{"id":"626f9463-83ba-4734-95ff-356cd1c0c249","arxiv_id":"2510.14160","paper_version":2,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":8.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"For q-local Hamiltonians with bounded change, an initial eigenstate remains exponentially concentrated in a macroscopic energy window under arbitrary time-dependent perturbations.","lead":"The paper proves a universal bound: a quantum system evolving under any local Hamiltonian with bounded change stays exponentially concentrated in an energy window of the instantaneous spectrum. It then uses this to show that certain error-correcting codes and optimization problems remain stable (or trapped) for exponentially long times.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Theorem 1 appears sound, but the optimization stable-phase claim rests on an explicitly unproven clustering/OGP input for the p-spin model (Sec. IV A).","rationale":"The reader's weakest_assumption correctly identifies the optimization clustering property as the main unsupported link. The central Theorem 1 proof is extensive and, taken on its own, plausible: the moment/commutator argument gives a concrete exponential leakage bound for Case 1. The applications to LDPC codes are on firmer ground because linear soundness is a standard, proven property for the cited code families. By contrast, the p-spin/optimization example explicitly relies on a result that the authors themselves describe as 'demonstrated, though not explicitly proved' in [62]. This is not an accusation of error; it is a missing proof that the paper acknowledges. My read does not change the verdict: CONDITIONAL remains appropriate. If the clustering/OGP property is later proven for the random regular q-uniform hypergraph model, the optimization application would follow; if it fails, the paper would still retain Theorem 1 and the LDPC applications but would lose the claimed algorithmic barrier for Hamiltonian optimization. Thus the same load-bearing concern and the same conditional verdict stand.","tokens_in":41239,"tokens_out":34108,"duration_ms":279310,"concrete_test":"Work out [62]'s OGP for H_L of Eq. (16) with q >= 4, specifying constants: show that all Z-basis states with L(z) <= E_g + B, B = Θ(n), split into clusters with diameter ν1 = o(n) and inter-cluster distance ν2 = Θ(n). If this derivation cannot be completed, run exact enumeration for n = 24, 32, 40 on an ensemble of random p-regular q-uniform hypergraphs: compute all Z-basis states in the window, form a graph with edges when Hamming distance <= ν1, and test whether the connected components have diameter <= ν1 and mutual distance >= ν2. A violation, or a window width that shrinks with n, would falsify the assumption behind Prop. IV.1.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The paper's own text (Sec. IV A) concedes the key input for the optimization application: the q-spin glass Hamiltonian (16) 'was demonstrated, though not explicitly proved, in [62]' to have the clustering property in a linear energy window. This is load-bearing because Theorem 1 only proves localization in energy space. To conclude that a Z-basis state stays near its initial configuration (Prop. IV.1) and that Hamiltonian-based algorithms are frozen, one needs Definition 1 with a barrier B ~ Θ(n) and inter-cluster separation ν2 ~ Θ(n). For LDPC codes this is rigorously supplied by linear soundness, Eq. (9). For the random regular q-uniform p-spin model, the cited OGP is not proved here: if the OGP window has sublinear width, or if ν2 is not Θ(n), the leakage bound of Theorem 1 no longer implies confinement to a cluster, and the claimed algorithmic barrier collapses. The paper itself flags a related hazard in footnote [46] (false vacua) when linear soundness/clustering is absent. Since the abstract presents the optimization trap as one of the two main applications, this missing proof is the weakest link in the central stable-phase claim, not in the energy-space localization theorem itself.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper proves a universal energy-space localization theorem for q-local, time-dependent Hamiltonians H(t) with bounded local norm. Theorem 1 states that an initial eigenstate of H(0) evolves so that its weight on instantaneous eigenstates with energy difference at least dn from E0 is at most exp[-n(λ/Δ)(d/λ−1−ln(d/λ)+o(1))] for d>λ, where Δ=2qM and ∫||dH||_X≤λn. A sharper bound is given when H(t) has a commuting core (Case 3). The result is extended to static perturbations by a quench limit, yielding eigenstate localization results for LDPC codes, and to quasi-q-local perturbations. Applications are developed for LDPC codes with linear soundness (exponential dynamical localization, eigenstate localization, slow mixing) and for classical optimization problems whose low-energy landscape satisfies a clustering property, where the authors argue that Hamiltonian-based algorithms become trapped in local minima.","tokens_in":41570,"tokens_out":13361,"duration_ms":113883,"significance":"If the main theorem is correct, it is a significant and general contribution: it gives parameter-free, rigorous leakage bounds for arbitrary time-dependent local driving, with no assumption on driving speed, and it provides a new route from energy-space localization to stability in systems with extensive energy barriers. The proof in Appendix B is detailed and the moment recursion is internally consistent; the Stirling asymptotics reproduce Eq. (3). The LDPC applications are well-supported by the linear soundness property (Eq. (9)) from the cited literature. The optimization application is the conditional part: it depends on a clustering property that, for the explicit p-spin example, is not proved in the manuscript. The paper is honest about this gap in Sec. IV A, but the gap is load-bearing for the advertised algorithmic-trap claim.","major_comments":[{"comment":"The optimization application depends on the clustering property (Definition 1) with a linear energy barrier B~Θ(n) and inter-cluster separation ν2~Θ(n). For the q-spin Hamiltonian (16), the paper states: \"It was demonstrated, though not explicitly proved, in [62] that the model has the clustering property in a linear energy window.\" This is a load-bearing input for Proposition IV.1 and the algorithmic-trap claim. Theorem 1 proves only energy-space localization; converting it to configuration-space localization requires the clustering property. If the OGP window is sublinear or ν2 is not Θ(n), the leakage bound of Theorem 1 does not imply confinement to one cluster, and the freezing conclusion collapses. Footnote [46] illustrates the same hazard for codes without linear soundness. Please provide a rigorous proof of the needed clustering for (16) or explicitly label the optimization applic","section":"Sec. IV A, Eq. (16)"},{"comment":"The freezing claim for the adiabatic algorithm is not rigorously derived from Theorem 1. Proposition IV.1 is stated for an initial Z-basis state inside a cluster, which is an eigenstate of H_L, but in the algorithmic evolution (17) the state at time s* is a general superposition, not a single Z-basis state. Theorem 1 controls leakage in the instantaneous eigenbasis of H(t), while the cluster decomposition is in the H_L eigenbasis. The paper's discussion after Eq. (17) asserts that states below E_B−2Λ are frozen and that the late-time evolution cannot improve or worsen the solution, but the change-of-basis argument and the treatment of superposed initial states are not given. Please supply the missing step or restrict the claim to evolutions with V(0)=0 and initial Z-basis states.","section":"Sec. IV A, Prop. IV.1 and Eq. (17)"}],"minor_comments":[{"comment":"The expression for the total variation Λ of H(t)=s(t)H_L+(1−s(t))H_M from s to 1 appears incorrect: the total variation is (1−s)||H_L−H_M||_X (assuming monotone s), not (1−s)/s ||H_M||_X. Please correct or define the rescaled perturbation Hamiltonian explicitly.","section":"Sec. IV A after Eq. (17)"},{"comment":"The notation E_j(d) is used before it is defined; define the energy window in the text preceding Eq. (5).","section":"Eq. (5)"},{"comment":"The notation \"quasi-q⋆-local\" and the star on q are used in tables and text before being formally defined in a way that distinguishes q from q⋆. Please add a definition near the first use.","section":"Tables I and II; Sec. II"},{"comment":"The condition [V(t), H'(t)]=0 in Case 3 is very restrictive in the dynamical setting (though automatically satisfied in the static reduction). The paper notes this, but it may be worth emphasizing that most natural time-dependent perturbations do not satisfy it, limiting the applicability of the improved bound (4).","section":"Sec. II, Case 3"}],"recommendation":"major_revision","confidential_remarks":"The main theorem appears sound and is a genuine contribution. The weakest point is the optimization application, where the clustering property for the p-spin model is explicitly not proved and the algorithmic argument is more heuristic than the LDPC part. The paper would be suitable after the authors either prove the needed clustering input or clearly mark the optimization claims as conditional, and after clarifying the initial-state/basis issue in Prop. IV.1. No concerns about novelty or citation practice."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Bottom line: this is a real result. Theorem 1 is new as far as I know—previous rigorous stability results for many-body systems were static or periodic—and the proof is substantial. I checked the moment recursion and the Stirling step; Eq. (3) reproduces correctly. The nested-commutator framework is a clean way to get the q-locality scaling, and the time-reparameterization invariance is genuinely useful. The paper deserves a serious referee.\n\nWhat it does well: the four cases are stated cleanly, the appendices carry the weight, and the LDPC applications are on solid ground where linear soundness is available. The quasi-q-local extension of [33] is a real plus, and the citation pattern looks honest. The static reduction via T→0 is plausible, though it is handled a little quickly; one sentence of justification in the main text would help.\n\nThe soft spot is exactly the one flagged in Sec. IV A: the clustering property for the random regular q-uniform p-spin model is imported from [62] as “demonstrated, though not explicitly proved.” That input is load-bearing. Theorem 1 only gives energy-space localization; to conclude configuration-space freezing and the claimed algorithmic trap, you need Definition 1 with a linear barrier B and linear inter-cluster separation ν2. The paper admits this indirectly and also flags the false-vacuum hazard in footnote [46]. So Proposition IV.1 is conditional, and the abstract somewhat oversells it. This is not a flaw in the main theorem, and it is not concealed—it is written in the text—but it matters for how the paper is read.\n\nMinor concern: the static reduction T→0+ is a little heuristic. I do not think it is wrong; the Schrödinger continuity argument is standard, but it deserves explicit justification rather than a parenthetical.\n\nNo circularity, no fitted parameters. The theorem is proved from first principles, with prior results used as inputs, and [33] and [14] are cited appropriately. No code or data, but none is needed for this kind of proof.\n\nWho gets value: anyone working on rigorous stability of quantum codes, prethermalization, or Hamiltonian algorithms for spin glasses. I would bring it to reading group, and I would cite Theorem 1 in my own work. I would not cite the optimization claim without the clustering assumption being proved.\n\nRecommendation: send it to peer review. Ask the authors to either prove or explicitly conditionalize the p-spin clustering input and to soften the abstract accordingly. The theorem itself is strong enough to justify the referee time.","headline":"A genuinely new and largely rigorous energy-space localization theorem for time-dependent q-local Hamiltonians, with LDPC applications solid and the optimization application conditional on an explicitly unproven clustering input.","tokens_in":41976,"tokens_out":1652,"would_cite":true,"duration_ms":17922,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["81Q10","81P70","82B44"],"pacs":[],"model":"deepseek-v4-flash","headline":"For any q-local Hamiltonian driven with bounded total variation, an initial eigenstate stays exponentially concentrated in a narrow instantaneous-energy window.","keywords":["energy-space localization","time-dependent perturbations","q-local Hamiltonians","total variation","instantaneous eigenstates","clustering property","LDPC codes","optimization barriers"],"falsifier":"Evolve an n-qubit system under H(t) with random all-to-all 2-local Pauli terms, local norm bounded by M=1, and drive it with a perturbation of total variation λn (say λ=0.1) for a time where d/λ=2. Numerically compute the weight on instantaneous eigenstates with energy |E−E0|≥2λn; if this weight decays slower than exp[-n(λ/Δ)(2−1−ln 2)] for n=10,…,20 after optimizing the o(1) correction, the theorem's exponential rate is violated or its constant is not tight at accessible sizes.","tokens_in":41182,"feed_emoji":"⚛️","tokens_out":6565,"duration_ms":59179,"temperature":0.7,"pith_summary":"This paper claims that a universal phenomenon, energy-space localization, governs the dynamics of generic local quantum systems: start in an eigenstate and drive with any time-dependent q-local perturbation whose total variation is at most λn; then, at every instant, the state's weight on instantaneous eigenstates whose energy differs by more than d n (for d>λ) is at most e^{-c n}, with c = (λ/Δ)(d/λ −1 − ln(d/λ)) up to o(1), where Δ=2qM. This is qualitatively stronger than the naive Markov bound λ/d and gives back classical energy conservation in the thermodynamic limit. The paper then shows that when the energy landscape has a clustering property with a linear energy barrier — as in good LDPC codes and some spin-glass optimization problems — energy-space localization becomes genuine configuration-space localization: information stays near its original codeword, and Hamiltonian algorithms get trapped in local minima. If correct, this would establish a broad class of provably stable quantum phases against generic time-dependent perturbations, and would reshape expectations for quantum memories and adiabatic optimization.","feed_headline":"Energy-space localization survives generic time-dependent driving","feed_subtitle":"An exponential bound keeps driven quantum states inside a narrow energy window, with consequences for LDPC memories and optimization algorit","key_machinery":"The engine is a moment estimate. Define g_k(t)=||(H(t)-E0)^k|ψ(t)⟩||; the paper bounds the growth of these 2k-th central moments in the instantaneous eigenbasis and uses Markov's inequality with a carefully chosen order k≈(d−λt)n/Δ to turn moment growth into exponential leakage decay. The growth is controlled by nested commutators ad^m_H(t)(H′(t)): for q-local H these scale at most factorially, m!Δ^m, while for a commuting core with [V,H′]=0 they scale exponentially, Δ^m, giving the improved bound. The second ingredient is the clustering property: if eigenstates below an energy E_B split into clusters with intra-cluster distance ≤ν1 and inter-cluster distance ≥ν2 ∼ n, then exponential energy","core_discovery":"The central discovery is a rigorous exponential leakage bound, not just an energy-expectation bound. For an initial eigenstate of H(0) evolving under a q-local H(t) with local norm ≤M and total variation ∫||dH||_X ≤λn, the probability that the state is found in instantaneous eigenstates with |E(t)−E0| ≥ d n is exponentially small in n for any d>λ; the rate is governed by λ/Δ times (d/λ −1 − ln(d/λ)), with Δ=2qM. When H can be split into a mutually commuting core plus a perturbation that commutes with H′, the bound improves to exp[-n (λ/Δ)((d/λ) ln(d/λ) − (d/λ −1))]. This is proven by controlling all 2k-th central moments of H(t)−E0 and applying Markov's inequality with an optimized moment or","pith_inferences":["Editorial extension: the moment-based method suggests a direct numerical protocol — simulate n∈[8,20] qubits with all-to-all random q-local Pauli driving, measure the instantaneous-energy leakage as a function of d/λ, and check whether the rate approaches the predicted (λ/Δ)(d/λ−1−ln(d/λ)); deviations would reveal where the o(1) correction and finite-size effects matter.","Because the bound depends only on the total variation, rapid pulses and slow ramps with the same ∫||dH|| should produce the same leakage; this time-rescaling invariance is a crisp, testable signature that distinguishes energy-space localization from mechanisms tied to specific speed limits.","If a model lacks the clustering property, energy-space localization should coexist with thermalization inside the window; a natural next step is to quantify how the window's internal mixing rate depends on the density of states, which the current theorem leaves open.","The nested-commutator formulation connects naturally to operator-growth complexity; for models with slower-than-factorial growth of ad^m(H′), such as certain mean-field models, the same proof could produce tighter windows or exact rates."],"forward_implications":["LDPC codes with linear soundness protect an encoded state for exponentially long times against generic time-dependent (quasi-)q-local noise, as long as the perturbation's total variation stays below a constant fraction of the linear energy barrier.","For static perturbations of commuting LDPC Hamiltonians, the eigenstates are exponentially localized in the unperturbed spectrum; the bound extends to quasi-q-local perturbations, which prior tridiagonal-matrix methods did not cover.","Hamiltonian-based quantum algorithms for clustered optimization problems cannot escape their starting cluster in polynomial time when the driving has small total variation, so late-time evolution neither improves nor ruins near-optimal outputs.","Local Gibbs samplers for these LDPC Hamiltonians have exponentially long mixing times even with weak quasi-q-local perturbations, extending thermodynamic-stability results to a broader noise class.","In the thermodynamic limit, the exponentially small high-energy leakage means the quantum dynamics obeys the classical energy-conservation bound, giving a provable quantum-to-classical correspondence for driven local systems."],"fun_headline_variants":["Universal energy-space localization beats time-dependent driving","Exponential bound keeps driven quantum states in a narrow energy window","Time-dependent perturbations can't destroy quantum energy-space localization","Stable quantum phases under time-dependent kicks: energy-space localization","Proven: quantum states stay in energy window under any time-dependent perturbation"],"cache_read_input_tokens":2304,"weakest_assumption_plain":"The stability applications all hinge on the clustering property holding in a linear energy window: the low-energy eigenstates must be cleanly separated into clusters whose mutual distance is O(n), enforced by an energy barrier of height B ∼ O(n) that grows linearly with system size; the paper cites linear soundness for LDPC codes but only a demonstrated-not-proved clustering for the spin-glass optimization example, and without such a barrier energy-space localization does not","fun_headline_variants_meta":{"raw":{"variants":["Universal energy-space localization beats time-dependent driving","Exponential bound keeps driven quantum states in a narrow energy window","Time-dependent perturbations can't destroy quantum energy-space localization","Stable quantum phases under time-dependent kicks: energy-space localization","Proven: quantum states stay in energy window under any time-dependent perturbation"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000535,"raw_usage":{"total_tokens":2430,"prompt_tokens":788,"completion_tokens":1642,"prompt_tokens_details":{"cached_tokens":256},"prompt_cache_hit_tokens":256,"prompt_cache_miss_tokens":532,"completion_tokens_details":{"reasoning_tokens":1560}},"tokens_in":532,"tokens_out":1642,"duration_ms":11719,"temperature":1.0,"reasoning_tokens":1560,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-04T09:38:13.266671+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Evolve an n-qubit system under H(t) with random all-to-all 2-local Pauli terms, local norm bounded by M=1, and drive it with a perturbation of total variation λn (say λ=0.1) for a time where d/λ=2. Numerically compute the weight on instantaneous eigenstates with energy |E−E0|≥2λn; if this weight decays slower than exp[-n(λ/Δ)(2−1−ln 2)] for n=10,…,20 after optimizing the o(1) correction, the theorem's exponential rate is violated or its constant is not tight at accessible sizes.","supporting_citations":[],"review_version":1}