REVIEW 2 major objections 5 minor 9 references
Large deviations at the edge for 1D gases and tridiagonal random matrices at high temperature
T0 review · 2 major / 5 minor · reviewed 2026-08-06 · deepseek-v4-flash
Pith's one-line read At high temperature $\beta_N = 2\alpha/N$, the largest particle of the 1D log gas obeys a full large deviation principle with the iid rate $I_d(x)=x^d-1$; tridiagonal matrices with Gaussian-tail entries give the same rate $x^2-1$.
desk verdict Solid high-temperature edge LDP paper; main theorem holds up, but fix the left-deviation hypothesis mismatch and double-check the imported PPP theorem. read the letter →
The pith
A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.
The reading
What carries the argument
The argument runs on four objects. First, the rate function $I_d(x)=x^d-1$ and the scaling $(\theta^{-1}\log N)^{1/d}$, which put the typical edge at $x=1$ and make a deviation to $x$ cost roughly $(x^d-1)\log N$. Second, a Wegner-type estimate on the one-point marginal, $\rho_N(u)\le C e^{-V(u)+2\alpha\log(1+|u|)}$ (from [Lam21, Prop. 2.1]), which by exchangeability and a union bound yields the right-deviation upper bound and exponential tightness. Third, the convergence of the edge point process to an inhomogeneous Poisson point process with intensity $e^{-x}$ (Lemma 2.10, imported from [Lam21, Thms 3.2, 3.4] and [DGM24, Thm 4.1]), which identifies the typical edge $E_N\sim(\log N)^{1/d}$ and supplies the lower bound: the event that one particle escapes to $x(\log N)^{1/d}$ while the other $N-1$ stay in $(-\infty,E_{N-1}]$ has probability at least a constant times $N\cdot(Z_{N-1}/Z_N)\cdot e^{-x^d\log N}$. Fourth, for left deviations, a pointwise comparison $p_N(\mathbf{x})\le e^{R_N}\prod_i\rho_{\mathrm{eq}}(x_i)$ obtained by energy estimates in the style of [MMS14] reduces the gas's left tail to that of $N$ iid samples from the equilibrium measure; here $R_N=O(\log N)$ in the log case and $O(N^{2s/(1+s)})$ in the Riesz case, which is why the full Riesz principle is left open in the intermediate window. For the matrix theorem the machinery is a truncation at level $\varepsilon\sqrt{2\log N}$: the top eigenvalue of the truncated matrix is exponentially equivalent to the full one (Lemma 3.4), the Gaussian-tail assumption forces the truncated matrix's blocks to have bounded size at the LD scale (Lemma 3.6), and a stability lemma (Lemma 3.3) says the Euclidean norm of finitely many independent Gaussian-tail variables again has Gaussian tail, so each block's top eigenvalue is controlled by the Euclidean norm of its entries.
What would settle it
A direct numerical test: for the log gas with $V(x)=x^2/2$ at $\beta_N=2\alpha/N$, simulate $N$ up to $10^5$ at $\alpha=0.5$ and $\alpha=5$ (for this $V$ one can use the tridiagonal matrix representation), and compare $(1/\log N)\log P(x_{\max}>x\sqrt{2\log N})$ at $x=1.5,2,2.5$ against $-(x^2-1)$. The theorem predicts the same limit for both $\alpha$ values; any visible $\alpha$-dependence, or a limit other than $x^d-1$, would refute the universality claim. A second check targets the infinite rate on the left: the theorem predicts that $P(x_{\max}<x(\log N)^{1/d})$ decays faster than every power of $1/N$ for $x<1$; observing polynomial decay (of order $N^{-c}$, say) would refute the claim that $I_d(x)=+\infty$ there.
Extended reading notes
Core claim
On the paper's own terms, Theorem 1.2 is the central claim: under Assumptions 1.1 on $V(x)=\theta|x|^d+\varphi(x)$, for the log gas $g=-\log$ at high temperature $\beta_N=2\alpha/N$, the law of $\bar{x}_{\max}=x_{\max}/(\theta^{-1}\log N)^{1/d}$ under $P_N$ satisfies a full large deviation principle at speed $\log N$ with good rate function $I_d(x)=x^d-1$ for $x\ge 1$ and $+\infty$ for $x<1$. This is the same rate function as for $N$ independent particles with density proportional to $e^{-V(x)}$, so the two-body repulsion leaves no trace at the exponential scale of the edge. Theorem 1.3 covers the Riesz case $g=|x|^{-s}$: matching upper and lower exponential bounds for $x>1$, and an infinite left-deviation rate for $x<((1-s)/(1+s))^{1/d}$, with the window in between left open. Proposition 1.5 refines the left-deviation upper bound in the log case to $\log\log P_N(\bar{x}_{\max}<1-x)^{-1}\ge(1-(1-x)^d)\log N$, and Proposition 1.7 gives moderate-deviation upper bounds with rate $x^d$. Theorem 1.10 extends the picture to the spectral edge of symmetric tridiagonal matrices, periodic or not, whose independent entries $a_i$ and $\sqrt{2}\,b_i$ have uniform Gaussian tails (tail probabilities of order $e^{-x^2/2}$): $\lambda_{\max}(J)/\sqrt{2\log N}$ satisfies a full large deviation principle with rate $I_2(x)=x^2-1$, recovering the gas result for quadratic $V$ through the tridiagonal representation of [DE02], and the proof shows that large deviations are produced by a few entries taking abnormally large values.
Load-bearing premise
The proof imports, rather than re-derives, the statement that at $N\beta_N\to 2\alpha$ the particle configurations near the edge converge to an inhomogeneous Poisson point process with intensity $e^{-x}$, with typical edge $E_N\sim(\log N)^{1/d}$ (Lemma 2.10, citing [Lam21, Thms 3.2, 3.4] and [DGM24, Thm 4.1]; for the Riesz gas the paper says the proof is identical). This Poisson convergence supplies both the typical edge scale and the LD lower bound, so if it failed for some interaction or potential, the universality claim — and with it the identification of the iid rate function — would collapse.
Editorial extensions
If this is right
- For $x>1$, Theorem 1.2 yields the sharp tail $P_N(x_{\max}>x(\theta^{-1}\log N)^{1/d})=N^{1-x^d+o(1)}$, identical to the iid case and independent of the interaction strength $\alpha$; this is the exact exponential scale at which truncation arguments can be built.
- For left deviations the rate is infinite: in the log case $P_N(\bar{x}_{\max}<1-\delta)$ is smaller than every power of $1/N$ for every $\delta\in(0,1)$, and Proposition 1.5 sharpens this to $\exp(-N^{1-(1-\delta)^d+o(1)})$, reflecting that a mesoscopic number of particles must move inward.
- Moderate deviations above the edge cost at most $\exp(-x^d(\log N)^{1-\gamma-1/d+o(1)})$ for $-1/d<\gamma<1-1/d$ (Proposition 1.7); matching lower bounds are not established in the paper.
- For any symmetric tridiagonal (or periodic tridiagonal) matrix whose independent entries $a_i$ and $\sqrt{2}\,b_i$ have uniform Gaussian tails, $\lambda_{\max}(J)/\sqrt{2\log N}$ satisfies the full LDP with rate $x^2-1$; this recovers the log-gas theorem for quadratic $V$ via the representation of [DE02] and covers the Lax matrix of the periodic Toda chain (an integrable system) at thermal equilib
- The matrix mechanism is explicit: large deviations of the top eigenvalue are created by a bounded number of neighbouring entries taking values of order $\sqrt{2\log N}$ or larger, all other entries being exponentially irrelevant.
Reading between the lines
- The Riesz window left open for left deviations, $((1-s)/(1+s))^{1/d}\le x<1$, is tied to the coarse error $O(N^{2s/(1+s)})$ in the density comparison: a sharpened comparison should deliver the full LDP with the same $I_d$, i.e. super-polynomial decay of left deviations for all $x<1$; the authors state they expect exactly this (Remarks 1.4 and 1.6).
- The matrix proof's ingredients — truncation, bounded blocks, Gaussian-tail stability under Euclidean norms — look generic, so the same rate $x^2-1$ should hold for banded random matrices with independent uniform-Gaussian-tail entries, including bandwidths growing with $N$ as long as blocks stay bounded at the LD scale; the paper does not state this extension.
- Conditioned on the event $\{\lambda_{\max}\approx x\sqrt{2\log N}\}$, the matrix should show a tight cluster of adjacent large entries (tightness follows from Lemma 3.6), but the law of the cluster size is not determined by the theorem and is a natural numerical target.
- The $\alpha$-independence of the rate suggests the equivalence with independent particles runs below the exponential scale — matching prefactors or a local limit law for $x_{\max}$ near its typical value — which the LDP alone cannot see but makes plausible.
Editorial analysis
A structured set of objections, weighed in public.
Referee Report
Summary. The paper studies Gibbs measures of N particles in R with pair interaction (2α/N)g, where g is either -log|x| or |x|^{-s}, and external potential V(x)=θ|x|^d+φ(x) with φ subleading. In the log case it proves (Theorem 1.2) that the normalized largest particle x_max/(θ^{-1} log N)^{1/d} satisfies a full large deviation principle at speed log N with good rate function I_d(x)=x^d-1 for x≥1 and +∞ for x<1. It also gives Riesz analogues (Theorem 1.3), refined left-deviation and moderate-deviation upper bounds (Propositions 1.5 and 1.7), and a large deviation principle for the top eigenvalue of tridiagonal and periodic tridiagonal random matrices with entries of uniform Gaussian tail (Theorem 1.10), recovering the quadratic log-gas result through the Dumitriu-Edelman representation. The gas proof combines a Wegner-type one-particle bound, a density comparison with the iid product of the equilibrium measure, and an imported Poisson point process convergence at the edge; the matrix proof truncates small entries and shows that only bounded-size blocks survive at the LDP scale.
Significance. If correct, Theorems 1.2 and 1.3 establish a clean universality statement: at the large-deviation scale the edge of a high-temperature one-dimensional log or Riesz gas is indistinguishable from that of N iid samples, with an explicit parameter-free rate function. The tridiagonal result is a useful addition to the sparse-matrix LDP literature and gives a transparent mechanism, namely one or a few abnormally large entries. The paper contains substantial detailed proofs, and the rate function is derived rather than fitted; the upper-bound technology and the block-truncation argument are valuable and mostly rigorous. The main reservations concern hypothesis management around an imported edge-PPP theorem and a mismatch between the assumptions of Proposition 2.7 and its use in the proof of Theorem 1.2.
major comments (2)
- [2.1.3, Proposition 2.7 and Corollary 2.9] Proposition 2.7 is stated under the additional assumption that φ is bounded, but Corollary 2.9, which supplies the left-deviation upper bound used in Theorem 1.2 for closed sets meeting (−∞,1), is stated under Assumptions 1.1 only and its proof invokes (28)–(29) from Proposition 2.7. Since Assumptions 1.1 allow unbounded φ as long as φ(x)=o(x^d), the proof of the full LDP in Theorem 1.2 is not complete as written. The gap is local: the first part of Proposition 2.7 only needs φ(u_N)=o(log N), which follows from Assumptions 1.1, so the authors should restate the proposition with the sharp hypothesis and re-derive (28)–(29) under Assumptions 1.1.
- [Lemma 2.10 and Proposition 2.12] Lemma 2.10 is the load-bearing input for the lower bound: Lemma 2.11 uses it to get a positive lower bound on the probability that all particles lie below E_{N-1}, and Proposition 2.12 relies on that event to build the one-particle deviation that yields the LD lower bound. The lemma is imported from Lambert (2021, Theorems 3.2 and 3.4) and DGM24 (Theorem 4.1); the paper only proves E_N∼(log N)^{1/d} and says the Riesz proof is 'identical'. Since the Riesz case is part of Theorem 1.3, the authors should state the precise hypotheses of the cited theorems (class of V, range of β_N, edge normalization) and confirm explicitly that they are satisfied under Assumptions 1.1 for both g=-log and g=|·|^{-s}; otherwise the lower bound rests on an unverified citation.
minor comments (5)
- [2.1.3, proof of Proposition 2.7] In the moderate-deviation part of the proof, the displayed denominator (log N)^{1+1/d−γ} is inconsistent with the statement's (log N)^{1−γ−1/d}; the correct exponent is 1−γ−1/d, as in Proposition 1.7.
- [Lemma 3.6] A block of size at least d yields at least d−1 consecutive large off-diagonal entries, not d; the union bound should sum over d−1 terms. The conclusion is unchanged, but the argument as written overcounts an event that is not implied by d_max(J^ε_N)≥d.
- [Remark 1.11 and Theorem 1.10] For the Dumitriu-Edelman model the variables √2 b_i have χ-law with parameter u_i=2α(N−i)/N; for i near N the parameter tends to 0 and the prefactor 1/Γ(u_i/2) grows like N, so these variables do not obviously satisfy Definition 1.9 with the o(x²) remainder uniform in i. The matrix LDP itself only needs certain upper-tail bounds and the diagonal lower bound, so the result is likely unaffected, but the verification in Remark 1.11 should be corrected or the tail assumption relaxed.
- [1.5.2] The outline refers to 'Theorems 1.2,1.5', but there is no Theorem 1.5 in the paper; Proposition 1.5 is presumably meant.
- [Lemma 2.10] The convergence statement in Lemma 2.10 is attributed to [Lam21, Theorems 3.2,3.4] and [DGM24, Theorem 4.1], but the proof only discusses the asymptotic of E_N; please add a sentence explicitly identifying which cited theorem proves the PPP convergence in the log case and which proves it in the Riesz case.
Circularity Check
No significant circularity: the LDP is derived from external edge Poisson convergence and explicit one-particle estimates, not from its own conclusion.
full rationale
Score 0. The central large deviation principle in Theorem 1.2 is obtained from three ingredients: (i) a one-particle Wegner-type upper bound imported from Lambert (Lam21, Proposition 2.1), (ii) a pointwise comparison with the iid equilibrium measure (Lemma 2.6, partially extending DGM24 Lemma 3.3), and (iii) the convergence of the edge point process to an inhomogeneous Poisson point process, imported via Lemma 2.10 from Lambert's Theorems 3.2 and 3.4. None of these inputs restates the target LDP, and no parameter is fitted to force the rate function I_d(x)=x^d-1. The rate function is computed explicitly from the assumed potential V(x)=|x|^d+o(x^d), and the lower and upper bounds are genuine estimates of the probabilities in question. The only same-author citation, [DGM24, Theorem 4.1], supplies the deterministic asymptotic E_N ~ (log N)^{1/d}, which is an elementary consequence of the defining equations (30)-(31) under Assumptions 1.1; it is not load-bearing for the rate function, and the Poisson convergence portion of Lemma 2.10 is attributed to Lambert, an external prior theorem. The tridiagonal matrix result in Theorem 1.10 is proved independently by a truncation procedure and uniform Gaussian-tail estimates. A skeptical concern that Lambert's hypotheses might not cover every case in Assumptions 1.1 is a correctness or assumption-checking issue, not circular reasoning, and the paper explicitly states that the assumptions are chosen to allow use of Lam21. The derivation chain is therefore not circular. (Cite: Lemma 2.10, Proposition 2.12, Section 3.)
Assumptions & free parameters
assumptions (6)
- domain assumption The edge point process of the high-temperature gas converges to an inhomogeneous Poisson point process with intensity e^{-x} when N beta_N tends to 2 alpha (Lemma 2.10, from Lam21 Theorems 3.2 and 3.4 and DGM24 Theorem 4.1).
- domain assumption One-particle marginal Wegner estimate: rho_N(u) <= C e^{-V(u) + 2 alpha log(1 + |u|) 1_{log case}} (Lam21 Proposition 2.1, NT20 Proposition 3.6).
- domain assumption The equilibrium measure exists and solves V + 2 alpha U^{rho_eq} + log rho_eq = lambda_eq, with the variational identity E(mu) - E(mu_eq) = alpha D^2(mu, mu_eq) + relative entropy (equation (4) and Lemma 2.5, from GZ19, Lam21, DGM24).
- domain assumption Dumitriu-Edelman tridiagonal representation for the beta-ensemble at N beta_N = 2 alpha (DE02 Theorem 2.12).
- domain assumption Assumptions 1.1 on the potential V = theta |x|^d + phi with phi = o(x^d), phi' = o(x^{d-1}), and the stated phi'' control are sufficient for Lambert's edge results and for the paper's energy estimates.
- standard math Standard large deviation machinery: a weak LDP plus exponential tightness implies a full LDP (Dembo-Zeitouni Lemmas 1.14 and 1.15).
Cite this review
Pith. "Pith review of Large deviations at the edge for 1D gases and tridiagonal random matrices at high temperature." pith.science (2026). https://pith.science/paper/ROXDKD2M
@misc{pith2026250714008,
author = {Pith},
title = {Pith review of: Large deviations at the edge for 1D gases and tridiagonal random matrices at high temperature},
year = {2026},
howpublished = {\url{https://pith.science/paper/ROXDKD2M}},
note = {Machine review of arXiv:2507.14008}
}
abstract
We consider a model of a gas of $N$ confined particles subject to a two-body repulsive interaction, namely the one-dimensional log or Riesz gas. We are interested in the so-called \textit{high temperature} regime, \textit{i.e.} when the inverse temperature is given by $\beta_N=2\alpha/N$ for some $\alpha>0$. We establish, in the log case, a large deviation (LD) principle and moderate deviations estimates for the largest particle $x_\mathrm{max}$ when appropriately rescaled. Our result is in the continuity of [Ben Arous Dembo Guionnet 01', Pakzad 20'] where such estimates were shown for the largest particle of the $\beta$-ensemble at fixed $\beta_N=\beta>0$ and $\beta_N\gg N^{-1}$ respectively. We show that the corresponding rate function is the same as in the case of iid particles. We also provide LD estimates in the Riesz case. Additionally, we consider related models of symmetric tridiagonal random matrices with independent entries having Gaussian tails; for which we establish the LD principle for the top eigenvalue. In a certain specialization of the entries, we recover the result for the largest particle of the log-gas. We show that LD are created by a few entries taking abnormally large values.
Figures
Reference graph
Works this paper leans on
-
[6]
[NTT24] F. Nakano, H-D. Trinh, and K-D. Trinh. Classicalβ-ensembles and related eigenval- ues processes at high temperature and the Markov-Krein transform.arXiv preprint arXiv:2412.01015,
-
[7]
[Pak18] C. Pakzad. Poisson statistics at the edge of Gaussianβ-ensembles at high temperature. arXiv preprint arXiv:1804.08214,
-
[8]
Emergence of a Poisson process in weakly interacting particle systems
[PGPT24] D. Padilla-Garza, L. Peilen, and E. Thoma. Emergence of a Poisson process in weakly interacting particle systems.arXiv preprint arXiv:2405.02625,
-
[9]
Poisson Statistics for Coulomb Gases at Intermediate Temperature Regimes
[PGPT25] D. Padilla-Garza, L. Peilen, and E. Thoma. Poisson statistics for Coulomb gases at intermediate temperature regimes.arXiv preprint arXiv:2507.08198,
- [2015]
-
[2019]
[Han25] Y. Han. Deviation of top eigenvalue for some tridiagonal matrices under various moment assumptions. InAnnales Henri Poincar´ e, volume 26, pages 1907–1926. Springer,
work page 1907
-
[2020]
[CDG23] N.A. Cook, R. Ducatez, and A. Guionnet. Full large deviation principles for the largest eigenvalue of sub-gaussian wigner matrices.arXiv preprint arXiv:2302.14823,
-
[2024]
Asymptotic expansion of the partition function for $\beta$-ensembles with complex potentials
[GKL24] A. Guionnet, K. Kozlowski, and A. Little. Asymptotic expansion of the partition func- tion forβ-ensembles with complex potentials.arXiv preprint arXiv:2411.10610,
Show all 9 references
-
[2026]
Ducatez, A
[DGH24] R. Ducatez, A. Guionnet, and J. Husson. Large deviation principle for the largest eigenvalue of random matrices with a variance profile.arXiv preprint arXiv:2403.05413,
Reviewed August 6, 2026 · model on record in the stance chip above.
Discussion (0). Continue with ORCID to comment.