{"id":"cf617b31-0acd-41e5-b541-fc4156a7bb56","arxiv_id":"2607.20959","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":7.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":1,"one_line_summary":"In chaotic quantum dynamics, large-deviation distributions equilibrate on O(1), e^N, and exp(e^N) time scales depending on the definition, and the slowly drifting cutoffs provide a proposed measure of quantum complexity.","lead":"The paper studies three ways to define large deviations in many-body quantum dynamics, showing they relax on three very different time scales: a constant O(1) time, an exponential time e^N, and a double-exponential time exp(e^N). It proposes using the slow evolution of the cutoff in these distributions as a new measure of quantum complexity.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The exp(e^N) time scale rests on an unproven assumption that the quasiperiodic trajectory s_i(t)=e^{-i ε_i t} has rare-region hitting times fixed by the Mattis entropy; the numerics (N≤7) are too small to validate this.","rationale":"The reader's weakest_assumption correctly identifies the uniform-sampling/ergodicity premise. I agree with that diagnosis and sharpen it: the static Mattis calculation is likely correct (Appendix A gives an exact D→∞ saddle point, and the numerics support the static distribution), but the dynamic step from Haar measure to finite-time sampling is not derived. The trichotomy's most distinctive claim is the double-exponential scale for expectation values; that claim is exactly where the unproven assumption enters. Since the paper itself frames Section 3.3 as an analogy ('treat the evolution... as a trajectory... we then apply the argument of Section 2'), the missing piece is a justification that the deterministic linear flow on U(1)^D has rare-region hitting times governed by e^{D S}. This is not an internal inconsistency and the paper's numerics are suggestive, but the evidence is too small-scale to confirm the exponential-of-entropy exponent, and no code is provided. A targeted first-passage-time scaling test, with a random-phase control, would settle whether the concern lands. Because this is a verification gap rather than a demonstrated error, the conditional verdict is unchanged.","tokens_in":13572,"tokens_out":9136,"duration_ms":100324,"concrete_test":"For the kicked Ising model (Eq. 5), compute first-passage times τ(a)=min{t: r(t)>a} for several thresholds a and sizes N=6,...,11 (D=64,...,2048), with t_max as large as feasible (≥10^8 for small N). Plot D^{-1} ln τ(a) versus the Mattis entropy S(-a) from Eqs. (26)-(28); the rare-region assumption predicts a collapse onto the line y=-S(-a). Also run the same test with i.i.d. random eigenphases (or CUE eigenvalues) as a control: if the control collapses onto the predicted line but the kicked chain does not, the discrepancy is due to spectral correlations, and Eq. (29) is not generic; if both collapse, the uniform-sampling assumption is validated and the central time-scale claim survives.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim that the distribution of expectation values reaches its long-time limit at t~exp(e^N) (Section 3.3) depends entirely on the step from the static Mattis model to the time dynamics. Equations (22)-(29) compute the infinite-time large-deviation form from the Haar measure on U(1)^D, then assert that a finite-time trajectory samples this measure with a sharp cutoff at S(-a*) = -D^{-1} ln t_max. This is the classical rare-region argument of Section 2 applied to the variables s_i(t)=e^{-i ε_i t}. But those variables are not an ergodic many-body trajectory; they are a one-parameter linear flow on the torus. For such a flow, the time to first reach an atypical region {r>a} is controlled by Diophantine properties of the eigenphases ε_i, not simply by the Haar measure of the region. Level repulsion and spectral correlations in a real chaotic Floquet system can either delay or accelerate the appearance of large return probabilities/SFF values relative to the i.i.d.-phase model. The paper offers no derivation or numerical check of this hitting-time assumption; Figures 3-4 use N=5,7 (D=32,128) and t_max≤10^7, which cannot distinguish e^D~e^128 from other scalings and do not test the double-exponential regime. Section 4 sets the correlation time t_c=1; if t_c grows with N, the effective number of independent samples is smaller and the observed cutoff would drift more slowly than Eq. (29), changing the proposed complexity measure.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper studies large deviations in many-body quantum dynamics and proposes a trichotomy of time scales. It distinguishes (i) the full distribution of an extensive observable, argued to equilibrate at t ~ O(1) independently of N; (ii) the time distribution of monitored measurement outcomes, whose finite-time cutoff drifts on an e^N scale; and (iii) the time distribution of expectation values such as the return probability and the spectral form factor, whose cutoff drifts on an exp(e^N) scale. The latter is derived by mapping the unitary evolution to variables s_i(t)=e^{-i ε_i t}, treating them as XY spins, and computing the infinite-time distribution from an analytically solvable Mattis model (Appendix A). A classical rare-region argument is then used to predict a sliding cutoff. The paper proposes the evolution of these cutoffs as a measure of quantum complexity. Numerical support is provided for the kicked Ising model.","tokens_in":13912,"tokens_out":18070,"duration_ms":191566,"significance":"If the trichotomy holds, the paper gives a new, operationally defined notion of quantum complexity with time scales distinct from Nielsen and Krylov complexity, and connects large deviations to Hilbert-space recurrence. The analytic Mattis-model solution in Appendix A is a clear strength: it is parameter-free and directly yields the infinite-time distribution for the return probability and SFF. The numerical comparison in Figs. 3-4 supports the distribution prediction. The paper is clearly written and provocative. However, the central exp(e^N) time scale rests on a heuristic hitting-time assumption that is not derived or directly tested, and the mode of convergence to the longtime limit is left ambiguous.","major_comments":[{"comment":"The double-exponential time scale is the central claim, but it rests on the unproved assertion that the linear flow s_i(t)=e^{-i ε_i t} samples U(1)^D uniformly and that the first-passage time to {r>a} is exp[D|S(-a)|]. For a Kronecker flow on a torus, ergodicity and hitting-time laws are controlled by Diophantine properties of the eigenphases ε_i; they are not consequences of the Haar measure. The numerics in Figs. 3–4 use D≤128 and t_max≤10^7, which is many orders of magnitude below e^D; they test the shape of P_{t_max}(a) at moderate a but cannot test the claimed exp(e^N) scaling. Please either prove or derive the hitting-time estimate under explicit assumptions, or clearly label it as a conjecture and provide a test of the D-dependence of the cutoff position (e.g., collapse of a* versus D^{-1} ln t_max for several N).","section":"§3.3, Eqs. (22)–(29)"},{"comment":"The abstract states that (iii) 'reach[es] its longtime limit at t∼exp(e^N)'. However, Eq. (29) combined with Eq. (50) (S∼(1/2)ln(1−a)) implies S is unbounded below, so the cutoff a* satisfies a*<1 for every finite t_max and the support of P_{t_max} never equals that of P∞. The text itself notes that the cutoff 'always exists and approaches 1 as t_max→∞'. The mode of convergence is therefore unclear: pointwise convergence of the large-deviation rate on a fixed interval is different from convergence of the probability measure, which can occur much earlier because the missing tail has small probability. Please state precisely in what sense the longtime limit is attained and adjust the abstract accordingly.","section":"Abstract and §3.3, Eq. (29)"},{"comment":"The monitored-outcome result is presented through a single realization (N=18, Fig. 2) with no ensemble average or error bars. The central prediction of Eq. (15) is a cutoff with f*=N^{-1} ln t_max; to support this, the paper should extract the empirical cutoff and compare it with the predicted relation for more than one N. The same applies to the claim that the result is independent of measurement scheme, which is not tested.","section":"§3.2, Eq. (15)"}],"minor_comments":[{"comment":"The text writes S(−a)∼ln(1−a), but Appendix A Eq. (50) gives (1/2) ln(1−a). Correct the typo.","section":"§3.3, near Eq. (29)"},{"comment":"The notation '−h = r or SFF' is confusing; define h explicitly in the partition function.","section":"Eq. (24)"},{"comment":"Clarify whether panel (b) combines N=5 and N=7 data or shows one; the caption is ambiguous.","section":"Figures 3–4"},{"comment":"The assertion that t_c scales at most as a power law for generic operators and measurement schemes is not demonstrated; add a reference or a test.","section":"§4"},{"comment":"There are several typos ('atpyical', 'an double exponential', 'The R_ij' capital letter, 'Dlnt'). Please proofread.","section":"General presentation"}],"recommendation":"major_revision","confidential_remarks":"I found the paper stimulating and the Mattis-model part is solid. The key issue is the unproven Diophantine/hitting-time step underlying the exp(e^N) time scale, and the ambiguity about the mode of convergence. If the authors make the assumption explicit and add a scaling test of the cutoff with D, the paper could become acceptable."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Dear [Colleague],\n\nYou should know two things about this paper before reading it. First, the Mattis-model computation for the infinite-time large deviations of the return probability and spectral form factor is clean, parameter-free, and very likely correct. Second, the paper's headline time scale, t~exp(e^N), is a heuristic extrapolation from that static computation; the numerics (N≤7) are far too small to confirm it. Read it for the analytic result and the conceptual trichotomy, not for the double-exponential claim.\n\nWhat is actually new: The paper separates three notions of quantum large deviation — the full distribution of an observable, the distribution of measured outcomes, and the distribution of expectation values over a time window — and shows they equilibrate on very different time scales: O(1), e^N, and presumably exp(e^N). The distinction between \"potential\" and \"observed\" rare fluctuations via quantum parallelism is worth thinking about. The mapping of the long-time return-probability/SFF distribution to a Mattis model with D XY spins is an elegant trick, and Appendix A solves it exactly in D→∞ without free parameters. The prediction that finite-time distributions are the infinite-time ones with a sliding cutoff is concrete and testable, and the small-scale numerics in Figures 3–4 do support it.\n\nSoft spots. The load-bearing step is Eq. (29): the time distribution of a linear quasiperiodic flow on the U(1)^D torus is assumed to sample the Haar measure with a rare-region hitting time set by the Mattis entropy. For a genuinely chaotic or stochastic process that is the standard extreme-value argument. For a torus translation it is not automatic — Diophantine properties of the eigenphases can change hitting times to small sets. The paper gives no derivation and no direct check of the exp(e^N) scale. The numerics are small (D≤128, tmax≤1e7) and without error bars, so they cannot distinguish exp(e^N) from, say, e^{cN^p}. The general-operator section is also partly circular: the rate function is extracted from the same time series used to verify the cutoff. And the correlation time t_c is set to one without discussion of N-dependence, which would affect the cutoff drift. One more minor internal inconsistency: the abstract says the expectation-value distribution equilibrates at exp(e^N), but Section 3.3 notes the cutoff never disappears—it just approaches a=1.\n\nThat said, the central analytic result is independent of the dynamical assumption. If you work on return probabilities, SFFs, or large deviations in Floquet systems, this is worth citing.\n\nMy recommendation: send it to peer review. It is coherent, original, and honest about where its evidence stops. A referee should push for a direct test of the hitting-time assumption or at least a caveat that the exp(e^N) scale is a conjecture, and for larger-N numerics if feasible. But this is the kind of paper that generates useful discussion.","headline":"The Mattis-model large-deviation computation is solid and the sliding-cutoff picture is attractive, but the exp(e^N) time scale is a heuristic that the numerics are far too small to support.","tokens_in":14418,"tokens_out":5741,"would_cite":true,"duration_ms":63296,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"This paper argues that large deviations in many-body quantum dynamics split into three distinct phenomena whose equilibration times differ enormously: the full distribution of an observable settles in size-independent O(1) time, monitored m","keywords":["large deviations","quantum dynamics","equilibration timescales","quantum complexity","spectral form factor","return probability","Mattis model","kicked Ising model"],"falsifier":"Numerically compute the time distribution of the spectral form factor in an integrable (e.g., noninteracting) spin chain and check whether the cutoff a*(t_max) obeys S(-a*) = -D^{-1} ln t_max with a Mattis-model entropy; if the cutoff grows much slower or the distribution fails to match the truncated large-deviation form, the ergodicity assumption is violated and the paper's central prediction for expectation values collapses.","tokens_in":13411,"feed_emoji":"⚛️","tokens_out":4004,"duration_ms":38444,"temperature":0.7,"pith_summary":"This paper argues that 'large deviations' in many-body quantum dynamics split into three distinct phenomena with vastly different equilibration times. The full distribution of an extensive observable settles to its infinite-time form in a time of order one, independent of system size, because quantum parallelism visits rare and typical configurations simultaneously. Under continuous measurement, the distribution of outcomes over a time window settles only at t_max ~ e^N, as in classical systems. The distribution of expectation values—such as the return probability or spectral form factor—settles at the Hilbert-space recurrence time t_max ~ exp(e^N). Before each of these limits, the distribution matches the infinite-time large-deviation form below a sharp, slowly drifting cutoff, and the paper proposes tracking that cutoff as a measure of quantum complexity.","feed_headline":"Rare quantum spikes take a double-exponential time to appear","feed_subtitle":"Full distributions settle instantly; monitored outcomes wait e^N; expectation-value outliers wait exp(e^N).","key_machinery":"The argument rests on a statistical-mechanics mapping. The phase factors s_i(t)=e^{-iε_i t} built from the pseudo-energy eigenvalues of the unitary evolution are treated as XY spins living on the torus U(1)^D, where D~e^N is the Hilbert-space dimension. The return probability and the spectral form factor become Hamiltonians of a Mattis model with these spins as degrees of freedom, and the long-time distribution of their values is obtained from the model's entropy density S(-a). The finite-time distribution is then predicted by the sliding-cutoff formula P_{t_max}(a) ~ e^{D S(-a)} for a ≤ a*, with S(-a*) = -D^{-1} ln t_max, which forces the equilibration time to be double-exponential. The sam","core_discovery":"The paper's central claim is a trichotomy of equilibration time scales for large deviations in generic, non-integrable many-body quantum dynamics without conservation laws. For an extensive observable A, the full distribution ⟨δ(A(t)-Na)⟩ reaches its long-time (infinite-temperature) form at t ~ O(1), independently of system size N. The distribution of measurement outcomes of a continuously monitored extensive observable over 0 ≤ t ≤ t_max reaches its infinite-time form only at t_max ~ e^N. And the distribution of expectation values ⟨A⟩_t over 0 ≤ t ≤ t_max—including return probability, spectral form factor, autocorrelation function, and operator size—reaches its infinite-time form only at t_","pith_inferences":["If the mapping is robust, the same trichotomy should appear in other chaotic quantum models, e.g., random circuits, and the sliding cutoff could be measured in current quantum simulators over times much shorter than exp(e^N) by looking at the frontier position for moderately rare events.","The expectation-value complexity, with its exp(e^N) saturation time, is a natural candidate holographic dual to continued black-hole interior growth; the paper hints at this but does not develop a gravitational dictionary.","A testable corollary: in integrable or many-body-localized systems where the XY-spin trajectory does not sample U(1)^D uniformly, the exp(e^N) cutoff growth should fail or slow dramatically, providing a clean diagnostic of quantum chaos.","The proposed 'number of independent samples' M ≈ ln(t_max/t_c) offers an operator-independent complexity that could be compared across different observables and measurement schemes; verifying that M grows as predicted would distinguish this notion from Krylov/Nielsen complexity."],"forward_implications":["The full distribution of an extensive quantity is not a useful complexity measure: it equilibrates in size-independent O(1) time, unlike classical deterministic systems.","Monitored quantum large deviations reproduce the classical story: atypical measurement outcomes are observed only after exponentially long times t ~ e^N.","Expectation-value spikes—brief, one-shot-observable rare fluctuations—occur on the Hilbert-space recurrence scale t ~ exp(e^N), matching known recurrence bounds.","The slowly drifting cutoff furnishes a concrete, operator-dependent measure of quantum complexity, with a single-parameter (number of independent samples) that is nearly operator independent.","For Hamiltonian systems with a conserved energy, the double-exponential scale survives as long as the initial state has nonzero entropy density, with D replaced by the effective thermal Hilbert-space dimension."],"fun_headline_variants":["Quantum large deviations: three time scales, one complexity measure","Quantum rare events: O(1), e^N, exp(e^N) time scales","From O(1) to exp(e^N): quantum large deviation time scales","Quantum complexity via rare outcome frontiers","Three equilibration times for quantum large deviations"],"cache_read_input_tokens":2304,"weakest_assumption_plain":"The entire double-exponential time scale and sliding-cutoff prediction rest on the assumption that the sequence of phase factors s_i(t)=e^{-iε_i t} samples the full torus U(1)^D uniformly, i.e., that the time trajectory of the wavefunction is ergodic over the Hilbert-space configuration space; if this ergodicity fails, the exp(e^N) scale does not follow.","fun_headline_variants_meta":{"raw":{"variants":["Quantum large deviations: three time scales, one complexity measure","Quantum rare events: O(1), e^N, exp(e^N) time scales","From O(1) to exp(e^N): quantum large deviation time scales","Quantum complexity via rare outcome frontiers","Three equilibration times for quantum large deviations"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.00107,"raw_usage":{"total_tokens":4302,"prompt_tokens":707,"completion_tokens":3595,"prompt_tokens_details":{"cached_tokens":256},"prompt_cache_hit_tokens":256,"prompt_cache_miss_tokens":451,"completion_tokens_details":{"reasoning_tokens":3509}},"tokens_in":451,"tokens_out":3595,"duration_ms":22483,"temperature":1.0,"reasoning_tokens":3509,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-01T08:51:32.001933+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Numerically compute the time distribution of the spectral form factor in an integrable (e.g., noninteracting) spin chain and check whether the cutoff a*(t_max) obeys S(-a*) = -D^{-1} ln t_max with a Mattis-model entropy; if the cutoff grows much slower or the distribution fails to match the truncated large-deviation form, the ergodicity assumption is violated and the paper's central prediction for expectation values collapses.","supporting_citations":[],"review_version":1}