{"id":"8353b669-80b6-4ef6-8a5e-4fded951e9f0","arxiv_id":"2411.18921","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":2,"one_line_summary":"Approximate quantum many-body states from various variational methods exhibit a nearly exponential spectral tail characterized by a small inverse effective temperature, with a two-stage behavior when targeting imaginary-time evolved states.","lead":"This paper shows that approximate quantum states produced by neural networks, tensor networks, and quantum circuits have energy spectra shaped like a hot Boltzmann distribution with a small effective temperature. The authors propose this temperature as a diagnostic for how well a given numerical method can represent quantum states.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The inverse effective temperature β̃ is protocol-dependent: unspecified fit windows/weights and sector restrictions can flip its sign (Fig. 2b) and change its value (Fig. S13), so the claimed universal exponential decay is not yet established.","rationale":"The reader's weakest-assumption identifies the robustness of the fitted slope β̃ as the key methodological gap. My analysis confirms that this is the single most load-bearing concern: every quantitative statement in the paper—the universal exponential decay, the comparison of effective temperatures across ansatzes, and the beta* values—depends on β̃ being a well-defined, protocol-independent quantity. The paper's own figures show that this is not the case: negative slopes for converged ground states and large changes from sector selection demonstrate that the fitted slope is not an intrinsic property of the approximate state but an artifact of the regression choices. Without a fixed fitting protocol, the central claim is underdetermined.\n\nOther concerns, such as the small system sizes (L ≤ 16) and the overuse of terms like 'universal' and 'phase transition,' are valid but secondary. The ITES two-stage behavior is a genuine numerical finding, supported by additional diagnostics in the SM (MSE, Pearson r², L-BFGS checks), and could survive even if the exponential-decay claim is softened. However, the quantitative beta* values inherit the same protocol dependence, so the two-stage claim also needs the fitting protocol specified.\n\nThe paper is not internally inconsistent or circular; it reports honest numerical observations. The issue is that the main metric is not robustly defined. This is addressable by a re-analysis with explicit protocols and by releasing data/code, which is why the reader's CONDITIONAL verdict remains appropriate. My concern reinforces the need for those conditions rather than moving the verdict to reject.","tokens_in":18825,"tokens_out":5439,"duration_ms":50918,"concrete_test":"Obtain the raw (ε_i, |c_i|^2) data for the converged MPS ground state in Fig. 2(b) and the VQE state in Fig. S13, and recompute β̃ under four explicit protocols: (i) unweighted least squares on all eigenstates, (ii) unweighted least squares on states with |c_i|^2 > 10^{-8}, (iii) fit restricted to the 20 lowest excited states, and (iv) full-sector versus half-filling-sector regression. If the sign or magnitude of β̃ changes by more than 0.1 across these choices (as the current figures already suggest), then the paper must specify and justify a fixed, pre-registered fitting protocol before the universal exponential-decay claim can be evaluated.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim that approximate states exhibit a universal exponential spectral decay |c_i|^2 ∝ exp(−β̃ ε_i) (Eq. 4) rests entirely on an unspecified linear fit of log |c_i|^2 versus ε_i. The paper never states the fitting window, the weighting, or the treatment of degenerate or near-zero coefficients. This is not cosmetic: the same data can yield different slopes depending on the protocol. Fig. 2(b) and S10(b) show converged ground-state approximants with β̃ ≈ −0.114 and −0.165, i.e., the coefficients increase with energy, contradicting the exponential-decay framing. For VQE, fitting the full spectrum gives β̃ ≈ 0.477–0.799, while restricting to the half-filling sector gives β̃ ≈ 0.284–0.317 (Fig. S13); the paper reports only the sector-restricted values. Thus the reported β̃ depends on sector choice and fit window, so the claimed universality of a small positive β̃ is not established. The ITES two-stage behavior is more robust, but the quantitative β* values inherit the same protocol dependence. The SM's argument against numerical-precision artifacts does not address this fitting ambiguity.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper studies the spectral decomposition of approximate quantum many-body states (obtained from MPS, PEPS, NQS/VMC, VQE, and fully parameterized vector states) into the exact eigenstates of the target Hamiltonian. It claims that the squared overlaps |c_i|^2 as a function of eigenenergy ε_i roughly follow an exponential decay |c_i|^2 ∝ exp(−β̃ ε_i), defining β̃ as an inverse effective temperature. The authors report small positive β̃ for most approximate ground states, a non-monotonic training dynamics of β̃, and a two-stage behavior when approximating imaginary-time evolved states (ITES), with a critical β* separating a regime where β̃ ≈ β from a regime where β̃ < β. The findings are presented as universal across ansatzes, objectives, and lattice geometries.","tokens_in":18997,"tokens_out":3398,"duration_ms":31735,"significance":"If the exponential spectral-decay pattern is shown to be robust, the effective temperature could provide a useful, fine-grained diagnostic for variational quantum and classical methods, complementing global metrics such as fidelity. The paper has notable strengths: it covers a wide range of ansatzes and objectives, provides extensive supplemental material with alternative metrics (MSE, Pearson coefficient, fit uncertainty), and includes a credible argument against a pure machine-precision explanation of the spectral plateau. The use of open-source software (TensorCircuit-NG, QuSpin) supports reproducibility. However, the central claim is currently vulnerable to the underspecified fitting procedure used to define β̃, and the reported universality is not yet established.","major_comments":[{"comment":"The extraction of β̃ via a linear fit to log |c_i|^2 versus ε_i is never specified: the paper does not state the fitting window, the weighting of points, the treatment of degenerate eigenstates, or how near-zero overlaps are handled. This is a load-bearing omission because the same spectral data can yield different slopes under different protocols. Concretely, Fig. 2(b) and Fig. S10(b) report converged ground-state approximants with β̃ ≈ −0.114 and −0.165, respectively, meaning the fitted coefficients increase with energy, which contradicts the 'exponential decay' wording of Eq. (4) and the claim of a universal small positive β̃. The manuscript should specify the fitting protocol exactly and demonstrate that the reported β̃ values are stable with respect to the fit window and weighting.","section":"Results for approximate ground states, Eq. (4)"},{"comment":"The reported β̃ depends on the charge-sector restriction. For VQE on the 4×3 lattice, fitting the full spectrum gives β̃ ≈ 0.477–0.799, whereas restricting to the half-filling sector gives β̃ ≈ 0.284–0.317 (Fig. S13), and the main text reports only the sector-restricted values. Similarly, the 4×4 calculations restrict spectral decomposition to the half-filling sector for VQE and other methods. Since the sector choice changes the numerical value of the central metric, the claimed universality of small positive β̃ is not established unless the sector selection is justified on physical grounds and consistently applied. The authors should report full-spectrum fits alongside sector-restricted fits and discuss the sensitivity.","section":"Results for approximate ground states and Fig. S13"},{"comment":"The two-stage behavior for ITES and the critical value β* are derived from the same underspecified single-exponential fit, and for large β the spectrum is explicitly not exponential over the full energy range (Fig. 2(d) and Fig. S10(d)). A single fitted slope β̃ is therefore not a valid characterization of such spectra, and β* inherits the fitting ambiguity. In addition, the high-temperature regime where β̃ ≈ β is expected almost by construction when the fidelity objective directly targets |ϕ(β)⟩, so this regime does not independently test the exponential law. The manuscript should define β* via an explicit, reproducible criterion (for example, the point where the MSE or the fit uncertainty crosses a stated threshold) and show that the qualitative conclusions are robust to the criterion chosen.","section":"Results for approximate imaginary-time evolved states, Fig. 4"}],"minor_comments":[{"comment":"The phrase 'The universality and validness of the results' should read 'validity'.","section":"Results for approximate ground states"},{"comment":"The abbreviation 'aka.' should be written as 'a.k.a.' or 'also known as' in the main text and figure captions.","section":"Throughout"},{"comment":"In the sentence 'This objective reduces to the ground state target in the β → ∞limit', insert a space before 'limit'.","section":"Introduction"},{"comment":"The abstract states that the effective temperature 'shows phase transition behaviors', while the Discussion more cautiously calls it a 'putative phase transition'; the wording should be aligned to avoid overclaiming.","section":"Abstract and Discussion"},{"comment":"The text uses 'VMC' for the neural-network ansatz while clarifying that sampling is not used; to avoid confusion with standard variational Monte Carlo, consider using 'NQS' consistently in the main text and figures.","section":"Methods, Neural quantum states"}],"recommendation":"major_revision","confidential_remarks":"The manuscript's central claim rests on a fitting protocol that is not fully described. I would ask the editor to ensure the revision includes a precise, reproducible definition of the fitting procedure and a sensitivity analysis, since the current evidence for universality is not yet convincing. The paper is within scope for a quantum-information/many-body journal, and the question is scientifically interesting, but the presented analysis needs to be tightened before publication."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Two things worth knowing. The paper's real contribution is the two-stage behavior of the fitted inverse temperature β̃ when targeting imaginary-time evolved states, with a critical β* that appears to track ansatz expressiveness; that part is new and reasonably well supported. The broader claim that every variational approximate state has a universal exponential spectral decomposition is overstated, mainly because β̃ is extracted from an underspecified linear fit and changes with the chosen charge sector.\n\nThe numerical survey itself is honest work: MPS, PEPS, NQS, VQE, and VEC on small 2D XXZ ladders, with careful SM diagnostics (Pearson r², fit uncertainty, MSE, L-BFGS checks) that rule out the most naive machine-precision artifact. The links to Ref. 28's sampling fragility and the expressiveness-vs-optimization explanation for the fidelity curves in Fig. 4b are sensible. The paper credits the QAOA and MPS precursors correctly.\n\nThe soft spots are real but fixable. The fitting protocol is not stated: no energy window, no weighting, no treatment of degeneracies. Since β̃ is the slope of that fit, 'exponential' is partly an artifact of having fitted a line. Negative β̃ values appear for converged ground states (Fig. 2b, S10b) without discussion; they contradict the 'decay' framing, though the paper's own Fig. 1 caption concedes β̃ can be negative. The VQE sector restriction changes β̃ by roughly a factor of two (Fig. S13), so the metric is not protocol-independent; the paper reports the half-filling values transparently but the 'universal' language should be softened. The 'phase transition' is already hedged as 'putative,' so that's fine.\n\nFor someone working on variational quantum many-body diagnostics, this is a useful paper to engage with. I would send it to peer review and ask for a precise description of the fitting procedure, a comment on the negative β̃ cases, and a more careful statement of what 'universal' means. Accept after revision.","headline":"Useful diagnostic survey with a real ITES two-stage finding, but the 'universal exponential spectrum' claim is oversold until the fitting protocol is pinned down.","tokens_in":19624,"tokens_out":3897,"would_cite":true,"duration_ms":36958,"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":"Approximate many-body states from variational optimization share a universal spectral pattern: squared overlaps with exact eigenstates decay exponentially in energy, with a small decay factor β̃ the paper calls the inverse effective…","keywords":["effective temperature","spectral decomposition","tensor networks","neural quantum states","variational quantum eigensolver","imaginary-time evolution","quantum many-body systems","exponential decay"],"falsifier":"For the 4×4 XXZ model in the half-filling sector, take a converged MPS ground-state approximant optimized with the fidelity objective and fit log|ci|² against εi separately over the lower third and upper third of the spectrum; if the two fitted slopes differ by more than the fit uncertainty, or if one is negative while the other is positive, the claimed universal single-exponential law fails for that state.","tokens_in":18496,"feed_emoji":"⚛️","tokens_out":8684,"duration_ms":76662,"temperature":0.7,"pith_summary":"This paper tries to establish that approximate wavefunctions produced by variational optimization—matrix product states, projected entangled-pair states, neural quantum states, variational quantum circuits, and fully parameterized vectors—all show the same spectral fingerprint when decomposed into the exact eigenstates of the target Hamiltonian: the squared overlap with eigenstate i decays roughly exponentially with its energy, |c_i|^2 ∝ exp(−β̃ ε_i), with a small decay rate β̃ interpreted as an inverse effective temperature. The authors argue this pattern is universal across ansatz structure, optimization objective, and physical system, and that β̃ tracks ansatz expressiveness and training progress. They also introduce imaginary-time evolved target states with exact exponential spectra and show a two-stage behavior: for small inverse temperature β the approximate state matches the target, while beyond a critical β* the high-energy part of the spectrum flattens and the fitted β̃ drops below β. If the picture holds, β̃ becomes a practical diagnostic for when a variational simulation is effectively hot and for comparing the capacity of different ansatzes.","feed_headline":"Approximate quantum states share one exponential spectral law","feed_subtitle":"Across ansatzes, eigenstate overlaps decay exponentially with a small fitted beta, offering a new accuracy diagnostic.","key_machinery":"The machinery is the spectral decomposition |ψ⟩ = ∑_i c_i|ε_i⟩ and the exponential fit |c_i|² ∝ $e^{{−β̃ ε_i}}$; β̃ is obtained by linear regression of log|ci|² against εi over the full spectrum or, for VQE and the 4×4 lattice, the half-filling sector. The companion object is the imaginary-time evolved state |φ(β)⟩ = Z(β)^{−1}∑_i $e^{{−β ε_i/2}}$|ε_i⟩, a family whose exact spectra are exponential by construction and which provides a controlled test of whether an ansatz can represent a given inverse temperature. The transition value β* is defined by the point where the fitted slope begins to deviate from the target β.","core_discovery":"On the paper's own terms, the central discovery is an empirical rule: for approximate ground states obtained by minimizing energy or infidelity, plotting log|ci|² against eigenenergy εi yields an approximately straight line over a wide range of energies, corresponding to |ci|² ∝ $e^{{−β̃ ε_i}}$. The fitted slope β̃ is small—typically below about 0.3, and sometimes negative for converged fidelity-optimized states—so the excited-state spectrum of an approximate ground state is nearly flat. For imaginary-time evolved targets |φ(β)⟩ = $Z^{{-1}}$∑ $e^{{−β ε_i/2}}$|ε_i⟩, the optimized approximate states reproduce the target exponential spectrum for β below a critical β*, while for β > β* only the low-energy part decays exponentially and the high-energy overlaps form a plateau; the paper presents evidence that this plateau is not a machine-precision artifact and that β* is characteristic of the ansatz's expressive power.","pith_inferences":["If the exponential spectrum is a generic outcome of gradient-based variational optimization, the same fit should appear for random or badly converged approximate states; testing this would separate a property of the ansatz family from a property of the optimization dynamics.","The definition of β̃ depends on the chosen energy window and charge sector, so a practical benchmark would need a prespecified fitting protocol; the paper's reported negative slopes for converged ground states suggest the single-exponential description is not always stable.","The β* transition may be governed by optimization hardness rather than representational power alone; the paper's comparison with a stronger optimizer suggests that the apparent phase boundary could shift if the optimizer is changed, so β* should be interpreted as a property of the method-plus-optimizer pair.","A scaling study of β̃ with bond dimension, circuit depth, or network width—only touched in the paper—would sharpen the claim that β̃ is an expressiveness diagnostic."],"forward_implications":["A converged approximate ground state is never truly cold: its excited-state spectrum remains nearly flat, so any observable computed from the variational state carries a high-energy tail that is easy to underestimate.","Operator expectations of functions of the Hamiltonian, such as ⟨ψ|H²|ψ⟩, inherit fragility from this tail, which matters for sampling-based energy-variance estimators.","The critical inverse temperature β* provides an ansatz-dependent ranking of how low an effective temperature each method can represent, and it correlates with fidelity.","Training dynamics of β̃ first rise and then fall, so the maximum β̃ reached during optimization acts as an upper bound tied to the ansatz's capacity.","Energy and fidelity objectives give similar β̃ dynamics within the same ansatz family, with energy typically yielding larger β̃ because it penalizes high-energy states unevenly."],"supporting_citations":[{"why":"Supplies the automatic-differentiation simulator used for all variational training runs in the numerical experiments.","marker":"[23]"},{"why":"Supplies exact diagonalization of the Hamiltonians, giving the exact eigenstates and energies used for the spectral decompositions.","marker":"[24]"},{"why":"Provides the phase diagram of the XXZ model used to choose parameter regimes in the spin-flipping and antiferromagnetic phases.","marker":"[25]"},{"why":"Reports pseudo-Boltzmann states in QAOA with β̃ around 0.2–0.3, the closest previously observed instance of the exponential law.","marker":"[26]"},{"why":"Analyzes approximate Boltzmann distributions in QAOA, supporting the interpretation of the fitted slope as a temperature.","marker":"[27]"},{"why":"Reports nearly flat energy spectra for matrix product states; the paper explains the apparent flattening via density-of-states compensation, making this a direct predecessor.","marker":"[28]"}],"fun_headline_variants":["Quantum states share a universal exponential spectral law","Small effective temperature governs approximate quantum states","Exponential decay reveals universal picture in quantum approximations","Approximate quantum states obey one exponential rule","Effective temperature: a new diagnostic for quantum ansatzes"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The universal law rests on assuming that a single fitted slope of log|ci|² versus eigenenergy, extracted over a chosen set of eigenstates, is a stable and representative description of the whole spectrum, independent of the fitting window, charge sector, and optimization run.","fun_headline_variants_meta":{"raw":{"variants":["Quantum states share a universal exponential spectral law","Small effective temperature governs approximate quantum states","Exponential decay reveals universal picture in quantum approximations","Approximate quantum states obey one exponential rule","Effective temperature: a new diagnostic for quantum ansatzes"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000197,"raw_usage":{"total_tokens":1341,"prompt_tokens":897,"completion_tokens":444,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":513,"completion_tokens_details":{"reasoning_tokens":375}},"tokens_in":513,"tokens_out":444,"duration_ms":4400,"temperature":1.0,"reasoning_tokens":375,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-12T10:44:58.345485+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"For the 4×4 XXZ model in the half-filling sector, take a converged MPS ground-state approximant optimized with the fidelity objective and fit log|ci|² against εi separately over the lower third and upper third of the spectrum; if the two fitted slopes differ by more than the fit uncertainty, or if one is negative while the other is positive, the claimed universal single-exponential law fails for that state.","supporting_citations":[{"cited_title":"Zhang, J","cited_arxiv_id":null,"evidence_quote":"Supplies the automatic-differentiation simulator used for all variational training runs in the numerical experiments."},{"cited_title":"Yunoki, Numerical study of the spin-flop transition in anisotropic spin-1/2 antiferromagnets, Phys","cited_arxiv_id":null,"evidence_quote":"Provides the phase diagram of the XXZ model used to choose parameter regimes in the spin-flipping and antiferromagnetic phases."},{"cited_title":"Díez-Valle, D","cited_arxiv_id":null,"evidence_quote":"Reports pseudo-Boltzmann states in QAOA with β̃ around 0.2–0.3, the closest previously observed instance of the exponential law."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Analyzes approximate Boltzmann distributions in QAOA, supporting the interpretation of the fitted slope as a temperature."}],"review_version":1}