Fluctuations of the free energy in the REM and the \(p\)-spin SK models (Q1872272)

A central limit theorem for the free energy of the Sherrigton-Kirkpatrick (SK) model was proven first by \textit{M. Aizenman, J. L. Lebowitz} and \textit{D. Ruelle} [Commun. Math. Phys. 112, 3-20 (1987)], using cluster expansion techniques, and later by \textit{F. Comets} and \textit{J. Neveu} [ibid. 166, 549-564 (1995; Zbl 0811.60098)], making use of martingale methods and stochastic calculus. A large class of natural generalizations of SK model, the so-called \(p\)-spin SK models and their \(p\uparrow +\infty\) limits, the random energy model (REM) are investigated. Using the martingale approach combined with truncation techniques inspired by a recent paper by M. Talagrand, on the \(p\)-spin version it is proven that the random corrections to the free energy are on a scale \(N^{(p-2)/2}\) only and, after proper rescaling, converge to a standard Gaussian random variable. This is shown to hold for all values of the inverse temperature, \(\beta\), smaller than a critical temperature \(\beta_p\). Also it is shown that \(\beta_p\to \sqrt{2\ln 2}\) as \(p\uparrow +\infty\). The precise limit theorem for the (properly rescaled) partition function at all temperatures is computed. For \(\beta <\sqrt{2\ln 2}\), fluctuations are found at an exponentially small scall, with two distinct limit laws above and below a second critical value \(\sqrt{\ln 2 /2}\): for \(\beta\) up to that value the rescaled fluctuations are Gaussian, while below that there are non-Gaussian fluctuations driven by the Poisson process of the extreme values of the random energies. For \(\beta\) lager than the critical \(\sqrt{\ln 2 /2}\), the fluctuations of the logarithm of the partition function are on a scale of 1 and are expressed in terms of the Poisson process of extremes. At the critical temperature, the partition function divided by its expectation converges to \(1/2\).