🔢 mathematics

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

This paper establishes the fifth-order asymptotic behavior of the constrained Parisi measure near zero, thereby rigorously confirming the predicted $\exp(-Na\delta^{6/5})$ decay rate for upper deviations of the Sherrington-Kirkpatrick free energy and characterizing the onset of constrained measures across all inverse temperatures $\beta > 1$ .

Original authors: Yan Ru Pei

Published 2026-10-06

📖 7 min read🧠 Deep dive

CC BY 4.0

Original authors: Yan Ru Pei

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

In the vast landscape of statistical physics, researchers study how collections of simple parts behave when they interact in complex, often messy ways. Imagine a crowd of people, each holding a sign that says either "up" or "down." In a calm, orderly room, everyone might agree to face the same way. But in a chaotic environment where neighbors pull in opposite directions, the group gets stuck in a state of constant tension, unable to settle into a single, perfect arrangement. This is the world of spin glasses, a class of materials where magnetic atoms are frozen in random orientations, creating a landscape of energy that is rugged and full of local traps. For decades, physicists have used a powerful mathematical tool called the Parisi formula to predict the average energy of these systems. This formula acts like a map, showing the most likely state the system will occupy when it is cooled down. However, maps usually show the average terrain, not the rare, extreme peaks that might exist far above the average. Understanding how likely it is for a system to suddenly jump to these high-energy states is crucial for understanding the stability of complex materials and the limits of information storage.

A new study by Yan Ru Pei tackles one of the most stubborn questions in this field: how rare are the extreme high-energy fluctuations in the Sherrington-Kirkpatrick model, a famous mathematical representation of a spin glass? For years, theoretical predictions suggested that the probability of finding the system in a state significantly higher than its average energy drops off in a very specific, unusual way. The researchers set out to prove this prediction and, more importantly, to calculate the exact number that governs this drop-off. They focused on the "upper tail" of the energy distribution—the rare events where the system's energy is much higher than normal. By analyzing the mathematical structure of the system's most likely path to these high energies, the team demonstrated that the probability of such an event decays according to a precise power law. Specifically, they found that the likelihood of the energy exceeding its typical value by a small amount decreases at a rate proportional to that amount raised to the power of six-fifths. This is a distinct, non-integer exponent that had been predicted by earlier theories but never rigorously confirmed with the exact constant.

The key to this discovery lies in understanding how the system's internal order changes when it is forced into these rare, high-energy states. The researchers discovered that as the system climbs toward these higher energies, its internal structure undergoes a subtle but critical transformation near the bottom of its energy landscape. Instead of changing smoothly, the system's order parameter—a measure of how the spins align—develops a small, flat region, or a "plateau," right at the start of its journey. This plateau is not just a flat spot; it is a specific, marginal state where the system is balanced on the edge of stability. The study shows that this marginal state is the bottleneck that controls the entire process. The system must pass through this specific configuration to reach higher energies, and the shape of this configuration dictates the rate at which the probability of reaching those energies falls.

The author calculated the exact mathematical constant that determines this rate. This constant depends on the density of the system's states near zero, a value that had been known to exist but not precisely quantified in this context. By combining advanced techniques from probability theory with the established Parisi formula, they proved that the rate of decay is governed by a fifth-order law in the underlying variables, which translates to the six-fifths power law for the probability itself. This result confirms a long-standing hypothesis made by physicists using a method called replica theory, which had predicted this specific behavior but lacked a rigorous mathematical proof. The study also clarifies that this behavior is universal for a wide class of similar models, provided their internal structure accumulates near zero in a specific way.

Furthermore, the paper explores what happens when the system is subjected to other small disturbances, such as a tiny external magnetic field or when the system naturally possesses a specific type of symmetry. In the case of a small external field, the researchers found that the system develops a gap in its energy structure, and the energy cost to create this gap follows a different power law, specifically related to the ten-thirds power of the field strength. This finding adds to the understanding of how these complex systems respond to external forces, showing that their response is not linear but follows these intricate, fractional power laws. The work also addresses the behavior of the system near a critical temperature, where the material transitions between different phases. Here, the researchers showed that the marginal state they identified earlier eventually gives way to a different, more stable configuration, marking a clear boundary known as the de Almeida-Thouless threshold.

The significance of this work extends beyond just confirming a formula. It provides a complete and rigorous description of the "shape" of the system when it is in these rare, high-energy states. The researchers showed that the system's path to these states is not random but follows a highly structured, deterministic route defined by the geometry of its energy landscape. They proved that the system's internal order parameter, which describes how the spins are arranged, develops a specific cubic shape near the bottom of the landscape, and this shape is what drives the unusual power-law behavior. This level of detail allows physicists to predict not just the average behavior of these materials, but also the likelihood of extreme fluctuations, which is essential for understanding their reliability and stability in real-world applications.

The study also clarifies the limits of previous theories. It shows that certain conditions, which were previously thought to be sufficient for stability, actually fail in the specific range of energies where these rare fluctuations occur. This means that some existing models for predicting the behavior of these systems may need to be adjusted when considering extreme events. The author also demonstrated that their results hold true for a wide variety of models, not just the specific one they analyzed, suggesting that this behavior is a fundamental feature of this class of complex systems. By proving that the probability of these extreme events follows a precise, predictable pattern, the study offers a new level of confidence in our understanding of how complex, disordered systems behave under stress.

In the end, this paper does more than just solve a mathematical puzzle; it reveals the hidden architecture of rare events in complex systems. It shows that even in a world of randomness and disorder, there are strict, elegant rules governing the most extreme possibilities. The researchers have mapped out the exact path a system takes when it defies the odds, showing that this path is defined by a delicate balance at the very edge of stability. This work bridges the gap between theoretical predictions and rigorous mathematical proof, providing a solid foundation for future studies of complex materials and the limits of their stability. The findings confirm that the universe of spin glasses, while chaotic, is governed by precise, quantifiable laws even in its most extreme moments.

Technical Summary: Upper deviations of the Sherrington–Kirkpatrick free energy

Problem Statement
This paper investigates the large deviation behavior of the free energy in the Sherrington–Kirkpatrick (SK) model and general mixed $p$ -spin models. Specifically, it addresses the probability that the normalized free energy $N^{-1} \log Z_N$ (or $N^{-1} F_{N,\beta}$ ) exceeds its equilibrium value $f_{eq}$ by a small amount $\delta > 0$ . While the equilibrium free energy is governed by the Parisi formula, the rate of decay for upper deviations is determined by the fractional moments of the partition function. Previous work by Kondor (near criticality) and Parisi and Rizzo (at all temperatures) predicted via replica theory that this probability decays as $\exp(-N a \delta^{6/5})$ . However, rigorous proofs for the SK model had only established bounds of this order without identifying the precise constant $a$ . This paper aims to rigorously derive the exact constant and the underlying mechanism governing these deviations.

Methodology
The analysis relies on the connection between large deviations and the constrained Parisi variational problem. The rate function $\Sigma(f)$ for the upper tail is identified as the Legendre transform of the function $\Lambda(\theta) = \theta p(\theta)$ , where $p(\theta)$ is the minimum of the Parisi functional constrained to measures with an atom of mass at least $\theta$ at zero.

The core methodology involves a "plateau calculus" near the origin of the support of the Parisi measure. The author analyzes the behavior of the constrained minimizer $\zeta_\theta$ as $\theta \downarrow 0$ . Key technical tools include:

Envelope Identity: A variational formula expressing the derivative $p'(\theta)$ as an integral involving the "obstacle" function $O_\theta$ and the difference between the function $\Gamma_\theta$ (related to the second moment of the derivative of the Parisi PDE solution) and the identity function.
Plateau Calculus: The author demonstrates that for small $\theta$ , the constrained measure $\zeta_\theta$ develops a "plateau" (an interval of zero mass) between zero and a first positive contact $q_\theta$ . On this plateau, the function $\Gamma_\theta - \text{id}$ behaves asymptotically as a universal cubic polynomial.
Universal Sign and Gap Exclusion: A crucial step involves proving a strict sign for the third spatial derivative of the Parisi PDE solution ( $\Phi_{\zeta,xxx}$ ) and using it to show that no "gaps" (intervals of zero mass) can exist between two contacts near zero for measures close to the equilibrium state. This forces the first positive contact to be "marginal" (satisfying the Plefka inequality with equality).
Fractional Moment Formula (FM): The paper utilizes the fractional moment formula, which relates the growth rate of fractional moments $E Z_N^\theta$ to the constrained Parisi minimum, to translate the variational results into large deviation probabilities for the spin system.

Key Contributions and Results

The Fifth-Order Law and Six-Fifths Constant:
For the SK model at every inverse temperature $\beta > 1$ , the paper proves that the constrained minimum $p(\theta)$ exceeds the equilibrium free energy $f_{eq}$ by a term of order $\theta^5$ :
$p(\theta) - f_{eq} = c_\beta \theta^5 (1 + o(1))$
where the constant is explicitly given by $c_\beta = \frac{9}{640} \beta^3 \rho_\beta(0)^{-3}$ . Here, $\rho_\beta(0)$ is the density of the Parisi measure at zero.
Consequently, the large deviation rate function $\Sigma(f_{eq} + \delta)$ behaves as:
$\Sigma(f_{eq} + \delta) = -\frac{5}{6} (6c_\beta)^{-1/5} \delta^{6/5} (1 + o(1))$
This rigorously confirms the exponent $6/5$ and identifies the exact prefactor, matching the prediction of Parisi and Rizzo.
Structure of the Constrained Minimizer:
The paper characterizes the shape of the constrained minimizer $\zeta_\theta$ for small $\theta$ . It shows that $\zeta_\theta$ places mass $\theta$ at zero, has no mass in the interval $(0, q_\theta)$ , and carries an atom of mass approximately $\theta/2$ at the first positive contact $q_\theta \approx \frac{3\theta}{2\rho_\beta(0)}$ . The contact is marginal, meaning the Plefka condition holds with equality. The function $\Gamma_\theta$ differs from the identity by a universal cubic on the interval $(0, q_\theta)$ .
Generalization to Mixed Models:
The results are extended to general mixed $p$ -spin models where the covariance function $\xi(t)$ satisfies $\xi'''(0) = 0$ and the Parisi measure accumulates at zero (Condition H). The constant $c$ depends on the model parameters and the density of the equilibrium measure at zero.
Perturbations of the Measure Bottom:
The paper analyzes two other perturbations of the bottom of the Parisi measure:
- Small External Field ( $h$ ): A small field opens a gap of order $|h|^{2/3}$ at the bottom of the support. The free energy acquires a correction term of order $|h|^{10/3}$ , and the magnetization follows a specific expansion.
- Atom at Zero: If the equilibrium Parisi measure has an atom at zero (e.g., in pure $p$ -spin models with $p \ge 3$ ), the large deviation rate function leaves the equilibrium level linearly, with a slope equal to the negative of the atom's mass.
Near-Criticality:
For models close to the critical covariance (including SK with $\beta$ close to 1), the paper proves that the marginal regime ends at a de Almeida–Thouless threshold $\theta_c$ . Above this threshold, the first positive contact becomes strictly Plefka stable. The constant $c_\beta$ converges to Kondor's value $9/5120$ as $\beta \downarrow 1$ .

Significance and Claims
The paper claims to provide the first rigorous derivation of the exact constant in the sixth-fifths law for the upper tail of the SK free energy, resolving a long-standing prediction from replica theory. It establishes that the mechanism driving these deviations is local to the origin of the Parisi measure's support and is governed by a "universal cubic" profile of the order parameter.

The author explicitly states that their results at every $\beta > 1$ rely on the theorem (by Lopatto or a computer-assisted proof by Pei) that the Parisi measure of the SK model accumulates at zero. They note that while their method identifies the leading order terms, it does not provide rates for the convergence of the $o(1)$ terms, nor does it address deviations that shrink with $N$ (finite $N$ effects) or the zero-temperature limit. Furthermore, the paper clarifies that the levels carrying the six-fifths law lie outside the scope of existing theorems regarding the existence of TAP states, as the strict Plefka condition fails in this regime.

The work unifies the understanding of various perturbations (constraint, field, atom) of the Parisi measure's bottom, showing they are all governed by the same underlying calculus involving the universal cubic and the gap exclusion principle.

More like this