{"id":"e4a638cb-58ac-4cd3-a0ae-9f603ee07659","arxiv_id":"2505.23910","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":5.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":1,"one_line_summary":"The exponent of the long-time power-law decay of the time-averaged survival probability equals the box-counting fractal dimension D2 in random matrix ensembles, the interacting Aubry-André model, and the disordered Heisenberg chain.","lead":"This paper shows that the power-law decay of a time-averaged survival probability gives the correlation fractal dimension D2 of quantum states, including in interacting many-body systems. The result offers a dynamical route to detecting extended nonergodic states without finite-size scaling.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The equality ν = Dbox_2 in the interacting many-body models rests on a single system size (L=16) and manually selected fitting windows, so finite-size contamination of the extracted power-law exponent is not ruled out.","rationale":"The reader's weakest assumption is exactly the point I find most load-bearing: the universal identification ν = Dbox_2 is assumed for interacting systems and tested only at L=16. My concern sharpens this by identifying the specific mechanism that could break the identification — spectral discreteness and correlation-hole contamination of the fitted power-law regime — and the concrete finite-size test that would settle it. I do not see a deeper flaw in the GOE derivation or in the random-matrix benchmarks, where N=12,000 provides a clean scaling regime. The manuscript is internally consistent; the issue is whether the many-body demonstration is representative of the thermodynamic limit. Since the authors themselves acknowledge ongoing uncertainty about the Heisenberg model's localized phase (Sec. IV B), the lack of a larger-size check is especially consequential. My proposed test is computationally feasible and would either confirm the equality at larger L or reveal that the L=16 agreement is a finite-size artifact. Given the reader's verdict was already CONDITIONAL with moderate confidence, my analysis supports that verdict rather than changing it.","tokens_in":18101,"tokens_out":1958,"duration_ms":22908,"concrete_test":"Recompute ν for the interacting Aubry-André model at h=1.4 (extended nonergodic phase) for L=14, 16, and 18 using Krylov or Chebyshev time evolution, with a pre-registered fitting-window rule (e.g., start after the first minimum of ⟨SP(t)⟩ and end when the running average deviates by more than 2% from a pure power law). If ν changes by more than 0.05 between L=14 and L=18, or if the usable fitting window shrinks to less than one decade at L=18, the L=16 agreement between ν and Dbox_2 does not support the thermodynamic claim.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim requires that the long-time power-law exponent ν of the time-averaged survival probability in Eq. (5) equals the box-counting fractal dimension Dbox_2 of Eq. (8) for interacting many-body systems. This is established for the Aubry-André and Heisenberg models only at L=16, N=12,870 (Secs. IV A and IV B, Figs. 6-8). The analytical justification that time averaging filters out spectral correlations is derived only for GOE matrices (Appendix A); no analogous argument is given for interacting systems. At L=16, the spectrum is strongly discrete, and the correlation-hole time scale may exceed the fitted window, so ν could reflect the pre-saturation decay of the survival probability rather than the asymptotic fractal structure. Moreover, the fitting windows are chosen 'the same for both exponents' but no procedure, error bars, or stability checks are reported (Figs. 2, 8, 9). Because Dbox_2 is also extracted from the same finite system using a hand-picked linear region of P(l), the observed agreement ν ≈ Dbox_2 could be a finite-size coincidence: both quantities may be renormalized in the same way at L=16 while both deviate from the thermodynamic D2. This would invalidate the advertised scalable extraction of D2 from dynamics.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The manuscript proposes that the long-time power-law exponent ν of the time-averaged survival probability, Eq. (5), equals the box-counting fractal dimension Dbox2 of Eq. (8), and that this equality allows D2 to be extracted from dynamics alone, without finite-size scaling. The authors test this claim on GOE random matrices, the Rosenzweig-Porter (RP) model, power-law banded random matrices (PBRM), the interacting Aubry-André model, and the disordered Heisenberg spin chain. For GOE they provide an explicit analytical derivation of the full time-averaged survival probability in Appendix A and verify it numerically. For the other models they compare γ, ν, DIPR0_2, and Dbox2 as functions of the relevant control parameter and report excellent agreement between ν and Dbox2 in all cases, while γ and DIPR0_2 deviate in extended nonergodic and chaotic regimes. The conclusion is that the time-averaged survival probability is a robust and scalable dynamical probe of multifractality in many-body systems.","tokens_in":18333,"tokens_out":4325,"duration_ms":45401,"significance":"If the central equality ν = Dbox2 holds for interacting many-body systems, the paper offers a genuinely useful tool: it would replace multi-size scaling analyses of eigenstates with a single dynamical simulation, and it connects to tensor-network time-evolution methods that reach larger sizes than exact diagonalization. The GOE derivation is a concrete strength: the analytical expression in Appendix A is explicit, the asymptotic ν = 1 is derived, and the numerical curves agree with it. The RP and PBRM results are also valuable benchmarks because they scan ergodic, extended nonergodic, and localized regimes. However, the many-body part of the claim, which is the advertised main advance, rests on a single system size and on fitted exponents without uncertainty estimates; this is the load-bearing gap that prevents the paper from being accepted in its present form.","major_comments":[{"comment":"The central many-body claim, ν ≈ Dbox2, is demonstrated only at L = 16 (N = 12,870), a single system size. Since both ν from Eq. (5) and Dbox2 from Eq. (8) are extracted from the same finite system with hand-picked fitting windows, the agreement could be a common finite-size renormalization rather than the thermodynamic equality asserted in the abstract and conclusions. Please provide results for at least two or three additional sizes (e.g., L = 12, 14, 16) with error bars, showing that ν and Dbox2 converge to the same value, or demonstrate the dynamical method on larger systems using the tensor-network/MPO time evolution advertised in the Introduction. Without this, the scalable extraction claim is not supported by the data.","section":"Sec. IV A/B, Figs. 6–8"},{"comment":"The filtering property that time averaging removes spectral correlations and leaves only correlations among eigenstate components is derived analytically only for GOE matrices (Eq. (12) and Appendix A). For the interacting Aubry-André and Heisenberg models (Secs. IV A/B) no analogous argument is given, and at L = 16 the discrete spectrum and finite correlation-hole time scale can contaminate the fitted power-law window. The manuscript should state a concrete criterion for choosing the fitting windows for γ, ν, and the P(l) scaling region, and should quantify sensitivity to window endpoints for the many-body data. As written, ν from Eq. (5) may reflect the pre-saturation dynamics of a finite system rather than the asymptotic fractal dimension D2.","section":"Sec. II A and Appendix A"},{"comment":"No error bars or statistical uncertainties are reported for the fitted exponents γ, ν, DIPR0_2, or Dbox2, and the fitting windows are described only as 'the same for both exponents' with no procedure for their selection. Because every conclusion in the paper is a comparison of fitted exponents, the absence of uncertainty quantification makes it impossible to judge whether the reported agreement between ν and Dbox2 is significant or within expected numerical scatter. Please report bootstrap or realization-to-realization errors and explicitly list the fitting intervals used in each figure.","section":"Figs. 2, 8, 9"},{"comment":"The paper's stated practical advantage is that the dynamical approach 'avoids the need for scaling procedures and enables access to larger systems than those typically reachable via exact diagonalization.' This is not demonstrated for the many-body models: all many-body data in Secs. IV A/B come from exact diagonalization at L = 16, the same scale as standard finite-size scaling studies. A concrete demonstration on a larger system, even for one parameter point, using the tensor-network methods referenced in the Introduction would be needed to substantiate this claim.","section":"Introduction and Conclusions"}],"minor_comments":[{"comment":"The same angle brackets ⟨·⟩ are used for the disorder average and for the time average in Eq. (4), which is confusing; consider a different notation such as an overline for the time average.","section":"Eq. (4)"},{"comment":"In the sentence 'The figures also reiterates that' the verb should agree with the plural subject; it should read 'The figures also reiterate that.'","section":"Sec. IV B"},{"comment":"The text says 'The bottom left and right Fig. 8(d) show the analysis of ⟨P(l)⟩,' but the P(l) panels are labeled as Fig. 8(e); please correct the cross-reference.","section":"Appendix B"},{"comment":"The color assignments for the α values in the caption of Fig. 3 are ambiguous; labeling the curves directly or using distinct markers would improve readability.","section":"Fig. 3 caption"},{"comment":"References [106] and [107] appear to be the same article by Xu et al. with slightly different titles; please merge or remove the duplicate.","section":"Refs. [106] and [107]"},{"comment":"The phrase '10 4 samples' appears twice and should be written as '10^4 samples'.","section":"Sec. II A"}],"recommendation":"major_revision","confidential_remarks":"For the editor: the manuscript is from an experienced group and the reference list is appropriate. My recommendation is based on evidence, not on any suspicion of impropriety. The GOE derivation and the random-matrix benchmarks are solid, but the many-body central claim—the paper's main novelty—rests on one system size and on fits without error bars. This is fixable within the scope of the manuscript if the authors add finite-size checks and uncertainty estimates; hence major revision rather than rejection."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Core message: the random-matrix part is solid, the GOE analytic result is a genuine addition, and the many-body part is a promising but unproven extrapolation from a single system size. Read it for the RP/PBRM comparisons and the derivation; treat the interacting-model claim with caution.\n\nWhat's new: the identity nu = Dbox2 was already explored for noninteracting quasiperiodic systems, and the paper says so. The new content is the systematic comparison across GOE, RP, PBRM, and two interacting spin chains, plus the analytical expression for the time-averaged survival probability in the GOE case. The GOE derivation in Appendix A is explicit, the numerical curves match it, and the RP/PBRM results at N=12000 show convincing agreement between nu and Dbox2 across the phases. The distinction from gamma and from DIPR0 is also well demonstrated—it makes the case that time averaging removes the spectral correlations that corrupt the survival-probability exponent.\n\nWhere it is soft: the interacting many-body results are all at L=16, with no error bars on the fitted exponents and no stated criterion for choosing the fitting windows. The box-counting dimension is also extracted from a hand-picked linear region of P(l). The filtering argument is only proven for GOE; for the spin models it is an assumption. Given that both nu and Dbox2 are extracted from the same finite system, the observed equality could be a finite-size coincidence, with both quantities renormalized in the same way at L=16 while both deviate from the thermodynamic D2. The scalability promise—accessing larger systems via tensor networks—is not demonstrated; the paper only suggests it. No code or data are provided, which makes the fitting choices hard to audit.\n\nNone of this kills the core idea. The random-matrix evidence is strong, and the many-body observation is plausible. But the central universal claim needs more support before it can be taken as established. A serious referee should ask for error bars, a fitting-window protocol, and at least one check at larger size or with tensor-network methods.\n\nRecommendation: send it to peer review. The random-matrix section and the GOE derivation justify referee time, and the many-body section is addressable with additional analysis rather than being fundamentally wrong.","headline":"Useful dynamical probe of D2 in random-matrix settings with a solid GOE benchmark, but the many-body extension rests on single-size hand-picked fits and needs stability checks before it can be trusted.","tokens_in":18893,"tokens_out":3688,"would_cite":true,"duration_ms":32653,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":["05.45.Mt","71.30.+h"],"model":"deepseek-v4-flash","headline":"The paper argues that the exponent of the long-time power-law decay of the time-averaged survival probability equals the correlation fractal dimension $D_2$, so fractal dimensions of many-body quantum states can be read off directly from…","keywords":["fractal dimension","survival probability","time-averaged survival probability","multifractality","extended nonergodic states","many-body localization","Aubry-André model","Rosenzweig-Porter model"],"falsifier":"Compute $\\nu$ from $\\langle SP(t)\\rangle$ and $D^{\\mathrm{box}}_2$ from the same initial states in the interacting Aubry-André model at $L=18$ or $20$ with a time-evolution algorithm: a systematic discrepancy, or the disappearance of a clean power-law window in the intermediate regime $0.7<h<1.7$, would falsify the claimed equality.","tokens_in":17890,"feed_emoji":"⚛️","tokens_out":10022,"duration_ms":86030,"temperature":0.7,"pith_summary":"Fractal dimensions of quantum states are usually extracted by scaling the inverse participation ratio across many system sizes, a costly procedure that struggles near critical points. The paper argues that the exponent $\\nu$ of the long-time power-law decay of the time-averaged survival probability — the fidelity between an initial state and its time-evolved version, averaged over time — equals the correlation fractal dimension $D_2$ of the initial state in the energy eigenbasis. This equality is demonstrated analytically for Gaussian orthogonal random matrices and numerically for Rosenzweig-Porter matrices, power-law banded random matrices, the interacting Aubry-André model, and the disordered Heisenberg chain, where $\\nu$ matches the box-counting value $D^{\\mathrm{box}}_2$ across ergodic, extended nonergodic, and localized regimes. Because a single time trace replaces finite-size scaling, the method opens access to larger systems and gives an unambiguous basis in which the fractal dimension is defined. A practical consequence is that extended nonergodic phases can be identified dynamically, without state tomography.","feed_headline":"Time-averaged survival probability decays with exponent D2","feed_subtitle":"No finite-size scaling needed: one time trace gives a direct probe of extended nonergodic quantum states.","key_machinery":"The central object is the time-averaged survival probability $\\langle SP(t)\\rangle = \\frac{1}{t}\\int_0^t \\langle SP(\\tau)\\rangle\\, d\\tau$, whose power-law tail exponent $\\nu$ is claimed to equal $D_2$. Its role is to act as a filter: time averaging removes the spectral contributions that contaminate the raw survival probability — the Bessel-function oscillations from the edges of the energy distribution and the correlation hole from level repulsion — leaving only the correlations among the components $C^{(0)}_n$ of the initial state in the Hamiltonian eigenbasis. The independent definition of $D_2$ used for comparison is the box-counting method, $P(l)\\propto l^{D_2}$, with $P(l)$ built from the squared projections of the initial state inside energy boxes of linear size $l$. In the Gaussian orthogonal ensemble case the machinery is fully explicit: integrating the analytical survival probability (Eq. 10) yields Eq. (12), whose large-time limit is $\\langle SP(t\\to\\infty)\\rangle \\approx \\frac{8}{9\\pi}\\frac{N}{\\Gamma t} SP - \\frac{1}{6}\\frac{N}{\\Gamma t} SP + SP$, giving $\\nu = 1$ with no correlation hole.","core_discovery":"The paper's central claim is that the long-time power-law exponent $\\nu$ of the time-averaged survival probability, $\\langle SP(t)\\rangle \\propto t^{-\\nu}$, coincides with the correlation fractal dimension $D_2$ obtained by box counting, $P(l)\\propto l^{D_2}$, where $P(l)$ sums the squared amplitudes of the initial state inside energy boxes of size $l$. The claim is tested in five settings: full random matrices from the Gaussian orthogonal ensemble, the Rosenzweig-Porter ensemble, power-law banded random matrices, the interacting Aubry-André model, and the disordered spin-$1/2$ Heisenberg chain. In every case, $\\nu$ agrees with $D^{\\mathrm{box}}_2$ across the ergodic, extended nonergodic, and localized phases, even where the raw survival-probability exponent $\\gamma$ or the IPR-based fractal dimension $D^{\\mathrm{IPR}_0}_2$ deviate. For the Gaussian orthogonal ensemble the entire evolution of $\\langle SP(t)\\rangle$ is derived analytically (Eq. 12): Bessel oscillations and the correlation hole are averaged away, and the asymptotic decay is exactly $t^{-1}$, so $\\nu = D_2 = 1$. The conclusion is that the time-averaged survival probability filters out spectral correlations and leaves a decay controlled solely by correlations among eigenstate components, making it a direct dynamical observable for $D_2$.","pith_inferences":["A testable extension is to push the equality $\\nu = D_2$ to larger system sizes in the interacting Aubry-André and Heisenberg models using time-evolution algorithms, checking whether a clean power-law window persists near the many-body localization crossover; the paper tests only $L=16$ for the interacting models.","If the equality holds across the transition region, the time-averaged survival probability could become a dynamical order parameter for many-body localization, locating the transition by where $\\nu$ (and hence $D_2$) extrapolates to zero.","The faster decay of $\\nu$ and $D^{\\mathrm{box}}_2$ relative to $D^{\\mathrm{IPR}_0}_2$ in intermediate regimes hints that box-counting and dynamics are more sensitive to weak multifractality than IPR scaling; a possible reason is that energy-box sums weight clusters of nearby components, an interpretation the paper does not develop."],"forward_implications":["If $\\nu = D_2$ holds, the fractal dimension of an initial state can be extracted from a single long-time dynamical trace, replacing the need to diagonalize many system sizes and perform finite-size scaling.","Because the method uses time evolution rather than full exact diagonalization, it is compatible with large-scale evolution algorithms and can reach larger system sizes than IPR-based scaling studies.","The basis is fixed to the energy eigenbasis, so the extracted $D_2$ is not affected by the choice of real-space or momentum-space basis that complicates eigenstate-based IPR analyses.","In extended nonergodic phases, $\\nu$ and $D^{\\mathrm{box}}_2$ agree even when the IPR-based $D_2$ does not, so the dynamical exponent is a sharper indicator of nonergodicity.","Since the survival probability is a fidelity, the approach may be usable in experiments without state tomography, as a direct probe of multifractal and extended nonergodic phases."],"supporting_citations":[{"why":"Supplies the foundational result that the correlation dimension of a quantum state controls the slow decay of temporal correlations; the paper's $\\nu = D_2$ claim is a direct extension of this relation.","marker":"[4]"},{"why":"Establishes power-law decay of the survival probability at the many-body localization transition, the context the paper's dynamical method generalizes to larger systems.","marker":"[19]"},{"why":"Provides the analytical form of the survival probability under full random matrices that the paper integrates to obtain the time-averaged expression in Eq. (12).","marker":"[21]"},{"why":"Documents strong fluctuations and slow convergence of IPR-based fractal dimensions near the many-body localization transition, motivating the dynamical alternative.","marker":"[23]"},{"why":"Shows scale-invariant survival probability at eigenstate transitions, supporting the use of survival-probability exponents as probes of multifractality.","marker":"[51]"},{"why":"Gives the generic dynamical features for quenched systems, including the random-matrix formula behind the GOE analytical analysis.","marker":"[68]"},{"why":"Identifies power-law decays with exponent set by spectral bounds, explaining why $\\gamma$ in the raw survival probability carries spectral rather than fractal information.","marker":"[89]"}],"fun_headline_variants":["Fractal dimension from a single time trace","Dynamical probe of nonergodic quantum states","Survival probability yields fractal dimension directly","One time trace gives D2 in many-body systems"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The method assumes that in the time window chosen for the fit, the time-averaged survival probability has already lost all spectral correlations (level repulsion, spectral edges, and the correlation hole), so its decay exponent is governed only by correlations among the eigenstate components; the paper proves this filtering only for the GOE and checks it numerically at $L=16$ for the interacting models.","fun_headline_variants_meta":{"raw":{"variants":["Fractal dimension from a single time trace","Dynamical probe of nonergodic quantum states","Survival probability yields fractal dimension directly","One time trace gives D2 in many-body systems"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000588,"raw_usage":{"total_tokens":2812,"prompt_tokens":1051,"completion_tokens":1761,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":667,"completion_tokens_details":{"reasoning_tokens":1702}},"tokens_in":667,"tokens_out":1761,"duration_ms":13369,"temperature":1.0,"reasoning_tokens":1702,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-07T12:38:35.318581+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Compute $\\nu$ from $\\langle SP(t)\\rangle$ and $D^{\\mathrm{box}}_2$ from the same initial states in the interacting Aubry-André model at $L=18$ or $20$ with a time-evolution algorithm: a systematic discrepancy, or the disappearance of a clean power-law window in the intermediate regime $0.7<h<1.7$, would falsify the claimed equality.","supporting_citations":[{"cited_title":"Hopjan and L","cited_arxiv_id":null,"evidence_quote":"Shows scale-invariant survival probability at eigenstate transitions, supporting the use of survival-probability exponents as probes of multifractality."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Gives the generic dynamical features for quenched systems, including the random-matrix formula behind the GOE analytical analysis."}],"review_version":1}