REM universality for linear random energy

Francesco Concetti; Simone Franchini

arxiv: 2604.06122 · v1 · submitted 2026-04-07 · 🧮 math.PR · cond-mat.stat-mech

REM universality for linear random energy

Francesco Concetti , Simone Franchini This is my paper

Pith reviewed 2026-05-10 18:51 UTC · model grok-4.3

classification 🧮 math.PR cond-mat.stat-mech

keywords random energy modelREM universalityPoisson point processlinear Hamiltoniansspin configurationsGibbs weightsextreme values

0 comments

The pith

Energy levels from linear random Hamiltonians converge in distribution to a Poisson point process with exponential intensity when exponentially many configurations are sampled.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper proves that a linear random Hamiltonian, built as a sum of independent random coefficients times spin deviations, has its energy levels behave exactly like those in the classic Random Energy Model when an exponential number of random spin configurations are examined. This convergence to a Poisson point process with exponential intensity measure captures both the extreme values and the order-1 fluctuations around them. A reader would care because the result extends REM universality to a wider family of models while allowing far more samples than prior dilution bounds permitted, and it also supplies the limiting law for the associated Gibbs weights.

Core claim

For the Hamiltonian H_n(h, sigma) = sum_{i=1}^n h_i (sigma_i - m) with i.i.d. h_i, the point process formed by the energy levels of e^{O(n)} randomly chosen configurations sigma in {-1,1}^n converges in distribution to a Poisson point process whose intensity measure is exponential. This establishes REM universality for the linear model and yields the asymptotic distribution of the Gibbs weights.

What carries the argument

Convergence in distribution of the point process of sampled energy levels to a Poisson point process with exponential intensity measure.

If this is right

The linear model displays full REM universality, including precise control of O(1) fluctuations around the extremes.
The allowable number of sampled configurations reaches e^{O(n)}, which is exponentially larger than the e^{o(sqrt(n))} bound obtained in earlier dilution studies.
The asymptotic distribution of the Gibbs weights follows directly from the Poisson point process limit.
Local REM universality results are strengthened by the global characterization of the entire collection of energy levels.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

The same Poisson convergence may hold for other linear or sparse interaction structures in high-dimensional random landscapes.
Numerical sampling of energies for moderate n could measure the speed of convergence to the predicted Poisson process.
The Gibbs-weight limit supplies a concrete prediction for the partition function fluctuations in this linear setting.

Load-bearing premise

The independent random variables h_i must satisfy the tail or moment conditions that make the Poisson point process limit hold.

What would settle it

If the empirical point process of energies from e^{c n} sampled configurations, for some fixed c>0 and large n, fails to match the exponential intensity Poisson process (for example by showing a different spacing distribution), the universality claim would be false.

read the original abstract

We consider a sequence of random Hamiltonians $H_n(h,\sigma)=\sum^n_{i=1}h_i(\sigma_i-m)$, and study the asymptotic ($n\to \infty$) distribution of the energy levels $(H_n(h,\sigma))_{\sigma\in \{-1,1\}^n}$, where $h_1,h_2,\cdots$ are i.i.d. random variables. We show that, when $e^{O(n)}$ configurations are sampled at random, the corresponding collection of energy levels converges in distribution to a Poisson point process with exponential intensity measure. This establishes the Random Energy Model (REM) universality for the present model. Our results strengthen earlier works on local REM universality by characterizing the distribution of $O(1)-$order fluctuations of $H_n$. In addition, we improve upon the REM universality by dilution studied by Ben Arous, Gayrard, Kuptsov by allowing an exponentially large number $e^{O(n)}$ of sampled configurations, instead of $e^{o(\sqrt{n})}$. Finally, we derive the asymptotic distribution of the Gibbs weight.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Desk Editor's Note private letter to a colleague

They extend REM universality to linear Hamiltonians with exp(O(n)) samples and get the full point process plus Gibbs weights, but the hidden constant in O(n) needs explicit control against overlap correlations.

read the letter

The main takeaway is that this paper gets Poisson point process convergence for the energies in a linear random Hamiltonian when you draw exponentially many configurations, improving the earlier sub-square-root sampling bound, and it also derives the limiting Gibbs weights. That combination of linear structure, larger sampling, and O(1) fluctuation detail is the concrete step forward from the cited prior work on dilution and local universality. The approach relies on i.i.d. assumptions for the h_i and standard point-process tools, which fits the setting cleanly once the tails are right. The linear form H = sum h_i (sigma_i - m) is handled directly without extra machinery, and the Gibbs asymptotic follows as a natural consequence of the point process limit. This is useful technical progress for people tracking extreme-value behavior in disordered systems. The soft spots are modest but worth noting. The abstract leaves the precise tail or moment conditions on the h_i implicit, so a reader has to wait for the body to see exactly what is assumed. The stress-test point on correlations is fair to raise: with M = exp(c n) the number of pairs is exp(2c n), and while typical overlaps are O(1/sqrt n), the tail on atypically large overlaps could matter if c is not small. The paper likely restricts the implicit constant or shows the dependence is negligible on the relevant scale, but if that control is not stated sharply it leaves the admissible range of O(n) a bit vague. No load-bearing circularity or invented objects appear. This is aimed at specialists in probability and statistical mechanics who work on spin glasses or random media. A reader who cares about precise limits for energies and Gibbs measures in these models will get value from the strengthened sampling regime. I would send it for peer review; the core claim is new enough and the methods standard enough that referees can tighten the assumptions and constants without starting over.

Referee Report

2 major / 2 minor

Summary. The manuscript studies the linear random energy model H_n(h,σ) = ∑_{i=1}^n h_i (σ_i - m) with i.i.d. h_i. It proves that the rescaled point process formed by the energies of M = e^{O(n)} uniformly random configurations σ converges in distribution to a Poisson point process with intensity measure e^{-x} dx. The result is presented as establishing REM universality for this model, strengthening prior local universality statements by allowing exponentially many samples (instead of e^{o(√n)}) and by characterizing O(1)-order fluctuations; the asymptotic distribution of the associated Gibbs weights is also derived.

Significance. If the central convergence holds under the stated conditions, the work meaningfully extends the range of REM universality results in high-dimensional disordered systems and supplies a concrete description of extreme-value statistics for this linear Hamiltonian. The derivation from i.i.d. assumptions via standard point-process techniques is a positive feature.

major comments (2)

[Abstract / Theorem 1.1] The abstract and theorem statements do not record the precise tail or moment assumptions imposed on the common law of the h_i. These conditions are load-bearing for the Poisson convergence and must be stated explicitly (e.g., in the main theorem) so that the scope of the result can be assessed.
[Main convergence theorem (presumably §3 or §4)] The claim that the Poisson limit holds for M = e^{O(n)} samples requires control of pairwise correlations induced by atypical overlaps. For M = exp(c n) the number of pairs is exp(2 c n); if the probability of overlap exceeding any fixed positive level decays only as exp(-d n) with d < 2 c, dependence may survive in the limit. The manuscript must either restrict the hidden constant in O(n) to a sufficiently small value or supply a uniform argument that rules out this obstruction for any fixed c.

minor comments (2)

[Abstract] The parameter m appearing in the definition of H_n is not defined in the abstract; its role (mean, centering constant, etc.) should be clarified at first appearance.
[Introduction] The citation to Ben Arous–Gayrard–Kuptsov on diluted REM universality should be given with a precise reference number in the introduction.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the careful reading and constructive comments. We address the major points below and will revise the manuscript to improve clarity and precision.

read point-by-point responses

Referee: [Abstract / Theorem 1.1] The abstract and theorem statements do not record the precise tail or moment assumptions imposed on the common law of the h_i. These conditions are load-bearing for the Poisson convergence and must be stated explicitly (e.g., in the main theorem) so that the scope of the result can be assessed.

Authors: We agree that the precise assumptions on the common law of the h_i must be stated explicitly, as they are essential for the Poisson point process convergence. The manuscript assumes i.i.d. h_i with finite exponential moments (specifically, E[exp(t |h_1|)] < infinity for some t > 0, ensuring the necessary large-deviation controls). We will add these conditions explicitly to the abstract and to the statement of Theorem 1.1. revision: yes
Referee: [Main convergence theorem (presumably §3 or §4)] The claim that the Poisson limit holds for M = e^{O(n)} samples requires control of pairwise correlations induced by atypical overlaps. For M = exp(c n) the number of pairs is exp(2 c n); if the probability of overlap exceeding any fixed positive level decays only as exp(-d n) with d < 2 c, dependence may survive in the limit. The manuscript must either restrict the hidden constant in O(n) to a sufficiently small value or supply a uniform argument that rules out this obstruction for any fixed c.

Authors: We acknowledge this potential obstruction for large c. The proof already uses a large-deviation estimate showing that the probability of atypical overlaps (exceeding any fixed positive level) decays as exp(-d n) with d > 0 fixed and independent of c. To ensure the Poisson limit is unaffected, we will restrict the hidden constant in e^{O(n)} to sufficiently small c (specifically 2c < d). We will update the theorem statement and add a clarifying remark on this restriction. revision: partial

Circularity Check

0 steps flagged

No circularity: standard probabilistic convergence proof

full rationale

The derivation establishes distributional convergence of rescaled energy levels for M = e^{O(n)} random configurations to a Poisson point process with intensity e^{-x} dx, starting from i.i.d. h_i satisfying tail/moment conditions and using standard point-process techniques. No step reduces by construction to a fitted parameter, self-definition, or load-bearing self-citation; the cited prior works (Ben Arous et al.) are external and the central claim remains an independent limit theorem. The O(n) sampling range is a theorem statement, not an input that is renamed as output.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The central claim rests on the i.i.d. property of the h_i together with unspecified regularity conditions that guarantee the Poisson limit; no free parameters or new entities are introduced.

axioms (1)

domain assumption h_i are i.i.d. with distribution satisfying the tail/moment conditions required for Poisson convergence
Standard hypothesis in extreme-value theory for the emergence of Poisson point processes from sums of independent terms.

pith-pipeline@v0.9.0 · 5485 in / 1200 out tokens · 65054 ms · 2026-05-10T18:51:39.845081+00:00 · methodology

discussion (0)

Reference graph

Works this paper leans on

2 extracted references · 2 canonical work pages

[1]

Number partitioning as a random energy model.Journal of Statistical Mechanics: Theory and Experiment, 2004:04003,

[BFM04] Heiko Bauke, Silvio Franz, and Stephan Mertens. Number partitioning as a random energy model.Journal of Statistical Mechanics: Theory and Experiment, 2004:04003,

work page 2004
[2]

Springer, Berlin ;, 2003

work page 2003

[1] [1]

Number partitioning as a random energy model.Journal of Statistical Mechanics: Theory and Experiment, 2004:04003,

[BFM04] Heiko Bauke, Silvio Franz, and Stephan Mertens. Number partitioning as a random energy model.Journal of Statistical Mechanics: Theory and Experiment, 2004:04003,

work page 2004

[2] [2]

Springer, Berlin ;, 2003

work page 2003