{"id":"7e1c0f90-f76e-4f37-86db-2b37ad1bdd8d","arxiv_id":"1908.03868","paper_version":3,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":5.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":2,"one_line_summary":"In two weakly coupled chaotic Bose-Hubbard trimers, quantum saturation profiles depend on the initial energy imbalance, a weak dynamical localization effect that is absent in the corresponding classical dynamics.","lead":"Two weakly coupled quantum chaotic subsystems can fail to fully thermalize and instead retain a memory of their initial preparation. The authors demonstrate this in a minimal pair of three-site Bose-Hubbard units, offering a new way to detect dynamical localization without comparing to classical dynamics.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The non-perturbative claim rests on participation numbers M computed from x0-window-averaged saturation profiles; perturbative M≈2 states would be broadened by the averaging, so a single-state M check is required.","rationale":"The reader's weakest assumption was finite-size scaling, which is a legitimate generality concern. I depart from it because the more immediate failure mode is internal to the presented diagnostic: the quantity that encodes \"non-perturbative\" in the strongest claim is M, and the way M is computed is confounded by the x0-local averaging used for smoothness. A single-state recomputation is cheap and decisive. If M is inflated, the nontriviality claim collapses even at N=24, so this concern is load-bearing regardless of finite-size behavior. I would keep the overall verdict conditional rather than reject because Fig. 3 shows broad saturation profiles that may survive single-state analysis, and the authors should be allowed to provide the direct M data. The finite-size check remains worth adding, but it is secondary to the diagnostic issue. The paper otherwise has independent support: the classical diffusion coefficient in Eq. (17) is validated against semiclassical simulations in Fig. 10, and the intrinsic, correspondence-free definition of localization is a useful methodological contribution.","tokens_in":18559,"tokens_out":10580,"duration_ms":124175,"concrete_test":"Recompute M for the E0=0.741, v=0.1 and v=0.2 data of Fig. 8 without any x0/E0 averaging: evaluate Psat(X) from Eq. (14) for each individual initial state |E0,x0⟩, apply Eq. (18) to that single-state profile, and then report the median and scatter of the single-state M values across the x0 window. If the single-state M is about 2 in the regions claimed as non-perturbative while the averaged M shown in Fig. 8(a) is much larger than 2, the non-perturbative claim is an artifact of local averaging; if single-state M is also much larger than 2, the concern is resolved and Fig. 8 should be replotted with single-state M.","verdict_should_be":"CONDITIONAL","load_bearing_attack":"The central claim in Sec. VII is that localization persists \"even when the coupling is not trivially small, so that M is large.\" The only quantitative support is the participation number M of Eq. (18), shown in Fig. 8(a). The caption states that M is evaluated from saturation profiles that were \"locally averaged over states within E=E0±0.02 and x=x0±0.025.\" This averaging is done before Eq. (18) is applied. In a perturbatively localized regime each exact initial state has a saturation profile peaked at its own x0, with a mirror peak at −x0, so the single-state M is close to 2. Averaging over the x0 window mixes several such peaks into one distribution, producing an M that can be substantially larger than 2 even though no non-perturbative mixing occurred. The window width 0.025 is not negligible compared with the claimed quantum-ergodic interval |x0|<0.25, so the artifact could affect exactly the region used to conclude non-perturbative behavior. The paper also states in Sec. VI that the break-time mechanism is not re-derived for this system, so there is no independent argument that rules out the artifact. Consequently the headline distinction from trivial perturbative localization is not established by the data as presented.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper studies dynamical localization in a closed system of two weakly coupled Bose-Hubbard trimers, each of which is classically chaotic in a substantial part of its spectrum. The authors define ergodicity intrinsically, as the loss of memory of the initial trimer energy imbalance x0 in the long-time saturation profile, and propose this as a notion of quantum localization that does not require comparison with classical dynamics. Using exact diagonalization for N=24 particles per trimer, they find that quantum saturation profiles retain x0 memory in most of the chaotic sea, even when the participation number M of Eq. (18) is large. The central claim, stated in Section VII, is that this constitutes a nontrivial many-body dynamical localization, distinct from perturbative localization (M≈2), from classical self-trapping, and from classical KAM-type localization. The paper also derives a semiclassical diffusion coefficient D(x) in Eq. (17) and validates it against independent semiclassical simulations in Fig. 10.","tokens_in":18734,"tokens_out":4188,"duration_ms":45220,"significance":"If established, the result would provide a correspondence-free, intrinsic diagnostic of quantum dynamical localization in a high-dimensional composite chaotic system, and would support the relevance of such localization for thermalization in larger arrays of weakly coupled chaotic subunits. The paper has clear strengths: the classical diffusion coefficient in Eq. (17) is a parameter-free prediction that is checked against semiclassical simulations in Fig. 10; the quantum saturation profiles are obtained from exact diagonalization via Eq. (14), avoiding uncertainties of finite-time averaging; and the use of an initial-condition-memory measure is a principled way to avoid trivializing quantum-to-classical correspondence failures. However, the headline distinction from perturbative localization rests primarily on the participation number M, which is computed from locally averaged saturation profiles, and all quantum results are obtained at a single system size. Both points need to be addressed before the central claim can be considered established.","major_comments":[{"comment":"The participation number M in Eq. (18) is evaluated from saturation profiles that are locally averaged over states within E=E0±0.02 and x=x0±0.025, as stated in the captions of Figs. 5 and 8. If each individual initial state is perturbatively localized, its single-state saturation profile is peaked at its own x0 and, by parity, at -x0, giving M close to 2. Averaging over the x0 window mixes such peaks from different initial states into one broader distribution, and the resulting M can be substantially larger than 2 even though no non-perturbative mixing occurred. Since the x0-window width 0.025 is not negligible compared with the claimed quantum-ergodic interval |x0|<0.25, this artifact could affect exactly the region used to conclude non-perturbative behavior. The paper should present M computed separately for each single initial state, without the x0-window averaging, and show that M is large for those individual states before claiming non-perturbative quantum mixing.","section":"Section VI A, Fig. 8(a)"},{"comment":"All quantum simulations are performed with NL=NR=24 particles, and the effective Planck constant is ℏ_eff=1/Nα=2/N. The paper states in Section I that 'the thermodynamic limit is not an issue' for this interaction-induced dynamical localization, but no finite-size scaling is provided. Without varying N (for example, N=18, 24, 30, 36), one cannot exclude the possibility that the observed x0-memory and the small quantum-ergodic region diminish or disappear as ℏ_eff decreases, which would change the interpretation of the result as a robust many-body effect. Given that the title and abstract generalize beyond N=24, a finite-size analysis of M, f∞, and the size of the x0-independent region is necessary.","section":"Section I and Section II, Fig. 5"},{"comment":"The conclusion that the observed localization is nontrivial rests on the statement that M is large while x0-memory persists (Fig. 8 and Section VII). Section VI B explicitly states that the break-time analysis is not repeated for the double-trimer system, so there is no independent semiclassical argument that rules out an averaging artifact in M. The paper should supply a direct diagnostic of non-perturbative mixing that does not depend on the locally averaged profiles—for example, the overlap of the coupled-system eigenstates with the unperturbed product basis for a single initial state, or the spreading of a single-state saturation profile in x—so that the distinction from perturbative localization is established by the data rather than by an unverified expectation.","section":"Section VI B and Section VII"}],"minor_comments":[{"comment":"The phrase 'Fokker-Plank' appears twice and should be 'Fokker-Planck'.","section":"Throughout"},{"comment":"There is a typographical error in 'Whenever the the initial state' on page 10; it should read 'Whenever the initial state'.","section":"Section VI C"},{"comment":"In the definition of the combined subspaces, 'A = As + As' appears to be a typo; based on the four representations Ss, Sa, As, and Aa introduced earlier, it should likely read 'A = As + Aa'.","section":"Appendix D"},{"comment":"The definition of ft in Eq. (15) is missing a closing brace or a display alignment; the formula should be typeset so that ft = sum over positive Δt(x) is unambiguous.","section":"Eq. (15)"}],"recommendation":"major_revision","confidential_remarks":"The paper is within the scope of the journal and the intrinsic-ergodicity framework is a valuable contribution, but the referee is not convinced that the central claim—non-perturbative dynamical localization—is supported by the current data. The participation-number argument in Fig. 8 is vulnerable to the x0-window averaging artifact, and the absence of any finite-size scaling weakens the claim that the thermodynamic limit is irrelevant. If the authors can provide single-state participation numbers and a small finite-size study, the paper could become publishable; without those, the headline distinction from trivial perturbative localization is not established."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Dear colleague,\n\nBottom line: this is a careful numerical study of two weakly coupled chaotic Bose-Hubbard trimers, and it finds real initial-condition memory in the quantum saturation profiles. The effect is weak, and the paper is honest about that. The non-perturbative headline, however, rests partly on an averaged participation number, and that part does not survive close reading.\n\nWhat's genuinely new: the extension from the authors' trimer-monomer model to trimer-trimer, where both subsystems are chaotic. The intrinsic, correspondence-free definition of localization—memory of x0 rather than comparison to classical spreading—is a useful methodological step. The classical diffusion coefficient in Eq. (17) is derived from correlation functions and validated against independent semiclassical simulations (Fig. 10), which is the strongest piece of evidence in the paper. The authors also state clearly that they are not re-deriving the break-time mechanism here, so the argument is set up honestly, just incomplete.\n\nSoft spots, in order. First, no finite-size scaling. All quantum results use N=24 per trimer, and the paper says the thermodynamic limit 'is not an issue' without a scaling study. That is an assertion, not evidence. Second—and this is where I think the stress-test note lands—the participation number M in Fig. 8(a) is computed from saturation profiles that were locally averaged over E and x0 windows. In a perturbatively localized regime each exact state would have M≈2, and that sharp localization can be smeared out by the x0 averaging. So the summary claim that localization persists 'even when the coupling is not trivially small, so that M is large' is not actually established by the data as presented. A single-state M diagnostic, or an alternative measure not based on window-averaged profiles, would fix this. The other evidence (f∞ and ⟨|x|⟩∞ memory) does indicate x0-dependence, so the localization effect itself is not a mirage; what's weakened is the distinction from trivial perturbative localization. Third, no code or data and no error bars on the quantum f values. Minor relative to the above, but worth noting.\n\nWhom is this for? People working on thermalization in small Bose-Hubbard systems, quantum chaos, and the interplay of ETH with finite-size effects. The methodological point about intrinsic ergodicity measures is worth citing even if the central claim needs reinforcement.\n\nRecommendation: send it to peer review. The questions it raises are real and the paper is built with enough care that a serious referee can push for the missing checks. It should not be desk-rejected, but it should not be accepted in this form.","headline":"Careful numerical study of initial-condition memory in coupled chaotic trimers; the localization signal is real, but the non-perturbative claim leans on an averaged participation number and N=24 without scaling.","tokens_in":19321,"tokens_out":3012,"would_cite":true,"duration_ms":33934,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"Two weakly coupled chaotic quantum subsystems can retain memory of their initial state, so quantum thermalization is suppressed by nontrivial dynamical localization.","keywords":["dynamical localization","many-body thermalization","Bose-Hubbard trimer","memory of initial conditions","mobility edge","ergodicity","participation number","interaction-induced chaos"],"falsifier":"One decisive check is exact diagonalization at larger subsystem sizes, such as 30, 40, and 50 particles per trimer, with the same parameters: if the saturation profile becomes independent of the initial imbalance and the overlap measure drops toward the semiclassical value as the particle number grows, the many-body dynamical localization is a finite-size effect rather than a genuine phenomenon.","tokens_in":18281,"feed_emoji":"⚛️","tokens_out":6757,"duration_ms":73143,"temperature":0.7,"pith_summary":"This paper argues that quantum dynamical localization can appear in the thermalization of two weakly coupled chaotic subsystems, even when the coupling is not perturbatively small. The model is a pair of 3-site Bose-Hubbard trimers, the minimal experimentally relevant units that are chaotic on their own. Because the composite six-site system is high-dimensional, the localization is too weak to see through quantum-versus-classical spreading; instead the authors define localization intrinsically as memory of the initial distribution and show that the long-time saturation profile retains such memory inside the chaotic sea. The observed loss of intrinsic ergodicity with large participation number is the nontrivial many-body dynamical localization the paper aims to establish. If correct, it means weak coupling does not guarantee thermalization even when both subsystems are chaotic and quantum mixing is non-perturbative.","feed_headline":"Chaotic quantum partners keep memory of their start","feed_subtitle":"A six-site Bose-Hubbard pair shows dynamical localization inside the chaotic sea, suppressing thermalization.","key_machinery":"The key object is the coupled double-trimer Hamiltonian, with the energy imbalance between the two trimers as the reaction coordinate for energy exchange and an effective Planck constant set by the inverse particle number per trimer. The argument runs through the exact quantum saturation profile obtained by diagonalization without time propagation, so the absence of ergodicity is a property of the eigenstates rather than a finite-time simulation effect. Localization is then judged by intrinsic ergodicity measures: the overlap between the saturation profile and the ergodic distribution, and especially the dependence of the mean absolute imbalance on the initial condition. The participation number is used to separate nontrivial localization, with large participation, from perturbative localization, where only about two states participate. What carries the argument is that the same saturation-profile machinery is applied to classical clouds and to quantum states, allowing a direct comparison that does not presuppose a classical limit.","core_discovery":"On its own terms, the paper's central discovery is that in a double trimer with inter-trimer coupling that is classically weak but quantum mechanically large, the quantum saturation profile remains dependent on the initial energy imbalance, while the semiclassical cloud ergodizes. The quantum memory is quantified by the overlap measure between the saturation profile and the microcanonical distribution, and by the dependence of the mean absolute imbalance on the initial condition. This memory persists for participation numbers much larger than 2, ruling out trivial perturbative localization. The boundary between a small central region of near independence from the initial condition and the localized periphery forms a mobility edge in the energy-imbalance plane. The paper therefore claims to demonstrate a quantum mechanical loss of intrinsic ergodicity despite non-perturbative quantum mixing, an effect it distinguishes from self-trapping, from classical localization due to phase-space barriers, and from perturbative localization.","pith_inferences":["Beyond the paper: the intrinsic memory measure could be used as an order parameter for a disorder-free many-body localization transition in longer chains of weakly coupled chaotic trimers, since it requires no comparison with classical dynamics.","Beyond the paper: the near-degeneracy timescale suggests a two-stage relaxation, with spreading followed by slow mirror-image tunneling, that could be observed as a plateau in the survival probability before final saturation; the paper does not exploit this as a standalone signature.","Beyond the paper: if finite-size scaling were performed, a testable expectation is that the width of the small ergodic central region grows with subsystem size; a power-law or logarithmic growth would distinguish a true dynamical localization from a pure finite-size effect."],"forward_implications":["If the central claim is right, weak coupling between chaotic subsystems is not a sufficient condition for thermalization; quantum memory can persist inside the chaotic sea.","The mobility edge in the energy-imbalance plane means that, for a fixed total energy, only central initial imbalances ergodize, so the outcome of a thermalization experiment depends on preparation.","Because the effect is defined intrinsically by initial-condition memory, the same criterion can be applied to systems with no meaningful classical limit, provided a saturation profile can be computed.","The effect is distinct from disorder-driven many-body localization: it arises from interaction-induced chaos in a clean, finite system, and the paper asserts that the thermodynamic limit is not an issue for this mechanism."],"supporting_citations":[{"why":"Provides the previous trimer-monomer model and the break-time mobility-edge theory that this paper extends to two chaotic subsystems.","marker":"[13]"},{"why":"Gives the Fokker-Planck thermalization picture and the quasistochastic scaled-time behavior used to compare classical simulations.","marker":"[12]"},{"why":"Introduces dynamical localization in low-dimensional chaotic systems, the phenomenon being generalized here to a many-body setting.","marker":"[1]"},{"why":"Supplies the recurrence-based notion of weak localization and the exploration-volume measure that motivates the intrinsic definition.","marker":"[8]"},{"why":"Establishes the large-particle-number classical limit of the Bose-Hubbard Hamiltonian used for the semiclassical clouds.","marker":"[20]"},{"why":"Provides the level-spacing statistics criterion used to locate the chaotic energy range of a single trimer.","marker":"[21]"},{"why":"Supplies the ratio statistic used to distinguish Gaussian-orthogonal-ensemble from Poisson spectral statistics.","marker":"[23]"},{"why":"Marks the disorder-induced many-body localization phenomenon that the paper contrasts with interaction-induced dynamical localization.","marker":"[14]"},{"why":"Provides a further reference for disorder-induced many-body localization, used as the comparison class for the present clean-system effect.","marker":"[15]"}],"fun_headline_variants":["Quantum memory survives chaos in coupled Bose-Hubbard trimers","Dynamical localization halts thermalization in chaotic double trimer","Chaotic subsystems keep quantum memory of initial state","Mobility edge separates ergodic and localized in 6-site model"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is that the memory of the initial state seen for 24 particles per trimer persists at larger particle numbers; the paper asserts the thermodynamic limit is not an issue, but shows no scaling test to back that claim.","fun_headline_variants_meta":{"raw":{"variants":["Quantum memory survives chaos in coupled Bose-Hubbard trimers","Dynamical localization halts thermalization in chaotic double trimer","Chaotic subsystems keep quantum memory of initial state","Mobility edge separates ergodic and localized in 6-site model"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000153,"raw_usage":{"total_tokens":1145,"prompt_tokens":823,"completion_tokens":322,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":439,"completion_tokens_details":{"reasoning_tokens":252}},"tokens_in":439,"tokens_out":322,"duration_ms":4278,"temperature":1.0,"reasoning_tokens":252,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-14T13:59:12.357873+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"One decisive check is exact diagonalization at larger subsystem sizes, such as 30, 40, and 50 particles per trimer, with the same parameters: if the saturation profile becomes independent of the initial imbalance and the overlap measure drops toward the semiclassical value as the particle number grows, the many-body dynamical localization is a finite-size effect rather than a genuine phenomenon.","supporting_citations":[{"cited_title":"Khripkov, A, Vardi, and D","cited_arxiv_id":null,"evidence_quote":"Provides the previous trimer-monomer model and the break-time mobility-edge theory that this paper extends to two chaotic subsystems."},{"cited_title":"Tikhonenkov, A","cited_arxiv_id":null,"evidence_quote":"Gives the Fokker-Planck thermalization picture and the quasistochastic scaled-time behavior used to compare classical simulations."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supplies the recurrence-based notion of weak localization and the exploration-volume measure that motivates the intrinsic definition."},{"cited_title":"coherent state","cited_arxiv_id":null,"evidence_quote":"Establishes the large-particle-number classical limit of the Bose-Hubbard Hamiltonian used for the semiclassical clouds."},{"cited_title":"Dey and A","cited_arxiv_id":null,"evidence_quote":"Provides the level-spacing statistics criterion used to locate the chaotic energy range of a single trimer."},{"cited_title":"Mossmann and C","cited_arxiv_id":null,"evidence_quote":"Supplies the ratio statistic used to distinguish Gaussian-orthogonal-ensemble from Poisson spectral statistics."},{"cited_title":"Basko, Ann","cited_arxiv_id":null,"evidence_quote":"Marks the disorder-induced many-body localization phenomenon that the paper contrasts with interaction-induced dynamical localization."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Provides a further reference for disorder-induced many-body localization, used as the comparison class for the present clean-system effect."}],"review_version":1}