{"id":"be124f29-8853-4b6f-9c90-d453244c4857","arxiv_id":"2411.09687","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"For high-dimensional superpositions of random plane waves, the authors derive annealed complexity formulas and a Parisi-type variational principle for the mean ground-state energy, with replica-symmetric, one-step, and full replica-symmetry-broken phases.","lead":"This paper derives formulas for the number of critical points and the deepest minimum in high-dimensional random landscapes made by superposing random plane waves in a quadratic trap. It gives a Parisi-style variational problem for the ground state energy and maps glassy phases for varying trap strength and wave count.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The annealed complexity formulas and the Parisi ground-state functional all inherit the unproved strong self-averaging property (Eq. 54) for signed diagonal T; a numerical test of this identity should precede treating the formulas as established.","rationale":"The reader correctly identifies Eq. (54) as the load-bearing assumption. The paper is otherwise careful: the finite-N Kac-Rice expressions are exact, the Gaussian/spherical wavevector equivalence is shown, and the replica computation is standard conditional on the Parisi ansatz. But every closed-form complexity and all phase boundaries depend on the conjecture. The paper explicitly calls the results 'well-grounded conjectures' and cites Ref. 61 for positive T only. A signed T is not a cosmetic extension: the matrix can be indefinite, the determinant can have either sign, and the absolute value prevents applying the positive-T theorems. The conditional large-deviation form used in Eq. (55) needs the determinant to concentrate at the spectral integral for each tilted T density, which is even stronger than the displayed Eq. (54). No contradiction with existing rigorous results was found, so the correct action is to keep the verdict CONDITIONAL and ask for the numerical check. I also note a display discrepancy: Proposition 3 Eq. (31) shows '-alpha ln E_phi[f(0,0)]', while the derived Eq. (149) and RS limit Eq. (37) use '-alpha E_phi[f(0,0)]'; the derivation supports the latter, so the proposition display should be corrected, but this is not the main obstacle.","tokens_in":43764,"tokens_out":9080,"duration_ms":100739,"concrete_test":"Fix alpha=0.5, mu=1, p0(t)=0.5 delta(t+1)+0.5 delta(t-0.5) (signed, finite support). For N=50,100,200,400, M=alpha N, Monte-Carlo sample K (iid N(0,1/N)) and T. Compute A_N=(1/N) ln E|det(mu I + K T K^T)| and B_N=(1/N) E ln|det(...)|, and C_N=integral rho(lambda) ln|mu+lambda| dlambda using rho from Eqs. (17)-(18). If |A_N-C_N| and |B_N-C_N| do not decrease toward 0 with N (or A_N differs from B_N in the limit), Eq. (54) fails and the annealed complexity formulas (16)-(21) need revision. A direct check on (1/N) ln E[N_tot] via Kac-Rice sampling for the same parameters would confirm the impact on Eq. (16).","verdict_should_be":"UNCHANGED","load_bearing_attack":"The finite-N Kac-Rice expressions (14)-(15) reduce the two annealed complexities to expectations of the absolute determinant of mu I + K T K^T. The passage to exponential rates uses Eq. (54): lim (1/N) ln E|det| = lim (1/N) E ln|det| = integral rho(lambda) ln|mu+lambda| dlambda. The authors state this is natural to conjecture but goes beyond the cases proved in Ref. 61, which require positive T. The subsequent large-deviation functional (55) and stationarity equations (60)-(64) for Sigma_tot, and the constrained analogue (89)-(95) for Sigma_min, replace the determinant by this spectral integral under a tilted empirical distribution of T. Hence the closed forms (16)-(18) and (19)-(21), the topological-trivialization threshold (24), and the quadratic coefficients (27)-(28) all carry this assumption. If the annealed rate is enhanced by rare (K,T) configurations with atypical spectra, the claimed exponential rates are not established; the ground-state analysis in Sec. V inherits the same issue through the annealed complexity/AT matching in Appendix C. This is a genuine gap, not an internal error, and the paper labels it a conjecture; but it is the load-bearing point.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper studies high-dimensional random landscapes of the form H(x) = (μ/2)|x|^2 + Σ_{l=1}^M φ_l(k_l·x), where the φ_l are i.i.d. stationary random processes and the k_l are random wavevectors, in the joint limit N,M→∞ with α = M/N fixed. The main results are (i) formulas for the annealed complexity of all critical points and of local minima, expressed via the spectral density of the random matrix KTK^T where K is the N×M wavevector matrix and T is diagonal with i.i.d. curvatures φ''_l, and (ii) a Parisi-type variational formula for the ground-state energy e0, given in terms of the ground-state energy of the one-dimensional disordered Hamiltonian H_{ν,h}(z) = (ν/2)z^2 - hz + φ(z). The paper analyzes replica-symmetric, one-step, and full replica-symmetry-broken solutions, derives AT, Gardner, and RFOT criteria, and works out a solvable example with φ(z) = γ cos(z - θ).","tokens_in":44154,"tokens_out":8543,"duration_ms":86048,"significance":"If the results are correct, the paper provides one of the few analytically tractable families of high-dimensional random landscapes beyond the Gaussian toy model, with explicit dependence on the full distribution p0(t) of the second derivative of the random potential. The finite-N Kac-Rice expressions (14)-(15) are exact, and the reduction to a one-dimensional disordered Hamiltonian is elegant and likely to be influential. The paper is transparent about its conjectural steps, but the central complexity formulas rest on an unproved strong self-averaging conjecture (Eq. 54) for matrices with signed diagonal entries, and the ground-state result relies on the replica trick and an assumed zero-temperature scaling. The main claims are therefore plausible but not fully established; they are well-grounded conjectures rather than proven theorems.","major_comments":[{"comment":"The strong self-averaging identity is the only bridge between the exact finite-N determinant in Eq. (52) and the deterministic spectral integral used in all subsequent complexity formulas. It is applied directly in the large-deviation functional (55), in the minima calculation (89)-(95), and in the stationarity equations (60)-(64) and (93)-(95), and it is inherited by Propositions 1 and 2 as well as by the ground-state analysis through the AT matching in Appendix C. The paper itself notes that the property is proved only for positive defnite T, while the present case has T_l of arbitrary sign. Since rare (K,T) configurations with atypical spectra could in principle enhance the annealed rate, the exponential growth rates (16)-(21) and the threshold (24) are conditional on this conjecture. Please provide numerical evidence for a concrete signed-T model (for example Gaussian φ with p0(t) = (2π)^{-1/2}e^{-t^2/2}), or prove the identity for symmetric i.i.d. T, or explicitly phrase Propositions 1, 2, and 3 as conditional on Eq. (54) and clearly separate conditional from unconditional statements.","section":"§III.A, Eq. (54)"},{"comment":"The summary formula for the ground-state energy contains the term \"-α ln E_φ[f(0,0)]\", which is not well-defined because f(0,0) is negative for typical realizations (it equals -ε_min(ν,0)). The derivation in §IV.C yields instead \"-α E_φ[f(0,0)]\" (see Eq. (149) and also Eq. (152)). Since Eq. (31) is the central result of Proposition 3, this is a load-bearing inconsistency: as printed, the main formula cannot be evaluated. Please remove the logarithm (or redefine f so that it is positive) and ensure that the abstract and the summary section match the derivation.","section":"§II.B, Eq. (31)"},{"comment":"The double equality e0 = lim_{N→∞} N^{-1} min_x H(x) = lim_{N→∞} N^{-1} E[min_x H(x)] is asserted without proof and is used to describe the ground-state energy as both the typical and the mean value. The Parisi-type computation in §IV gives the quenched free energy (the typical value), not the annealed mean, unless concentration of min_x H(x)/N holds. For the general class of stationary φ_l studied here, such concentration is not established and is not a consequence of the replica calculation. Please either state explicitly that the result concerns the quenched (typical) ground-state energy and remove the claim that it equals the annealed mean, or provide a concentration argument or citation that justifies the equality in Eq. (13).","section":"§II.B, Eq. (13)"}],"minor_comments":[{"comment":"The equation defining μ_c(α) is typeset with two consecutive equal signs in a way that is easy to misread; please split it into two separate equalities or add an explicit definition of the condition.","section":"§II.A, Eq. (24)"},{"comment":"The equivalence between Gaussian wavevectors and vectors uniformly distributed on the sphere is stated for N→∞. Please state explicitly that all subsequent results are derived for the limiting Gaussian model, since the finite-N equivalence is only asymptotic.","section":"§IV.A, Eq. (117)"},{"comment":"The expansion ξ(ν,h) = Σ_{k≥1} h^k C_k(ν)/k! is used in the stability analysis (Eqs. (40)-(42)); please indicate whether this series is formal or whether a finite radius of convergence is known for the classes of φ considered.","section":"§II.B, Eq. (36)"},{"comment":"The paper would benefit from at least one figure illustrating the predicted phase diagram in the (μ, α) plane, especially for the exactly solvable example of Appendix D; the lack of any figure makes the rich phase structure harder to follow.","section":"General"}],"recommendation":"major_revision","confidential_remarks":"The paper is a substantive contribution to the physics of high-dimensional random landscapes, and the authors are transparent about the conjectural status of several key steps. However, the strong self-averaging property (Eq. 54) supports the entire complexity analysis, and the summary formula Eq. (31) contains a notational error that obscures the main result. I recommend major revision rather than rejection: the authors should either prove or numerically test Eq. (54) for signed T, correct Eq. (31), and clarify the mean/typical distinction in Eq. (13). If the journal is receptive to heuristic derivations with clearly labeled conjectures, acceptance after these revisions would be reasonable."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"The two things you should know: this is a genuine generalization of the authors' earlier Gaussian-curvature result to arbitrary stationary random functions, and it adds a Parisi-type variational problem for the ground-state energy. The catch is that every complexity formula and the ground-state analysis rest on the unproved strong self-averaging property in Eq. (54), which the authors honestly label as a conjecture. If that property fails, the claimed exponential rates are not established.\n\nWhat is new and good: the extension from the Gaussian p0(t) to general p0(t) is real, not cosmetic. The finite-N Kac-Rice expressions are exact, the large-deviation/Marchenko-Pastur machinery is standard and clearly presented, and the paper derives concrete predictions: topology trivialization thresholds, quadratic coefficients near the transition, and a phase diagram with RS, 1RSB, and FRSB phases separated by AT and Gardner lines. The matching of the AT criterion with the topology-trivialization criterion (Appendix C) is a nice, non-obvious result. The exactly solvable trigonometric model in Appendix D is a useful check and gives explicit formulas. The paper is also unusually transparent about its own limitations.\n\nThe soft spots, in proportion: Eq. (54) is load-bearing, and the cited rigorous results (Ref. 61) cover positive T, not the signed T relevant here. The stress-test note is right that rare atypical spectra could change the annealed rates if the self-averaging fails. This is a genuine gap, not an internal error, and it is labeled as a conjecture, but it carries the entire complexity calculation. The ground-state energy also relies on the replica trick and analytic continuation in n, standard physics but not rigorous. The unusual quadratic vanishing of the minima complexity--versus the expected cubic--is flagged by the authors themselves as a possible indication that annealed and quenched complexities differ; that deserves scrutiny.\n\nWho this is for: researchers in random landscapes, spin glasses, and high-dimensional optimization, plus anyone interested in Berry's conjecture connections. It is a serious paper with substantial new results and an honest statement of what is proven and what is conjectured. It deserves a serious referee. My recommendation: send to peer review, and ask the referee to push for numerical checks of Eq. (54) for signed T, and to comment on whether the quadratic vanishing of the minima complexity undermines the annealed approach.","headline":"Generalizes the Gaussian plane-wave landscape to arbitrary stationary curvatures with a Parisi variational principle, but the load-bearing determinant self-averaging is a conjecture, not a theorem.","tokens_in":44527,"tokens_out":1917,"would_cite":true,"duration_ms":20956,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["82B44","60G60","15B52"],"pacs":[],"model":"deepseek-v4-flash","headline":"This paper claims that high-dimensional landscapes built by superposing random plane-wave-like components have exactly computable annealed complexities and a Parisi-type ground-state energy that changes phase with the confinement…","keywords":["random landscape","annealed complexity","ground-state energy","replica symmetry breaking","Parisi functional","Kac-Rice formula","random matrix determinant","plane waves"],"falsifier":"Take a concrete curvature distribution $p_0(t)$ with mixed signs, for example a Gaussian or a two-point distribution, and compute both sides of Eq. (54) numerically at increasing $N$ and $M=\\alpha N$: if $\\tfrac{1}{N}\\ln E[\\det(\\mu I+KTK^T)]$ and $\\tfrac{1}{N}E[\\ln\\det(\\mu I+KTK^T)]$ fail to converge to the same limit for some $\\mu$, the annealed complexity formulas (16)--(21) collapse. A second, more direct check is to simulate the Kac\\textendash{}Rice count of critical points and minima at moderate $N$ and extrapolate its exponential rate, comparing the result with the integrals in Eqs. (16) and (19).","tokens_in":43581,"feed_emoji":"🌊","tokens_out":5352,"duration_ms":53975,"temperature":0.7,"pith_summary":"The paper studies random landscapes of the form $H(\\mathbf{x})=\\frac{\\mu}{2}\\mathbf{x}^2+\\sum_{l=1}^M \\phi_l(\\mathbf{k}_l\\cdot\\mathbf{x})$ in the limit $N,M\\to\\infty$ with $\\alpha=M/N$ fixed, where the functions $\\phi_l$ and wavevectors $\\mathbf{k}_l$ are random. It aims to establish exact expressions for the exponential growth rates of the expected number of critical points and local minima, and for the mean depth of the global minimum, the ground-state energy. The ground-state energy is expressed through a Parisi-type variational functional, whose optimizer can be replica-symmetric, one-step, or full replica-symmetry-broken depending on $\\alpha$ and the curvature $\\mu$. A sympathetic reader would care because the model offers a tractable, non-Gaussian family of random landscapes with a rich glassy phase picture, including transitions whose location can be compared with the vanishing of the annealed complexity.","feed_headline":"Random plane-wave landscapes get exact ground-state formula","feed_subtitle":"Critical-point counts and the deepest minimum follow from one Parisi-style variational problem in the high-dimensional limit.","key_machinery":"The central objects are the random matrix $\\mu I + K T K^T$, where the columns of $K$ are the random wavevectors and $T$ is the diagonal matrix of curvatures $\\phi_l''(z)$, together with its limiting spectral density $\\rho(\\lambda)$. The argument converts determinant expectations into spectral integrals through the strong self-averaging property of Eq. (54), then uses the Marchenko\\textendash{}Pastur equation (57) to reduce the complexity formulas to integral equations for the Stieltjes transform. For the ground state, the machinery is the replica method with a Parisi function $w(\\tau)$; in the zero-temperature limit the relevant object is the one-dimensional disordered Hamiltonian $H_{\\nu,h}(z)$ and its minimum $\\varepsilon_{\\min}(\\nu,h)$, whose statistics enter the Parisi PDE and the final variational problem.","core_discovery":"For the random landscape $H(\\mathbf{x})=\\frac{\\mu}{2}\\mathbf{x}^2+\\sum_{l=1}^M\\phi_l(\\mathbf{k}_l\\cdot\\mathbf{x})$ with i.i.d. stationary functions $\\phi_l$ and random wavevectors, the paper derives the annealed total complexity $\\Sigma_{\\rm tot}(\\mu,\\alpha)=\\int_\\mu^\\infty\\big(\\tfrac{1}{\\nu}+m_r(-\\nu;\\alpha)\\big)d\\nu$ and the annealed complexity of minima $\\Sigma_{\\rm min}(\\mu,\\alpha)=\\int_\\mu^\\infty\\big(\\tfrac{1}{\\nu}+m(-\\nu;\\alpha)\\big)d\\nu$, where the Stieltjes-transform variables satisfy the integral equations (17)--(21). The mean ground-state energy is claimed to be the supremum in Eq. (31) over $l\\ge 0$ and non-decreasing $w(\\tau)$, with the auxiliary function $f(t,h)$ solving the Parisi PDE of Eq. (32) and the boundary condition (33) set by the ground-state energy of the one-dimensional disordered Hamiltonian $H_{\\nu,h}(z)=\\tfrac{\\nu}{2}z^2-hz+\\varphi(z)$. The qualitative discovery is that the phase structure depends on the support of the curvature distribution $p_0(t)$: for bounded support there is a finite topology-trivialization threshold $\\mu_c(\\alpha)$ where both annealed complexities vanish, and the continuous replica-symmetry-breaking line coincides with that threshold; for unbounded support the landscape is always complex and always in a replica-symmetry-broken phase.","pith_inferences":["Beyond the paper: the matching of the de Almeida\\textendash{}Thouless line with the topology-trivialization transition suggests that, for continuous transitions, the yet-unknown quenched complexity may vanish at the same threshold even if its value differs from the annealed complexity elsewhere.","The reduction to a one-dimensional disordered Hamiltonian hints at a broader principle: statistics of these high-dimensional landscapes may be controlled by the spectral density of a curvature matrix and by the ground-state properties of a low-dimensional effective operator, possibly extending to other superpositions of random plane waves.","A testable extension is to use finite-$N$ numerical Kac\\textendash{}Rice counts for an asymmetric curvature distribution and compare their $N\\to\\infty$ growth rates with the integral equations (16)--(21), which would directly probe whether the strong self-averaging assumption holds beyond the known proofs."],"forward_implications":["If the claims hold, the model provides exact annealed complexity formulas for a whole class of non-Gaussian random landscapes, not just for the Gaussian curvature case treated earlier.","The ground-state energy is determined by a finite-dimensional variational problem built from the statistics of a one-dimensional disordered Hamiltonian, giving a practical route to phase boundaries for any stationary $\\phi_l$ with finite first absolute moment of curvature.","When the curvature distribution has bounded support, the replica-symmetry-breaking transition coincides with the topology-trivialization threshold $\\mu_c(\\alpha)$, so the onset of glassy behaviour can be read from the vanishing of annealed complexity.","For unbounded curvature support, the model is predicted to be always replica-symmetry-broken and always topologically complex, in analogy with the Sherrington\\textendash{}Kirkpatrick model in a field.","The conditioned Hessian spectrum at stationary points is generally not a shifted version of the unconditioned spectrum, and near its edges it shows square-root behaviour, providing testable signatures of the landscape geometry."],"supporting_citations":[{"why":"Supplies the random-matrix treatment of landscape complexity for the Gaussian-curvature special case on which this paper builds.","marker":"[4]"},{"why":"Introduces the annealed-complexity and topology-trivialization approach that this paper generalizes to a wider class of models.","marker":"[20]"},{"why":"Originates the replica method and the ground-state formalism used as the starting point for the Parisi-type variational problem.","marker":"[34]"},{"why":"Provides the Parisi variational framework, the zero-temperature scaling, and the stability analysis techniques used in the derivation.","marker":"[35]"},{"why":"Gives the de Almeida\\textendash{}Thouless replicon-stability criterion that fixes the continuous replica-symmetry-breaking transition.","marker":"[58]"},{"why":"Provides the Gardner-stability framework used to locate the transition between one-step and full replica symmetry breaking.","marker":"[60]"},{"why":"Supplies the strong self-averaging theorem for random determinants that Eq. (54) assumes and extends to arbitrary signs of the diagonal entries.","marker":"[61]"},{"why":"Gives the Marchenko\\textendash{}Pastur spectral equation that converts the determinant problem into the integral equations for the complexity.","marker":"[63]"}],"fun_headline_variants":["Deepest minimum formula for random plane-wave landscapes","Exact ground state in high-dimensional random landscapes","Complexity and minimum of random plane waves solved","Parisi variational principle gives random landscape minimum"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The whole computation of the exponential growth rates assumes the strong self-averaging property in Eq. (54): that the logarithm of the determinant of $\\mu I+KTK^T$, divided by $N$, converges to the same limit whether or not it is averaged first, even when the random diagonal entries of $T$ have arbitrary signs; the authors state this is natural to conjecture but is proved only for a narrower class of cases.","fun_headline_variants_meta":{"raw":{"variants":["Deepest minimum formula for random plane-wave landscapes","Exact ground state in high-dimensional random landscapes","Complexity and minimum of random plane waves solved","Parisi variational principle gives random landscape minimum"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.00097,"raw_usage":{"total_tokens":4222,"prompt_tokens":1142,"completion_tokens":3080,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":758,"completion_tokens_details":{"reasoning_tokens":3022}},"tokens_in":758,"tokens_out":3080,"duration_ms":20903,"temperature":1.0,"reasoning_tokens":3022,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-12T20:22:51.269482+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Take a concrete curvature distribution $p_0(t)$ with mixed signs, for example a Gaussian or a two-point distribution, and compute both sides of Eq. (54) numerically at increasing $N$ and $M=\\alpha N$: if $\\tfrac{1}{N}\\ln E[\\det(\\mu I+KTK^T)]$ and $\\tfrac{1}{N}E[\\ln\\det(\\mu I+KTK^T)]$ fail to converge to the same limit for some $\\mu$, the annealed complexity formulas (16)--(21) collapse. A second, more direct check is to simulate the Kac\\textendash{}Rice count of critical points and minima at moderate $N$ and extrapolate its exponential rate, comparing the result with the integrals in Eqs. (16) and (19).","supporting_citations":[{"cited_title":"Lacroix-A-Chez-Toine , author Y","cited_arxiv_id":null,"evidence_quote":"Supplies the random-matrix treatment of landscape complexity for the Gaussian-curvature special case on which this paper builds."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Introduces the annealed-complexity and topology-trivialization approach that this paper generalizes to a wider class of models."},{"cited_title":"M \\'e zard , author G","cited_arxiv_id":null,"evidence_quote":"Originates the replica method and the ground-state formalism used as the starting point for the Parisi-type variational problem."},{"cited_title":"Parisi , author P","cited_arxiv_id":null,"evidence_quote":"Provides the Parisi variational framework, the zero-temperature scaling, and the stability analysis techniques used in the derivation."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Gives the de Almeida\\textendash{}Thouless replicon-stability criterion that fixes the continuous replica-symmetry-breaking transition."},{"cited_title":"Gardner ,\\ title title Spin glasses with p-spin interactions , \\ @noop journal journal Nuclear Physics B \\ volume 257 ,\\ pages 747--765 ( year 1985 ) NoStop","cited_arxiv_id":null,"evidence_quote":"Provides the Gardner-stability framework used to locate the transition between one-step and full replica symmetry breaking."},{"cited_title":"Ben Arous , author P","cited_arxiv_id":null,"evidence_quote":"Supplies the strong self-averaging theorem for random determinants that Eq. (54) assumes and extends to arbitrary signs of the diagonal entries."},{"cited_title":"Pastur \\ and\\ author V","cited_arxiv_id":null,"evidence_quote":"Gives the Marchenko\\textendash{}Pastur spectral equation that converts the determinant problem into the integral equations for the complexity."}],"review_version":1}