Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Geometry of spherical spin glasses

Eliran Subag

TL;DR

This paper articulates a unified geometric picture of spherical spin glasses by linking the high-dimensional landscape of the spherical $p$-spin Hamiltonian to two complementary analyses: (i) concentration of the Gibbs measure near high-energy critical points in pure models, via Band decompositions around these points, and (ii) a generalized TAP framework for general mixtures built on free-energy functionals computed over bands using many orthogonal replicas. It rigorously characterizes the complexity of critical points, derives an ultrametric, tree-like organization of pure states, and establishes TAP-like representations that connect band energies to the global free energy, including explicit Onsager corrections. These geometric insights yield concrete optimization implications, including polynomial-time Hessian-descent strategies along orthogonal increments and connections to Parisi’s solution and ultrametricity, with extensions to Ising spins and Smale’s 17th problem over the reals. Overall, the results illuminate how high-dimensional random landscapes organize into band-localized, hierarchically structured pure states and how this structure can be exploited algorithmically in non-convex optimization.

Abstract

Spherical spin glasses are canonical models for smooth random functions in high dimensions. In this review, we survey several interrelated lines of research on their geometric structure. We begin with results concerning critical points and their relationship to the Gibbs measure. For the pure models, the measure concentrates on spherical bands around critical points that approximately maximize the energy at a particular radius. Next, we present another approach in which a similar picture is derived for general mixed models. At the core of this approach is a free energy functional computed over bands using multiple orthogonal replicas, satisfying a strong concentration of measure. We discuss several implications of this method for a generalized Thouless-Anderson-Palmer (TAP) approach. Finally, we explain how these geometric insights inform optimization algorithms, and briefly relate them to Smale's 17th problem over the real numbers.

Geometry of spherical spin glasses

TL;DR

This paper articulates a unified geometric picture of spherical spin glasses by linking the high-dimensional landscape of the spherical -spin Hamiltonian to two complementary analyses: (i) concentration of the Gibbs measure near high-energy critical points in pure models, via Band decompositions around these points, and (ii) a generalized TAP framework for general mixtures built on free-energy functionals computed over bands using many orthogonal replicas. It rigorously characterizes the complexity of critical points, derives an ultrametric, tree-like organization of pure states, and establishes TAP-like representations that connect band energies to the global free energy, including explicit Onsager corrections. These geometric insights yield concrete optimization implications, including polynomial-time Hessian-descent strategies along orthogonal increments and connections to Parisi’s solution and ultrametricity, with extensions to Ising spins and Smale’s 17th problem over the reals. Overall, the results illuminate how high-dimensional random landscapes organize into band-localized, hierarchically structured pure states and how this structure can be exploited algorithmically in non-convex optimization.

Abstract

Spherical spin glasses are canonical models for smooth random functions in high dimensions. In this review, we survey several interrelated lines of research on their geometric structure. We begin with results concerning critical points and their relationship to the Gibbs measure. For the pure models, the measure concentrates on spherical bands around critical points that approximately maximize the energy at a particular radius. Next, we present another approach in which a similar picture is derived for general mixed models. At the core of this approach is a free energy functional computed over bands using multiple orthogonal replicas, satisfying a strong concentration of measure. We discuss several implications of this method for a generalized Thouless-Anderson-Palmer (TAP) approach. Finally, we explain how these geometric insights inform optimization algorithms, and briefly relate them to Smale's 17th problem over the real numbers.
Paper Structure (13 sections, 7 theorems, 71 equations)

This paper contains 13 sections, 7 theorems, 71 equations.

Table of Contents

  1. Introduction.
  2. Additional Background: the Parisi solution and the Gibbs measure.
  3. Critical points: complexity and the Gibbs measure.
  4. Annealed and quenched complexity and the extremal process.
  5. Decomposition of the Gibbs measure around critical points.
  6. Free energies on bands and a generalized TAP approach.
  7. Using many orthogonal replicas.
  8. Multi-samplable overlaps.
  9. Generalized TAP correction.
  10. Generalized TAP approach.
  11. Energies on the tree of pure states.
  12. Optimization algorithms.
  13. Polynomial systems of equations.

Key Result

Theorem 3.1

For any interval $B\subset\mathbb{R}$,

Theorems & Definitions (8)

  • Theorem 3.1: Annealed complexity, Auffinger-Ben Arous ABA2
  • Theorem 3.2: Quenched complexity 2nd
  • Theorem 3.3: Limiting extremal process, Subag-Zeitouni pspinext
  • Theorem 3.4
  • Theorem 3.5: Ben Arous-Subag-Zeitouni geometryMixed
  • Theorem 4.1: Uniform concentration, FElandscape
  • Definition 4.2
  • Lemma 4.3