A spin glass is an alloy, like copper laced with a few iron atoms, where neighboring tiny magnets (spins) sometimes want to align and sometimes oppose each other, causing “frustration.” In 1979 Parisi analyzed a mathematical trick that handles many copies (replicas) of the system and uncovered a hidden hierarchical structure, introducing “replica symmetry breaking.” The result revealed an energy landscape of countless valleys and hills, applicable to glasses, granular avalanches and more. The ideas also inspired brain memory models and optimization techniques in machine learning.

In the replica method disordered couplings are averaged by duplicating the system n times and taking the n→0 limit. Earlier solutions assumed replica symmetry, but Parisi (1979) showed that this symmetry breaks spontaneously in a hierarchical fashion; he introduced a full-RSB solution for the overlap matrix q_{ab}. This resolved the unphysical negative entropy of the Sherrington–Kirkpatrick model and quantified the multitude of metastable states and algorithmic complexity in the low-temperature phase. The dynamical transition in the p-spin model matches the mode-coupling description of the structural glass transition and has been extended to jamming criticality and correlation analysis of bird flocks.

In Parisi’s RSB theory, a hierarchical block matrix representation q(x) is inserted into the n-replica free-energy functional; solving the Parisi differential equation ∂_xP(h,x)=½ q̇(x)(∂²_hP−2x∂_hP) yields P(h,x) satisfying self-consistency. The static solution relates to the TAP free energy, and the 1RSB-to-full-RSB crossover coincides with the de Almeida–Thouless instability line. Dynamically, the Cugliandolo–Kurchan equations describe non-equilibrium aging, with the fluctuation–dissipation ratio X(C) interpreted as an effective temperature. These concepts generalize to complex systems—protein folding, SAT phase transitions, swarm intelligence—and underpin statistical-mechanical Bayesian inference.