi) is com- we will use L1 and L2 to identify subsystems of fermionic puted as the von Neumann entropy S(ρA) of the reduced state modes with subsystems of bosonic modes, and vice versa. ρA, namely The maps L1 and L2 are the canonical choices for the iden- tification maps because they preserve the Kahler¨ structures of SA(|ψi) = S(ρA) = − Tr ρA log ρA . (64) the fermionic ground state |Jf i and the bosonic ground state |Jbi. That is, if we consider the fermionic 2-point function Calculating this quantity in practice is notoriously hard, as 1   it requires to compute the spectrum of ρA that demands vast Cαβ = hJ | ξˆαξˆβ |J i = Gαβ + iΩαβ (69) f,2 f f 2 f f and the bosonic

9 Not to confuse with the quadrature operator qˆ which carries at most one 1 i Cab = hJ | ξˆaξˆb |J i = Gab + iΩab
, (70) index. b,2 b b 2 b b 10 then one can show that we have (Af ∪A˜f ) is the smallest subsystem containing Ab (Af ) which is in a pure partial state, i.e., shares no entanglement with the ab a αβ | b 1 a αβ | 1 b Gb = (L2) αGf (L2 )β = (L1− ) αGf (L1 − )β , rest of the system. (71) Furthermore, Ab (Af ) shares the same amount of entangle- ment with the rest of the system as does A˜ (A˜ ). This follows as well as (dropping the indices for a better readability) b f from the fact that the restricted linear complex structures b JAb 1 1 b f f Ω = L Ω L| = L− Ω L− |. (72) and J (J and J ) have the same spectrum. To see this, f 1 b 1 2 b 2 A˜b Af A˜f consider the decomposition of the phase space into the direct Thus, the identification maps L1 and L2 preserve both the sum Vb = Ab ⊕ Bb according to (58). We define by PAb the symmetric and the antisymmetric forms of the Kahler¨ struc- projector onto A with respect to this decomposition: ture, and exactly map the bosonic and fermionic 2-point func- tions of the ground state onto each other. Interestingly, we see PAb (Ab) = Ab,PAb (Bb) = 0. (76) 1 that it makes no difference whether we use L1 and (L1)− , 1 The restriction of J to A is then J b = P J P . Anal- or L2 and (L2)− for this purpose. The reason for this is that b b Ab Ab b Ab 2 ˜ ˜ both maps are closely related. In fact, since Jf = −1 and ogously, considering the decomposition Vb = Ab ⊕ Bb, we J 2 = −1, it follows that find that the projector onto A˜ is P = −J P J , and b b A˜b b Ab b 1 − − b b (L1)− = L2Jf = JbL2 , J ˜ = PA˜ JbPA˜ = −JbJA Jb . (77) (73) Ab b b b 1 (L )− = −J L = −L J . 2 f 1 1 b 1 b b Since J − = −Jb, J and J are represented by similar b A˜b A˜b matrices and, hence, have the same spectrum. In fact, if v ∈ C. Dual supersymmetric subsystems is an eigenvector of b with b ± , then is Ab JAb JAb v = iλv Jbv an eigenvector of J b with the same eigenvalue. A˜b Since the identification maps L1 and L2 preserve the Kahler¨ structures, subsystems in one part (bosonic/fermionic) of a su- persymmetric Gaussian state, can be identified with subsys- D. Duality for Gaussian states and their entanglement tems in the other part (fermionic/bosonic). If A ⊂ Vb corresponds to a bosonic subsystem, then In the previous section, we analyzed the structure of subsys- 1 both L1(A) and L2− (A) are even-dimensional subspaces of tems in supersymmetric Gaussian states. In particular, we dis- the fermionic phase space Vf , hence they correspond to a cussed how the identification maps L1 and L2 relate bosonic fermionic subsystem, as defined in Sec. III A. If, on the other subsystems to fermionic subsystems, and vice versa. We can hand, A ⊂ Vf corresponds to a fermionic subsystem, then now use this background structure to derive the following du- 1 L2(A) and L1− (A) only correspond to a bosonic subsystem, ality between bosonic and fermionic Gaussian states. if the restriction of Ωf to A is non-degenerate. Following (72), The setting is as follows. We consider a classical phase this condition ensures that Ωb is non-degenerate as required space V ' R2N with Kahler¨ compatible structures (G, Ω,J) 1 for L2(A) and L1− (A) to yield a bosonic subsystem. and a choice of a subspace A ⊂ V with dim A = 2NA. We How does the subsystem which A is mapped to, depend on can associate two distinct quantum theories, namely a bosonic whether we use the identification map L1 (and its inverse) or Hilbert space Hb with Gaussian state |Jib and a fermionic the map L2? If the subsystem A is in a pure state, there is no Hilbert space Hf with Gaussian state |Jif . In both quan- difference, both identification maps identify A with the same tum theories, we can construct a reduced density operator subsystem. For example, if A ⊂ Vb is a bosonic subsystem ρA whose spectrum is determined by the a restricted complex which is in a pure state, then we have Jb(A) = A, thus structure. Crucially, however, the restriction of J to A is different de- 1 (74) L1(A) = L1(Jb(A)) = L2− (A). pending on whether we consider a bosonic system and use a However, for an entangled subsystem, we have J(A) =6 A symplectic decomposition of the phase space, or consider a and are led to the following commutative diagram. fermionic system and use an orthogonal decomposition, ac- cording to (58). This is due to the fact that the 2NA-by-2NA L1 subblock of the matrix J associated to the subspace A de- Ab Af L2 pends also on the basis elements that are not contained in A. Jb Jf (75) In particular, we choose two different bases for the bosonic L2 ˆ ˆA ˆB ˆ ˆA ˆB and fermionic case ξb = (ξb , ξb ) and ξf = (ξf , ξf ), such A˜b A˜f L1 that ˆA ˆA Here we have chosen Ab ⊂ Vb as a bosonic subsystem, span(ξb ) = A = span(ξf ) , (78) ˜ defined Af = L1(Ab), and denoted Ab = Jb(Ab) and ˆB ˆB span(ξb ) = Bb =6 Bf = span(ξf ) , (79) A˜f = Jf (Af ). Whereas Ab and A˜b (Af and A˜f ) define different bosonic where Bb and Bf are the respective bosonic and fermionic (fermionic) subsystems, they are intimately related: Ab ∪ A˜b complements defined in (58). Consequently, the restrictions 11 of J to the subspace A can be different on the bosonic and Corollary 1. Given a supersymmetric system with super- b the fermionic side, which therefore are denoted by JA and charge operator Qˆ, we have a supersymmetric ground state f 2 JA, respectively. Equipped with this, we can now prove the |Jbi⊗|Jf i of Hˆ = Qˆ and identification maps L1 : Vb → Vf , following proposition. 1 1 L2 : Vf → Vb, and their inverses L1− and L−2 , as above. Then Proposition 1 implies the following. Proposition 1 (Entanglement duality). We consider a super- Let S ⊂ Vb be a bosonic subsystem and L(S) ⊂ Vf , with symmetric system with phase space V ' Vb ' Vf equipped 1 L = L1 or L = L2− , be a dual fermionic subsystem. Then with Kahler¨ structures (G, Ω,J), which simultaneously de- b f the restricted linear structures JS and JL(S) of these two sub- scribe a bosonic and a fermionic Gaussian state, namely systems are such that |Jib ∈ Hb and |Jif ∈ Hf . We now choose a subsystem A ⊂ V . This leads to two inequivalent decompositions of b 1 1 f JS = L− Jf L|S = L− JL(S)L, V , namely V = A ⊕ Bb and V = A ⊕ Bf , where the com- (85) f 1 b 1 plementary subsystems Bb and Bf are defined in (58). The JL(S) = LJbL− |L(S) = LJS L− . b f associated reduced states ρA and ρA are both Gaussian and fully described by the restricted complex structure J b and J f , Let R ⊂ Vf be a fermionic subsystem and L(R) ⊂ Vb, with A A 1 respectively, which satisfy the following relation: L = L2 or L = L1− , be a dual bosonic subsystem. Then the f b restricted linear structures JR and JL(R) of these two subsys- f b 1 JA = −(JA)− . (80) tems are such that f 1 1 b b b JR = L− JbL|R = L− JL(R)L, In particular, this implies that the eigenvalues ±iλi of JA are f f b f b f 1 f 1 related to the eigenvalues ±iλi of JA via λi = 1/λi. JL(R) = LJRL− |L(R) = LJRL− .

Proof. The decompositions V = A ⊕ Bb and V = A ⊕ Bf Thus the eigenvalues of the dual restricted complex structures define projectors, such that Pb : V → A, Pf : V → A, P¯b : are inverses of each other, and their entanglement spectra are V → Bb and P¯f : V → Af , such that 1 = Pb +P¯b = Pf +P¯f . accordingly related by (63). The restricted complex structures are then defined as While our result applies to any identification where a b bosonic and a fermionic phase space are related, supersym- JA = PbJ|A : A → A, (81) metric systems with the identification maps L and L as dis- f 1 2 JA = Pf J|A : A → A. (82) cussed in section II C are the prime examples where such an identification is naturally chosen. f b 1 We need to show JA = −(JA)− which is equivalent to The entanglement duality implies an intimate relation of a f b JAJA = −1A. To show the latter, we take a vector a ∈ A subsystem’s entanglement entropy with that of its dual sub- and calculate system, because both the von Neumann entropy (65), as well

2 as the Renyi entropies (67) are functions of the restricted com- −a = −Pf a = Pf J a = Pf J(Pb + P¯b)Ja (83) plex linear structure’s spectrum. For the simplest possible b ¯ f b ¯ = Pf JJAa + Pf JPbJa = JAJAa + Pf JPbJa . case, where the subsystems each consist of a single mode only, Fig. 3 shows the relation between the von Neumann entropy of The second term in (83) vanishes since for an arbitrary vector the fermionic mode and the bosonic mode. Here, the restricted v ∈ V , the inner product complex structures have one pair of imaginary eigenvalues, 1 ±iλ for for the fermionic and ±iλ− for the bosonic system, 1 1 G− (v, Pf JP¯ba) = G− (Pf v, JP¯bJa) which with the formula for the von Neumann entropies (65) 1 yields the relation plotted in Fig. 3. = −Ω− ( Pf v , P¯bJa) = 0 , (84) |{z} | {z } Evidently, the bosonic and the fermionic entanglement be- A Bb ∈ come asymptotically equal when the corresponding modes ap- proach a pure partial state and consequently the entanglement 1 1 where we have used the relationship G− (·,J·) = −Ω− (·, ·) approaches zero (λ → 1). In the opposite direction, however, following from (11). In matrix notation, we would write the entanglement in the bosonic mode grows without a bound 1 1 a b G− (v, w) = (G− )abv w and so on. That the inner prod- as λ → 0, whereas the entanglement in the dual fermionic 1 uct Ω− (·, ·) in (84) vanishes follows from the definition of mode tends to saturate at the maximal value of log 2. Bb in (58), and therefore, proves the identity in (80). This relation between the SUSY partner single modes read- ily extends to multiple modes because, as is evident from (65), At first glance, this result is a simple statement about re- the total entanglement entropy of a subsystem is given by the stricting a complex structure J : V → V to a subspace A ⊂ V sum of the entanglement entropies over the individual normal in two inequivalent ways. However, its application to bosonic modes of that subsystem. This is related to the fact, that a and fermionic Gaussian states implies a rather complicated re- mixed Gaussian state always can be expressed as the product b f lationship of the spectra ρA and ρA via (63) and (80), which state of its normal modes, which are given by the eigenmodes can be made precise in the following corollary relating the re- of the restricted linear complex structure [62]. As a conse- stricted complex structures of the dual subsystems. quence of the entanglement duality, the identification maps 12

3.0 b f eigenvalues ±iλi of Jb are related to the eigenvalues ±iλi of J via λb = 1/λf . 2.5 f i i Proof. Our identification of the phase spaces V ' V ' V 2.0 b f gives rise to a single Lie algebra generator K : V → V 1.5 for the Hamiltonian βHˆ from (25). The spectrum of K agrees with that of the bosonic generator Kb : Vb → Vb 1.0 as well as the fermionic generator Kf : Vf → Vf defined as (K )a = β Ωachb and (K )α = β Gαγ qf respectively. 0.5 b b cb f δ γδ βHˆ We can compare with (60) to identify that ρ = e− /Z gives 0.0 b β b β 1 c f β f 0.0 0.2 0.4 0.6 0.8 1.0 rise to qab = 2 hab = 2 Ωac− (Kb) b and qαγ = 2 hαγ = β 1 δ 2 Gαδ− (Kf ) γ . We can invert (61) to find

b FIG. 3. Entanglement entropies for dual subsystems each consisting Jb = − cot Ωq = − cot(Kb/2) ≡ − cot(K/2) , (87) of a single mode. The solid line shows the entanglement entropy of a f Jf = tan Gq = tan(Kf /2) ≡ tan(K/2) , (88) single fermionic mode for which the restricted complex structure has eigenvalues ±iλ. Due to the entanglement duality (80), the restricted from which (86) readily follows. complex structure of the dual bosonic mode has eigenvalues ±iλ−1, and the dashed line plots the resulting entanglement entropy. E. Application: two-mode system identify normal modes with reciprocal eigenvalues of the re- In this section, we study some consequences of the en- stricted complex linear structures. tanglement duality in a basic two-mode example where the At this stage, it is an important question if the entangle- SUSY Hamiltonian is given by a fermionic and a bosonic two- ment duality is merely an interesting observation or to what mode squeezing Hamiltonian. While this is a minimal exam- extent it matters for physical systems. In the following two ple, it explains certain basic relations which are important for sections, we therefore investigate the entanglement duality in our analysis of a lattice Hamiltonian in the next subsection. two concrete applications. First, we consider the toy model of ˆ a supersymmetric system with two bosonic and two fermionic Consider the following supercharge operator Q, which is modes in Sec. III E, before we then move on to the recently parametrized by real numbers rb ≥ 0 and 0 ≤ rf < π/4, proposed SUSY Kitaev honeycomb model in Sec. III F. corresponding to squeezing parameters. Before proceeding, we note that mixed states that arise as Qˆ = (cosh(r ) cos(r ) − sinh(r ) sin(r )) (ˆγ qˆ +η ˆ pˆ ) a thermal state of a supersymmetric Hamiltonian also come b f b f 1 1 1 1 under the ambit of our duality. In detail, we ask if a relation, + (cosh(rb) cos(rf ) + sinh(rb) sin(rf )) (ˆγ2qˆ2 +η ˆ2pˆ2) similar to what applies to the mixed states arising from a re- + (cosh(rb) sin(rf ) − sinh(rb) cos(rf )) (ˆγ1qˆ2 − ηˆ1pˆ2) duction of a pure state to a subsystem A, holds for a thermal + (cosh(rb) sin(rf ) + sinh(rb) cos(rf )) (−γˆ2qˆ1 + η2pˆ1) . ˆ ˆ2 ˆ ˆ state of a supersymmetric Hamiltonian H = Q = Hb + Hf , (89) 1 βHˆ 1 βHˆb βHˆf i.e., ρ = Z e = Z e ⊗ e . The following proposition answers this question in the affirmative. It generates a SUSY Hamiltonian Hˆ = Hˆb + Hˆf which con- sists of the two-mode Hamiltonians: Proposition 2 (Thermal state duality). We consider a super- symmetric system with phase space ' ' equipped cosh(2rb) X V Vb Vf Hˆ = qˆ2 +p ˆ2
+ sinh(2r ) (ˆp pˆ − qˆ qˆ ) , with Kahler¨ structures (G, Ω,J), for which we have a Hamil- b 2 i i b 1 2 1 2 i=1,2 ˆ ˆ2 ˆ ˆ 1 βHˆ tonian H = Q = Hb + Hf . The thermal state ρ = Z e (90) at inverse temperature β is a tensor product of two Gaussian i cos(2rf ) X states ρ = ρb ⊗ ρf with associated restricted complex struc- Hˆ = (ˆγiηˆi − ηˆiγˆi) + i sin(2r ) (ˆγ η − γ η ) . f 2 f 1 2 2 1 tures Jb and Jf related by i=1,2 (91) 1 Jb = −(Jf )− , (86) The ground states of these Hamiltonians are two-mode which exactly resembles the entanglement duality, but now ap- squeezed states. Accordingly, the identification maps L1 and plies to the whole system. In particular, this implies that the L2 are represented by 13

 cos(rf ) cosh(rb) − sin(rf ) sinh(rb) sin(rf ) cosh(rb) − cos(rf ) sinh(rb) 0 0  − sin(rf ) cosh(rb) − cos(rf ) sinh(rb) sin(rf ) sinh(rb) + cos(rf ) cosh(rb) 0 0 L1 ≡ 0 0 cos(rf ) cosh(rb) − sin(rf ) sinh(rb) cos(rf ) sinh(rb) − sin(rf ) cosh(rb) , 0 0 cos(rf ) sinh(rb) + sin(rf ) cosh(rb) sin(rf ) sinh(rb) + cos(rf ) cosh(rb)

 0 0 cos(rf ) cosh(rb) − sin(rf ) sinh(rb) cos(rf ) sinh(rb) + sin(rf ) cosh(rb)  0 0 cos(rf ) sinh(rb) − sin(rf ) cosh(rb) sin(rf ) sinh(rb) + cos(rf ) cosh(rb) L2 ≡ sin(rf ) sinh(rb) − cos(rf ) cosh(rb) cos(rf ) sinh(rb) + sin(rf ) cosh(rb) 0 0 . cos(rf ) sinh(rb) − sin(rf ) cosh(rb) − sin(rf ) sinh(rb) − cos(rf ) cosh(rb) 0 0 (92)

They lead to the complex structures tator, we can make use of (72) to find

hJf | L˜2(ˆq1)L˜2(ˆp1) − L˜2(ˆp1)L˜2(ˆq1) |Jf i  0 0 cosh(2rb) sinh(2rb)  b 0 0 sinh(2rb) cosh(2rb) 1 i J ≡   = hJb| qˆ1pˆ1 − pˆ1qˆ1 |Jbi = . (98)  − cosh(2rb) sinh(2rb) 0 0  cosh(2rb) cosh(2rb) sinh(2r ) − cosh(2r ) 0 0 b b Hence, the restriction of J has eigenvalues ±i(cosh(2r )) 1, (93) f b − as predicted by the entanglement duality.  0 0 cos(2rf ) sin(2rf )  The opposite identification of the fermionic mode R = 0 0 − sin(2r ) cos(2r ) J f ≡  f f  (ˆγ1, ηˆ1) with a bosonic mode is completely analogous, but  − cos(2rf ) sin(2rf ) 0 0  it additionally highlights an important effect in the limit of − sin(2rf ) − cos(2rf ) 0 0 rf → π/4. The restriction of Jf to (ˆγ1, ηˆ1) is represented by (94)   f 0 cos(2rf ) JR ≡ , (99) which define pure two-mode squeezed states. − cos(2rf ) 0 Let us now study, how the identification maps act on the which has eigenvalues ±i cos(2rf ). Mapping γˆ1 and ηˆ1 via single site modes (ˆq1, pˆ1) and (ˆγ1, ηˆ1) respectively. It is clear the identification map L1, we obtain two bosonic observables from (92) that when both the bosonic and fermionic squeezing which obey vanish, i.e., rb = 0 = rf , these are trivially identified with each other. However, when either squeezing parameters takes hJb| [L1(ˆγ1),L1(ˆη1)] |Jbi = cos(2rf ). (100) a non-zero value, the identification maps will mix the modes Hence, we need to rescale the operators 1 and 2. Beginning with the bosonic mode S = (ˆq1, pˆ1), we find it ˜ L1(ˆγ1) ˜ L1(ˆη1) L1(ˆγ1) := p , L1(ˆη1) := p (101) has the restricted complex linear structure cos(2rf ) cos(2rf )   in order to obtain properly anti-commuting quadrature opera- b 0 cosh(2rb) JS ≡ , (95) tors, defining a bosonic mode. For these, we find − cosh(2rb) 0 ˜ ˜ 1 which has eigenvalues ± , signalling that for hJb|{L1(ˆγ1), L1(ˆη1)}|Jbi = , (102) i cosh(2rb) rb > cos(2rf ) 0, the mode is in a mixed Gaussian state, due to its entan- glement with mode . Now, we can use to associate this showing that the restriction of Jb has eigenvalues 2 L2 1 bosonic mode with a fermionic mode. Here, we need to take ±i (cos(2rf ))− . → into account that the fermionic observables L2(ˆq1) and L2(ˆp1) In the limit of rf π/4, the fermionic site mode (ˆγ1, ηˆ1) are not properly normalized Majorana operators. In fact, as a approaches maximal entanglement with the rest of the system, consequence of (71), we have corresponding to an entanglement entropy of one bit, i.e., log 2 in natural units. Consequently, also its bosonic dual system hJf | {L2(ˆq1),L2(ˆp1)} |Jf i = hJb| {qˆ1, pˆ1} |Jbi = cosh(2rb). approaches maximal entanglement. However, for the bosonic (96) mode this means that its entanglement entropy grows with- out bound, as shown in Fig. 3. In particular, at the point of Instead, the properly normalized Majorana operators, which rf = π/4, the fermionic mode 1 would represent a fermionic correspond to an orthogonal basis in the fermionic phase subsystem which is maximally entangled with mode 2. How- space, are ever, such a fermionic mode is not mapped to a valid bosonic subsystem by the identification maps. In fact, for rf = π/4, the identification map acts as ˜ L2(ˆq1) ˜ L2(ˆp1) L1 L2(ˆq1) := p , L2(ˆp1) := p , (97) cosh(2rb) cosh(2rb) cosh(rb) − sinh(rb) L1(ˆγ1) = √ (ˆq1 +q ˆ2) , 2 which we can use to calculate the restriction of Jf to this sub- e.g. cosh(rb) − sinh(rb) system, , in order to calculate its entanglement with the L1(ˆη1) = √ (ˆp1 − pˆ2) , (103) rest of the fermionic system. When calculating their commu- 2 14 which are commuting observables, and thus do not define Hamiltonian Hˆ = Qˆ2 is [11] a proper bosonic mode (cf. (58)), as also seen by the fact that (100) vanishes. N ˆ X
This example highlights, how, in general, fermionic Ma- Q = γˆiAijqˆj +η ˆiδijpˆj , (105) jorana operators that generate an almost maximally entan- i,j=1 gled mode are mapped to almost commuting bosonic opera- which implies a block-diagonal matrix representation of R . tors by the identification maps, which, in the limit of max- αa The bosonic part of this Hamiltonian imal fermionic entanglement, thus, fail to define a bosonic mode. The following Sec. III F showcases a peculiar conse- N N quence of this fundamental relationship between highly en- 1 X X 2 Hˆ = qˆi (A|A) qˆj + pˆ (106) b 2 ij i tangled fermionic modes and their bosonic counterparts in two i,j=1 i=1 dimensions. corresponds to a triangular lattice of harmonic oscillators as depicted in Fig. 4a or, in the appropriate classical limit, to a F. Application: supersymmetric Kitaev honeycomb model triangular network of balls and springs [11]. As previously mentioned, in the gapped phase, both the In this section, we demonstrate consequences of the de- fermionic and the bosonic lattices of this SUSY Hamiltonian rived entanglement duality in the example of the celebrated exhibit an area law scaling (2) in the entanglement entropy of Kitaev honeycomb model [60], a spin model with charac- lattice subregions. Accordingly, Fig. 4b shows good agree- teristic bond-directional exchanges on the honeycomb lat- ment of our numerical example with an area law scaling of tice (Fig. 4a), and its supersymmetric extension [11]. In the entanglement entropy. There, we consider a honeycomb their gapped phases, both the fermionic and the bosonic lat- lattice which is periodic, with equal side lengths, comprising tice of this supersymmetric system exhibit the entanglement- N = 45 × 45 = 2025 unit cells, of two sites each, in total. area law (2). Because the identification maps between the From this lattice, we cut out parallelogram-shaped subsystems fermionic and the bosonic lattice behave local and preserve of sidelength m, i.e., containing M = m × m unit cells, as the shape of subregions of a lattice very well, one may ex- indicated in Fig. 4a, and calculate their entanglement entropy f pect also the entropy of these dual subsystems to follow an S(ρA) with the rest of the lattice. We compare two different area law. However, we show that in mapping from fermionic combinations of the hoppings, ~j = (jx, jy, jz) = (2.5, 1, 1) subregions to bosonic ones, a peculiar phenomenon can arise and ~j = (1, 1, 2.5), with respect to the orientation of the where the entanglement entropy of the dual bosonic subsys- parallelograms. These two orientations differ in the type of tems scales much faster than its preimage in the fermionic lat- neighboring sites which the parallelogram’s boundary sepa- tice which follows the area law. This is attributed to the pres- rates. The boundary only cuts through links with hopping jy ence of almost maximally entangled modes in the fermionic and jz. Thus, in the first case, it only separates sites linked subsystem. by the two weaker hoppings, whereas in the second case, the The analytical solution of the Kitaev honeycomb model is links with the strongest hopping are included alongside those achieved by recasting it in terms of non-interacting Majorana with one of the weaker hoppings. The effect of this is an over- fermions hopping on the same honeycomb lattice (in the back- all higher entanglement entropy in the second case. However, ground of a classical (static) Z2 gauge field). The resulting both cases still exhibit the same power law predicted by the fermionic Hamiltonian reads entanglement area law. N In the periodic lattices considered here, the identification i X |
Hˆ = − ηˆiA γˆj − γˆiAijηˆj . (104) maps behave local in the sense that on-site operators in one f 2 ij i,j=1 lattice are mapped to exponentially localized operators on the supersymmetric partner lattice. Hence, the geometrical ˆ Expressed this way, Hf describes the hopping of Majorana appearance of subsystems is well preserved when they are fermions between the two types of sites of the honeycomb lat- mapped to their dual subsystems in the supersymmetric part- tice (Fig. 4a), where each of the Majorana operators γˆi and ner lattice by the identification maps. At first sight, this may ηˆi resides on one type of the lattice sites. The N × N-matrix seem to suggest that the entanglement entropy of the dual A corresponds to the connectivity matrix of the lattice as de- systems also should exhibit an area law scaling, since the picted in Fig. 4a, which we consider to be periodic. It involves entanglement-area relation of (2) holds in both the fermionic the hopping strengths along the three bonds around each site and the bosonic lattice we consider. of the honeycomb lattice, which we denote by jx, jy, jz. The Indeed, this is what we generally find for subsystems of inequality |jx| ≤ |jy| + |jz| and its cyclic permutations to- the bosonic lattice and their dual subsystems in the fermionic gether imply a gapless spectrum of Hˆf , otherwise Hˆf has a lattice: the entanglement entropies here often only differ by gapped spectrum. While the phenomena discussed below can a relatively small overall factor. However, Fig. 4c demon- arise in both phases, for our numerical results, we will focus strates, for the numerical example introduced before, that the on the gapped phase below. entanglement entropy of fermionic subsystems and their dual ˆ ˆα ˆa ˆ bosonic subsystems can scale very differently: depending on A supercharge operator Q = Rαaξf ξb that leads to Hf being identified with the fermionic part of the supersymmetric the orientation of ~j the dual entropy may scale in agreement 15

−1 (a) Fermionic honeycomb lattice of (104) and bosonic triangular lattice of (106). The map L1 identifies fermionic subsystems with bosonic dual subsystems.

**** 1000 50 ****** * "super area law" ********* area law scaling ****** *** ****** ***** *** *** **** *** ** *** ** * *** +++++ * * * * ++++++ + * * * * * +++++ 100 * 10 * +++++ * * ++++ * +++ * area law scaling * +++ * 5 * ++ + + * ++++++++++ * + ++++++++ + 10 * ++++++ * + +++++ + ++++ + + ++ * + + 1 * + + + * + * + 1 + 0.5 + + + + +

1 10 100 1000 1 10 100 1000 (b) Entanglement entropy of parallelogram-shaped subsystems (c) Entanglement entropy of dual bosonic subsystems FIG. 4. Supersymmetric Kitaev honeycomb model: The supercharge (105) generates the fermionic honeycomb Kitaev model (104) and the bosonic triangular lattice (106). Figures 4b and 4c show numerical calculations of the entanglement entropy of parallegram-shaped subsystems on the fermionic side, and of their dual bosonic subsystems on the other side, for two different orientations of the hopping parameters ~j = (jx, jy, jz). For the numerical examples, periodic lattices with a total number of N = 45 × 45 = 2025 unit cells were considered. The parallelograms of the subsystems contain M = m × m modes, i.e., M unit cells of the honeycomb lattice, and are oriented such that they do not cut through links with hopping parameter jx. The fermionic entropies show good numerical agreement with the area law, which they are known to follow in the thermodynamic limit. Depending on the orientation of the couplings ~j relative to the parallelogram, the dual entropies can follow the area law, or a scale much faster. with an area law scaling or they can scale much faster accord- tems considered in Fig. 4a for the two distinct orientations of ing to a “super area law”. the hopping mentioned before. In the first case, where the How does this phenomenon arise? First, let us note that, parallelograms do not cut through any strong links, the eigen- when the parallelogram in the original fermionic lattice only values roughly lie in the interval 0.9 < |λi| ≤ 1, as seen in cuts links with weaker hoppings, the dual entropy follows the inset of Fig. 5. As is evident from Fig. 3, in this regime f an area law scaling. The higher scaling of the dual entropy the entanglement entropy of each of the eigenmodes of JA, appears when the parallelogram cuts through links with the i.e., the normal modes of the subsystem, is almost the same as strongest hopping. This separation of strongly linked Ma- the entanglement entropy of their dual bosonic modes. Thus, jorona sites, however, heralds the presence of normal modes in the first case, the entanglement entropies for the fermionic in the fermionic parallelogram which are (almost) maximally subsystems and their bosonic duals are almost the same and entangled with the rest of the lattice. The presence of such follow, in particular, the same scaling. modes is the reason for the observed peculiar scaling of the f In contrast, in the second case, the spectrum of JA exhibits dual entropies. a certain number of eigenvalues which are very small or al- In fact, the mathematical explanation for the observed am- most zero. Note, that the number of these pairs of eigenvalues plified scaling of the dual entropies is rooted in the spectrum ±iλi, that fall below λi . 0.1, corresponds exactly to the f of the restricted fermionic linear complex structure JA. Fig. 5 side length of the parallelograms, i.e., is half of the number of plots the absolute values of the eigenvalues for the subsys- strong links which the parallelogram cut through. The normal 16

ery time its boundary cuts through a pair of strong links (on + +* +* +*** +****** +*************** +************************* +********** +*** +++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++***++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++ *********************************************************************************************************************************************************************************************************************************************************++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++******+++++++++++***************************************++++++++++++++++++++++++++++++++++++++++++******************************************++++++++++++++++++++++++++++++++++++++++++*****************************************+++++++++++++++++++++++++++++++++++++++++****************************************+++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++************************************++++++++++++++++++++++++++++++++++++++++++++++*************************************************************************************************************************************************************************************************++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++***************************************+++++++++++++++++++++++++++++++++++++++**************************************++++++++++++++++++++++++++++++++++++++++++******************************************++++++++++++++++++++++++++++++++++++++++++++++++++++++++*************************************************************+++++++++++++++++++++++++++++++++++++++++****************************************+++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++**************************************+++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++******************************************************+++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++*********************************++++++++++++++++++++++++++++++++++++++++++++++++++****************************************************************+++++++++++++++++++++++++++++++++++++++++++++++++++++++********************************************************************++++++++++++++++++++++++++++++++++++++++++++++++*****************************************************************************++++++++++++++++++++++++++++++++++++++++++++++++++************************************************************************************************++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++*********************************************************************************************++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++***********************************************************************++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++*******************************************************************************************+++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++************************************************************************************************************************+++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++******************************************************************************************************************************++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++************************************************************************************************++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++***********************************************************************+++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++**********************************************************************************+++++++++++++++++++++++++++++++++++++++*****************************************************************************++++++++++++++++++++++++++++++++++++++++++++++++++++************************************************ ***********************************************+++++++++++++++++++++++++++++++++++++++++++++********************************************************************************************++++++++++++++++++++++++++++++++++++++++++++++++++++++*************************************************+++++++++++++++++++++++++++++++++++++++++++++++++++ +******************************************************************************* +++++++++++++++++++++++++++++++++++++++++++++++++++ +++++ * * **** * * * * * ******** * * * * * * * * * ** ************ hibits a strongly entangled normal mode (corresponding to an * * * * * * * ******* ***************** -3 * * * * * ********************************* 10 * * * ******************************** almost vanishing eigenvalue of the restricted complex struc- * * * * * ******************************** * * * ******** ********************* * * ************ **************** ture). These normal modes are highly localized at the edge * * * * * * *********************** * ******** ************** * * * *** *************** of the subsystem and share no entanglement with any mode * * *** ******** ************** -8 * * * * ************ 10 * * * *** ************* inside the subsystem but with those lying on the complement * ** ****************** * ** ********** * * ** *********** of the subsystem. In fact, the normal modes of the subsys- * ** *** ***** * 1 + + + +++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++ ++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++ ++++++++++++++++++++++++++++++++++++++++++++++++++++++++++ ++++++++++++++++++++++++++++++++++++++++++++++++++++++++++ +++++++++++++++++++++++++++++++++++++++++++++++++++++ +++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++ ++++++++++++++++++++++++++++++++++++ +++++++++++++++++++++++++++++++++++++ +++++++++++++++++++++++++++++++++++ +++++++++++++++++++++++++++++++++++++++++++++++++++ +++++++++++++++ ++++++++++++++++++++++++++++++ +++++++++++++ ++++++++++++++++ +++++++++++++++++++++++++++++ ++++++++++ ++ +++ +++++++++ ++++ ++++++ +++++ ++++++++ +++++++ ++++++++++++++++++ * * ** ********** + + + + ++++++++++++++++++++++++++++++++++++ *** ** ****** tem, which carry entanglement, form pairs with those from + + + + ++++++++++++++++++++++++++++++++++++ ** ** *** -13 + + + ++++++++++++++++++++++++++++++++++++ ******** 10 + + ++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++ * *** the complement such that each normal mode is entangled with + + ++++++++++++++++++++++++++++++++++++++++++++++++++++++++ ** * ***** 0.9 + + + ++ ** * **** 10 + **** * exactly one partner (normal) mode of the complement. The ******** 1 10 100 1000 **** partner normal modes of the highly entangled subsystem nor- 1 10 100 1000 mal modes are localized right outside the subsystem. Thus a pair of partner modes (one inside the subsystem and one + * outside) forms a two-mode subsystem, localized in the imme- diate neighborhood of the subsystem’s boundary, which is not FIG. 5. Absolute values of the eigenvalues of the fermionic linear entangled with the rest of the system but in a pure two-mode f complex structure JA on the Kitaev honeycomb lattice restricted to squeezed state on its own. The identification maps now map the subsystems of Fig. 4. The inset zooms in on the eigenvalues of the each pair of such fermionic normal mode partners to a pair of ~ first case, j = (2.5, 1, 1) which hardly fall below |λi| ≈ 0.9. In the bosonic normal mode partners, one inside the dual subsystem second case, ~j = (1, 1, 2.5), the absolute values decay exponentially and one outside. Due to the locality properties of the identifi- with the subsystem size M, thus triggering the amplified scaling of cation maps, their joint support on the bosonic lattice sites is the dual entropies in Fig.4c, up until the subsystem exhausts the full closely related to the shape of the fermionic pair. lattice of N = 2025 modes. In this mapping, pair by pair, the same mechanism as dis- cussed in (103) takes place. The Majorana operators of the fermionic subsystem normal mode are mapped to a pair of modes corresponding to these eigenvalues thus share almost bosonic observables which are almost commuting, thus de- maximal entanglement with the rest of the system, i.e., the fine a highly entangled bosonic mode. Such almost commut- complement of the region surrounded by the parallelogram. ing bosonic observables need not be spatially separated on the In terms of their entanglement entropy, as discussed before lattice, but they can have equal support on the same lattice and also evident from Fig. 3, the fermionic normal modes ap- sites, as (103) demonstrates: there both bosonic observables proach the maximum value of one bit entanglement entropy have equal support on both of the two modes, however one as λi → 0, whereas the entanglement entropy of their dual quadrature is proportional to qˆ1 +q ˆ2 but the other to pˆ1 − pˆ2, 1 bosonic modes diverges as λi− → ∞. thus they commute. As a result, the total entanglement entropy of the dual Because of such localized and highly entangled bosonic bosonic subsystem scales much faster with its subsystem size modes, it is possible for the dual bosonic subsystems, despite than the original fermionic system does. This effect is vi- being well localized, to exhibit a scaling of entanglement en- sualized in Fig. 6, whose stacked plots show the mode-wise tropy that exceeds the area law of the original fermionic lat- contribution of the normal modes to the total entropy of the tice. The entanglement-area law assumes the subsystem di- fermionic subregions and their duals in the bosonic lattice. On vision being a direct sum of individual lattice sites, i.e., in the fermionic side, the individual contributions are bounded a bosonic system, the quadarature operators qˆi and pˆi either by one bit per mode, thus their summed contribution still both belong to the subsystem or they both do not. In con- result in a growth linear in the perimeter of the parallelo- trast, the boundary between the dual bosonic subsystems and gram. However, on the bosonic side, the individual contribu- the rest of the (bosonic) lattice considered in this example tion from each normal mode continues to grow as the system may well separate different linear combinations of the onsite size increases, resulting in a higher scaling of the entangle- bosonic operators. ment entropy than that predicted by the area law. Let us em- phasize that the√ total number of low-lying fermionic eigenval- ues scaling as M (with M being the subsystem size) alone is IV. DISCUSSION not sufficient to give rise to a “super area law” on the bosonic side, but also that these low-lying values actually decay to- In this article, we study the entanglement properties of wards zero. If they were bounded by some λmin, such that bosonic and fermionic Gaussian states that are related via su- λi ≥ λmin > 0, the entropy of each dual bosonic mode would persymmetry, in other words, belong to Hamiltonians which be upper bounded√ by sb(1/λmin), resulting again in the con- are supersymmetric partners of each other. After reviewing a ventional M scaling of the area law. The peculiar phenomenon observed above can be viewed as a direct physical instance of the minimal two-mode example 10 in Sec. III E taking place at the edge of the subsystem: ev- These pairs are connected by the complex structure Jf of the ground state. 17

25 500

20 400

15 300

10 200

5 100

0 0 0 5 10 15 20 0 5 10 15 20

(a) Fermionic hexagonal subsystem Af (b) Dual bosonic subsystem L2 (Af ) FIG. 6. Stacked plots of the mode-wise contribution to the total entanglement entropy, for the setup of Figs. 4b and 4c, for the case ~j = (1, 1, 2.5). The difference between the (i − 1)-th and i-th line gives the contribution of the subsystem’s i-th normal mode to the total entanglement entropy.√ Thus, the upper-most line coincides with the respective plot in Figs. 4b and 4c. Note, that here we have changed√ the horizontal axis to M, i.e., the sidelength of the parallelogram-shaped fermionic subsystems. The fermionic entropy is dominated by the M normal modes that are almost maximally entangled, and the bosonic entropy by their dual modes. unified framework to describe these states in terms of Kahler¨ ance of “super area law” behaviour in the entanglement en- structures, we prove the main result of this article in Propo- tropy of bosonic subsystems dual to subsystems with certain sition 1, which relates the bosonic and the fermionic entan- shapes on the fermionic side, seen in Sec. III F. This phe- glement spectrum of a chosen subsystem in a supersymmetric nomenon is related to the appearance of almost maximally Gaussian state. The result is based on supersymmetric iden- entangled modes in the fermionic subsystem, for which the tification maps that are constructed from the supercharge op- spectrum of the fermionic linear complex structure nearly van- erator Qˆ. They enables us to uniquely identify subsystems ishes. The entanglement duality then implies an unbounded both bosonic and fermionic, which we refer to as dual to each growth of entanglement for the dual bosonic system. Since it other. In Proposition 2, we extend the said duality to include is well-known [61] that the entanglement entropy associated thermal states associated with supersymmetric Hamiltonians, to a ground state of a gapped and local Hamiltonian (bosonic for which we find the same relationship between the bosonic or fermionic) satisfies an area law, this raises the question and the fermionic thermal states as for the reduced states in of how occasions where our entanglement duality relates an the subsystems. area law on the fermionic side with a “super area law” on the The rest of the article illustrates this result and its implica- bosonic side for the respective ground states of a local super- tions in supersymmetric lattice models. In particular, we in- symmetric Hamiltonian (such as the honeycomb model con- vestigate to what extent identification maps constructed from sidered in III F) can appear. The answer to this question lies a local supercharge operator preserve this locality, i.e., to what in the type of bosonic subsystem that arises under the duality extent a local subsystem on the bosonic side is identified with for a fermionic subsystem with large entanglement. We saw a local subsystem on the fermionic side and vice versa. This is in Sec. III E how there can be an arbitrary amount of entan- important to explain why our abstract duality is of relevance glement associated to a single bosonic mode due to choosing when studying physical systems: it shows a simple relation a subsystem that effectively separates a quadrature operator qˆi between the entanglement of bosonic and fermionic subsys- from its canonically conjugate operator pˆi. Such type of sub- tems that can both be thought local in a precise way. systems are typically not considered in the context of studying area laws and it is not surprising that the standard results on The examples in this work suggest that, for SUSY lat- area laws for the ground states of gapped local Hamiltonians tice Hamiltonians, the locality properties of the identifica- do not apply to them. Hence, there can be situations as seen tion maps are related to the boundary conditions of the lat- in Sec. III F, in which the bosonic side of the entanglement tice and the presence of edge modes. For example, in the duality shows such “super area law”, but we want to empha- one-dimensional open chain of Sec. II D, the identification size that this is a peculiarity of the specific choice of geom- maps featured highly non-local behaviour in the topological etry and couplings (namely one exhibiting maximally entan- phase. However, in the same system with periodic boundary gled fermionic modes, thus separating canonically conjugate conditions (obtained by extending the supercharge (54) to be bosonic variables), but by no means the typical behavior. In translation-invariant) the identification maps behave rather lo- fact, we saw that most natural choices of local fermionic sub- cal, even deep in the topological phase. systems (and the standard choices of local bosonic subsys- In the context of localized subsystems, a particular pecu- tems) lead to area laws under the supersymmetric duality, as liar consequence of the entanglement duality is the appear- one would expect. 18

In keeping with the study of topological properties of tanglement properties hitherto unexplored; our dualities find a translation-invariant SUSY lattice Hamiltonians in arbitrary straightforward application therein. dimensions, a generalization of the identification maps to higher dimensional lattices certainly constitutes a promising avenue to explore. Here we would like to highlight [71] where a spin-fermion correspondence, very much in the same ACKNOWLEDGMENTS spirit of our SUSY map, has been worked out engaging three- dimensional lattice models as well. Entanglement properties RHJ gratefully acknowledges support by the Wenner Gren of a three-dimensional generalization of the Kitaev honey- Foundations. LH gratefully acknowledges financial support comb model have also been studied [72]. Other variants of by the Alexander von Humboldt Foundation. KR thanks the three-dimensional Kitaev spin liquids exist [73, 74] with en- sponsorship, in part, by the Swedish Research Council.

[1] Yuri A Gol’fand and Evgeny P Likhtman. Extension of the alge- 2018. bra of poincare´ group generators and violation of p invariance. [18] Jan Derezinski´ and Christian Gerard.´ Mathematics of Quanti- In Supergravities in diverse dimensions. Volume 1. 1989. zation and Quantum Fields. Cambridge Monographs on Math- [2] Pierre Ramond. Dual theory for free fermions. Physical Review ematical Physics. Cambridge University Press, 4 2013. D, 3(10):2415, 1971. [19] Joshua M. Deutsch. Quantum statistical mechanics in a closed [3] Andre´ Neveu and John H Schwarz. Factorizable dual model of system. Physical Review A, 43(4):2046–2049, February 1991. pions. Nuclear Physics B, 31(1):86–112, 1971. [20] M. Srednicki. Chaos and quantum thermalization. Physical [4] Nicolas Sourlas. Introduction to supersymmetry in condensed Review E, 50(2):888–901, August 1994. matter physics. Physica D: Nonlinear Phenomena, 15(1- [21] M. Rigol, V. Dunjko, and M. Olshanii. Thermalization and 2):115–122, 1985. its mechanism for generic isolated quantum systems. Nature, [5] Howard Baer and Xerxes Tata. Weak scale supersymmetry: 452:854, 2008. From superfields to scattering events. Cambridge University [22] Luca D’Alessio, Yariv Kafri, Anatoli Polkovnikov, and Mar- Press, 2006. cos Rigol. From quantum chaos and eigenstate thermaliza- [6] Fred Cooper and Barry Freedman. Aspects of supersymmetric tion to statistical mechanics and thermodynamics. Advances quantum mechanics. Annals of Physics, 146(2):262–288, 1983. in Physics, 65(3):239–362, 2016. [7] Fred Cooper, Avinash Khare, and Uday Sukhatme. Supersym- [23] Christian Gogolin and Jens Eisert. Equilibration, thermalisa- metry and quantum mechanics. Physics Reports, 251(5-6):267– tion, and the emergence of statistical mechanics in closed quan- 385, 1995. tum systems. Reports on Progress in Physics, 79(5):056001, [8] A Kirchberg, JD Lange,¨ PAG Pisani, and A Wipf. Algebraic 2016. solution of the supersymmetric hydrogen atom in d dimensions. [24] Joshua M Deutsch. Eigenstate thermalization hypothesis. Re- Annals of Physics, 303(2):359–388, 2003. ports on Progress in Physics, 81(8):082001, 2018. [9] Asim Gangopadhyaya, Jeffry V Mallow, and Constantin Rasi- [25] S. Goldstein, J. L. Lebowitz, R. Tumulka, and N. Zangh`ı. nariu. Supersymmetric quantum mechanics: An introduction. Canonical typicality. Physical Review Letters, 96(5):050403, World Scientific Publishing Company, 2017. February 2006. [10] Michael J Lawler. Supersymmetry protected topological phases [26] S. Popescu, A. J. Short, and A. Winter. Entanglement and the of isostatic lattices and kagome antiferromagnets. Physical Re- foundations of statistical mechanics. Nature Physics, 2:754, view B, 94(16):165101, 2016. 2006. [11] Jan Attig, Krishanu Roychowdhury, Michael J Lawler, and [27] H. Tasaki. From quantum dynamics to the canonical distribu- Simon Trebst. Topological mechanics from supersymmetry. tion: General picture and a rigorous example. Physical Review Physical Review Research, 1(3):032047, 2019. Letters, 80(7):1373–1376, February 1998. [12] Edward Witten. Constraints on supersymmetry breaking. Nu- [28] Anatoli Polkovnikov, Krishnendu Sengupta, Alessandro Silva, clear Physics B, 202(2):253–316, 1982. and Mukund Vengalattore. Colloquium: Nonequilibrium dy- [13] CL Kane and TC Lubensky. Topological boundary modes in namics of closed interacting quantum systems. Reviews of Mod- isostatic lattices. Nature Physics, 10(1):39–45, 2014. ern Physics, 83(3):863, 2011. [14] Xiang-Bin Wang, Tohya Hiroshima, Akihisa Tomita, and [29] Lev Vidmar, Lucas Hackl, Eugenio Bianchi, and Marcos Rigol. Masahito Hayashi. Quantum information with gaussian states. Entanglement entropy of eigenstates of quadratic fermionic Physics reports, 448(1-4):1–111, 2007. hamiltonians. Physical Review Letters, 119(2):020601, 2017. [15] Christian Weedbrook, Stefano Pirandola, Raul´ Garc´ıa-Patron,´ [30] Chunxiao Liu, Xiao Chen, and Leon Balents. Quantum entan- Nicolas J Cerf, Timothy C Ralph, Jeffrey H Shapiro, and Seth glement of the Sachdev-Ye-Kitaev models. Physical Review B, Lloyd. Gaussian quantum information. Reviews of Modern 97(24):245126, 2018. Physics, 84(2):621, 2012. [31] Lev Vidmar, Lucas Hackl, Eugenio Bianchi, and Marcos Rigol. [16] Gerardo Adesso, Sammy Ragy, and Antony R Lee. Continu- Volume law and quantum criticality in the entanglement en- ous variable quantum information: Gaussian states and beyond. tropy of excited eigenstates of the quantum ising model. Phys- Open Systems & Information Dynamics, 21(01n02):1440001, ical Review Letters, 121(22):220602, 2018. 2014. [32] Lucas Hackl, Lev Vidmar, Marcos Rigol, and Eugenio Bianchi. [17] Tao Shi, Eugene Demler, and J Ignacio Cirac. Variational study Average eigenstate entanglement entropy of the xy chain in a of fermionic and bosonic systems with non-gaussian states: transverse field and its universality for translationally invariant Theory and applications. Annals of Physics, 390:245–302, quadratic fermionic models. Physical Review B, 99(7):075123, 19

2019. 2015. [33] Lev Vidmar and Marcos Rigol. Entanglement entropy of eigen- [55] R. D. Sorkin. On the entropy of the vacuum outside a hori- states of quantum chaotic hamiltonians. Physical Review Let- zon. In Tenth International Conference on General Relativity ters, 119(22):220603, 2017. and Gravitation (held in Padova, 4-9 July, 1983), Contributed [34] Eugenio Bianchi and Pietro Dona.` Typical entanglement en- Papers, volume 2, pages 734–736, 1983. tropy in the presence of a center: Page curve and its variance. [56] I. Peschel. Calculation of reduced density matrices from corre- Physical Review D, 100(10):105010, nov 2019. lation functions. J. Phys. A, 36(14):L205, 2003. [35] Patrycja Łydzba,˙ Marcos Rigol, and Lev Vidmar. Eigenstate [57] Ryszard Horodecki, Paweł Horodecki, Michał Horodecki, and entanglement entropy in random quadratic hamiltonians. Phys- Karol Horodecki. Quantum entanglement. Reviews of modern ical Review Letters, 125(18):180604, 2020. physics, 81(2):865, 2009. [36] Charles H. Bennett and Peter W. Shor. Quantum information [58] Nicolas Laflorencie. Quantum entanglement in condensed mat- theory. IEEE transactions on information theory, 44(6):2724– ter systems. Physics Reports, 646:1–59, 2016. 2742, 1998. [59] Ingemar Bengtsson and Karol Zyczkowski.˙ Geometry of quan- [37] Jens Eisert. Entanglement in quantum information theory. tum states: an introduction to quantum entanglement. Cam- arXiv preprint quant-ph/0610253, 2006. bridge university press, 2017. [38] Patrick Hayden, Debbie W. Leung, and Andreas Winter. As- [60] Alexei Kitaev. Anyons in an exactly solved model and beyond. pects of generic entanglement. Communications in Mathemat- Annals of Physics, 321(1):2–111, 2006. ical Physics, 265:95–117, July 2006. [61] Jens Eisert, Marcus Cramer, and Martin B Plenio. Colloquium: [39] Patrick Hayden and John Preskill. Black holes as mirrors: Area laws for the entanglement entropy. Reviews of Modern Quantum information in random subsystems. JHEP, 09:120, Physics, 82(1):277, 2010. 2007. [62] Lucas Hackl and Robert H Jonsson. Minimal energy cost of [40] Yasuhiro Sekino and Leonard Susskind. Fast Scramblers. entanglement extraction. Quantum, 3:165, 2019. JHEP, 10:065, 2008. [63] Lucas Hackl and Eugenio Bianchi. Bosonic and fermionic [41] Pavan Hosur, Xiao-Liang Qi, Daniel A. Roberts, and Beni gaussian states from kahler¨ structures. arXiv preprint Yoshida. Chaos in quantum channels. JHEP, 02:004, 2016. arXiv:2010.15518, 2020. [42] Daniel A. Roberts and Beni Yoshida. Chaos and complexity by [64] Bennet Windt, Alexander Jahn, Jens Eisert, and Lucas Hackl. design. JHEP, 04:121, 2017. Local optimization on pure gaussian state manifolds. arXiv [43] Yuya O. Nakagawa, Masataka Watanabe, Sho Sugiura, and preprint arXiv:2009.11884, 2020. Hiroyuki Fujita. Universality in volume-law entanglement [65] Roger Penrose and Wolfgang Rindler. Spinors and space-time: of scrambled pure quantum states. Nature Communications, Volume 1, Two-spinor calculus and relativistic fields, volume 1. 9(1):1635, 2018. Cambridge University Press, 1984. [44] Tsung-Cheng Lu and Tarun Grover. Renyi Entropy of Chaotic [66] Eugenio Bianchi, Lucas Hackl, and Nelson Yokomizo. Linear Eigenstates. Physical Review E, 99(3):032111, 2019. growth of the entanglement entropy and the kolmogorov-sinai [45] Hiroyuki Fujita, Yuya O. Nakagawa, Sho Sugiura, and Masa- rate. Journal of High Energy Physics, 2018(3):25, 2018. taka Watanabe. Page Curves for General Interacting Systems. [67] A Yu Kitaev. Unpaired majorana fermions in quantum wires. JHEP, 12:112, 2018. Physics-Uspekhi, 44(10S):131, 2001. [46] Alexei Kitaev and John Preskill. Topological entanglement en- [68] Bryan Gin-ge Chen, Nitin Upadhyaya, and Vincenzo Vitelli. tropy. Physical review letters, 96(11):110404, 2006. Nonlinear conduction via solitons in a topological mechanical [47] Michael Levin and Xiao-Gang Wen. Detecting topological or- insulator. Proceedings of the National Academy of Sciences, der in a ground state wave function. Physical review letters, 111(36):13004–13009, 2014. 96(11):110405, 2006. [69] Lucas Hackl, Tommaso Guaita, Tao Shi, Jutho Haegeman, Eu- [48] Shunsuke Furukawa and Gregoire´ Misguich. Topological en- gene A Demler, and J Ignacio Cirac. Geometry of varia- tanglement entropy in the quantum dimer model on the triangu- tional methods: dynamics of closed quantum systems. SciPost lar lattice. Physical Review B, 75(21):214407, 2007. Physics, 9(4), 2020. [49] Hong Yao and Xiao-Liang Qi. Entanglement entropy and entan- [70] Eugenio Bianchi, Lucas Hackl, and Nelson Yokomizo. Entan- glement spectrum of the kitaev model. Physical review letters, glement entropy of squeezed vacua on a lattice. Physical Re- 105(8):080501, 2010. view D, 92(8):085045, 2015. [50] Sergei V Isakov, Matthew B Hastings, and Roger G Melko. [71] Jan Attig and Simon Trebst. Classical spin spirals in frustrated Topological entanglement entropy of a bose–hubbard spin liq- magnets from free-fermion band topology. Physical Review B, uid. Nature Physics, 7(10):772–775, 2011. 96(8):085145, 2017. [51] Stefan Depenbrock, Ian P McCulloch, and Ulrich Schollwock.¨ [72] Ian Mondragon-Shem and Taylor L Hughes. Entanglement of Nature of the spin-liquid ground state of the s= 1/2 heisen- a 3d generalization of the kitaev model on the diamond lat- berg model on the kagome lattice. Physical review letters, tice. Journal of Statistical Mechanics: Theory and Experiment, 109(6):067201, 2012. 2014(10):P10022, 2014. [52] Hong-Chen Jiang, Zhenghan Wang, and Leon Balents. Iden- [73] Kevin O’Brien, Maria Hermanns, and Simon Trebst. Classifi- tifying topological order by entanglement entropy. Nature cation of gapless z 2 spin liquids in three-dimensional kitaev Physics, 8(12):902–905, 2012. models. Physical Review B, 93(8):085101, 2016. [53] Shou-Shu Gong, Wei Zhu, and DN Sheng. Emergent chiral spin [74] Tim Eschmann, Petr A Mishchenko, Kevin O’Brien, Troels A liquid: Fractional quantum hall effect in a kagome heisenberg Bojesen, Yasuyuki Kato, Maria Hermanns, Yukitoshi Mo- model. Scientific reports, 4(1):1–6, 2014. tome, and Simon Trebst. Thermodynamic classification of [54] Krishanu Roychowdhury, Subhro Bhattacharjee, and Frank three-dimensional kitaev spin liquids. Physical Review B, Pollmann. Z 2 topological liquid of hard-core bosons on a 102(7):075125, 2020. kagome lattice at 1/3 filling. Physical Review B, 92(7):075141,