WorksPhysics and Astronomy
Odd-Even Effects on Liouvillian Spectra and Relaxation Dynamics in Boundary-Driven Spin Chains
Abstract
We investigate odd-even effects on Liouvillian spectra and relaxation dynamics in boundary-driven XY and cluster chains. With the bulk couplings and boundary dissipation rule fixed, the parity of chain length determines whether non-decaying oscillatory modes exist. Consequently, odd- and even-sized chains exhibit qualitatively different dynamical behaviors. We trace this mechanism to the coupling between bulk Majorana paths and the boundary jump operators. An isolated path supports undamped modes, and products of an even number of its Majorana operators span a non?decaying subspace of even operators. This structure yields nonzero purely imaginary eigenvalues of the Liouvillian and persistent contributions to bond correlations or magnetization after preparation in product states. The mechanism extends to cluster interactions of even range under a fixed boundary dissipation rule. The existence and frequencies of the persistent modes are independent of the magnitudes of the positive boundary dissipation rates. Our results unveil how chain parity controls the existence of persistent dynamics.
Introduction
Understanding relaxation and information retention in quantum many-body systems is a central problem in nonequilibrium physics [1, 2]. Advances in the control of dissipation have stimulated extensive interest in open quantum systems and their collective behavior beyond thermal equilibrium [3, 4, 5, 6]. A particularly important class is boundary-driven systems, where the environment acts only near the boundaries and the bulk interactions determine how dissipation propagates through the system [7]. Such systems exhibit rich stationary and dynamical phenomena, including long-range correlations and nonequilibrium transitions [8, 9]. Exactly solvable models and analytical approaches have provided valuable insights into their Liouvillian spectra and relaxation dynamics [10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20].
Much of the discussion of open many-body systems has focused on phases in the thermodynamic limit and on the size dependence of relaxation times [21, 22, 23, 24, 25]. From the dynamical perspective, considerable attention has been paid to the Liouvillian spectrum, the structure of its eigenmodes, and their overlaps with initial states and observables [26, 27, 28, 29, 30]. Related studies have established routes to information retention through decoherence-free structures and symmetries [31, 32, 33, 34], spatially restricted dissipation [35, 36], and boundary modes [37, 38, 39, 40, 41]. Dissipative spectral structure and selective coupling to the environment further influence the survival of boundary excitations [42, 43, 44, 45], while dynamical symmetries can support persistent coherent motion and synchronization [46, 47, 48, 49, 50].
These results also highlight the importance of finite size in interpreting non-decaying dynamics [51, 52]. However, the role of odd-even effects in connecting Liouvillian spectra to relaxation dynamics has received comparatively less attention. Focusing on the thermodynamic limit or on smooth finite-size scaling does not by itself resolve the distinction between odd and even chain lengths, while an analysis of spectral and initial-state contributions at a given size does not directly address how adding one site changes the coupling to the boundaries. Previous work considered a boundary-dephased XY chain and showed that the relaxation timescales obey different scaling laws for odd and even chain lengths in the weak-dissipation limit, with the difference washed out by strong dissipation [53]. Odd-even differences have also been reported in the steady-state density profiles of boundary-driven double spin chains [54]. Exact persistent Liouvillian modes have been constructed in boundary-dissipative spin chains [50]. Nevertheless, a criterion, in terms of the parity of chain length, for the existence of an entire exactly non-decaying operator sector, under fixed bulk couplings and a fixed boundary dissipation rule, has not been established. It therefore remains unclear whether changing the chain length from odd to even can control whether persistent oscillations exist, and what physical mechanism underlies this odd-even distinction.
In this work, we investigate odd-even effects on Liouvillian spectra and relaxation dynamics in boundary-driven XY and cluster chains. We show that changing the chain length alters the connectivity between the bulk degrees of freedom and the dissipative boundaries, thereby determining whether an entire operator sector remains decoupled from dissipation. The same endpoint spin flip channels lead to opposite odd-even effects in the two models: non-decaying oscillatory modes occur in even XY chains and odd cluster chains of sufficient length. This distinction is reflected in the presence or absence of nonzero purely imaginary Liouvillian eigenvalues and can be observed through bond correlations and magnetization following simple product-state preparations. We identify the underlying mechanism through the connectivity of Majorana paths to the dissipative contacts and extend it to cluster interactions of even range under a fixed boundary dissipation rule. The existence and frequencies of the persistent modes are independent of the magnitudes of the positive boundary dissipation rates, including unequal rates. Our results establish the parity of chain length as a discrete control of dissipation-free dynamics and connect odd-even effects in the spectrum directly to measurable relaxation behavior.
The paper is organized as follows. Section II presents the XY chain, its Liouvillian spectrum, and spin correlation dynamics. Section III establishes the reversed odd-even effect in cluster chains and its magnetization signature. Section IV summarizes the results.
Boundary-driven XY chain
Model
We begin with a zero-field XY chain of spins, with Pauli operators () and real exchange couplings,
Here we take and . The density matrix obeys the Lindblad master equation with Liouvillian [55, 56]
The two boundary jump operators are
Both jumps are Hermitian, so , and is a steady state.
Define the Hermitian Majorana operators by
In the ordering , these operators obey . We also define the spin parity and the operator parity map , with eigenvalue . Here operator parity is distinct from the odd-even character of the chain length or site indices. The Hamiltonian is even and the jumps odd under , so the Liouvillian preserves operator parity [34].
Here a Majorana path denotes a sequence of Majorana operators coupled by the quadratic Hamiltonian; an isolated path is one on which boundary dissipation does not act in the even operator subspace. The Hamiltonian couples to and to , forming two independent Majorana paths, and . The left jump is proportional to , while at the right boundary generates the same dissipative action as on even-parity operators. For even , both boundary dissipators act on in the even operator subspace, leaving isolated from dissipation, as shown in Figure 1(b). For odd , boundary dissipation acts on both paths, leaving neither path isolated, as shown in Figure 1(a). The explicit Majorana couplings are given in Appendix A.1. This odd-even dependence of the action of boundary dissipation accounts for the distinct Liouvillian spectra and long-time dynamics.

Figure 1. XY chain at , , and , with in the left and right columns. (a,b) Majorana paths: blue and gray links denote and , rings mark Majorana coordinates acted on by boundary dissipation, and green nodes are isolated. (c,d) Rapidity spectra. (e,f) Liouvillian spectra: blue and red denote the even and odd operator sectors. Green diamonds mark eigenvalues with zero real part.
Liouvillian spectrum
We write the quadratic Hamiltonian as , where is the vector of Majorana operators and is a real antisymmetric matrix. The effect of boundary dissipation is encoded in the positive semidefinite damping matrix , giving the rapidity spectrum matrix . For the endpoint channels considered here,
where and is the th standard coordinate vector. We denote the eigenvalues of by (rapidities) and those of the Liouvillian by . The matrix determines the evolution of even Majorana bilinears, as shown below; its eigenvalues are not themselves the full Liouvillian spectrum.
For an eigenmode satisfying , the antisymmetry of yields
Hence, an undamped mode must satisfy and . This condition leads to a simple criterion for undamped modes: for independent finite open Majorana paths with nonzero couplings between adjacent Majorana coordinates and independent dissipation with positive rates acting only at selected endpoints, a path supports undamped modes if and only if neither endpoint is coupled to a dissipative channel. Indeed, on any path subject to boundary dissipation, forces the amplitude at the damped endpoint to vanish, and the eigenvalue equation then recursively forces all amplitudes along that path to vanish. By contrast, an isolated path retains its purely imaginary spectrum because its rapidity spectrum matrix is real and antisymmetric.
Consequently, the number of undamped rapidity modes equals the total number of Majorana coordinates on the isolated paths. The criterion for undamped modes is derived in Appendix A 1; its connection to non-decaying spin operators is discussed in Appendix A 1.
For even , the isolated XY path contains Majorana coordinates, whereas neither path is isolated for odd . For , , and positive boundary dissipation rates in Eq. (3), the criterion for undamped modes yields
Liouvillian eigenvalues on the imaginary axis correspond to non-decaying modes. Eigenoperators satisfying with span the non-decaying operator
subspace: zero eigenvalues describe stationary dynamics, while the imaginary parts of nonzero purely imaginary eigenvalues determine persistent oscillation frequencies. We reconstruct the physical Liouvillian spectrum using the method with projection onto operator parity sectors of Ref. [18], with the operator parity constraint in Eq. (25) and the XY blocks in Eq. (34).
Figure 1(c,Figure 1d) show the rapidity spectra, and Figure 1(e,Figure 1f) show the full Liouvillian spectra, for at , , and . For odd , both paths are subject to boundary dissipation and every rapidity has a strictly negative real part in Figure 1(c). For even , one path is isolated and six rapidities lie exactly on the imaginary axis in Figure 1(d). Their number agrees with in Eq. (7). The remaining rapidities have negative real parts and describe damped modes.
The full Liouvillian spectra show the corresponding distinction between stationary and oscillatory dynamics. For odd , only stationary modes remain non-decaying in Figure 1(e). There are no nonzero purely imaginary eigenvalues. For even , nonzero purely imaginary eigenvalues appear alongside the stationary modes at the origin in Figure 1(f). They supply non-decaying oscillatory modes, while eigenvalues with negative real parts describe decaying contributions. A nonzero purely imaginary eigenvalue contributes a persistent oscillation only when the total coefficient at that frequency, determined by the initial state and measured observable, is nonzero. The spin correlation dynamics below provides this connection.
Spin correlation dynamics
The magnetization and bond observables studied here are even under . Their expectations depend only on , since for even and . We therefore obtain their dynamics from the even bilinears. With , define the covariance matrix and its evolution by [7, 10, 12]
A local connects the two XY paths, whereas a bond correlation can probe one path directly. For even chains, odd-indexed bonds access bilinears within the isolated path. Using the covariance evolution in Eq. (9), we prepare , with , and measure
Initially the independent nonzero covariances are , so .
For even , odd-indexed bonds lie within the isolated path and even-indexed bonds lie within the path subject to boundary dissipation . Define , where specifies whether the bond’s left site has an odd or even index, , and . Then
For the coefficients are and . For even , the first term is independent of the boundary dissipation rates, whereas the second decays. At there is only one bond. Its average is and no average is defined. For odd both bond components decay.
Figure 2(a) shows the dynamical consequence of the spectra at the same parameters: odd-chain bond correlations decay, while the even chain retains an oscillatory contribution. The comparison in (b) exhibits the same non-decaying dynamics for even-length chains. The logarithmic-time comparison reaches . In its final linear-axis inset, the transient envelope is below for both sizes. The even-chain oscillations persist while the odd-chain bond correlation has relaxed.
Figure 2. XY spin correlation dynamics from at , , and . (a) . Inset: . (b) . Inset: . Blue solid/red dashed curves denote even/odd lengths.
For the XY chain with , the non-decaying contribution has the explicit form
It exhibits both a constant contribution and a contribution at a nonzero frequency, each independent of the boundary dissipation rates. At the isolated component instead supports only stationary bond correlations. For , the correlation averaged over the chain is , with the second term decaying. The same covariance construction gives the curves in Figure 2(a). Its effective two-level derivation is in Appendix A.3. Results for are given in Appendix A.4.
The same boundary jump operators can reverse which chain lengths support non-decaying dynamics when the bulk Hamiltonian couples the Majorana operators differently. We demonstrate this reversal in the cluster chain.
Boundary-driven cluster chains
Model
Cluster interactions provide a second realization of the mechanism of isolation from boundary dissipation, with an odd-even effect opposite to that of the XY chain. Cluster states are resources for measurement-based quantum computation 1, and transverse-field cluster Hamiltonians can be reduced to two Ising sectors 2. Long-lived boundary qubits have also been studied in perturbed, dimerized cluster chains 3. We consider a cluster family with even interaction ranges and use the cluster chain with to relate the Majorana paths to the Liouvillian spectrum and magnetization dynamics. The open-chain Hamiltonian is
where is a fixed even interaction range, is a positive integer, is the cluster coupling, is the transverse field, , and . The string has intermediate spins. The boundary construction follows the paths generated by the bulk coupling, each linking sites separated by . For even , each path contains only odd-indexed or only even-indexed sites. Applying dissipation to alternating endpoint Majorana coordinates therefore acts on one half of the paths at each boundary. We use the same offsets from the left and right ends for every allowed , without changing the form of the boundary jump operators when the length changes. In the Lindblad equation (2), the independent boundary channels are, for ,
All rates are positive and empty products are the identity. The channels enter independently in the dissipator sum. For fixed , the jumps are supported within sites of each end. At , they reduce to the same two endpoint channels in Eq. (3). For each fixed , there are channels per end: their spin strings produce dissipation on individual endpoint Majorana coordinates in the even operator subspace. As changes, the Majorana coordinates acted on by the right boundary dissipators follow the same rule relative to the chain ends, while their assignment to the bulk paths changes. This is what produces the odd-even effect.
The cluster chain with has the Hamiltonian
with the channels in Eq. (3). Under the Majorana transformation Eq. (4), the Hamiltonian for becomes
More generally, the interaction in couples to . The Majorana graph separates into paths, each containing sites with the same index modulo . Because is even, each path contains physical sites whose indices are all odd or all even. This structure allows the action of the boundary channels to be determined before diagonalizing the generator.
Liouvillian spectrum
In the even operator sector, the boundary dissipators act on and , respectively, since . The corresponding damping matrix is in Eq. (51).
The left boundary dissipators act on the endpoints of all paths on odd sites. The right boundary dissipators act on the same paths for odd , leaving the even-site paths isolated, but the complementary paths for even . The criterion for undamped modes therefore gives
For odd , the isolated even sites contain Majorana operators. For even , every path is subject to boundary dissipation and all rapidity modes decay. This reverses the odd-even effect in the XY chain despite the identical endpoint jumps in the XY chain and the cluster chain with . Whether odd or even chains support undamped modes therefore depends on how the boundary jump operators act on the Majorana paths generated by the bulk Hamiltonian. Importantly, the isolated-path block of the rapidity matrix is independent of the magnitudes of the positive boundary dissipation rates. Varying these rates modifies the spectra of the boundary-coupled paths but leaves the undamped rapidities and the corresponding purely imaginary Liouvillian eigenvalues unchanged.
We now specialize to and compare the rapidity and full Liouvillian spectra in Figure 3. The full spin spectrum is reconstructed using Eq. (25) and the cluster blocks in Eq. (47). Their derivation is given in Appendix B. Figures 3(c,d) show the rapidity spectra, and

Figure 3. Cluster chain at , , , and , with in the left and right columns. (a,b) Majorana paths: blue and gray links denote and , red rings mark Majorana coordinates acted on by boundary dissipation, and green nodes are isolated. (c,d) Rapidity spectra. (e,f) Liouvillian spectra.
Figure 3(e,Figure 3f) show the full Liouvillian spectra, for at , , and . In the rapidity spectra, the odd chain has four purely imaginary eigenvalues in Figure 3(c), exactly as predicted by in Eq. (17). The even chain has no imaginary-axis rapidities: every mode has a strictly negative real part in Figure 3(d). This reverses the odd-even dependence of the XY rapidities under the same endpoint channels.
The full Liouvillian spectra display the corresponding reversal. At odd , nonzero purely imaginary eigenvalues coexist with stationary modes at the origin in Figure 3(e), supporting persistent oscillations. At even , only stationary modes remain non-decaying in Figure 3(f), and all nonstationary modes decay. The magnetization dynamics below probes the persistent frequencies of the odd chain. The corresponding spectra and dynamics for are presented in Appendix C. The robustness of this parity-dependent spectral and dynamical behavior against unequal positive boundary dissipation rates is demonstrated in Appendix B § 1.3.5.
Magnetization dynamics
We first express the magnetization dynamics for the even-range family using Eq. (9) with , and then evaluate the chain shown in Figure 4.
Figure 4. Cluster magnetization from at , , and . (a) . Inset: . (b) . Inset: . Blue solid/red dashed curves denote even/odd lengths.
An on-site magnetization probes the cluster paths directly because and lie on the same path. We prepare , with , and measure
The initial covariance has and no other independent nonzero entries. Define , where labels odd- and even-indexed sites, , and . For odd , these averages satisfy
where evolves independently of dissipation for every fixed even with the boundary jump operators in Eq. (14). For even , all paths are subject to boundary dissipation and both sublattice magnetizations decay. The following explicit dynamical examples use . At , the two weights are and .
Figure 4(a) uses exactly the same sizes and parameters as the spectral comparison. The odd chain retains oscillations, whereas the even signal decays, directly linking non-decaying Liouvillian modes to a measurable spin response. Figure 4(b) extends the comparison to . The small-chain window already resolves the oscillatory contrast. For the larger chains, the logarithmic axis follows the transient to , and the final linear-axis inset resolves the remaining oscillations. At that window the estimated absolute transient envelope is below for both curves, so the odd-chain signal is governed by its non-decaying Liouvillian modes rather than a slowly decaying contribution (see Appendix A 3). The non-decaying contribution is carried by the even sublattice and is independent of the boundary dissipation rates. The corresponding dynamics is shown in Appendix B 4 with the same layout.
At , the even-sublattice magnetization is
This expression contains both a stationary part and an oscillatory part. It describes the even-site sublattice decoupled from dissipation, whereas Figure 4(a) shows the average over the whole chain: at , Eq. (19) gives , with at long times. At , however, the spin observable is conserved, so the signal retains memory without oscillation. The derivation is given in Appendix B 3. For larger even ranges, the isolated paths obey the same dynamics with alternating couplings. Together with the XY bond response, this magnetization signal shows how different bulk paths determine both which chain lengths support non-decaying dynamics and which spin observables detect it.
Conclusions
We have investigated odd-even effects on Liouvillian spectra and relaxation dynamics in boundary-driven XY and cluster chains. With the bulk couplings and the boundary-dissipation rule fixed, the parity of the chain length determines which Majorana paths couple to the boundary jump operators. An isolated path supports undamped modes, and products of an even number of Majorana operators along such a path span a non-decaying subspace of even-parity operators. The same endpoint spin-flip channels therefore lead to opposite parity-dependent behavior in the two models. For sufficiently long chains, even-length XY chains and odd-length cluster chains with support nonzero purely imaginary Liouvillian eigenvalues, giving rise to persistent oscillations in bond correlations and magnetization, respectively.
These results connect odd-even effects to the existence of non-decaying oscillatory modes and extend the discussion beyond parity effects on relaxation timescales. We establish a criterion for undamped modes that moves beyond previous case-specific constructions of persistent modes and decoherence-free dynamics, showing precisely when a fixed boundary-dissipation rule isolates bulk degrees of freedom. The same mechanism applies to even-range cluster interactions with the prescribed boundary channels. Varying the positive boundary dissipation rates, including making them unequal, modifies the decay rates of boundary-damped modes but leaves the isolated paths unchanged. Consequently, the existence and frequencies of the persistent modes are independent of the magnitudes of the boundary dissipation rates, provided all prescribed rates remain positive. Thus, the parity of the chain length provides discrete control over non-decaying dynamics, with signatures accessible through state preparation in product states and spin measurements.
Acknowledgments. S.C. was supported by the National Key Research and Development Program of China (Grant No. 2023YFA1406704) and the NSFC under Grants No. 12474287 and No. T2121001.
Data availability. The data that support the findings of this article are openly available[59].
XY chain
Majorana representation and boundary dissipation
Writing for the th Majorana on path defined in Sec. II, the physical decomposition is
For odd , each path contains an odd number of physical Majoranas, so its operators cannot be paired into an independent set of complex fermions. The two paths must therefore be retained when reconstructing the physical spin Liouvillian spectrum.
For , with real antisymmetric and , multiplying by and taking the real part gives (6). Positive semidefiniteness implies precisely when . Hence, for any eigenpair with ,
Conversely, an eigenvector of satisfying is unaffected by damping and is also an undamped eigenvector of .
For independent finite open paths with nonzero adjacent couplings and diagonal damping with positive rates at selected endpoints, sets the amplitude at each endpoint subject to dissipation to zero. The endpoint eigenvalue equation then sets the neighboring amplitude to zero, and recursion eliminates the entire path subject to boundary dissipation. An isolated path has an antisymmetric block of the rapidity spectrum matrix and purely imaginary eigenvalues. This establishes the criterion for undamped modes used in the main text.
For the specified boundary jump operators, products of an even number of isolated Majorana operators commute with all jumps, and the Hamiltonian preserves the subspace they span. Within each even-degree Majorana subspace, the eigenvalues of the adjoint Liouvillian are sums of distinct rapidities. Products of isolated eigenmodes therefore evolve without decay, whereas sums containing a damped eigenmode have negative real part. Zero and nonzero imaginary sums give stationary and oscillatory contributions, respectively. This construction concerns even operators; the full spin Liouvillian spectrum also requires the operator parity projection derived below.
The absence of decay in the even operator subspace follows without replacing the full right jump by a linear fermion. For an operator satisfying , with ,
Here is the adjoint dissipator acting on observables. For even XY chains, even products of Majoranas on the isolated path commute with both boundary jumps and evolve within that path. (23) also explains why a linear-Majorana dissipator is adequate for the even bilinears, but not for arbitrary parity-odd spin operators.
Liouvillian spectrum
For the XY chain, fermionization in the doubled space in a fixed sector gives [18]
Here contains auxiliary complex fermions with and . The auxiliary rapidity spectrum matrix is complex symmetric. In this representation, is the matrix of the operator parity superoperator defined in Sec. II. The XY blocks are given in Eq. (34). Let be their eigenvalues, with labeling a block and a mode within that block. These auxiliary rapidities are distinguished from the rapidities of : in the even operator sector, is similar to , so , as derived in Eq. (39). The physical spin spectrum is
The numbers are mode occupations. The constraint is imposed on their total across both blocks. The vectorization below derives both the parity constraint and the shift determined by the sector.
With vectorization by rows , the map becomes for operators . Denote replica spins by and , with . Then
Vectorization also underlies the non-Hermitian Kitaev ladder and Ashkin–Teller representations of other dissipative spin chains [15, 16]. Those mappings involve different bulk jumps and sector structures. The boundary identities below are specific to the present models.
For the XY Hamiltonian, . The dissipative part is . Define , , and . To distinguish auxiliary and physical Majoranas, use
Distinct operators in this set of Majoranas anticommute, and each squares to the identity. This explicit phase convention fixes the signs of the auxiliary hoppings and should be used together with the pairing
With , the boundary identities are
Consequently the constant is . There is no additional multiplying the right potential.
The number-conserving form of Eq. (24) follows from the boundary identities above, as in Ref. [18]. The reconstruction on physical operator space is
For the auxiliary Majorana vectors with entries and , the canonical anticommutation relations and give
With , an explicit antisymmetric matrix form is
The matrix is generally complex, and is a generator rather than a Hermitian physical Hamiltonian. The constant must be retained: for the stated boundary potentials.
For a path with coordinates, define the alternating-hopping matrix
The corresponding coordinate projector is .
In the auxiliary space let and for odd , interchanging the assignments for even . The two blocks are
Here equals one for odd and zero for even . The left boundary is always , whereas the right one is for even and for odd . These statements determine the isolated path without inspecting numerical eigenvalues.
Substituting into Eq. (26) gives
The nonzero off-diagonal entries of are and , together with their transposes. The endpoint diagonal entries are and .
For diagonalizable , write , where is diagonal and is an invertible matrix of right eigenvectors. Define and . Then and , although generally . The projection remains necessary after diagonalization, yielding Eq. (25). The same eigenvalue formula remains valid by continuity when eigenvectors coalesce and the matrix is no longer diagonalizable.
Spin correlation dynamics
For , the Majorana anticommutation relations give . The Hamiltonian contribution is consequently . For , conjugation of by a boundary Majorana changes its sign exactly when one index equals :
Using Eq. (23) in the even sector gives . Together these results prove Eq. (9).
In the interleaved physical ordering, the independent nonzero Hamiltonian entries are
The opposite entries follow by antisymmetry, and all bonds are restricted to the chain. With , the state has covariance
It pairs with , and the bond correlation is .
In the path basis, the covariance blocks obey , where label the paths.
For the XY chain, a phase transformation along each open path relates to the auxiliary rapidity spectrum matrix in the even operator sector:
Here labels modes across both auxiliary blocks, so combines the block eigenvalues defined in Appendix A. For diagonalizable , write and . A Hermitian quadratic observable , with real and a real antisymmetric matrix , obeys
Here is a column of . The transpose, rather than the Hermitian conjugate, is required. We use , and . The symbol in Eq. (39) denotes matrix similarity. These pair eigenvalues describe the bilinear subspace, not the full even sector. The identity is separate. The odd-sector scalar shift is essential in Eq. (25) but does not enter these even observables.
For each distinct pair sum, the weights must first be added over all pairs with that eigenvalue. A nonzero purely imaginary pair sum contributes an oscillation only if this total weight is nonzero; zero pair sums give stationary contributions. Pair sums with negative real parts give the decaying remainder.
To choose the large-system plotting windows, we separate the non-decaying contribution from the transient . (40) gives the observable-dependent estimate
The final insets for the large systems begin after this envelope falls below in the normalized observable for both compared sizes. Isolation is identified by boundary connectivity, not a numerical cutoff on small decay rates. The logarithmic main panels display sampled dynamics over the relaxation interval. The linear insets resolve the surviving oscillations.
For diagonalizable , antisymmetry of gives
Contracting with yields Eqs. (40) and (42). For , the independent entries are . To establish Eq. (39), let the upper hopping of a physical component be and that of be , with . A diagonal similarity , with and successive phase ratio maps the upper entry to and the lower entry to the same value. The diagonal entries already obey . If a bond vanishes, this construction applies separately on each remaining connected segment. When is not diagonalizable, the matrix-exponential covariance solution remains valid even when Eq. (44) must be replaced by a Jordan expansion [11, 12].
The distinction between a stationary contribution and oscillation can be seen without numerical diagonalization. For XY at , define . In the initially populated sector , where , set and . The Hamiltonian of the isolated path is , the all- preparation has , and . The effective two-level evolution gives the normalized signal,
with . The constant is a stationary contribution. The remaining term is a contribution at a nonzero frequency when . Both are independent of .
Results for additional parameters
Figure 5 gives the corresponding comparison. As in the main text, the even chain supports the non-decaying bond correlation. Changing the coupling changes its waveform but not the connectivity responsible for the non-decaying contribution.
Figure 5. XY spin correlation dynamics at , , and . The initial state is . (a) . Inset: . (b) . Inset: .
For the large chains at this coupling ratio, the time window ends at , and the transient envelope in each final inset is below .
Cluster chain
The general construction in Appendix A also applies to the cluster chain with the same endpoint channels. We give only model-dependent steps below.
Majorana representation and odd-even effects
For , the physical Hamiltonian couples to through the field and to through the cluster interaction. The Majorana graph therefore separates into one path containing both Majoranas on every odd-indexed site and another containing both Majoranas on every even-indexed site. In the even operator sector, the boundary dissipators act through and , exactly as in the XY chain. The Majorana operator always belongs to the odd-site path. For odd , belongs to that same path, leaving the entire even-site path isolated; for even , it belongs to the even-site path, so both paths are subject to boundary dissipation.
This explains the reversal relative to the XY chain. The XY paths alternate Majorana species between adjacent sites: and lie on the same path for even . The cluster paths instead connect sites separated by two lattice spacings, so these two dissipative Majorana coordinates lie on the same path for odd . Applying the criterion for undamped modes of Appendix A1 therefore gives an isolated path for even XY chains and odd cluster chains, despite identical endpoint jump operators. For the latter, Eq. (23) ensures that even products of Majoranas on the isolated even-site path remain unaffected by dissipation. The odd-even reversal thus follows from the different bulk connectivity and requires no change in the boundary channels.
Liouvillian spectrum
For the physical Majoranas, and . Using the vectorization and auxiliary fermions of Appendix A 2 gives
The intersite hopping initially has sign . Set and , . Since , this changes to while leaving , the occupations, the endpoint terms, and the global operator parity unchanged.
Reorder by , and , . Each physical site contributes one bond. Set , , and . The reordered blocks are
The operator-parity projection and spectral reconstruction derived in Appendix A 2 are unchanged, so Eq. (25) applies directly to these blocks. Their boundary terms encode the distribution of boundary dissipation between the paths established in Appendix B 1: for odd , has no dissipative boundary potential; for even , each block has one.
Magnetization dynamics
The covariance derivation and spectral expansion in Appendix A 3 apply without change. Only the Hamiltonian entries, initial covariance, and measured observable differ:
The remaining entries of follow by antisymmetry. The initial state is , and . For the mean magnetization, the observable matrix has . Substituting these quantities into Eqs. (40)–(42) gives the magnetization response; Eq. (43) bounds its decaying remainder.
For the cluster chain at , restrict to the initially populated sector and set and . The Hamiltonian of the isolated path is , and the all- preparation has . The measured even-sublattice magnetization is . These effective Pauli operators are representations within the indicated sectors.
The effective Hamiltonian has the same form as in the XY derivation, so Eq. (45) gives the result with .
Results for additional parameters
Figure 6 repeats the size comparison of Figure 4 at . The isolated-path assignment is unchanged: odd chains retain a non-decaying contribution, while even-chain magnetization relaxes. The changed coupling modifies the mean and oscillation amplitude while preserving the isolation from boundary dissipation.
Figure 6. Cluster magnetization at , , and . The initial state is . (a) . Inset: . (b) . Inset: .
For the large chains at this coupling ratio, the time window ends at , and the transient envelope in each final inset is below .
Unequal boundary dissipation rates
The mechanism of isolation from boundary dissipation requires positive endpoint dissipation rates, but not equality between them. We illustrate this with the cluster chain with at , , , and . Changing the rates leaves the Majorana coordinates subject to boundary dissipation unchanged and therefore preserves which paths are isolated. For odd , the isolated path and its Hamiltonian are unchanged, so its persistent oscillation frequencies and non-decaying magnetization contribution coincide with those at equal rates. The paths subject to boundary dissipation acquire different decay rates.
Figure 7(a,Figure 7b) show the rapidity spectra for . The odd chain retains four purely imaginary rapidities, consistent with , whereas every rapidity of the even chain has negative real part. Figure 7(c,Figure 7d) show the corresponding full Liouvillian spectra. The odd chain supports nonzero purely imaginary eigenvalues alongside stationary modes at the origin, whereas only stationary modes remain non-decaying in the even chain. Thus unequal positive rates preserve the same spectral distinction as equal rates.
Figure 7. Cluster spectra at , , , and . Columns: . (a,b) Rapidity spectra. (c,d) Liouvillian spectra.
Figure 8 gives the corresponding dynamics from the all- product state for both small and large chains. The signal still contains a visible decaying contribution in the short-time inset. For , the logarithmic time window extends to . The transient envelope in the final inset is below for both sizes. The odd-chain oscillations survive because the isolated paths remain unaffected by the unequal boundary dissipation rates.
Figure 8. Cluster magnetization at the unequal boundary dissipation rates of Figure 7. The initial state is . (a) . Inset: . (b) . Inset: .
Cluster chains with even interaction ranges
Model and boundary dissipation
For Eq. (13), the physical Jordan–Wigner transformation gives
The path labeled contains sites , , up to , with alternating on-site and intersite links. For even , each path contains only odd-indexed or only even-indexed physical sites, which underlies the choice of boundary jump operators used here. For odd , successive sites on a path alternate between odd and even indices, and the endpoint assignment depends on the residue classes modulo . The same construction therefore does not directly yield the odd-even dependence established above. We restrict the extension of the cluster family to even and leave the classification of boundary channels for odd ranges to future work.
The jumps of Eq. (14) obey
The right jump can therefore be replaced by the associated Majorana only when evaluating the dissipative action on operators with even operator parity. The rapidity spectrum matrix below describes the evolution of even bilinears. Even products of its isolated Majorana coordinates generate non-decaying observables. The covariance equation retains the form of Eq. (9), with
Here , select the corresponding physical Majorana coordinates, and is defined by . In particular, and , with the opposite entries fixed by antisymmetry.
Spectrum and dynamics
For odd , both sets of boundary dissipators act on paths on odd sites. For even , they act on complementary sets of paths. Equation (6) applies to with . Since , endpoint recursion excludes every path subject to boundary dissipation from the undamped subspace, yielding Eq. (17).
For these jumps, the Heisenberg-picture evolution preserves the number of Majorana factors in an even product. For even , the adjoint Liouvillian restricted to the subspace spanned by products of distinct Majorana operators has eigenvalues given by sums of distinct rapidities. Any sum involving a damped mode has negative real part. Thus the non-decaying operators with even operator parity are built entirely from isolated modes, together with the identity.
Figure 9 shows the rapidity spectra for at . For each range, all rapidities of the even chain have negative real parts, whereas the odd chain has 50 purely imaginary rapidities, counting multiplicities. These counts agree with and , respectively. Even products of the isolated modes evolve without dissipation and can contribute to the magnetization response.

Figure 9. Cluster rapidities for even interaction ranges at , , and . Rows: . Columns: . Blue/green denote damped/undamped modes. Here is the number of undamped modes, including degeneracies.
Figure 10 supplies the corresponding dynamical evidence at identical parameters, using the all- initial state. As in the cluster chain with , even-site magnetization accesses the isolated paths of odd chains. The large-system windows extend to for and for . In each final inset the absolute transient envelope is below for both sizes, exposing the odd-chain oscillations after the damped contribution has become negligible.
Figure 10. Cluster magnetization for even interaction ranges from at the parameters of Figure 9, comparing . (a) . Inset: . (b) . Inset: .
References
References
- [1]J. Eisert, M. Friesdorf, and C. Gogolin, Quantum many-body systems out of equilibrium, Nat. Phys. 11, 124 (2015).DOI
- [2]C. Gogolin and J. Eisert, Equilibration, thermalisation, and the emergence of statistical mechanics in closed quantum systems, Rep. Prog. Phys. 79, 056001 (2016).DOI
- [3]H.-P. Breuer and F. Petruccione, The Theory of Open Quantum Systems (Oxford University Press, Oxford, 2002), doi:10.1093/acprof:oso/9780199213900.001.0001.
- [4]D. Manzano, A short introduction to the Lindblad master equation, AIP Adv. 10, 025106 (2020).DOI
- [5]R. Fazio, J. Keeling, L. Mazza, and M. Schirò, Many-body open quantum systems, SciPost Phys. Lect. Notes 99 (2025).DOI
- [6]L. M. Sieberer, M. Buchhold, J. Marino, and S. Diehl, Universality in driven open quantum matter, Rev. Mod. Phys. 97, 025004 (2025)
- [7]G. T. Landi, D. Poletti, and G. Schaller, Nonequilibrium boundary-driven quantum systems: Models, methods, and properties, Rev. Mod. Phys. 94, 045006 (2022).DOI
- [8]T. Prosen and I. Pižorn, Quantum Phase Transition in a Far-from-Equilibrium Steady State of an XY Spin Chain, Phys. Rev. Lett. 101, 105701 (2008).pubmed.ncbi.nlm.nih.gov/18851227
- [9]M. Yao, A. Lingenfelter, R. Belyansky, D. Roberts, and A. A. Clerk, Hidden Time Reversal in Driven X X Z Spin Chains: Exact Solutions and New Dissipative Phase Transitions, Phys. Rev. Lett. 134, 130404 (2025).DOI
- [10]T. Prosen, Third quantization: a general method to solve master equations for quadratic open Fermi systems, New J. Phys. 10, 043026 (2008).DOI
- [11]T. Prosen, Spectral theorem for the Lindblad equation for quadratic open fermionic systems, J. Stat. Mech. (2010) P07020.DOI
- [12]T. Barthel and Y. Zhang, Solving quasi-free and quadratic Lindblad master equations for open fermionic and bosonic systems, J. Stat. Mech. (2022) 113101.DOI
- [13]F. Thompson and A. Kamenev, Field theory of many-body Lindbladian dynamics, Ann. Phys. 455, 169385 (2023).DOI
- [14]K. Yamanaka and T. Sasamoto, Exact solution for the Lindbladian dynamics for the open XX spin chain with boundary dissipation, SciPost Phys. 14, 112 (2023).arxiv.org/abs/2104.11479
- [15]N. Shibata and H. Katsura, Dissipative spin chain as a non-Hermitian Kitaev ladder, Phys. Rev. B 99, 174303 (2019).DOI
- [16]N. Shibata and H. Katsura, Dissipative quantum Ising chain as a non-Hermitian Ashkin-Teller model, Phys. Rev. B 99, 224432 (2019).DOI
- [17]N. Shibata and H. Katsura, Quantum Ising chain with boundary dephasing, Prog. Theor. Exp. Phys. (2020) 12A108.DOI
- [18]Z.-Y. Zheng, X. Wang, and S. Chen, Exact solution of the boundary-dissipated transverse field Ising model: Structure of the Liouvillian spectrum and dynamical duality, Phys. Rev. B 108, 024404 (2023).DOI
- [19]A. Teretenkov and O. Lychkovsky, Exact dynamics of quantum dissipative XX models: Wannier–Stark localization in the fragmented operator space, Phys. Rev. B 109, L143002 (2024).DOI
- [20]X. Wang and S. Chen, Closed dynamical recursion equations for correlation functions and their application for the construction of the Liouvillian spectrum in Lindbladian systems, Phys. Rev. B 111, 094305 (2025).DOI
- [21]F. Minganti, A. Biella, N. Bartolo, and C. Ciuti, Spectral theory of Liouvillians for dissipative phase transitions, Phys. Rev. A 98, 042118 (2018).DOI
- [22]M. Žnidarič, Relaxation times of dissipative many-body quantum systems, Phys. Rev. E 92, 042143 (2015).DOI
- [23]T. Mori and T. Shirai, Resolving a Discrepancy between Liouvillian Gap and Relaxation Time in Boundary-Dissipated Quantum Many-Body Systems, Phys. Rev. Lett. 125, 230604 (2020).DOI
- [24]T. Mori and T. Shirai, Symmetrized Liouvillian Gap in Markovian Open Quantum Systems, Phys. Rev. Lett. 130, 230404 (2023).DOI
- [25]B. Zhou, X.Wang, and S. Chen, Exponential size scaling of the Liouvillian gap in boundary-dissipated systems with Anderson localization, Phys. Rev. B 106, 064203 (2022).DOI
- [26]V. V. Albert and L. Jiang, Symmetries and conserved quantities in Lindblad master equations, Phys. Rev. A 89, 022118 (2014).DOI
- [27]V. V. Albert, B. Bradlyn, M. Fraas, and L. Jiang, Geometry and Response of Lindbladians, Phys. Rev. X 6, 041031 (2016).DOI
- [28]F. Song, S. Yao, and Z. Wang, Non-Hermitian Skin Effect and Chiral Damping in Open Quantum Systems, Phys. Rev. Lett. 123, 170401 (2019).DOI
- [29]T. Haga, M. Nakagawa, R. Hamazaki, and M. Ueda, Liouvillian Skin Effect: Slowing Down of Relaxation Processes without Gap Closing, Phys. Rev. Lett. 127, 070402 (2021).DOI
- [30]F. Yang, Q.-D. Jiang, and E. J. Bergholtz, Liouvillian skin effect in an exactly solvable model, Phys. Rev. Research 4, 023160 (2022).DOI
- [31]P. Zanardi and M. Rasetti, Noiseless Quantum Codes, Phys. Rev. Lett. 79, 3306 (1997).DOI
- [32]D. A. Lidar, I. L. Chuang, and K. B. Whaley, Decoherence-Free Subspaces for Quantum Computation, Phys. Rev. Lett. 81, 2594 (1998).DOI
- [33]D. Nigro, On the uniqueness of the steady-state solution of the Lindblad–Gorini–Kossakowski–Sudarshan equation, J. Stat. Mech. (2019) 043202.DOI
- [34]B. Buča and T. Prosen, A note on symmetry reductions of the Lindblad equation: transport in constrained open spin chains, New J. Phys. 14, 073007 (2012).DOI
- [35]A. Pocklington, Y.-X. Wang, Y. Yanay, and A. A. Clerk, Stabilizing volume-law entangled states of fermions and qubits using local dissipation, Phys. Rev. B 105, L140301 (2022).DOI
- [36]S. Andréys, Repeated Interaction Processes in the Continuous-Time Limit, Applied to Quadratic Fermionic Systems, Ann. Henri Poincaré 21, 115 (2020).DOI
- [37]A. Yu. Kitaev, Unpaired Majorana fermions in quantum wires, Phys. Usp. 44, 131 (2001).
- [38]S. Diehl, E. Rico, M. A. Baranov, and P. Zoller, Topology by dissipation in atomic quantum wires, Nat. Phys. 7, 971 (2011).DOI
- [39]P. Fendley, Strong zero modes and eigenstate phase transitions in the XYZ/interacting Majorana chain, J. Phys. A: Math. Theor. 49, 30LT01 (2016).DOI
- [40]J. Kemp, N. Y. Yao, C. R. Laumann, and P. F. Fendley, Long coherence times for edge spins, J. Stat. Mech. (2017) 063105.DOI
- [41]J. Kemp, N. Y. Yao, and C. R. Laumann, Symmetry-Enhanced Boundary Qubits at Infinite Temperature, Phys. Rev. Lett. 125, 200506 (2020).DOI
- [42]S. Lieu, M. McGinley, and N. R. Cooper, Tenfold Way for Quadratic Lindbladians, Phys. Rev. Lett. 124, 040401 (2020).DOI
- [43]Z.-Y. Zheng, X. Wang, and S. Chen, Characterizing dynamical behavior in topological open systems with boundary dissipation, Phys. Rev. B 111, 214304 (2025).DOI
- [44]F. Yang, P. Molignini, and E. J. Bergholtz, Dissipative boundary state preparation, Phys. Rev. Research 5, 043229 (2023).DOI
- [45]Y. T. Wang and X. Z. Zhang, From local weight selection to Zeno slowdown in an open Su-Schrieffer-Heeger chain with a single local loss, Phys. Rev. A 114, 012457 (2026).arxiv.org/abs/2607.10218
- [46]B. Buča, J. Tindall, and D. Jaksch, Non-stationary coherent quantum many-body dynamics through dissipation, Nat. Commun. 10, 1730 (2019).DOI
- [47]J. Tindall, C. Sánchez Muñoz, B. Buča, and D. Jaksch, Quantum synchronisation enabled by dynamical symme tries and dissipation, New J. Phys. 22, 013026 (2020).
- [48]B. Buća, C. Booker, and D. Jaksch, Algebraic theory of quantum synchronization and limit cycles under dissipation, SciPost Phys. 12, 097 (2022).DOI
- [49]C. W. Wächtler and J. E. Moore, Topological Quantum Synchronization of Fractionalized Spins, Phys. Rev. Lett. 132, 196601 (2024).
- [50]C. Matsui and N. Tsuji, Boundary dissipative spin chains with partial solvability inherited from system Hamiltonians, Phys. Rev. B 110, 224306 (2024).DOI
- [51]C. Booker, B. Buća, and D. Jaksch, Non-stationarity and dissipative time crystals: spectral properties and finite-size effects, New J. Phys. 22, 085007 (2020).DOI
- [52]S. Dutta, S. Zhang, and M. Haque, Quantum Origin of Limit Cycles, Fixed Points, and Critical Slowing Down, Phys. Rev. Lett. 134, 050407 (2025).DOI
- [53]S.-Y. Zhang, M. Gong, G.-C. Guo, and Z.-W. Zhou, Anomalous relaxation and multiple timescales in the quantum XY model with boundary dissipation, Phys. Rev. B 101, 155150 (2020).DOI
- [54]A. Lingenfelter, M. Yao, A. Pocklington, Y.-X. Wang, A. Irfan, W. Pfaff, and A. A. Clerk, Exact Results for a Boundary-Driven Double Spin Chain and Resource-Efficient Remote Entanglement Stabilization, Phys. Rev. X 14, 021028 (2024).DOI
- [55]V. Gorini, A. Kossakowski, and E. C. G. Sudarshan, Completely positive dynamical semigroups of N-level systems, J. Math. Phys. 17, 821 (1976).DOI
- [56]G. Lindblad, On the generators of quantum dynamical semigroups, Commun. Math. Phys. 48, 119 (1976).DOI
- [57]R. Raussendorf and H. J. Briegel, A One-Way Quantum Computer, Phys. Rev. Lett. 86, 5188 (2001).
- [58]A. C. Doherty and S. D. Bartlett, Identifying Phases of Quantum Many-Body Systems That Are Universal for Quantum Computation, Phys. Rev. Lett. 103, 020506 (2009).pubmed.ncbi.nlm.nih.gov/19659193
- [59]Z.-Y. Zheng, Dataset, Zenodo (2026); 10.5281/zenodo.24334198.






