Free energy subadditivity for symmetric random Hamiltonians

Mark Sellke

TL;DR

This paper introduces a powerful, general subadditivity property for the expected free energy of random Hamiltonians that possess specific distributional symmetries. By leveraging transitive group actions, the author provides a simple upper bound for the free energy of combined systems that applies to a vast array of models, from classical spin glasses and branching random walks to quantum SYK models.

What Problem Does This Paper Address?

In statistical physics, calculating the "free energy" of a disordered system—where the energy function (Hamiltonian) itself is random—is a central but notoriously difficult task. Traditionally, proving that free energy behaves nicely (e.g., has a well-defined limit as the system size grows) requires proving "subadditivity."

Existing methods for proving subadditivity often rely on complex interpolation techniques or specific convexity properties that are unique to certain models (like mean-field spin glasses). This paper addresses the gap by asking: Can we find a more universal source of subadditivity? The author identifies "symmetry" as that source, providing a framework that doesn't require the system size to change, but rather looks at how free energy behaves when you sum independent random energy components.

Methods

The core of the paper is a mathematical proof (Theorem 1.1) based on two main ingredients:

  1. Transitive Symmetry: The state space must allow a group of transformations (like rotations or bit-flips) that can move any state to any other state.
  2. Invariance: The "law" or distribution of the random Hamiltonian must stay the same under these transformations.

The author uses Jensen’s Inequality—a standard tool in probability—to show that if one Hamiltonian ($H_1$) is symmetric, the expected free energy of the sum of $H_1$ and another independent Hamiltonian ($H_2$) is always less than or equal to the sum of their individual expected free energies.

For the quantum version, the author extends this logic using the Golden-Thompson inequality, which handles the mathematical headaches that arise when operators (matrices) don't commute.

Datasets

As a theoretical mathematical physics paper, this work does not use empirical datasets. Instead, it validates its findings by applying the theorem to well-known theoretical models, including:

  • Branching Random Walks (directed polymers on trees).
  • Sherrington-Kirkpatrick (SK) Model and Mixed p-spin glasses.
  • Random Constraint Satisfaction Problems (like k-SAT).
  • Random Field Ising Model (RFIM).
  • Quantum SYK Model (relevant to black hole physics and holography).

Key Insights and Ideas

  • Symmetry as a Proxy for Convexity: While previous breakthroughs in spin glass theory (like those by Guerra) relied on the mathematical structure of the model's covariance, this paper shows that simple geometric symmetry is often enough to get a useful bound.
  • The "Annealed" Connection: The paper points out that this subadditivity bound is "asymptotically tight" at high temperatures (weak disorder). In these regimes, the "quenched" free energy (what we usually want) behaves almost like the much simpler "annealed" free energy.
  • Quantum Generality: The symmetry conditions for the quantum case are surprisingly flexible. The author defines "M-symmetrizing" groups, showing that even non-diagonal Hamiltonians (like those in the SYK model) obey these subadditivity rules.
  • Simplifying Existing Proofs: The framework provides a much shorter, more intuitive proof for known bounds in the Random Field Ising Model.

Key Results

  • The Main Inequality: $F(L_1 + L_2) \le F(L_1) + F(L_2)$. This means the "disorder" in a combined system is always "less than the sum of its parts" in terms of free energy.
  • $\beta^2$ Subadditivity: For Gaussian processes (like most spin glasses), the free energy is subadditive in the square of the inverse temperature ($\beta^2$). This relates free energies at different temperatures in a concrete way.
  • SYK and Quantum SK: The paper rigorously proves that these high-profile quantum models satisfy the subadditivity property, which is a crucial step for understanding their thermodynamic limits.

Limitations and Future Work

  • Low-Temperature Slack: The author acknowledges that the bound is generally not tight at low temperatures (strong disorder). In these "frozen" states, the specific structure of the energy landscape matters more than the broad symmetries.
  • Zero Temperature Triviality: At absolute zero ($T=0$), the bound reduces to a trivial statement about maximum values ($\max(A+B) \le \max(A) + \max(B)$), losing the nuanced probabilistic insights available at finite temperatures.
  • Extension to Tensors: While the results for matrices (2-spin models) are very "clean," the author notes that extending some of the more specific results to tensors (p-spin models for $p \ge 3$) remains an open challenge that might require new tools from "free probability" theory.

Summary updated

Abstract

We consider a random Hamiltonian H:Σ→R defined on a compact space Σ that admits a transitive action by a compact group G. When the law of H is G-invariant, we show its expected free energy relative to the unique G-invariant probability measure on Σ, which obeys a subadditivity property in the law of H itself. The bound is often tight for weak disorder and relates free energies at different temperatures when H is a Gaussian process. Many examples are discussed, including branching random walks, several spin glasses, random constraint satisfaction problems, and the random field Ising model. We also provide a generalization to quantum Hamiltonians with applications to the quantum Sherrington–Kirkpatrick and Sachdev–Ye–Kitaev models.
BibTeX
@article{sellke2023free,
  author = {Mark Sellke},
  title = {{Free energy subadditivity for symmetric random Hamiltonians}},
  journal = {Journal of Mathematical Physics},
  year = {2023},
  doi = {10.1063/5.0124718},
}
Figure