{"id":"1e9083a5-8953-4e76-9cc4-532a086ccf62","arxiv_id":"2509.19944","paper_version":3,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":5.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":1,"one_line_summary":"In the PXP spin chain, the theta-symmetric and blockaded initial states keep their single-site density matrices close to the initial value (average local fidelity above 0.99) even though global dynamics stay irregular, a decoupling the paper calls local reminiscence.","lead":"This paper asks whether a small piece of a many-particle quantum system can remember its starting configuration while the whole system keeps evolving in a complex, non-thermal way. Studying a model of Rydberg atom arrays, the authors find two special starting states whose local pieces stay almost unchanged over time, offering a new way to detect memory that global probes miss.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Thermodynamic-limit claim rests on L≤20 and t=100 without extrapolation control; the L-trend in <F_1-site>_t and σ could reverse at larger L.","rationale":"The reader's verdict is CONDITIONAL, and the weakest assumption identified is the finite-size extrapolation from L≤20 to the thermodynamic limit. I agree that this is the most load-bearing concern for the central claim, because the claim is explicitly an asymptotic one ('as L grows') and the paper provides no analytic bound or extrapolation control. The two other weaknesses noted by the reader—the mismatch between the scar-overlap mechanism and the Néel data, and the fact that Wiener's theorem applies only to global fidelity—are real but secondary: they weaken the proposed explanation without directly falsifying the numerical observation that the two specific states have high local fidelity at L≤20. A direct test at larger L is therefore the decisive check. I also give credit for the analytically verifiable cross-checks (Eqs. (11), (13), (14), (15)) and for the internally consistent main definitions, but no code is shipped and convergence in time is undocumented, which supports keeping the verdict CONDITIONAL rather than ACCEPT.","tokens_in":20563,"tokens_out":13090,"duration_ms":93450,"concrete_test":"Recompute Figs. 4(f,g) and 5(g,h) for L=22,24,26 (and L=28 if feasible) using the same exact-diagonalization/Krylov time evolution for |Θ^{symm}_+(π/4)> and |φ_L>, evaluating <F_1-site>_{t=100} and σ with Eqs. (7)-(8). If <F> drops below 0.99 or σ rises above 0.05 at any larger L, the claimed thermodynamic-limit trend is not established. In addition, for L=20 extend the Cesàro average to t=1000 to check whether <F_1-site>_t has actually converged; if it decays after t=100, the 'robust' local reminiscence is a finite-time artifact.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim is asymptotic: robust local reminiscence as L grows. The numerical basis is exact diagonalization for L≤20 and Cesàro averages at t=100 (Figs. 4(f,g), 5(g,h)), with no extrapolation procedure, no convergence check, and no analytic bound on the local fidelity. Since the constrained Hilbert-space dimension grows as F_{L+2}, t=100 may be far shorter than the time needed for the Cesàro average to approach its infinite-time limit; the observed monotone rise of <F_1-site>_t and fall of σ with L could be a finite-size/finite-time artifact that reverses for L>20. Wiener's theorem, which is the paper's only spectral tool, applies to the global fidelity |ν(t)|², not to the local fidelity F_1-site(t); therefore there is no theoretical guarantee that the pure-point spectral weight (which controls the global fidelity) also controls the persistence of single-site reduced density matrices. The proposed scar-overlap mechanism is also internally tested and failed by the paper's own Néel data: |Z2> has the largest scar overlap but (D^{max}_{1-site})_{t=100}≈0.681, far above the 0.065 and 0.095 reported for the reminiscent states, so high scar overlap is not sufficient for local reminiscence. If the L-trend is not asymptotic, the central claim 'approaches above 0.99 / below 0.05' is unsupported.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper introduces the concept of \"local reminiscence\" — the temporal stability of reduced density matrices of small subsystems — and studies it in the PXP model. Three families of initial states are examined: the product state |Θ+⟩, its spatially symmetrized version |Θ_symm+⟩, and the equal-weight blockaded state |φ_L⟩. The central numerical claim is that, for |Θ_symm+⟩ at θ=π/4 and for |φ_L⟩, the time-averaged single-site fidelity ⟨F_{1-site}⟩_t approaches values above 0.99 while its standard deviation falls below 0.05 as L grows (Figs. 4(f,g) and 5(g,h)), even though the global dynamics remain complex and non-ergodic. Analytic results include the normalization of |Θ_symm+⟩, a closed-form expression for the single-site reduced density matrix of |φ_L⟩ based on Fibonacci numbers, and a golden-ratio limit for the initial magnetization. The paper proposes a spectral interpretation via Wiener's theorem, arguing that local reminiscence is linked to a large component of the initial state on the pure-point spectrum.","tokens_in":20875,"tokens_out":8964,"duration_ms":79281,"significance":"If established, the finding that local memory can persist even when global fidelity is irregular and entanglement entropy grows would be a useful conceptual addition to the quantum many-body scars literature. The analytic derivations for |φ_L⟩ (Eqs. (13)-(15), Appendix B) are explicit, internally consistent, and correct: the trace identity F_{j+1}F_{L-j+2}+F_jF_{L-j+1}=F_{L+2} indeed yields a normalized reduced density matrix, and the golden-ratio limit Z_1(0)≈1/φ^2−1/φ checks out. The paper has no fitted parameters in the central numerical observation; the PXP Hamiltonian is fixed and the initial states are prescribed. However, the asymptotic claim and the proposed spectral mechanism are not yet demonstrated at the level required by the paper's conclusions.","major_comments":[{"comment":"The central claim that local reminiscence becomes robust as L grows — ⟨F_{1-site}⟩_t above 0.99 and σ below 0.05 — is based on exact diagonalization for L≤20 and Cesàro averages at t=100. The constrained Hilbert-space dimension grows as F_{L+2}; for L=20 this is ≈17711, so t=100 may be much shorter than the time needed for the Cesàro average to approach its infinite-time limit. No extrapolation procedure, convergence check, or analytic bound on the local fidelity is provided. The monotone trend in Figs. 4(f,g) and 5(g,h) could reverse for L>20. Please add a finite-size scaling analysis (e.g., tensor-network calculations for larger L), an error estimate for the t=100 averages, or at minimum a precise statement of the assumed asymptotic behavior.","section":"§IV, Figs. 4(f,g); §V, Figs. 5(g,h); §VII"},{"comment":"The spectral mechanism in Sec. VI connects local reminiscence to the pure-point part of the spectral measure via Wiener's theorem. However, Wiener's theorem controls the Cesàro average of the global fidelity |ν(t)|², not the single-site local fidelity F_{1-site}(t). Persistence of reduced density matrices does not follow from the pure-point weight of the global survival amplitude. Moreover, the paper's own data undercut the scar-overlap mechanism: the Néel state |Z2⟩ has the largest scar overlap among the states studied, yet (D^max_{1-site})_{t=100}≈0.681, far above the values 0.065 and 0.095 reported for the reminiscent states. High scar overlap is therefore not sufficient. The concluding assertion that 'having a large overlap with the scarred eigenstates results in a good local reminiscence' is not supported.","section":"§VI, Eq. (19); §III.1"},{"comment":"The standard deviation in Eq. (8) is defined with ⟨F_{1-site}⟩_{t'} inside the integrand, i.e., with a running mean that depends on the upper limit of the inner average. This does not measure fluctuations around the long-time average of the signal; for a non-stationary signal it mixes the drift of the running mean with the actual fluctuations. The quantity should be defined with ⟨F_{1-site}⟩_t (the average up to the same final time t) subtracted. Since Figs. 4(g) and 5(h) are central to the 'suppressed fluctuations' claim, the reported values should be recomputed with the corrected definition.","section":"§II.2, Eq. (8)"},{"comment":"The final paragraph contains an incomplete and overreaching statement: 'Although the connection between the overlap with the scarred eigenstates and the local reminiscence, we can certainly say that having a large overlap with the scarred eigenstates results in a good local reminiscence of the system.' This both acknowledges that the connection is not established and then asserts it. Given the |Z2⟩ counterexample and the Wiener-theorem mismatch discussed above, the assertion should be removed or substantially qualified.","section":"§VII, Conclusions"}],"minor_comments":[{"comment":"The text refers to Eq. (19) as an 'upper bound' when the displayed inequality is a lower bound (≥). Please correct the terminology for consistency with the equation and the surrounding discussion.","section":"§VI"},{"comment":"The coefficient of |1,0,φ_{L-2}⟩ appears to be sqrt(F_{L+1}/F_{L+2}) in Eq. (B4), but from Eq. (13) and the Fibonacci counting it should be sqrt(F_L/F_{L+2}). Please check and correct.","section":"Appendix B, Eq. (B4)"},{"comment":"Minor typographical issues: 'Lesbegue' should be 'Lebesgue' (Sec. VI); 'ad eventually' should be 'and eventually' (Sec. VI); 'We first analyze the the overlap' (Sec. V); inconsistent accent usage for 'Néel'.","section":"Throughout"},{"comment":"No data or code availability statement is provided. For reproducibility of the exact-diagonalization results, consider adding such a statement or depositing the numerical data.","section":"General"}],"recommendation":"major_revision","confidential_remarks":null},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Here's my read on Perciavalle et al. The one thing to know: the paper reports a real, non-obvious dynamical distinction — two PXP initial states whose single-site reduced density matrices stay stubbornly close to their initial values even though global fidelity shows no clean revivals — and it includes one very clean analytic result: the single-site reduced density matrix of the blockaded state, Eq. (14), with its Fibonacci coefficients and golden-ratio asymptotics. I checked the normalization, the trace identity (F_{j+1}F_{L-j+2}+F_jF_{L-j+1}=F_{L+2}), and the Z_1(0) golden-ratio limit; all fine. The numerical trend toward <F_1-site>_t > 0.99 and σ < 0.05 is visible for L up to 20 and is consistent with the advertised picture. Credit where due: the analytic Appendix B result is genuinely new and correct, and the framing of local reminiscence as a diagnostic separate from global revivals is worth having.\n\nThe soft spots are real but mostly addressable. First, the thermodynamic claim rests on L ≤ 20 and Cesàro averages at t=100, with no extrapolation, no convergence check, and no analytic handle on the local fidelity. Because the constrained Hilbert space grows as F_{L+2}, t=100 is not obviously in the asymptotic time regime. The observed L-trend could in principle reverse. I don't find that likely, but the paper doesn't give me a reason beyond the plot. Second, the spectral argument leans on Wiener's theorem, which constrains the global survival amplitude |ν(t)|²; it says nothing directly about single-site fidelity. So the chain from pure-point spectral weight to local reminiscence is an inference, not a derivation. Third, and most damaging: the paper's own mechanism — 'large overlap with scarred eigenstates results in good local reminiscence' — is contradicted by its Néel-state data, where |Z2> has the largest scar overlap but the worst local memory (D^max ~0.68 vs 0.065 for the reminiscent states). The authors note the difference but do not reconcile it. That's an overclaim that should be softened or replaced by a more nuanced statement: scar overlap may be necessary but is clearly not sufficient; the structure of the overlap distribution matters. There's also a typo in Eq. (B4) where the second coefficient is printed as sqrt(F_{L+1}/F_{L+2}) instead of sqrt(F_L/F_{L+2}); the main-text Eq. (13) is correct.\n\nWho is this for: anyone working on quantum many-body scars, Rydberg arrays, or diagnostics of non-ergodicity. The analytic blockaded-state result alone is worth a citation. The paper deserves a serious referee — not a desk reject — but I would send it back for revisions: add finite-size extrapolation or at least a discussion of the time-scale issue, fix the mechanism overclaim, and correct the typos. The core observation is likely right; the interpretation needs tightening.","headline":"A useful new local-memory diagnostic with one clean analytic result, but the thermodynamic-limit claim is not yet nailed down and the scar-overlap mechanism is contradicted by the paper's own Néel data.","tokens_in":21446,"tokens_out":4686,"would_cite":true,"duration_ms":30715,"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 claims that two special states of the PXP model retain near-perfect local memory (single-site fidelities above 0.99) even as global dynamics stay complex and non-ergodic, decoupling local memory from global revival strength.","keywords":["local reminiscence","PXP model","quantum many-body scars","local fidelity","Rydberg atom arrays","non-ergodic dynamics","spectral measure","constrained spin systems"],"falsifier":"Compute the time-averaged single-site fidelity for the θ-symmetric state at L=22–30 (e.g., using tensor-network time evolution, since exact diagonalization is prohibitive); if ⟨F_{1-site}⟩_t drops appreciably below 0.99 or its standard deviation rises above 0.05 as L grows, the claimed thermodynamic-limit local reminiscence would fail.","tokens_in":1439,"feed_emoji":"⚛️","tokens_out":2430,"duration_ms":54267,"temperature":0.7,"pith_summary":"The paper studies whether a quantum many-body system can keep stable memory at the level of single sites while its global state evolves in a highly nontrivial way. It introduces the concept of local reminiscence, measured by the fidelity between the initial and time-evolved reduced density matrices of small subsystems. The central finding is that two initial states of the PXP model—the θ-symmetric superposition and the blockaded state—exhibit robust local reminiscence: as the system size grows, the time-averaged single-site fidelity rises toward above 0.99 and its standard deviation falls below 0.05. This happens even though the global fidelity and entanglement entropy show irregular, size-dependent, non-ergodic behavior. The authors connect this local memory to a large overlap of the initial state with the pure-point (discrete) part of the Hamiltonian spectrum, which includes the quantum many-body scar states.","feed_headline":"Two states keep local quantum memory near perfect as systems grow","feed_subtitle":"In the PXP model, single-site fidelities exceed 0.99 while global evolution stays irregular.","key_machinery":"The central objects are the single-site local fidelities F_j(t) between the reduced density matrix of site j at time t and at time 0, their spatial average F_{1-site}(t), and the Césaro time average ⟨F_{1-site}⟩_t with standard deviation. These quantify the stability of local memory. The explanatory mechanism is the spectral measure induced by the initial state: via Wiener's theorem, the long-time averaged global fidelity is bounded below by the total weight on the pure-point spectrum, and the two locally reminiscent states have large overlap with the scarred eigenstates, which the paper identifies as belonging to the discrete part of the spectrum of the PXP Hamiltonian.","core_discovery":"The paper's core claim is that the θ-symmetric state |Θ_symm+⟩ (at θ=π/4) and the blockaded state |φ_L⟩ generate dynamics in which the reduced density matrices of single sites remain close to their initial forms over long times. Quantitatively, the time-averaged single-site fidelity ⟨F_{1-site}⟩_t approaches values above 0.99 and its standard deviation falls below 0.05 as L increases (see Figs. 4(f,g) and 5(g,h)), while the long-time averaged maximum trace-distance bound (D̄^max_{1-site})_{t=100} is about 0.065 and 0.095, respectively. Meanwhile the global fidelity and entanglement entropy exhibit non-ergodic, oscillatory, volume-law-scaling behavior that does not settle to a simple revival","pith_inferences":["A natural testable extension is to scan other constrained models (deformed PXP, Fibonacci anyon chains) using the local-reminiscence criterion; any state with sufficiently large pure-point spectral weight should show the effect even if global fidelity revivals are weak.","Because local memory persists while entanglement grows volume-law, the preserved information may live in local coherences rather than local populations—a direct experimental probe would be measuring single-site transverse magnetization (X or Y) alongside populations.","The connection to non-Markovianity, mentioned in the paper as related, could be sharpened: local reminiscence is essentially the fidelity-side signature of non-Markovian local dynamics, and the same Césaro-averaged quantities could quantify both.","If the pure-point overlap is the causal mechanism, then engineering superpositions with controlled overlap onto scar states should allow one to tune local reminiscence on and off, which could be verified in Rydberg-array experiments."],"forward_implications":["Local fidelity becomes a diagnostic that can reveal memory retention even when global fidelity and entanglement entropy suggest complex, non-thermal dynamics.","The previously identified quantum many-body scar states are part of the pure-point spectrum of the PXP Hamiltonian, supporting their role in non-ergodic dynamics.","Initial states with larger overlap with scar states—such as the θ-symmetric and blockaded states—show stronger and more stable local reminiscence than the Néel state.","In the simulated sizes, local reminiscence is enhanced, not destroyed, as the system size increases, with time-averaged fidelity growing and fluctuations shrinking.","Local reminiscence is most pronounced for single-site subsystems and fades as the block size grows, indicating it is a genuinely local phenomenon."],"fun_headline_variants":["Two PXP states preserve single-site memory near perfectly","Local fidelity above 0.99 in two PXP states","PXP model: two states shield local info while global chaos unfolds","θ-symmetric and blockaded states exhibit robust local reminiscence","Single-site memory persists in PXP despite global non-ergodicity"],"cache_read_input_tokens":22656,"weakest_assumption_plain":"The claim rests on the observed monotone finite-size trends at L up to 20—that the time-averaged single-site fidelity keeps rising toward 0.99 and its fluctuations keep falling below 0.05—being representative of the thermodynamic limit, since no analytic bound or extrapolation procedure is provided and the pure-point spectral weight could in principle melt into the continuum at larger L.","fun_headline_variants_meta":{"raw":{"variants":["Two PXP states preserve single-site memory near perfectly","Local fidelity above 0.99 in two PXP states","PXP model: two states shield local info while global chaos unfolds","θ-symmetric and blockaded states exhibit robust local reminiscence","Single-site memory persists in PXP despite global non-ergodicity"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.00025,"raw_usage":{"total_tokens":1381,"prompt_tokens":724,"completion_tokens":657,"prompt_tokens_details":{"cached_tokens":256},"prompt_cache_hit_tokens":256,"prompt_cache_miss_tokens":468,"completion_tokens_details":{"reasoning_tokens":569}},"tokens_in":468,"tokens_out":657,"duration_ms":4847,"temperature":1.0,"reasoning_tokens":569,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-04T15:17:02.439557+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Compute the time-averaged single-site fidelity for the θ-symmetric state at L=22–30 (e.g., using tensor-network time evolution, since exact diagonalization is prohibitive); if ⟨F_{1-site}⟩_t drops appreciably below 0.99 or its standard deviation rises above 0.05 as L grows, the claimed thermodynamic-limit local reminiscence would fail.","supporting_citations":[],"review_version":1}