(A)Symmetric Complexity and the Quantum Mpemba Effect Cameron Beetar, Jeff Murugan and Hendrik J.R. Van Zyl Abstract The Quantum Mpemba Effect (QME) — the counter-intuitive phenomenon where states further from equilibrium can relax faster than those closer to it — challenges standard expectations of quantum thermalization. In this work, we introduce Krylov complexity as a sensitive diagnostic for the QME. We show that Krylov spread complexity encodes the asymmetry essential to the effect, and we define a new class of projective (a)symmetric complexities that sharpen this connection. Strikingly, the structure of these projective complexities at the initial moment (t=0) already carries predictive power for the onset of Mpemba-like inversions, obviating the need for explicit time evolution. Our results suggest that the geometry of states in Krylov space captures deep information about non-monotonic relaxation and provides a powerful framework for diagnosing and anticipating anomalous thermalization phenomena in quantum systems. 1 Introduction The paradoxical phenomenon where hot liquid can, under certain conditions, freeze faster than cold liquid – known as the Mpemba effect Mpemba1969 – has long confused scientists and the lay public alike. Recently though, interest in its quantum analogue, the Quantum Mpemba Effect (QME) Ares:2025onj , has surged, driven by its potential to probe nontrivial relaxation dynamics in complex quantum systems. Unlike classical counterparts, where intuitive notions of thermal gradients drive equilibration, quantum systems carry additional layers of structure, such as conserved quantities and sectorial symmetries, which can intricately modulate the route to equilibrium. Capturing such phenomena requires tools sensitive not only to population imbalances but to coherence structures between symmetry sectors. This paper explores how quantum complexity measures, particularly Entanglement Asymmetry Ares2023 and Krylov Complexity Parker:2018yvk ; Nandy:2024evd ; Rabinovici:2025otw ; Baiguera:2025dkc , serve as diagnostic tools for the QME, revealing hidden layers of dynamical behavior. The entanglement asymmetry, introduced in Ares2023 , has emerged as one way to quantify the distance between a reduced density matrix \rho_{A} of a subsystem A and its symmetry-projected counterpart \rho_{A,Q} , defined via the relative entropy \Delta S=S(\rho_{A,Q})-S(\rho_{A}), (1) where S(\rho)=-\text{Tr}(\rho\ln\rho) is the von Neumann entropy of the system and the symmetry-projected density matrix, \rho_{A,Q}=\sum_{q}\Pi_{q}\rho_{A}\Pi_{q}, (2) is obtained by projecting \rho_{A} onto the charge sectors labeled by q associated with a conserved symmetry, with projectors \Pi_{q} . If \rho_{A} is already block-diagonal in these sectors, \Delta S vanishes. Otherwise, \Delta S measures the asymmetry or coherence across these sectors. Importantly, \Delta S is a divergence, rather than a true distance—it is non-negative, zero only when \rho_{A}=\rho_{A,Q} , and lacks both the symmetric property and the triangle inequality Nielsen:2012yss . In the context of the Quantum Mpemba Effect Ares:2025onj , the entanglement asymmetry serves as a dynamic probe to track how coherences across symmetry sectors evolve over time. Specifically, for two initial states \rho_{1} and \rho_{2} , we say that a Mpemba-like swap occurs if: 1. Initially, \Delta S(\rho_{1})<\Delta S(\rho_{2}) at t=t_{0} , but 2. There exists a later time t_{M}>t_{0} where \Delta S(\rho_{1})>\Delta S(\rho_{2}) . Note that this inversion of the entanglement asymmetry ordering implies that the plots of \Delta S(\rho_{1}(t)) and \Delta S(\rho_{2}(t)) cross at some point in time. A stronger version of the QME demands that this inversion persists for all t\geq t_{M} . Notably, vanishing entanglement asymmetry is necessary – but not sufficient – for thermal equilibrium, making it a subtle yet informative diagnostic for nonequilibrium dynamics. On the other hand, Krylov Complexity Parker:2018yvk ; Nandy:2024evd ; Rabinovici:2025otw , or more specifically “spread complexity” Balasubramanian:2022tpr , quantifies how a state spreads in its Krylov basis under time evolution. Originally applied to operator dynamics, its adaptation to state evolution has proven sensitive to phase distinctions, such as thermalization-localization transitions RevModPhys.91.021001 and \mathsf{PT} -symmetry breaking Beetar:2023mfn ; Bhattacharya:2024hto . Like entanglement asymmetry, Krylov complexity is non-symmetric and lacks a triangle inequality Aguilar-Gutierrez:2023nyk , but it faithfully measures the “distance” (in complexity space) between a fixed initial state and its time-evolved form. The interplay between entanglement asymmetry and Krylov complexity is nuanced. The former is sensitive to coherences between symmetry sectors, while the latter captures the overall spreading – both within and across sectors. This distinction becomes crucial in systems where dynamics exhibit both diffusive motion within symmetry sectors and hopping between them. Prior investigations Caputa:2025mii approached Krylov complexity by summing over sector complexities, accurately capturing intra -section diffusion but missing some of the nuances of inter -sector hopping – a crucial component for detecting Mpemba-like anomalous relaxation. In this work, we propose a complementary approach; while entanglement asymmetry naturally isolates inter-sector hopping dynamics, we develop a complexity formalism designed to appropriately weight intra-section diffusion contributions, aligning the sensitivity of Krylov complexity to the “asymmetry-hopping” dynamics detected by entanglement asymmetry. This alignment not only sharpens the former’s utility in probing QME but also bridges complexity-theoretic perspectives with symmetry-resolved quantum dynamics. As a concrete setting, we focus on the Aubry-André (AA) model Aubry1980 , a one-dimensional crystal with periodically varying onsite energies exhibiting a tunable transition between Eigenstate Thermalization Hypothesis Deutsch:2018ulr (ETH)-dominated phases and Many-Body Localized (MBL) Nandkishore:2014kca ; Altman:2014fhm regimes. The Hamiltonian is given by H=-\sum_{i}^{N}\left(\sigma_{i}^{x}\sigma_{i+1}^{x}+\sigma_{i}^{y}\sigma_{i+1}^{y}\right)+V\sum_{i=1}^{N}\sigma_{i}^{z}\sigma_{i+1}^{z}+\sum_{i=1}^{N}W_{i}\sigma_{i}^{z}, (3) where V controls nearest-neighbour interactions and W_{i}=W\cos(2\pi\alpha i+\phi) represents a quasiperiodic potential with \alpha=(\sqrt{5}+1)/2 and random phase \phi . The nearest-neighbour interaction strength is set to V=1/2 throughout this work. The model hosts a well-characterized ETH/MBL transition around W_{c}\approx 3.3 (see appendix A and liu2024 ). The global U(1) symmetry corresponding to total magnetization Q=\sum_{i}\sigma_{i}^{z} partitions the Hilbert space into charge sectors, furnishing a natural playground for exploring entanglement asymmetry and Krylov complexity in tandem. In this paper, we demonstrate that a suitably refined Krylov complexity can serve as order parameters for the quantum Mpemba effect in the AA model, much like entanglement asymmetry \Delta S liu2024 . We show that identical (pre-quench) initial states have differing asymmetry profiles due to the quench (tilt) operation, and can exhibit Mpemba-like inversions in their relaxation dynamics. Furthermore, we elucidate how the ETH/MBL transition modulates the prevalence and timescales of these inversions, with the MBL regime enhancing long-lived asymmetry-induced dynamical memory effects. The rest of this article is as follows: In Section 2 , we formalize the relationship between entanglement asymmetry and Krylov complexity, highlighting their complementary sensitivities. Section 3 outlines in detail the notion of symmetry-resolved Krylov complexity and the required modification of this measure, and presents our main finding: an Mpemba-like crossing in the symmetric complexity. The following section 4 investigates the initial values of the (a)symmetric complexity and Lanczos coefficients, and analyzes their usefulness as probes of Mpemba inversions. We conclude in Section 5 with a broader discussion of the implications for complexity-theoretic perspectives on nonequilibrium quantum dynamics. 2 Entanglement Asymmetry and Krylov Complexity Entanglement asymmetry has proven to be a valuable probe of the quantum Mpemba effect, particularly in situations where a broken symmetry is approximately restored under time evolution Ares:2025onj ; Ares2023 . Unlike its classical counterpart, however, the quantum version is more nuanced. At least two distinct definitions arise, depending on whether one prioritizes (i) the time required to reach a steady state 1 1 1 This differs from the classical case, where the final state is strict thermal equilibrium. Quantum systems may not exhibit such equilibration, but for our purposes we assume the existence of a steady state. or (ii) the extent of symmetry restoration. The strongest form of the QME would demand both: the more asymmetric (“hotter”) state not only crosses the less asymmetric (“colder”) one but also reaches its steady state more rapidly. In this work, however, we adopt a definition that emphasizes the crossing in entanglement asymmetry as the primary diagnostic, since in practice the steady states typically occur on very similar timescales 2 2 2 This definition is also employed in ares2025qme , where the authors state: “In terms of symmetries, the quantum Mpemba effect happens when the more the symmetry is broken, the faster it is restored.” . A weaker formulation is also conceivable, in which symmetry restoration plays no role and the effect is defined purely in terms of time-to-steady-state. These distinctions arise because, in the classical setting, trajectories converge to a common macroscopic equilibrium, while in the quantum case distinct initial states generally evolve toward different late-time configurations 3 3 3 Moreover, quantum steady states are subject to fluctuations. . Under unitary dynamics, the role of the asymmetry operator is to induce an effective temperature that depends on the interplay of the dynamics, the initial state, and the operator itself. Within this framework, entanglement asymmetry serves as a positive semi-definite quantity that measures the distance from a symmetric configuration. Nevertheless, the subtleties of defining and interpreting the QME in the quantum regime require careful consideration. Krylov Parker:2018yvk ; Nandy:2024evd ; Rabinovici:2025otw ; Baiguera:2025dkc , or more precisely, spread complexity Balasubramanian:2022tpr appears, at first sight, to be a natural probe of the QME for several reasons. Unlike entanglement asymmetry, which requires the arbitrary choice of a subsystem, Krylov complexity provides a global characterization of quantum dynamics: spread complexity measures how the entire wavefunction explores the full accessible Hilbert space. Its evolution is determined solely by the Lanczos coefficients a_{n} , the diagonal entries of the Krylov Hamiltonian which encodes local onsite potentials in the Krylov chain, and b_{n} , the off-diagonals of the Krylov Hamiltonian which encode the microscopic rate of this spreading Lanczos1950AnIM ; viswanath2008recursion . Moreover, the initial asymmetry quench produces a nontrivial superposition that serves as the seed of the Lanczos recursion. Since each initial state generates a distinct set of Lanczos coefficients and thus a unique trajectory in Krylov space, the corresponding complexity C_{K}(t) directly encodes the relaxation pathway. As we will show below, this makes Krylov complexity particularly sensitive to Mpemba-like inversions, which are defined by anomalous reorderings of relaxation trajectories between different initial conditions. Within the Aubry–André model ( 3 ), the presence of the ETH/MBL transition offers an ideal setting in which to compare entanglement asymmetry and Krylov complexity as complementary probes of the QME in both thermalizing and localizing regimes. To this end, we examined two extreme initial product states. The fully polarized ferromagnetic state \ket{\uparrow\uparrow\cdots\uparrow} , which lies at the edge of the symmetry spectrum ( S_{z}^{\mathrm{tot}}=N ), explores a sparse region of Hilbert space. Its constrained dynamics make it a natural candidate for exhibiting non-chaotic relaxation and enhanced susceptibility to anomalous effects such as the QME, across all values of the quasiperiodic potential strength W . By contrast, the Néel state \ket{\uparrow\downarrow\cdots\uparrow\downarrow} resides in the central sector ( S_{z}^{\mathrm{tot}}=0 ), the largest and densest region of Hilbert space. States in this sector are expected to display strongly chaotic behavior, thereby providing a complementary probe of how the QME manifests in the ergodic regime. To investigate the QME, it will be necessary to explicitly break the symmetry of the system. Toward this end, we employ the conventional tilt operation Ares2023 ; liu2024 , \ket{\psi(\theta)}=\exp\left(-i\frac{\theta}{2}\sum_{i}\sigma_{i}^{y}\right)\ket{\psi(0)}\equiv U(\theta)\ket{\psi(0)}, (4) with the initial state \ket{\psi(0)} taken to be either the ferromagnetic or Néel states above. The resulting one-parameter families of states will be referred to as Tilted Ferromagnetic States (TFS) and Tilted Néel States (TNS), respectively. The AA Hamiltonian preserves a global U(1) symmetry associated with the total magnetization, since [H,\sum_{i}\sigma_{i}^{z}]=0 . However, the tilt operator U(\theta) does not commute with this conserved charge. Consequently, in the presence of tilting, \theta functions as a continuous asymmetry-inducing parameter. Within the interval \theta\in[0,\pi/2] , it monotonically increases the initial entanglement asymmetry of the state. For clarity of presentation in all figures, we rescale \theta by a factor of \pi , so that \theta denotes a fraction of \pi (e.g., \theta=0.1 corresponds to a tilt angle of 0.1\pi ). 2.1 Entanglement Asymmetry The QME in the Aubry–André model has recently been studied in liu2024 . In our work, the entanglement asymmetry serves as the baseline probe for defining the QME, and reproducing its behavior is essential both for benchmarking against previous results as well as for comparison with Krylov complexity. For completeness, and to verify the accuracy of our numerics, we summarize the key features of the entanglement asymmetry-based analysis here. We prepare the system in two tilted initial states, \ket{\psi(\theta_{1})} and \ket{\psi(\theta_{2})} , and evolve them in time. A QME is said to occur whenever the entanglement asymmetry curves for the two states exhibit a crossing. Unless otherwise stated, our simulations use system size N=12 , with subsystem size N_{A}=3 (where required), averaged over 240 realizations of the random phase \phi , sampled from the interval [0,2\pi] using stratified sampling to ensure uniform coverage. Figure 1 demonstrates that, for tilted ferromagnetic states, entanglement asymmetry crossings are observed for all reasonable choices of tilt angles (e.g., \theta_{1}=0.2,\theta_{2}=0.5 ), independent of the quasiperiodic potential strength W . This robustness extends across the ETH/MBL transition, centered near the critical potential W_{c}\approx 3.3 . For WW_{c} it resides in the MBL regime. In contrast, the behavior of tilted Néel states is more subtle. Deep in the ETH regime (small W ), no QME is observed. As W increases toward the MBL regime – but remains below W_{c} – crossings appear for some pairs of initial tilts, though not universally (see Fig. 1d ). Once W exceeds the critical value, the QME is observed consistently for all pairs of tilted Néel states. In both ferromagnetic and Néel cases, increasing W delays the approach to steady state; however, given the persistence of the crossings over long time windows (typically \Delta t\approx 9\times 10^{4} ), we interpret them as robust indicators of the QME, even without full convergence to late times. To summarise: • For tilted ferromagnetic states, the QME is observed for all W . • For tilted Néel states, the QME is observed for all W>W_{c} (MBL regime). • For WW_{c} shows how MBL effects cause a reduction in the rate of spread of the state. Conversely, the ferromagnetic state never experiences the ‘encouraged’ spreading at W3.3 ) is highlighted when comparing the thermomajorization curves of Fig. 25 to the survival probabilities shown in Fig. 11 . The observed behavior of the thermomajorization curves in the MBL regime (at W=5.0 ) is that larger tilts create states nearer to the (hypothetical) Gibbs state. On the other hand, the survival probabilities indicate that states with larger tilts reach steady states with which they have a lower overlap, and which, unsurprisingly, are decidedly athermal. Conversely, the strength of this measure is that, in the ETH regime, it provides a measure of distance from the thermal state to which the system will generally converge. For the tilted ferromagnetic states, larger tilts (up to \theta=0.5 ) always correspond to a population distribution that is closer to that of the thermal one. The Néel state is different in that it exhibits non-monotonicity of the tilt operation for potential strengths W<1.0 , whereafter it is also monotonically decreasing as a function of the tilt. This data highlights the challenge of using entanglement asymmetry as a measure of distance from thermal (steady) state, especially in the ETH phase. While it is true that the tilt operation does increase the asymmetry as measured by the entanglement asymmetry, it is not necessarily true that this asymmetry is aligned with the population of states being more distant from the thermal one. What this analysis also highlights is an additional strength of the Krylov complexity as a probe of the QME in localizing systems; a region where thermomajorisation is not readily applicable as a meaningful measure, since the state will converge to a localized one, with a population distinct from the thermal one. D.7 Effects of averaging Lanczos Coefficients In sec. 4.2 , we present Fig. 10 , and study the behavior of the minimum and maximum of the b_{1} values in the (\theta,W) parameter space. In this appendix, we show how the behaviour of these minima and maxima are altered as more b_{n} are included, corresponding to studying the average spread in Krylov space over increasing timescales. We will focus on the tilted Néel states, as the tilted ferromagnetic states display comparatively simple behavior, being approximately totally ordered by n_{max}=10 . (a) TNS, n_{max}=5 (b) TNS, n_{max}=10 (c) TNS, n_{max}=14 (d) TNS, n_{max}=28 Figure 27: Figures showing the heatmaps of the average of the first n Lanczos coefficients in (\theta,W) parameter space. Each b_{i} is averaged over 240 realisations for an N=12 system, then the average of the first n_{max} of these values is computed. The effect of including additional Lanczos coefficients is clear – it drives down the location of the minimum, eventually aligning with the minimum tilt for 28\leq n_{max} . The stability of the location in the potential of the sudden maximum shift is interesting, lying between W=3.6 and W=4.1 for tested values of n_{max}\in[1,50] . It is interesting to note that the sudden shift of the minimum and the maximum overlaps at n=14 , where the jump occurs going from W=3.8 to W=3.9 .