Searcharxiv⌕ Search

arXiv · 2610.06480

Upper deviations of the Sherrington-Kirkpatrick free energy: the fifth-order law and the onset of constrained Parisi measures

Abstract

Let $Z_N$ be the partition function of the Sherrington-Kirkpatrick model at inverse temperature $β>1$. Replica theory (Kondor near the critical temperature, Parisi and Rizzo at all temperatures) predicts that the probability that $N^{-1}\log Z_N$ exceeds its typical value by $δ$ decays like $\exp(-Naδ^{6/5})$ with an explicit constant $a$; Aronow and Lopatto recently proved bounds of this order. The exponent is the Legendre transform of the Parisi problem constrained to carry an atom $θ$ at zero. We show that the constrained minimum exceeds the equilibrium free energy by $c_βθ^5(1+o(1))$, where $c_β=\frac{9}{640}β^3ρ_β(0)^{-3}$ and $ρ_β(0)>0$ is the density at zero of the Parisi measure. With the fractional-moment formula, valid for this model, this gives the exponent $\frac56(6c_β)^{-1/5}δ^{6/5}(1+o(1))$ as $N\to\infty$ and then $δ\to0$. The constrained minimizer has a marginal first positive contact, to leading order at $3θ/(2ρ_β(0))$ with an atom $θ/2$, below which a function equal to the identity at the contacts differs from it by a universal cubic to leading order. This plateau calculus applies to every mixture with $ξ'''(0)=0$ whose Parisi measure accumulates at zero; it also shows that a small field $h$ opens a gap of order $|h|^{2/3}$ and adds a term of order $|h|^{10/3}$ to the free energy, and that an atom of the Parisi measure at zero, present in every pure $p$-spin model with $p\ge3$, makes the rate leave zero linearly, with slope equal to the atom. Near criticality the constant tends to Kondor's value $9/5120$, the marginal regime ends at a de Almeida-Thouless threshold in the constraint, and above its first contact the constrained measure agrees with the Parisi measure up to $O(θ^5)$. The results at every $β>1$ use Lopatto's theorem that the Parisi measure accumulates at zero.

Explore related subjects

Keep this discovery

Explore connections, maps & timelines

BibTeXRIS

Yan Ru Pei. 2026-10-05. Upper deviations of the Sherrington-Kirkpatrick free energy: the fifth-order law and the onset of constrained Parisi measures. https://arxiv.org/abs/2610.06480

Cite the original work for its findings. Save a collection to share your selection of sources.

KEEP EXPLORING

Related papers

Highest Trees of Random Mappings

We study the heights of the trees of a random mapping of $n$ elements. Using singularity analysis of exact generating functions, we prove that the mapping has a unique highest tree with probability $1 - \sqrt{\fracπ{8}}\, n^{-1/2} + o(n^{-1/2})$. The property of having a unique highest tree plays a crucial role in the solution of the Road Coloring Problem [Trahtman, 2009]. More generally, we consider $c$-branches, the subtrees rooted at distance $c$ from the cycles. For all fixed $c \ge 0$ and $j \ge 1$ we show that the highest $c$-branch exceeds every other $c$-branch in height by at least $j$, except with probability asymptotic to $(2j-1) \sqrt{\fracπ{8}}\, n^{-1/2}$. We also show that, for any fixed $α> 1$, with probability $1 - Θ(n^{-1/2})$ the part of the highest $c$-branch that lies above the second highest one (its crown) has more than $α$ times as many vertices as roots. The last result is used in the author's proof that a random $2$-letter automaton with $n$ states is synchronizing with probability $1 - Θ(1/n)$.

math.PR↗

Rapid mixing in unimodal landscapes and efficient simulatedannealing for multimodal distributions

We consider nearest neighbor weighted random walks on the $d$-dimensional box $[n]^d$ that are governed by some function $g:[0,1] \ra [0,\iy)$, by which we mean that standing at $x$, a neighbor $y$ of $x$ is picked at random and the walk then moves there with probability $(1/2)g(n^{-1}y)/(g(n^{-1}y)+g(n^{-1}x))$. We do this for $g$ of the form $f^{m_n}$ for some function $f$ which assumed to be analytically well-behaved and where $m_n \ra \iy$ as $n \ra \iy$. This class of walks covers an abundance of interesting special cases, e.g., the mean-field Potts model, posterior collapsed Gibbs sampling for Latent Dirichlet allocation and certain Bayesian posteriors for models in nuclear physics. The following are among the results of this paper: \begin{itemize} \item If $f$ is unimodal with negative definite Hessian at its global maximum, then the mixing time of the random walk is $O(n\log n)$. \item If $f$ is multimodal, then the mixing time is exponential in $n$, but we show that there is a simulated annealing scheme governed by $f^K$ for an increasing sequence of $K$ that mixes in time $O(n^2)$. Using a varying step size that decreases with $K$, this can be taken down to $O(n\log n)$. \item If the process is studied on a general graph rather than the $d$-dimensional box, a simulated annealing scheme expressed in terms of conductances of the underlying network, works similarly. \end{itemize} Several examples are given, including the ones mentioned above.

math.PR↗

Picard Approximation and Martingale Integrand Estimation for Path-Dependent FBSDEs

This paper develops a Picard iteration method for path-dependent forward-backward stochastic differential equations (FBSDEs). We construct a Picard iteration scheme, prove its convergence to the exact solution, and derive an explicit convergence rate. Although the underlying problem is conceptually straightforward, establishing rigorous convergence results in the path-dependent setting is technically demanding. Path dependence introduces a logarithmic correction: the time-discretization error bound is of order $|π|\log(1/|π|)$, compared with the order $|π|$ obtained in the Markovian setting, where $|π|$ denotes the mesh size of the time partition. We also introduce an estimator for the martingale integrand of the FBSDE that is tailored to path dependence and establish, within a Monte Carlo framework, a concentration inequality that quantifies its statistical error. The applicability of the proposed method is demonstrated in two non-Markovian derivative-pricing settings. The primary objective of this paper is to provide a theoretical convergence analysis of the Picard iteration method in the path-dependent setting and illustrate its application to option pricing, rather than to develop a faster or more computationally efficient algorithm.

math.PR↗