Unconventional supersymmetric quantum mechanics in spin systems

1, 1 1, Amin Naseri, ∗ Yutao Hu, and Wenchen Luo † 1School of Physics and Electronics, Central South University, Changsha, P. R. China 410083 It is shown that the eigenproblem of any 2 × 2 matrix Hamiltonian with discrete eigenvalues is involved with a supersymmetric quantum mechanics. The energy dependence of the superal- gebra marks the disparity between the deduced supersymmetry and the standard supersymmetric quantum mechanics. The components of an eigenspinor are superpartners—up to a SU(2) trans- formation—which allows to derive two reduced eigenproblems diagonalizing the Hamiltonian in the spin subspace. As a result, each component carries all information encoded in the eigenspinor. We p also discuss the generalization of the formalism to a system of a single spin- 2 coupled with external fields. The unconventional supersymmetry can be regarded as an extension of the Fulton-Gouterman transformation, which can be established for a two-level system coupled with multi oscillators dis- playing a mirror symmetry. The transformation is exploited recently to solve Rabi-type models. Correspondingly, we illustrate how the supersymmetric formalism can solve spin-boson models with no need to appeal a symmetry of the model. Furthermore, a pattern of entanglement between the components of an eigenstate of a many-spin system can be unveiled by exploiting the supersymmet- ric quantum mechanics associated with single spins which also recasts the eigenstate as a matrix product state. Examples of many-spin models are presented and solved by utilizing the formalism.

I. INTRODUCTION sociated with single spins. This allows to reconstruct the eigenstate as a matrix product state (MPS) [2, 5]. Quantum entanglement is turned out to be a crucial An essential ingredient of our formalism is connected property in targeting relevant corners of the Hilbert space with the supersymmetric quantum mechanics (SUSY of a many-body system [1–5] rather than merely being a QM) [11–13]. Supersymmetry (SUSY) is conjectured in puzzle in foundations of quantum mechanics [6]. Entan- relativistic quantum field theory for unification of fun- glement is perceived as a correlation between constituents damental forces of nature. The prediction of superpart- of a composite system. Specifically, an entangled state ners of elementary particles, whose spins differ by a half- of a bipartite system is a quantum state which cannot integer, has not been observed experimentally. This im- be described as a product of states of subsystems [6]. A plies SUSY has to be spontaneously broken. SUSY was 1 initially introduced into non-relativistic quantum me- two-level system, i.e. a single spin- 2 , interacting with an external field provides a concrete example in this situa- chanics to study the non-perturbative breaking of SUSY tion. The setting is among the simplest quantum systems [14] and later became a framework on its own. In addition which models varieties of physical phenomena including to proposals for its realization in condensed matter and atom-field interactions and quantum optics [7, 8], meso- cold atom systems [15–20], SUSY QM is adopted in clas- scopic and molecular physics [9, 10], etc.. The eigenstates sifying analytically solvable potentials in quantum me- of such models are expected to describe an entanglement chanics in a close relation with the factorization method between the spin and the external field if no good spin [21]. quantum number exists. Let us briefly review SUSY QM and mention the asso- Our focus here is on models which can be partitioned ciated aspects which are of interest to our study. SUSY QM can be constructed by two supercharge operators arXiv:2012.00197v2 [quant-ph] 7 Feb 2021 into two parts and their eigenproblem (EP) is involved Q = F σ and a Hamiltonian H = Q+ + Q , where with a 2 2 matrix Hamiltonian. The aim of this paper is ± ± ± − × 2σ = σx iσy with σx,y,z being Pauli matrices and to show that there exist redundancies in degrees of free- ± ± dom (DOF) required to construct the eigenspinor of the F = F † are some external fields. The supercharges sat- system if the state is entangled. The entanglement also isfy± Q2 ∓= 0. The EP of the Hamiltonian is H φ = E φ ± | i | i allows to represent a two-component eigenspinor of the where system by means of the only one of the components. We   p φ generalize the formalism to a single spin- for p > 1 cou- φ = | +i (1) 2 | i φ pled with external fields. In these setups also the entan- | −i glement reduces DOF required to describe the eigenstate is an eigenspinor and E is the associated eigenenergy. of the system. One step further, we show a pattern of The square of the Hamiltonian H2 has a diagonal form entanglement can be revealed in an eigenstate of a many- and together with the supercharges close the superalge- spin system by taking into account the redundancies as- bra

2  2 H = Q+,Q , Q , H = 0, (2) { −} ± ∗ [email protected] where , is an anticommutator and [, ] is a commutator. † [email protected] H2 has{ a} good supersymmetry if its ground states φ | 0i 2 with E = 0 are vacuums of all supercharges Q φ0 = 0. results are as follows. In Sec. II, we introduce USUSY φ can be written as a product of an eigenspinor±| i of QM at a generic ground. In Sec. II A, we present the | 0i σz and the vacuum of either F+ or F , and hence, is USUSY QM pertaining to the EP of a generic 2 2 ma- a disentangled state. The components of− an eigenspinor trix Hamiltonian. In Sec. II B, we demonstrate that× the with E = 0 are superpartners F φ = E φ and by eigenproblem of a Hamiltonian describing interactions 6 ±| ∓i | ±i p knowing one of the components, the other can be found. between some fields and a single spin- 2 for p > 1 is If F are non-hermitian operators, then also endowed with a USUSY QM. In Sec. II C, we de- ± vise USUSY QM associated with single spins to reduce  1    E F+ φ φ+ DOF of a generic interacting many-spin system. φ = | −i = 1 | i (3) | i φ F φ+ | −i E −| i Examples are presented in Sec. III and Sec. IV to illus- trate how USUSY QM can be exploited in dealing with is clearly an entangled state since it cannot be written as the EP of spin systems. We set the direct exact diagonal- a product of an eigenspinor of σ and a state associated z ization (ED) as a benchmark. By direct ED we mean the with the fields F . The intertwining relationship between ± diagonalization of a Hamiltonian represented in a basis components leads to two EPs F F φ = E2 φ . ± ∓ ± ± constructed out of disentangled states. A single spin- Solving either of the EPs solves the EP| ofi H. We| gen-i 1 in a uniform magnetic field is the simplest Hamilto- eralize these considerations to a generic 2 2 matrix 2 nian endowed with an USUSY QM which is considered Hamiltonian and show these bipartite systems× are en- in Sec. III A. In Sec. III B, we solve the Jaynes-Cummings dowed with a SUSY QM. The associated superalgebra (JC) model [25, 26]. The treatment of Rabi-type models is defined in terms of energy-dependent operators, and [27] are presented in Sec. III C. In Sec. IV A, we start to hence, is dubbed unconventional supersymmetric quan- analyze systems including more than one spin by solving tum mechanics (USUSY QM) in order to distinguish it a model of interacting spins and mention how the formal- from the standard SUSY QM. The existence of super- ism can be engaged iteratively if the system is extended charges of the USUSY QM makes it possible to solve the by adding more spins. In Sec. IV B and Sec. IV C, we an- EP of the full system via solving one of the EPs asso- alyze the eigensolutions of Tavis-Cummings (TC) model ciated with its subsystems. That is a subsystem of the [28] in a MPS representation and solve generalized Dicke system contains all the information of the system due to models [29, 30], respectively, which describe many spins the correlation between DOF. interacting with a single boson mode. The formalism presented herein can also be regarded as an extension to the Fulton-Gouterman (FG) transfor- mation [22]. A two-level system interacting with multi oscillators can be diagonalized by the FG transforma- II. UNCONVENTIONAL SUPERSYMMETRIC tion in the spin subspace if the Hamiltonian possesses QUANTUM MECHANICS a mirror symmetry. Particularly in the context of spin- boson models, different generalizations of the Rabi model In this section, we construct the USUSY QM for the are solved by exploiting the FG transformation inspired EP of a 2 2 matrix Hamiltonian. Next, the scheme is by the approach introduced in the solution of the Rabi generalized× to a model describing interactions between a model [23]. The USUSY formalism relaxes the symmetry single spin- p with p > 1 and external fields. Finally, we constraint of the FG transformation. 2 discuss an interacting system of many spin- 1 and recon- The FG transformation is generalized to a multi-level 2 struct its eigenstates by means of USUSY QM associated system interacting with a multi-oscillatory system also with the single spins. restricted to Hamiltonians with a discrete symmetry [24]. The generalization of the FG transformation can also be covered by the USUSY formalism irrespective of the sym- 1 metry of the model. This means a Hamiltonian describ- A. Single Spin- 2 p ing interactions between a single spin- 2 , i.e. a (p+1)-level system, and external fields can be diagonalized in the We start from a generic 2 2 matrix Hamiltonian and spin subspace due to redundancies associated with the the associated EP × USUSY QM independent of symmetries of the Hamilto- nian. The EP can be then reduced to p+1 decoupled EPs. H F  H = + + ,H Ψ = E Ψ , (4) Moreover, solving one of p+1 EPs solves the full EP. This F H | i | i connects also our work to the parasupersymmetric quan- − − tum mechanics [11, 13] where supercharge operators are where H = H† , F+ = F † and the eigenstate is a two- generalized to parasupersymmteric charges in a diagonal ± ± − Hamiltonian associated with a single spin- p . However, component spinor Ψ = ( ψ+ ψ ). H are restricted 2 h | h | h −| (n) ± our approach here is to reduce DOF of aforementioned to those with discrete eigenvalues E , collected in sets systems by means of constraints associated with USUSY . If E/ , the components are certainly± connected E± ∈ E± QM. 1 The structure of the paper and a brief summary of our ψ = G F ψ ,G = (E H )− , (5) | ±i ± ±| ∓i ± − ± 3 which allow to construct the eigenstate from one of the induces new diagonal terms which can have different components sets of eigenenergies ˜ = . As far as H˜ have dis- E± 6 E± ±     crete eigenenergies, the singularity can be removed as U G+F+ ψ ψ+ Ψ = | −i = | i . (6) changes the eigenenergies of the diagonal terms continu- | i ψ G F ψ+ | −i − −| i ously. Consequently, the singular points are isolated and the USUSY QM can be realized for any eigenstate of H The EP of H is therefore decoupled into two EPs if U warrants E/ ˜ . ∈ E± h ψ = E ψ , h = H + F G F , (7) In retrospect, assume we solve h+ in order to find ±| ±i | ±i ± ± ± ∓ ∓ eigensolutions of H. Those eigenenergies of H which are which show components belong to subsystems governed in might be missing. That is E is a necessary by h and either carries all information encoded in Ψ . conditionE− for having a missing solution∈ E of− H though it is h do± not represent physical systems as they have eigen-| i ± not a sufficient condition. We note also that the excluded states which are associated with more than one eigenen- solutions are those states with no USUSY QM. If ergy. The induced potentials F G F compensate the ± ∓ ∓ decoupling of the components and can turn h into a (m) ± H Ψ = E Ψ , m ψ = 0, (14) non-local differential equation in problems having a po- | i − | i h | −i 6 sition space representation where H m = E(m) m , are satisfied, then solutions −| i (m)− | i r F G F r0 = r F G F r δ(r r0), (8) of h+ exclude E . In fact, Eq. (14) sets a constraint h | ± ∓ ∓| i 6 h | ± ∓ ∓| i − − on parameters of the model if E(m) is an eigenenergy of where δ(r r0) is the Dirac delta function. The EPs and − h in Eq.− (7) are addressed as reduced EPs and reduced H. A class of excluded solutions, which can be readily Hamiltonians,± respectively, in the following. Equiva- figured out, are those states satisfying lently, the original EP can be presented with a diagonal F+ 0 F = 0,H 0 F = E 0 F , (15) matrix | i −| i | i       h+ 0 ψ+ ψ+ that is the vacuum of F+ is an eigenstate of H . The | i = E | i . (9) − 0 h ψ ψ eigenstate can be chosen as Ψ = 0 F where σz = − | −i | −i . The spin and the field| i are|−i| disentangledi in|±i these ±|±i The fact that ψ are degenerate under the action of the states clearly and the states are an eigenspinor of σz. | ±i hermitian operators h implies a SUSY QM. To make it However, not all the excluded solutions are disentan- ± explicit, we define two supercharge operators gled states and not all disentangled eigensolutions are excluded. Excluded entangled eigenstates require a fine- = G F σ , (10) Q± ± ± ± tuning of parameters of the model and need to be looked which fulfil 2 = 0 independent of the nature of external for in models with no parametric analytical solutions. fields coupledQ± with the spin. The EP of H can now be In view of a perturbation theory, if off-diagonal terms of H are considered as a perturbation to the diagonal recast as Ψ = Ψ by a traceless matrix = + + which actsH| likei an| idempotenti operator H2 Ψ Q= QΨ−. terms, the excluded solutions of H are non-perturbative and 2 satisfy a superalgebra H | i H| i solutions as the sum of all non-zero order corrections has Q± H to vanish. In some problems, the states without a USUSY 2  2 +, = , , = 0. (11) QM can be shown to be among states with isolated exact {Q Q−} H Q± H solution, e.g. the Rabi model in the single-mode spin- The SUSY QM is characterized by energy-dependent op- boson representation [35]. Another approach to check erators, and therefore, it is dubbed USUSY QM in order whether there exists any missing solution is to compare to distinguish it from the standard SUSY QM. The exis- the solutions of different h˜ corresponding to some dif- tence of the USUSY QM, like a symmetry, reduces DOF + ferent unitary transformations of H in Eq. (12) which is of the problem as only one of the reduced EPs is needed practical if solutions within a finite range of energy are of to be dealt. interest. If h is chosen to be solved instead, the above The energy-dependent operators can lead to a break- − consideration can be applied as well by a swap in the down of the USUSY QM, which contrasts the difference notation + . between the standard SUSY QM and USUSY QM. This The existence↔ − of USUSY QM is a strong condition for can happen for a given eigenstate with E or E . + reducibility of an EP and it can be relegated to the exis- However, a unitary transformation of the∈ Hamiltonian E ∈ E− tence of either Q as it supports the existence of either ±  ˜ ˜  h , correspondingly. Specifically, one needs to transform H+ F+ ∓ H˜ = U †HU = ,U SU(2) (12) H to find a representation in which one of the intertwin- F˜ H˜ ∈ − − ing relationships ψ˜ = G˜ F˜ ψ˜ can be constructed | ±i ± ±| ∓i leading also to for those eigenstates of H lying within a given energy range of interest. In other words, the EP can be re-  ˜  ψ+ 1 duced if the components are maximally correlated at Ψ˜ = U † Ψ = | i , G˜ = (E H˜ )− , (13) | i | i ψ˜ ± − ± least through one way, e.g. ψ+ = G+F+ ψ , then | −i | i | −i 4 knowing ψ uniquely establishes ψ+ . If one of the to reduce DOF in a many-spin system. We illustrate | −i | i 1 intertwining relationships exists, we address it as a re- it for the case of a system of many spin- 2 and show ducible EP and if none of them exist, it is regarded as a MPS [2, 5] emerges automatically after removing the an irreducible EP. The components are considered to be redundancies. Furthermore, a pattern of entanglement superpartners if both intertwining relationships exist. also appears through reconstructing the many-spin eigen- state by means of the intertwining relationship of single spins. A Hamiltonian describing interactions between p B. Single Spin- 2 many spins and external fields reads

p N   We generalize the formalism to deal with a spin- 2 for (N) X (k) (k) (k) (k) (k) (k) H = H0 + ∆ σz + F+ σ+ + F σ , p > 1. A Hamiltonian describing interactions between a − − single spin- p and external fields has the following form k=1 2 (23) (k) H p = H0 + ∆Sz + F+S+ + F S , (16) where here also H0 does not include σx,y,z for k = 2 − − 1, ,N. The nature of fields coupled to the spins does ··· where H0 does not include spin operators. The fields not affect our results. For instance, a one-dimensional p satisfy F+ = F † and ∆ = ∆†. Sz and S are spin- chain of N + 1 spins can also be represented in a similar ± 2 generators of SU−(2) algebra [11, 13] form as H (N) if the fields are replaced by p (S ) = j(p j + 1)δ , (17) (k) (k+1) (k) (k) (k+1) + jk j+1,k F+ = σ = (F )† and ∆ = σz . (24) − − − Sz = diag (p, p 2, , p + 2, p) , (18) − ··· − − Spin models on a D-dimensional lattice can also be rep- for 1 6 j, k 6 p + 1, which satisfy resented similarly given k is replaced by a proper vector. In order to exercise USUSY QM of a single spin, H (N) [S+,S ] = Sz, [Sz,S ] = 2S . (19) (1) − ± ± ± can be represented, for instance, in eigenspinors of σz The EP is H p Ψ = E Ψ where E is the associated 2 (N 1) (1) ! | i | i H − F eigenenergy and Ψ has p + 1 components. The scheme H (N) = + + , (25) | i (1) (N 1) of making the Hamiltonian traceless by inverting the di- F H − − − agonal terms in the EP p Ψ = Ψ and then squaring 2 (N 1) 2 H | i | i − (N 1) (1) it p Ψ = Ψ results as well here in decoupling (al- where H = H − ∆ and H 2 | i | i ± ± most) half of the components of the eigenspinor from the N rest. This is because p can be written as a sum of two (N 1) X (k) (k) (k) (k) (k) (k)
H 2 H − = H0 + ∆ σz + F+ σ+ + F σ . supercharges − − k=2

( p )  ( p )2 (N) 2 2 An eigenspinor of H can be grouped into two compo- = I (S+ + S ) , with = 0, (20) ± − (N) (N 1) (N 1) Q± K Q± nents Ψ = ( ψ+ − ψ − ). The same procedure h | h | h − | (N 1) 1 − where I and are diagonal matrices with entries adopted for a single spin- 2 decouples ψ in the cor- ± K responding EP associated with the eigenenergy| ± i E and

(I+)jj = δj,2l+1, (I )jj = δj,2l, (21) renders two reduced Hamiltonians − (N 1) (N 1) (1) (N 1)
1 (1) given l is an integer number, and h − = H − + F E H − − F (26) ± ± ± − ∓ ∓ 1 if Ψ(N) is reducible. That is by solving an (N 1)- ( )jj = [E H0 ∆(p 2j + 2)]− . (22) K − − − spin| subsystem,i the system of N spins is solved.− It is We note that the supercharges in Eq. (20) reduce to the worthwhile to mention that in the case of N = 2, the 1 same procedure adopted for a single spin- p for p > 1 spin- 2 supercharges in Eq. (10) for p = 1. The scheme 2 can be repeated over and over on the components which works as well here in decoupling four components of the are not decoupled in order to achieve ultimately p+1 de- eigenspinor and achieving four reduced EPs, but it fails coupled EPs (an example is worked out in App. D). Once for N > 2. again, a unitary transformation can turn an irreducible Using the relation in Eq. (6), Ψ(N) can be constructed | i (N 1) eigenstate to a reducible one by lifting the degeneracy out of one of the components, for example ψ − . (N 1) | − i between H p and diagonal terms. ψ − can be expanded in terms of eigenstates of 2 | (−N 1) i (N 1) H+ − or H − since both Hamiltonians contain N 1 spins and provide− a proper basis. Suppose we choose− (N 1) 1 − C. Many spin- 2 the eigenstates of H+ as a basis. By representing (N 1) (2) H+ − in the eigenspinor of σz , once again the rela- It is possible to explore how the intertwining relation- tion in Eq. (6) can be used to reconstruct the reducible (N 1) ship between components of single spins can be exploited eigenstates of H+ − in terms of one of the components. 5

(N 3) H 3 where (N 2) H 2 (N `) ! (N 3) − (N 1) H Φj N `+1 H | − i (N 3) jN `+1 = (N `) . (30) − Θ − (N 2) H | i j | N `+1 i H (N 3) − H + H (N) The eigenstates are explicitly reconstructed by the rela- (N 3) tion in Eq. (6) as (N 2) H H (N 3) H + (N 1) (N `) ! H + Θ − (N 3) jN `+1 H + j = g | (N −`) i (31) (N 2) N `+1 jN `+1 − H +2 | − i − Θj (N 3) N `+1 H +3 | − i with

(N `) FIG. 1. The possible paths for decomposing the eigenstates − gj = diag( j F+, 1), (32) (N) (k) (k) N `+1 G N `+1 of H where F± = F± and ∆ = ∆ are assumed for sim- − − (N `) (N `) 1 (N) (1) − = (E H − `∆)− . (33) plicity. By representing H in eigenspinors of σz , eigen- j jN `+1 N `+1 − (N 1) G − − − states of H − +∆, for instance, can be used to represent an eigenstate of H (N) through the relation in Eq. (6). The same In Eq. (27), ςj is defined as (N 1) | i procedure can be carried out by representing H − + ∆ in (2) N 1 the eigenspinors of σz and expanding its eigenstates in terms Y− (N 2) of eigenstates of H − +2∆. The recursive procedure can be ςj = I` 1 gj j0 1 , (34) | i − ⊗ N ` | i| i followed until |Ψ(N)i finds an expansion in terms of eigenstate `=1 − of H (0) + N∆. The arrows show the path which is adopted ` ` for decomposing |Ψ(N)i into eigenstates of subsystems. where I` is a 2 2 identity matrix, j0 is an eigenstate (0) × | i of H+ = H0 + N∆, and

(N 1) The irreducible eigenstates of H − with a non-zero pro- N + O (k) (N 1) 1 = ( + + ) , where σ = . jection on ψ − are stuck and cannot be connected | i | ik |−ik z |±ik ±|±ik to smaller| spin− subsystems.i Similar to the single-spin k=2 case, a unitary transformation can alter the set of irre- The expansion in Eq. (27) can be understood in this ducible states and remove them from the representation. way that subsystems of the N-spin system provide in- The same procedure can be repeated N times to exploit trinsically a correlated orthonormal basis to represent the USUSY QM of all spins which attributes a fractal struc- N-spin eigenstate. The coefficients of the expansion in ture to Ψ(N) as presented in the following. We assume | i (k) Eq. (29) are specified by two indices which indicates they the spins are coupled with an identical field F = F (N `) − (k) ± ± are entries of some matrices Ω . Therefore in other and ∆ = ∆ as this simplification does not change the words, an N-spin eigenstate is represented as a MPS pattern of entanglement which we are interested to re- (N) by removing the redundancies associated with the sin- veal. According to the hierarchical structure of H , gle spins. No approximation is imposed in the derivation there is no unique path for decomposing Ψ(N) in terms | i of the representation. So the problem after removing the of eigenstates of the subsystems. We pursue the path redundancies is identical to the original problem modulo indicated in Fig. 1 by arrows. local transformations to remove irreducible eigenstates. If no irreducible state comes up in the course of repeat- It is shown in [36] that every state can be represented in ing the scheme down to the level where USUSY QM of all (N 1) a MPS representation. Our result is consistent with this spins are exploited, ψ − can be represented formally | − i statement in the case of eigenstates (opposed to an ar- as bitrary state). However, the MPS emerges here through N 1 − exploiting the correlations provided by the Hamiltonian (N 1) X Y jN ` ψ − = Ωj Ω − ςj , (27) N 1 jN ` 1 rather than applying Schmidt decomposition [36]. | − i − − − | i (N `) j `=1 If the dimension of the matrices Ω − is assumed to be . χ for instance by a truncation, the number of co- where j = j0, , jN 1 and jN ` is a collective quan- 2 { ··· − } − efficients to fix the N-spin eigenstate is χ N which tum number labelling the eigenstate jN ` and the as- ∝ × (N `)| − i grows linearly rather than exponentially with number of sociated eigenenergy E of H − + `∆. All quan- jN ` − spins. The correlated basis should be contrasted with a tum numbers designated by jN ` for 1 6 ` 6 N are − basis constructed out of disentangled states with a di- summed up in Eq. (27). The coefficient of the expan- mension of 2N M to diagonalize H (N), where M is the sion are grouped into tensors, i.e. one vector and N 1 × − number of states required to represent the fields coupled matrices, with elements to spins. If the correlated basis is used to diagonalize (N 1) (N 1) Ω = j ψ − , (28) the reduced Hamiltonian h − , a considerably smaller jN 1 N 1 − h − | − i (truncated) basis is required− to reproduce the results of jN `+1 (N `) − − Ωj = jN ` Θj , (29) a direct ED. This is plausible since correlations between N ` − N `+1 − h | − i 6

DOF are partially accumulated in the eigenbases of the indicates eigenstates of h are the eigenstates of the num- ± subsystems. Furthermore, the expansion in Eq. (27) re- ber operator a†a n = n n . The common eigenenergies veals the pattern of correlations between components of of h can be found| i by solving| i an eigenspinor of H (N), and can be used to construct a ± 2 relevant ansatz. A similar representation can be estab- (E n 1 + ∆)(E n ∆) = α (n + 1), n N, (38) p − − − − ∈ lished for a system of many spin- 2 due to the USUSY QM associated with the single spins. which gives two solutions for a given n. Parity quantum numbers η = are required to label the eigenenergies ± r 1 1 III. SINGLE-SPIN EXAMPLES E(n,η) = n + + η (∆ )2 + α2(n + 1) (39) 2 − 2 In this section, we solve some single-spin models by and the corresponding non-normalized eigenstate their reduced EPs in order to illustrate the formalism. First, we solve the EP of a spin in a uniform magnetic (n,η) !   (n,η) ψ+ n field. Next, JC model and Rabi-type models are studied Ψ = | (n,η)i = α | i . (40) | i ψ E(n,η) n 1+∆ n + 1 which are involved with boson creation (annihilation) op- | − i − − | i erator a† (a) with [a, a†] = 1. The unit of energy is set The full eigenstate can be constructed either by an eigen- equal to unity ~ω = 1 in all examples herein. state of h+ or the one of h since the components are superpartners. The only eigenstate− with no USUSY QM is the disentangled state Ψd = (0 0 ) with Ed = ∆ A. A single spin- 1 in a uniform magnetic field h | h | − 2 which is the vacuum of +, but does not exist for this state, and the stateQ has no parityQ− number. There- We begin by considering an elementary example of a fore, all the eigensolutions of JC model can be found by 1 single spin- interacting with a constant magnetic field solving h . Although HJC and H+ can be tuned to have 2 − B0 = (B1,B2,B3) which is described by the EP common eigenenergies, no eigenenergy of HJC is missed by solving h for α = 0. This is because − 6 B0 σ σ = E σ , (35) · | i | i H Ψ = (m + ∆) Ψ , m ψ = 0 (41) JC | i | i h | +i 6 where σ = (σx, σy, σz) and σ = (c+∗ c∗ ) with c C. h | − ± ∈ for a given m cannot be satisfied. It is notable that the As far as B1,2 = 0, either of the reduced EPs in Eq. (7) gives both eigenvalues6 normalized spin texture
 B2 + B2  Ψ(n,η)(r) † σ Ψ(n,η)(r) B + 1 2 c = Ec (n,η) x 3 ρ¯x (r) = (n,η) 2 (42) ± E B3 ± ± Ψ (r) q∓ | | 2 2 2 E = B1 + B2 + B3 . (36) (n,η) (n,η) ⇒ ± ± with Ψ (r) = r Ψ has a topological index [31, 32] defined by h | i Non-normalized eigenspinors then can be constructed by Eq. (6) as (n,η) (n,η) lim ρ¯x ( r )¯ρx ( r ) 1, (43) r →∞ −| | | | → −  B1 iB2  − E B3 d σ = ±− , (37) which is absent in Ψ (r). This implies the non-local as- | ±i 1 pect of the entanglement in the JC model. which are orthogonal σ σ+ = 0. h −| i C. Rabi-type models

B. Jaynes-Cummings model If H and F in Eq. (4) do not close an algebra, the reduced± EP can± be expanded in a given basis and the The reduced EPs in Eq. (7) can be algebraically solved eigenenergies are the roots of the determinant of the ma- if H and F close under commutation relations. The trix. If the Hilbert space is not finite, a truncated basis celebrated± JC± model [25, 26] is a suitable example to can be used to achieve low-energy eigensolutions which demonstrate it. The JC model describes interaction be- is equivalent to the approach of ED routine. This is par- tween a single spin and a single boson mode. The Hamil- ticularly proper Rabi-type models. tonian in Eq. (4) becomes the JC Hamiltonian HJC by We consider a generalization of the Rabi model [27] in choosing H = a†a ∆ and F+ = αa with ∆ and α the presence of a homogeneous in-plane magnetic field, ± ± being real numbers. The conserved operator of the JC specified by H = a†a ∆ and F+ = αa + βa† + γ 1 ± ± Hamiltonian C = a†a + σz induces [h , a†a] = 0 which in Eq. (4), in order to illustrate solving a spin-boson 2 ± 7

1.01 44 eigensolution would be excluded from the solutions of h˜(q)(E) = 0. We consider the possibility of having ex- 0.5 0.5 33 cluded solutions from different aspects. First, we check 7 ) 10

n ˜ (

/ for which ϑ the full Hamiltonian H and H can have a

) 2

0.0 ˜ 2 − 0.0 E E common eigenvalue within a given range of energy while ( ) q E, ( + 1 parameters α, β, γ and ∆ are fixed numerically. Low-

˜ 1 h 0-0.5.5 lying eigenenergies of H˜ together with the eigenenergies 00 of H are plotted versus−ϑ in Fig. 2 which shows there is

-1.0 no degeneracy between H and H˜ , and hence, no missing 1 0 1 2 3 4 0 1 2 3 4 5 6 0 1 2 3 4 0.0 1.0 2.0 3.0 solution within the range of energy,− for a transformation E # p with ϑ = 2 arctan β/α with α = 0.7 and β = 0.3. It FIG. 2. (color online) Generalized Rabi model’s results for is notable that at least for the ground state, minimizing (0) α = 0.7, β = 0.3, ∆ = 0.3 and γ = 0.2eiπ/3. Left: h˜(q)(E) E˜ with respect to ϑ provides a better approximation versus E is plotted for q = 12 and the transformation is in− comparison with a zero-order perturbation theory on p fixed by tan (ϑ/2) = β/α. Right: The horizontal lines are H, i.e. ϑ = 0, while taking the off-diagonal terms as a eigenenergies of the generalized Rabi model found by a direct perturbation to the diagonal terms. ˜(n) ED. The curvy lines are E− for n = 0, 1, ··· , 9 which are a On the other hand, it is possible to find out the inter- function of ϑ. At the crossing points, USUSY QM might not relationship between parameters of the model by diago- exist for the corresponding eigenstate of the generalized Rabi nalizing H˜ Ψ˜ = E˜(n) Ψ˜ , which provides the necessary model. The vertical dashed line is the angle associated with | i − | i α = 0.7, β = 0.3. condition for a missing solution. The results for some n are depicted in Fig. 3. Finally, exact isolated solu- tions [27, 34, 35] also can be constructed for the eigen- model without a closed algebra. Without loss of gen- states without USUSY QM. In this class of solutions the ˜ erality, we suppose α, β 0 while γ . For γ = 0, expansion of ψ in terms of the eigenstates of either > C | ±i the reduced Hamiltonians posses a parity∈ symmetry as H˜ have a finite number of terms and the eigenenergy ± (n) [h , eia†aπ] = 0 and the intertwining relationship implies is fixed to either E = E˜ , correspondingly, for a given ± ± two components have different parities ψ+ ψ = 0. If n, see App. C. The exact solution needs to be sought the reduced Hamiltonian has a tridiagonalh matrix| −i repre- self-consistently by deriving a set of equations which con- sentation, an analytical progress can be made as the de- strain the parameters of the model. terminant of the matrix representation can be achieved Braak derived the spectrum of the Rabi model by a three-term recurrence equation [33]. The reduced   a†a + ∆ g(a + a†) Hamiltonians of the generalized Rabi model are tridi- HR = (47) g(a + a†) a†a ∆ agonal if at least one of α, β, γ is zero. Otherwise, a − transformation by virtue of the Bargmann-Fock space [23] (some mod- 1 els solved by exploiting the same formalism are reviewed   2 iσ ϑ ϑ β ± U U(ϑ) = e y 2 with tan = (44) in [27]). The regular eigenenergies (contrasted with iso- ≡ 2 α lated exact solutions including Judd points [34]) are zeros of a transcendental function i.e. the so-called G-function. makes the representation of either h˜ tridiagonal, see Each summand in the definition of the G-function is asso- ± App. B. We proceed with h˜ and assume n˜ are the ciated with a sequence which needs to be evaluated sepa- + | i eigenstates of H˜ associated with the eigenenergy E˜(n) rately through a three-term recurrence relation which is − p − after rotating H with tan (ϑ/2) = β/α. Making the involved with the tridiagonality of the Rabi model. Find- matrix out of elements ing roots of the G-function also requires a truncation sim- ilar to the ED treatment applied here. Z2 symmetry of (m,n) the Rabi model is invoked through the FG transforma- f˜ = m˜ (h˜+ E) n˜ , (45) h | − | i tion [22, 24, 37] to reduce DOF in Braak’s approach. the j-th continuant h˜(j)(E) obeys a recurrence relation The same results can be reproduced by employing the [33] USUSY QM of the Rabi model as is presented briefly in the following. The reduced Hamiltonian of the Rabi (j+1) (j,j) (j) (j,j 1) 2 (j 1) h˜ (E) = f˜ h˜ (E) f˜ − h˜ − (E) (46) Hamiltonian − | | g2 ˜( 1) ˜(0) h = a†a + ∆ + (a + a†) (a + a†) (48) with initial values h − (E) = 0 and h (E) = 1. Given R E a a + ∆ the desired convergence is achieved by keeping q + 1 − † ˜ in the associated reduced EP h ψ = E ψ needs to states, the eigenvalues of h+ are the roots of the char- R| Ri | Ri acteristic polynomial h˜(q)(E) = 0, see Fig. 2. be solved and then the eigenstate of HR would be (m) ˜   As discussed earlier, if E = E˜ and m˜ ψ = 0, ψR ˜ ˜ ˜ − h ˜| −i 6 ΨR = |
i1 . (49) where Ψ = ( ψ+ ψ ) is an eigenstate of H, then the | i g E a†a + ∆ − (a + a†) ψR h | h | h −| − | i 1.0 1.0 1.0 8 1.0 1.0 0.5 1 in order to exemplify how the maximal information can 0.5 0.5 n=0 be extracted from the reduced EPs. Next, we explain 0.5 0.5 n=0 how the achieved eigensolutions can be exploited to solve 0.5 n=10 n=10 the reduced EPs of a three-spin system. The procedure 0.0
n=1 Out[ ]= 0.0 n=2 can be iteratively carried on to include more spins one Out[
]= 0.00.0 n=21 0.0
by one and construct larger systems. A remark on the Out[ ]= 0.0 n=32 Out[
]= n=32 notation is that the superscript of the Hamiltonians and -0.5 n=43 -0.5 0.5 n=43 corresponding eigenvalues/eigenstates denotes how many -0.5 = - n 4 spins are included in them while the superscript of the 0.5 n=4 (2) -0.5 spin operators labels the single spin. Representing H -1.01 (2) -1.01 -0.50.5 0.00.0 0.50.5 1.01 in eigenspinors of σz -1.0 ⇣ -1.0 -0.5 0.0 0.5 (1)! -1.0 -0.5 0.0 0.5 0 σ -1.0 - (2) FIG. 3.0.5 (color online)0.0 Generalized0.5 Rabi model’s results. The H = (1) − , (51) - relation0.5 between ∆0.0 and ζ (defined0.5 by α = ζ cosh τ and β = σ+ 0 ζ sinh τ with tanh τ = 3/7) is given if H˜ |Ψ˜ i = E˜(n)|Ψ˜ i is − the associated reduced Hamiltonians are satisfied for n = 0, 1, 2, 3, and γ = 0.2eiπ/3 within a direct ED treatment. 1 h(1) = σ(1)σ(1). (52) ± E(2) ∓ ±

(2) In order to exploit the Z2 symmetry of HR, the reduced The square of H has a standard SUSY QM, however, EP needs to be expanded either in the even-parity basis we follow to analyze the associated USUSY QM. H(2) ne = 2n or the odd-parity basis no = 2n + 1 while has four eigensolutions: three triplet states n| i .| Thei matrix representation| of thei reduced| i EP is N (2) (2) tridiagonal∈ in either parity basis and its determinant can Ψ = , Ψ = ++ , | 1 i | −−i | 2 i | i be achieved by a three-term recursive equation similar to 1 Ψ(2) = ( + + + ) (53) the one in Eq. (46). Once again a truncation of the basis | 3 i √2 | −i | − i is required in order to achieve a characteristic equation (2) (2) whose roots are the eigenenergies of the Rabi model. The with energies E1,2 = 0, E3 = 1 and one singlet state errors then stems from the truncated basis and can be reduced by increasing the size of the basis. The possible (2) 1 Ψ = ( + + ) (54) excluded solutions can be searched for through examining | 4 i √2 | −i− | − i HR Ψ = (n ∆) Ψ . | i − | i (2) Other spin-boson models can also be treated in the with energy E4 = 1, where the local basis is chosen as same way. Although the tridiagonality of the matrix (k) − (1) σz = . Therefore, h do not exist for states representation facilitates the construction of the deter- with|±i a zero±|±i energy, and hence,± two zero-energy states minant, it is not essential in solving the problem. The would be missing from the solutions of either reduced difficulty arises in finding the roots of the characteris- Hamiltonian. We note that zero-energy eigenstates are tic equation where the graphical root-finding approach is also disentangled states. If the two-spin Hamiltonian is adopted in Rabi-type models [27], similar to the plot in rotated by Fig. 2. iσ(2)θ/2 U U(θ) = e y , 0 < θ < π, (55) ≡ IV. MANY-SPIN EXAMPLES we find  ˜ (1) ˜  (2) (2) H F+ In this section, the USUSY QM of single spins are ex- H˜ = U †H U = − , (56) F˜ H˜ (1) ploited in some models with more than one spin. Exam- − ples include a model of spin-spin interactions, TC model and generalized Dicke models. Specially here we inter- where pret the solutions of the TC model from a MPS point of 1 H˜ (1) = sin θσ(1), (57) view. 2 x θ θ F˜ = σ(1) cos2 ( ) σ(1) sin2 ( ). (58) ± ∓ 2 − ± 2 A. A model of spin-spin interactions ˜(1) (2) ˜ Expanding the reduced EP (h+ E ) ψ+ = 0 in the ˜ (1) − | i We first solve the EP of two interacting spins eigenbasis of H and showing its matrix representation ˜(1) by M(θ), eigenenergies of h+ are zeros of det [M(θ)] (2) (1) (2) (2) (2) (2) (2) (2) H = σ+ σ + h.c., H Ψ = E Ψ (50) as a function of E . The eigenenergies of the reduced − | i | i 9

Hamiltonian cover now the full eigenenergies of H(2) if θ which gives three eigenenergies is not an integer multiple of π, see Fig. 4. The drawback   (3) (3) 1 p 2 of the transformation is that the reduced Hamiltonians E1 = τ, E2 = τ 8 + τ , h˜(1) do not have the conserved quantity of the initial − 2 − ± (3) 1  p  reduced Hamiltonians [h(1), σ(1)] = 0. Clearly, the good E = τ + 8 + τ 2 . (64) z 3 2 quantum number of the± full Hamiltonian is maintained (2) after the transformation There exists one state with zero-energy E4 = 0 which has m = 2. The characteristic equation for the eigen- (2) (1) (2) ˜ (2) (1) (2) z U †[H , σz +σz ]U = [H , σz +U †σz U] = 0. (59) (2) states of h+ with mz = 0 needs to be constructed by taking the determinant 2 2 2 ✓ =0 (1+τ) 1 τ 0 < ✓ 6 ⇡/2 1 E + 2E −2E 2 = 0, (65) − 1 τ 2 (1 τ) 1 − 1 E + − 1 2E − − 2E )] ✓

( which renders the same set of solutions as those in 0 (3) M 0 Eq. (64). At τ = 1 for instance, eigenenergies of H become 1, 0, 2, which are four-, two-, two-fold degener- det [ − -1 ate, respectively. Only one of the solutions with energy 1 is missing from the equation in Eq. (63) which re- − duces the number of solutions of h(2) to six. In order to -2 + -1.51.5 -11.0.0 -0.50.5 00.0.0 00.5.5 11.0.0 11.5.5 find out the excluded state, we consider E(2) τ 1 F (τ) Ψ(2) = − Ψ(2) , (66) FIG. 4. (color online) Plots of det [M(θ)] for different θ. The | 4 i √2 | 1 i characteristic equation of the original reduced Hamiltonian θ = 0, shown by the blue line, excludes two zero-energy solu- which vanishes if τ = 1 and allows to construct an eigen- tions of H(2) while the characteristic equations corresponding (3) (2) state of H as Ψ4 in which the subsystem of to the transformed Hamiltonian with θ 6= nπ for n ∈ Z include spin 1, 2 is disentangled|−i ⊗ | fromi the third spin. Also we all the solutions. note that the other eigenstates of H(3) with energy 1 (2) (2) − do not fulfil Ψ4 ψ = 0 and so their solutions are The same situation recurs if we extend the model to h | − i 6 (3) (2) include one extra spin, for instance, not excluded. The degeneracy between H and H can be lifted by a SU(2) transformation on the third (3) (2)  (1) (3) (2) (3)  spin. Subsequently, the eigensolutions of the associated H = H + τσ+ σ + σ+ σ + h.c. , (60) (3) − − reduced Hamiltonian cover all the solutions of H . In brief, the full Hamiltonian needs to be transformed in or- where τ is chosen to be a non-zero real number for the (3) der to find a representation in which the components of sake of simplicity. H has the following form in the the eigenstates are intertwined from one way e.g. (3) eigenspinors of σz ˜(2) 1 ˜ ˜(2)  (2)  ψ = F †(τ) ψ+ , (67) (3) H F (τ) (1) (2) | − i E H˜ (2) | i H = (2) ,F (τ) = τσ + σ . (61) F †(τ) H − − − see the discussion at the end of Sec. II A. H(3) has two disentangled eigenstates as far as The procedure can be iterated in order to include more (2) | ± ± ±i spins and achieve a targeted system. The iteration can- τ = 1. h+ does not exist for and the solution | | 6 | − − −i (2) not run forever as at some stage the dimension of the of this state would be missing from the solutions of h+ . Hilbert space exceeds the available resources required to Other eigenenergies of H(3) are captured by the reduced solve the model. At this point, truncating the basis is (2) Hamiltonian. h+ commutes with the z-component of required which allows to achieve a low-energy descrip- total angular momentum and the eigenstates can be la- tion of the model. The decimation procedure is elab- beled by its eigenvalues orated in methods like the numerical renormalization group (NRG) [38] or the density matrix renormalization T (1) (2) T (2) (2) σz = σz + σz , σz ψ+ = mz ψ+ . (62) group (DMRG) [39, 40]. The common ground between | i | i the formalism presented here and NRG and DMRG is the (2) essential role of the subsystems in constructing the full- The characteristic equation for the eigenstates of h+ system solution. DMRG minimizes the loss of informa- with mz = 2 is − tion of an approximated solution by finding a truncated (1 + τ)2 (1 τ)2 basis which maximizes the entanglement entropy [3, 40] + − E(3) = 0. (63) E(3) 1 E(3) + 1 − of the wavefunction. A similar variational approach can − 10

be combined with the scheme of USUSY QM to over- =1 N =7 N come the limitations of the ED versus the growth of the Hilbert space with number of spins.

B. Tavis-Cummings model

TC model [28] describes interactions of a single bo- N 1 (N) N son mode and N spin- 2 . The TC Hamiltonian HTC is (k) achieved by replacing H0 = a†a and F+ = αa in Eq. (23) FIG. 5. (color online) Results of genralized Dicke models with for k = 1, ,N although the following treatment works α = 0.15, β = 0.1, γ = 0.1 and ∆ = 0.2. Models are solved ···(k) (k) (k) as well if F+ = α a with a site-dependent α . All via diagonalizing the reduced Hamiltonian in a N -dimensional the subsystems of the TC model need to be solved in basis and then outcomes are compared with results of a direct order to construct the inner products in Eq. (29) and ED engaging a 2N × M dimensional basis, where the number subsequently a MPS representation similar to the one in of lowest boson states is fixed by M = 20. Left: deviations Eq. (27). However, the symmetry of the TC model in the ground-state energy versus number of spins found by N = 1. Right: deviations in seven distinct low-lying eigenen- " N # ergies versus N for a model with N = 7. The results of each (N) 1 X (k) eigenstate are connected by a line. The numbers in the legend H , a†a + σ = 0, (68) TC 2 z from 1 to 7 stand for low to high energies, respectively. k=1 allows to construct a MPS ansatz Ψ(N) = (N) ς | n,ηi Mn | n,ηi with assume α is increased adiabatically from zero for ∆ > 0, | | (N) N 1 then the ground state is shifted from Ψ0 to one of the −   (N) | i (N) (N) Y aIN ` 1 0 states in Ψ1,η and then to one of the states in Ψ2,η n = I` − − IN n , (69) | i | i M ⊗ 0 IN ` 1 | i and so on. `=0 − − which imitates the entanglement pattern of the state in ` ` Eq. (27) and specifically Eq. (34). I` is 2 2 identity N × C. Generalized Dicke model matrix, ςn,η is a spinor with 2 unknown parameters as its components,| i n stands for the boson occupation number while η 1, , 2N is a quantum number as- The last example is the generalized Dicke model ∈ { ··· } (N)† (N) (N) [29, 30, 41, 42] characterized by replacing H0 = a†a and sociated with spin part. Because HTC = 0 Mn0 Mn (k) for n = n0, the boson DOF is fixed by the ansatz and the F = αa + βa† + γ in Eq. (23) for k = 1, ,N. In 6 + ··· problem is reduced to determining ςn,η . Fig. 5, we show the results of the EP of N-spin model The eigensolutions can be grouped| intoi two classes re- through diagonalizing the associated reduced Hamilto- (N 1) (N 1) garding n > N or n < N. In the first class, the as- nian h − . The basis for representing h − is con- − (N− 1) sociated eigenenergies are roots of the determinant of structed out of lowest eigenstates of H+ − . The re- (N) (N)† (N) N N (N) n (HTC E) n and 2 parameters need to be sults are compared with outcomes of a direct ED on H Mcalculated. In− theM second class, the states have less pa- by means of a basis with a dimension of 2N M. It is rameters to be determined. For instance, no parameter the merits of the intrinsic correlated basis that× even by needs to be found for the disentangled state with n = 0 shrinking the size of the truncated basis down to = 1, and with the eigenenergy N∆, where the eigenstate is the ground-state energy can be achieved quite accuratelyN (N) − denoted by Ψ0 . The number of unknown coefficients in comparison with the direct ED, see also App. F. In are N + 1 and| Ni(N + 1)/2 + 1 for n = 1 and n = 2, this case, the equation respectively. Generally, the number of parameters is de- (N 1) (N 1) termined by − − Ψ0 h Ψ0 E, (71) h | −| i ≈ n X N P (N, n) = , (70) where the expectation value is taken by the ground state j (N 1) j=0 of H+ − , needs to be solved for E and the smallest so- lution is a candidate for the ground state of H (N). We see App. E for a demonstration. It can be readily shown note that no optimization in Eq. (71) needs to be per- that there is an upper bound to the number of parameters formed and the source of errors is the truncated basis. As n P (N, n) 6 (1 + N) for arbitrary n. Hence, if N n, it can be seen in Fig. 5, the results become more accu- then P (N, n) N n which shows the number of parame- rate by increasing the number of spins N. The latter can ters increases∼ polynomially with N. be attributed to the pattern of correlations in Eq. (34): If the spin and the fields are not correlated strongly the higher number of spins induces more correlations into α < ∆ < 1, the ground state is in the second class: characteristic equations. 11

V. DISCUSSION where  iφ θ iφ θ  e cos 2 e sin 2 The EP of a Hamiltonian describing interactions be- U U (φ, θ) = . (A2) p   ≡ iφ θ iφ θ tween a single spin- 2 and external fields provides in- e− sin e− cos tertwining relationships between the components of the − 2 2 eigenstate. The existence of these correlations implies Given H = H0 ∆, the explicit form of the transformed ± ± there exist degrees of redundancy in fixing the quantum Hamiltonian is written in terms of the following operators state of the system. In this work, we exploited these ˜ 1 iφ iφ
relations to reduce DOF required to solve the EP and H = H0 ∆ cos θ e F + e− F+ sin θ,(A3) ± ± ∓ 2 − also to derive USUSY QM. The formalism reveals that iφ 2 θ iφ 2 θ a subsystem of the system can carry all information en- F˜+ = ∆ sin θ + F+e− cos F e sin . (A4) 2 − − 2 coded in the full eigenstate. The scheme is non-trivial if the Hamiltonian does not bear a good spin quantum We proceed with the generalized Rabi model specified + by H0 = a†a and F+ = αa + βa† + γ for α, β R and number. In such cases the eigenstates are expected to ∈ γ C. The entries of the transformed Hamiltonian are be entangled states. It can be concluded that the entan- ∈ glement provides correlations between constituents of the   H˜ = a†a aλ∗ + a†λ + ∆˜ , (A5) system and can be exercised to reduce DOF in spin mod- ± ± els. It is curious to find out whether the same conclusion F˜+ =αa ˜ + βa˜ † +γ ˜ = F˜† (A6) can be made in a wider class of interacting problems or − not which we leave it for future studies. where The importance of the entanglement in exploring rel- 1 iφ iφ
λ = sin θ αe + βe− , (A7) evant sectors of Hilbert spaces is elevated in tensor net- −2 work (TN) methods [2, 4, 5] developed under the influ- 1 iφ iφ
∆˜ = ∆ cos θ sin θ γe− + γ∗e , (A8) ence of the quantum information science. A TN state is − 2 a trial wavefunction which allows to control the amount iφ 2 θ iφ 2 θ α˜ = αe− cos βe sin , (A9) of the entanglement in the state. Our study can also be 2 − 2 of relevance to this realm since it reveals that a system iφ 2 θ iφ 2 θ β˜ = βe− cos αe sin , (A10) of interacting spins offers its own intrinsic basis with a 2 − 2 tractable pattern of entanglement that can be deduced iφ 2 θ iφ 2 θ directly from the Hamiltonian. The connection becomes γ˜ = ∆ sin θ + γe− cos γ∗e sin . (A11) 2 − 2 tangible by noticing that the generic representation of a many-spin eigenstate in Eq. (34) is a MPS [40], which is H˜ satisfy eigenvalue problems H˜ n˜; = E˜(n) n˜; , ± ±| ±i ± | ±i the simplest TN state. Moreover, the fractal structure of where n˜; = D†( λ) n is defined by action of the dis- the states in Eq. (34) mimics the way by which entangle- | ±i ± | i λa† λ∗a placement operator D(λ) = e − on the eigenket of ment is implemented in some TN ansatz [1, 43]. It can be the number operator a†a n = n n , and the eigenener- subjects of further researches to connect the study pre- gies are | i | i sented here and TN methods in order to construct new families of ansatz in the study of spin systems. E˜(n) = n λ 2 ∆˜ . (A12) ± − | | ± The commutation relations between the fields in the Hamiltonian change as ACKNOWLEDGMENT h i H˜ , F˜+ = β˜(a† λ∗) α˜(a λ), (A13) ± ± − ± We thank Reinhold Egger for useful discussions. This h i ˜ ˜ 2 ˜ 2 2 2
work is supported by NSF-China under Grant No. F+, F = α˜ β = α β cos θ, (A14) − | | − | | − 11804396, and is carried out in part using computing re- h i ˜ ˜
sources at High Performance Computing Centre of Cen- H+, H = 2 λ∗a λa† . (A15) − − tral South University.

Appendix B: Tridiagonalizing Matrix Representations of Generalized Rabi Models Appendix A: Unitary Transformations of Generalized Rabi Models Assume neither of α, β and γ is zero. Based on the We consider intially a 2 2 matrix Hamiltonian and commutation relation in Eq. (A13), setting eitherα ˜ = 0 ˜ its transformation under a rotation× in SU(2) or β = 0 turns the representation of reduced Hamiltoni- ans     H+ F+ H˜+ F˜+ 1 H = , H˜ = = U †HU, (A1) h˜ = H˜ + F˜ G˜ F˜ , G˜ = , (B1) F H F˜ H˜ ± ± ± ∓ ∓ ∓ E H˜ − − − − − ∓ 12 into a tridiagonal form when represented in the basis of baseline of parameters in Eq. (C1) for JC model β = γ = H˜ . To illustrate it, we choose β˜ = 0 which is satisfied 0 in [44] and for generalized Rabi models with a parity ∓ θ p symmetry γ = 0 in [17]. by tan 2 = β/α and φ = 0. Concerning the eigenvalue problem (h˜+ E) ψ˜+ = 0, the matrix elements in the base ket of n˜−; become| i | −i (m,n) f˜ = m˜ ; (h˜+ E) n˜; 2. Second Class h −| − | −i = m + 3 λ 2 + ∆˜ + (m + 1) α˜ 2˜(m+1) | | | | − If β˜ = 0 and λα˜ =γ ˜, then [H˜+, F˜+] = F˜+. That 2 (m) i ˜ − + γ˜ +αλ ˜ ˜ E δm,n is F+ is the annihilation operator in an algebra diago- | | − − nalizing the Hamiltonian H˜+ = F˜ F˜+ + const.. The h (n)i − + 2λ∗ +α ˜(˜γ∗ +α ˜∗λ∗)˜ √nδm+1,n constraints on the parameters are − h (m)i + 2λ +α ˜∗(˜γ +αλ ˜ )˜ √mδm,n+1, (B2) ( q − tan θ = η β for η = , 2 α ± (C2) (m) ˜(m) 1 β2 α2 where ˜ = (E E )− . The determinant of the −2 = ∆ + γ cot θ for γ R. tridiagonal− matrix− can− be worked out by means of the ∈ three-term recurrence relation in Eq. (46). The anzatz

M (m) ! Appendix C: Exact Isolated Solutions of Genralized X c+ m˜ ; + Ψ˜ M = | i (C3) Rabi Models | i (m) m=0 c m˜ ; + − | i We consider three classes of exact isolated solutions with M 0, 1, 2, , where m˜ ; + is the eigenstate of according to the commutation relations between H˜ , F˜+ ∈ ··· | i ± ˜ ˜(m) 2 ˜ ˜ H+ with the eigenenergy E = m λ + ∆, leads to and F for non-zero α and β. We set φ = 0 and treat θ + − | | as a free− parameter to be fixed restricted to 0 < θ < π. two coupled recurrence equations The USUSY QM does not exist for the states with the (p) ˜ 2 (p+1) p exact solution. Two first classes are about genralized c+ (p + ∆ λ E) + c α˜ p + 1 = 0, (C4) − − − Rabi models α = β and the third class is concerned with (p) (p+1) p 6 c (p ∆˜ + 3λ2 E) c 2λ p + 1 the Rabi model α = β. − − − − − (p 1) (p 1) c − 2λ√p + c+ − α˜√p = 0, (C5) − − 1. First Class by implementing it in H˜ Ψ˜ = E Ψ˜ . The ansatz | M i | M i has 2(M + 1) unknown coefficients c(p) for 0 p M. If β˜ = 0 and λα˜ = γ˜, then [H˜ , F˜+] = F˜+. That is 6 6 − − − Since the coefficients of the largest quantum± number are F˜+ is the annihilation operator in an algebra [F˜+, F˜ ] = − (M) (M) 2 ˜ ˜ ˜ linearly related 2λc =αc ˜ + , the set of equations can α˜ > 0 diagonalizing the Hamiltonian H = F F+ + − const.| | . The constraints read explicitly as − − be represented as a matrix equation of rank 2M + 1. The ansatz is an eigenstate of H˜ with E = E˜(M) if the ( q + tan θ = η β for η = , determinant of the matrix is zero which introduces an 2 α ± (C1) α2 β2 extra constraint on the parameters of the model. For − = ∆ + γ cot θ for γ R. 2 2 ∈ instance, one finds 2λ = ∆˜ for M = 0; and in the case Two solutions of θ in the limited range are labeled by of M = 1, the condition is set by a determinant η = . For either solution, if 0˜ is a vacuum of F˜ , then ˜ ± ˜ | i −    (0) Ψ = (0 0 ) is an eigenstate of the transformed Hamil- 1 0 α˜ c+ h | ˜ h | ˜ 2 − (0) tonian H. The state Ψ is disentangled and does not  0 1 + 2∆˜ 4λ 2λ  c  = 0 | i −   bear the USUSY QM since one of the supercharges does α˜ 2λ 2∆˜ 4λ2 (1)− − − c not exist. We note that if γ = 0, two disentangled eigen- − states are degenerate and therefore their superposition 1 0 α˜ − can entangle the boson state and the spin state. 0 1 + 2∆˜ 4λ2 2λ = 0. (C6) ⇒ − ˜ 2 Ifα ˜ = 0 and λβ˜ =γ ˜, then [H˜+, F˜ ] = F˜ . That is α˜ 2λ 2∆ 4λ − − − − − F˜ is the annihilation operator in an algebra [F˜+, F˜ ] = − 2 − β˜ < 0 diagonalizing the Hamiltonian H˜+ = F˜+F˜ + The eigenstates are degenerate if γ = 0 once again. const.−| | . The solutions in this case can be achieved by− a Within an equivalent path, one can follow to find the unitary transformation from the proceeding case. 4 4 solutions through settingα ˜ = 0 and λβ˜ = γ˜ which supersymmetric Hamiltonians are constructed along× the provide [H˜ , F˜ ] = F˜ . − − − − − 13

3. Third Class a n n 3 a a | i | i a a 1 n n 2 In this class we deal with Rabi models α = β in the | i | i a a n n 2 presence of an in-plane magnetic field. An ansatz 1 | i | i a n n 1 1 1 | i | i PM (m) ! a a n ! n 2 ˜ m=0 c+ m˜ ; + 1 | i | i ΨM = M 1 (m)| i (C7) | i P − a n n 1 m=0 c m˜ ; + 1 1 | i | i − | i a n n 1 1 1 | i | i  ˜ n with M 1, 2, 3, . Inserting ΨM in the EP, a 1 1 1 n ∈ ··· | i | i | i system of equations can be derived. Ψ˜ can be an | M i ˜ ˜(M) (N) eigenstate of H with E = E+ if the system of equa- FIG. A1. (color online) The vectorial representation of Mn tions can be solved self-consistently. It can come true by in Eq. (69) for N = 3. The column vectors represent the (3) settingα ˜ = β˜ = 0 which fixes the angle of rotation as diagonal terms of the matrices in Mn . The rows in which 2 θ the same number of a appears are shown by the same color. tan 2 = 1. To illustrate it, first, we consider M = 1 by which the set of equations can have non-trivial solutions if a condition on the parameters can be met where !  1   (0) G1 = (E H0 ∆)− , 1 γ˜ c+  − − −˜ 2 (0) = 0 1 γ˜∗ 1 + 2∆ 4λ c G2 = (E H0)− , (D3) − − −  − 1 G3 = (E H0 + ∆)− . 1 γ˜ − − 2 = 0, (C8) ⇒ γ˜∗ 1 + 2∆˜ 4λ Taking the square of the traceless matrix decouples one − − of the components from two others and the coefficient with the largest quantum number is (1) (0) (H0 + F G1F+ + F+G3F ) ψ2 = E ψ2 . (D4) fixed byγ ˜∗c+ = 2λc . For M = 2, one finds − − | i | i − Two other components are coupled by a 2 2 matrix  (0) ×   c 2 γ˜ 0 0 + G F G F G F G F   ψ   ψ  − 2 (0) 1 + 2 1 + 2 + 1 1  γ˜∗ 2(1 + ∆˜ 2λ ) 0 2λ  c  − | i = | i , (D5)   = 0 G3F G2F G3F G2F+ ψ3 ψ3 −0 0− 1 γ˜   (1)−  − − − | i | i   c+  −˜ 2 and can be decoupled by repeating the scheme to find 0 2λ γ˜∗ 1 + 2∆ 4λ c(1) − − two extra reduced eigenvalue problems. − 2 γ˜ 0 0 − 2 γ˜∗ 2(1 + ∆˜ 2λ ) 0 2λ − − = 0, (C9) ⇒ 0 0 1 γ˜ Appendix E: Number of Parameters in a MPS − 2 representation of TC Model 0 2λ γ˜∗ 1 + 2∆˜ 4λ − − (2) (1) Let us first illustrate how the number of non-zero com- accompanied withγ ˜∗c+ = 2√2λc . Here also the states are degenerate if γ = 0 as two− solutions of θ do ponents of an eigenstate of TC model in the MPS repre- not generate distinct models. sentation can be counted for N = 3, and then generalize (N) the result for arbitrary N. n in Eq. (69) is a multipli- cation of diagonal matricesM whose non-zero elements can be depicted by column vectors as shown in Fig. A1. The Appendix D: Diagonalizing a spin-1 model in the spin subspace number of states n j with j = 0, 1, 2, 3 in the ansatz | − i 3 is the binomial coefficient . In fact, this number cor- j Hamiltonian of a single spin-1 coupled with an external responds the number of ways to distribute j operators field in the associated eigenvalue problem reads in three consecutive entries of a row in Fig. A1. For in- stance, one (two) operator(s) can be distributed in three H + ∆ F 0   ψ   ψ  0 + | 1i | 1i different ways which are shown in blue (green) in Fig. A1.  F H0 F+   ψ2  = E  ψ2  . (D1) − | i | i The state n j is vanished if j > n. If n < 3, then 0 F H0 ∆ ψ3 ψ3 | − i − − | i | i n X 3 P (3, n) = , (E1) Making the Hamiltonian traceless, we find j j=0       0 G F 0 ψ ψ represents the number of all non-zero components. The 1 + | 1i | 1i G2F 0 G2F+  ψ2  =  ψ2  , (D2) same arguments applies for arbitrary N which leads to − | i | i 0 G3F 0 ψ3 ψ3 the Eq. (70). − | i | i 14

(N 1) Appendix F: Exact Diagonalization of N-spin h − is represented in a matrix (N) by means of a - Generalized Dicke models dimensional± basis. If the truncation is the same for dif-N N 1 ferent spins, then max ( ) = 2 − M. In Fig. A2, it is N Direct ED of the generalized Dicke model H = a a shown that engaging eigenstates of (N 1)-spin sys- 0 † N − and F+ = αa + βa† + γ tem (to diagonalize the reduced Hamiltonian) results in more than eigenstates of N-spin system with energy (N 1) ! (N 1) ! N (N) (N) H+ − F+ ψ+ − H Ψ = (N 1) | (N 1)i | i F H − ψ − − − | − i  (N 1)  H − + ∆ F+ (N) = (N 1) Ψ F H − ∆ | i − − = E Ψ(N) (F1) | i is about to diagonalize the matrix representation of H(N) in the truncated basis n, Si where n [0,M 1] is the boson occupation number| andi S = (s∈ , s , . .− . s ) , i i 1 2 N i ∈ 1, 2N  is the collective spin index of the N-spin system while si is the state of a single spin. The matrix element of the Hamiltonian is simply given by FIG. A2. (color online) Results of generalized Dicke model m,n  (N) (N) of N = 7 with α = 0.15, β = 0.1, γ = 0.1 and ∆ = 0.2. H = n, Si H m, Sj , i,j h | | i The reduced Hamiltonian is diagonalized by means of Nin lowest eigenstates of H(N−1), and the number of extracted N − and the dimension of the matrix is 2 M. (N) × eigenstates of H , having energy errors smaller than 4% in Instead, the reduced Hamiltonian of the N-spin gener- comparison with a direct ED, is shown by Nout. The dashed alized Rabi model in the associated eigenproblem line corresponds to Nout/Nin = 2 which is given as a matter of comparison. (N 1) (N 1) h+ − ψ+ − =  | i  (N 1) 1 (N 1) H − + ∆ + F+ F ψ+ − errors < 4%. Particularly, each eigenstate of N 1 spins E + ∆ H(N 1) − | i − − leads to two eigenstates of N-spin system when− 4. (N 1) N ≤ = E ψ+ − (F2) In fact, we note that eigenstates of the (N 1)-spin | i subsystem input in the N matrix lead to 2 eigenstates− can be diagonalized by means of a truncated basis con- F N (N 1) of the N-spin system. The reason why out < 2 in with structed out of eigenstates of H − . The eigenenergies N N − in > 4 in Fig. A2 is due to the fact that the eigenstates are then given by the roots of the determinant of the withN error > 4% are not counted. matrix We give a physical argument on the relation between  (N) (N 1) (N 1) (N 1) input and output numbers of states in the diagonalization = Ψj − (h+ − E) Ψk − , (F3) F j,k h | − | i of the reduced Hamiltonian. Assume α = β = 0, then the matrix elements in Eq. (F3) just have diagonal non-zero (N 1) where Ψi − is the eigenstate of the (N 1)-spin terms | i (N 1) − Hamiltonian with its eigenenergy Ei − . The eigen- 2  (N 1)  γ state in a truncated basis is formally written as ( ) = E − + ∆ E + | | . (F5) F j,j j − (N 1) E + ∆ Ej −  PM 1 i,1  − n=0− cn n PM 1 i,2 | i − c n Q  n=0 n  The eigenenergies are the roots of = jN=1 ( )j,j = 0 N 1  PM 1 i,3 | i  |F| F Ψ −  − c n  , (F4) which has 2 solutions. Now, turning on α and β adia- i ≈  n=0 n  N  ... | i  batically, the eigenenergies are redistributed and no state M 1 N 1 (N) (N 1) P i,2 − is generated/annihilated (as far as H and H − ∆ n=0− cn n | i are not degenerate). Therefore, exploiting states to− di- i,1 N where cn can be obtained by direct ED of (N 1)-spin agonalize (N 1)-spin system gives rise to 2 eigenstates system or obtained by solving its reduced eigenproblem.− of N-spin system.− N

[1] G. Vidal, Entanglement Renormalization. Phys. Rev. states, projected entangled pair states, and variational Lett. 99, 220405 (2007). renormalization group methods for quantum spin sys- [2] F. Verstraete, V. Murg, and J.I. Cirac, Matrix product tems. Adv. Phys. 57, 143-224 (2008). 15

[3] J. Eisert, M. Cramer, and M. B. Plenio, Colloquium: to the beam maser. IEEE Proc. 51, 89-109 (1963). Area laws for the entanglement entropy. Rev. Mod. Phys. [26] A. D. Greentree, J. Koch, and J. Larson, Fifty years 82, 277 (2010). of Jaynes-Cummings physics. Journal of Physics B: [4] G. Evenbly and G. Vidal, Algorithms for entanglement Atomic, Molecular and Optical Physics 46, 220201 renormalization. Phys. Rev. B 79, 144108 (2009). (2013). [5] R. Or´us,A practical introduction to tensor networks: [27] Q. Xie, H. Zhong, M. T. Batchelor, and C. Lee, The Matrix product states and projected entangled pair quantum Rabi model: solution and dynamics. Journal states. Ann. Phys. 349, 117-158 (2014). of Physics A: Mathematical and Theoretical 50, 113001 [6] R. Horodecki, P. Horodecki, M. Horodecki, and (2017). K. Horodecki, Quantum entanglement. Rev. Mod. Phys. [28] M. Tavis and F. W. Cummings, Exact solution for an 81, 865-942 (2009). N-molecule-radiation-field hamiltonian. Phys. Rev. 170, [7] I. I. Rabi, Space Quantization in a Gyrating Magnetic 379-384 (1968). Field, Phys. Rev. 51, 652 (1937). [29] R. H. Dicke, Coherence in Spontaneous Radiation Pro- [8] M. O. Scully and M. S. Zubairy, Quantum Optics (Cam- cesses. Phys. Rev. 93, 99 (1954). bridge University Press, Cambridge, England, 1997). [30] P. Kirton, M. M. Roses, J. Keeling and E. G. Dalla Torre, [9] M. Wagner, Unitary Transformations in Solid State Introduction to the Dicke Model: From Equilibrium to Physics (North-Holland, Amsterdam, 1986). Nonequilibrium, and Vice Versa. Adv. Quantum Technol. [10] I. Thanopulos, E. Paspalakis, and Z. Kis, Laser-driven 2, 1800043 (2019). coherent manipulation of molecular chirality, Chem. [31] A. Naseri, S. Peng, W. Luo, and J. Sirker, Spin vor- Phys. Lett. 390, 228 (2004). tices and skyrmions of a single electron in inhomogeneous [11] F. Cooper, A. Khare, and U. Sukhatme, Supersymme- magnetic fields. Phys. Rev. B 101, 115407 (2020). try and quantum mechanics. Phys. Rept. 251, 267-385 [32] S. Weinberg, The Quantum Theory of Fields. Vol. 2. (1995). (Cambridge University Press, 2005). [12] G. Junker, Supersymmetric Methods in Quantum and [33] R. A. Horn and C. R. Johnson, Matrix Analysis. (Cam- Statistical Physics (Springer-Verlag Berlin Heidelberg, bridge University Press, 2nd edition, 2012). 1996). [34] B. R. Judd, Exact solutions to a class of Jahn-Teller sys- [13] B. K. Bagchi, Supersymmetry In Quantum and Classical tems. Journal of Physics C: Solid State Physics 12, 1685- Mechanics (Chapman and Hall/CRC, 2000). 1692 (1979). [14] E. Witten, Dynamical breaking of supersymmetry. Nucl. [35] C. Emary and R. F. Bishop, Bogoliubov transforma- Phys. B 188, 513-554 (1981). tions and exact isolated solutions for simple nonadiabatic [15] Z. F. Ezawa, Quantum Hall Effects: Field Theoretical Hamiltonians. Journal of Mathematical Physics 43, 3916 Approach and Related Topics (World Scientific Publish- (2002). ing Company, 2nd edition, 2008). [36] G. Vidal, Efficient Classical Simulation of Slightly En- [16] Y. Yu and K. Yang, Supersymmetry and the Goldstino- tangled Quantum Computations. Phys. Rev. Lett. 91, Like Mode in Bose-Fermi Mixtures. Phys. Rev. Lett. 100, 147902 (2003). 090404 (2008). [37] A. Moroz, Generalized Rabi models: Diagonalization in [17] M. Tomka, M. Pletyukhov, and V. Gritsev, Supersymme- the spin subspace and differential operators of Dunkl try in quantum optics and in spin-orbit coupled systems. type. Euro. Lett. 113, 50004 (2016). Sci. Rep. 5, 13097 (2015). [38] K. G. Willson, The renormalization group: critical phe- [18] T. H. Hsieh, G. B. Hal´asz,and T. Grover, All Majorana nomena and the Kondo problem. Rev. Mod. Phys. 47, Models with Translation Symmetry are Supersymmetric. 773 (1975). Phys. Rev. Lett. 117, 166802 (2016). [39] Steven R. White, Density-matrix algorithms for quantum [19] B. Bradlyn and A. Gromov, Supersymmetric waves in renormalization groups. Phys. Rev. B 48, 10345 (1993). Bose-Fermi mixtures. Phys. Rev. A 93, 033642 (2016). [40] U. Schollw¨ock, The density-matrix renormalization [20] M. Lahrz, C. Weitenberg, and L. Mathey, Implement- group in the age of matrix product states. Annals of ing supersymmetric dynamics in ultracold-atom systems. Physics 326, 96-192 (2011). Phys. Rev. A 96, 043624 (2017). [41] D. Braak, Solution of the Dicke model for N = 3. J. [21] S. H. Dong, Factorization Method in Quantum Mechanics Phys. B 46, 224007 (2013). (Springer Netherlands, 2007). [42] S. He, L. Duan, and Q. H. Chen, Exact solvability, non- [22] R. L. Fulton and M. Gouterman, Vibronic coupling. I. integrability, and genuine multipartite entanglement dy- Mathematical treatment for two electronic states. The namics of the Dicke model. New J. Phys. 17, 043033 Journal of Chemical Physics 35, 1059 (1961). (2015). [23] D. Braak, Integrability of the Rabi model. Phys. Rev. [43] L. Tagliacozzo, G. Evenbly, and G. Vidal, Simulation Lett. 107, 100401 (2011). of two-dimensional quantum systems using a tree tensor [24] M. Wagner, Generalised Fulton-Gouterman transforma- network that exploits the entropic area law. Phys. Rev. tion for systems of abelian symmetry. Journal of Physics B 80, 235127 (2009). A: Mathematical and General 17, 2319-2333 (1984). [44] V. A. Andreev and P. B. Lerner, Supersymmetry in the [25] E. T. Jaynes and F. W. Cummings, Comparison of quan- Jaynes-Cummings model. Phys. lett. A 134, 507 (1989). tum and semiclassical radiation theories with application