In a simple physical system, like a cooling crystal, atoms settle into a single, well-defined ground state. But in "frustrated" systems—such as spin glasses, neural networks, or even expanding universes—the landscape of possible states is a rugged mountain range with countless peaks and valleys.
The central discovery of this research direction is that these states are not scattered randomly. Instead, they are organized into a rigorous, tree-like hierarchy known as ultrametricity. This structure is the physical manifestation of Replica Symmetry Breaking (RSB), providing a universal grammar for complexity across disparate fields.
The Geometry of a Tree
To understand why these systems are "hierarchical," we must look at how we measure the distance between two states. In a spin glass, distance is often defined by "overlap"—how many spins point in the same direction.
In a normal Euclidean space, if state A is near B, and B is near C, then A and C are also relatively near. But as Giorgio Parisi and M. A. Virasoro revealed in Random free energies in spin glasses, disordered systems obey a much stricter rule: the strong triangle inequality. For any three states, the distances between them must form either an equilateral triangle or an isosceles triangle with two equal, longer sides.
This isn't just a mathematical curiosity; it is the defining property of a tree (a taxonomic hierarchy). If you pick three leaves on a tree, the distance between any two is determined by their "common ancestor" node. This organization implies that the system's configuration space is partitioned into nested clusters, where statistical rules remain invariant across different scales.
Why Hierarchies Emerge
Why does nature prefer a tree over a flat landscape? The answer lies in how these systems break ties when they have too many competing constraints—a phenomenon called frustration.
Jean‐Philippe Bouchaud showed in Universality classes for extreme-value statistics that this hierarchical structure is the physical equivalent of the Gumbel universality class. When a system has independent energy states with fast-decaying tails, it naturally "shatters" into a hierarchy at low temperatures.
This shattering is not just theoretical. Brice Huang and Mark Sellke recently provided a Constructive Proof of the Spherical Parisi Formula by explicitly building an "ultrametric tree" of states layer-by-layer. Each level of the tree corresponds to a specific level of similarity (overlap) between states, showing that the complex landscape is a tangible physical construction rather than a computational trick.
From Supercomputers to Quantum Cavities
For decades, researchers debated whether this "mean-field" hierarchy—originally solved for systems where every part interacts with every other part—actually exists in the real, 3D world.
- Numerical Evidence: Using the massive Janus supercomputer, R. A. Baños and colleagues demonstrated in Sample-to-sample fluctuations of the overlap distributions in the three-dimensional Edwards-Anderson spin glass that as system sizes grow, the data increasingly aligns with the ultrametric RSB scenario.
- Experimental Evidence: In Multimode Cavity QED Ising Spin Glass, Brendan Marsh and his team used ultracold atoms in optical cavities to simulate these systems. They found that even in "driven-dissipative" environments—where energy is constantly flowing in and out—the states still organize into an ultrametric hierarchy.
The Limits of Memory and Order
The reach of this hierarchical framework extends to the very edges of physics. Dan Roberts and Douglas Stanford explored this in On memory in exponentially expanding spaces, finding that an expanding universe can "remember" its initial conditions specifically because its possible states are organized ultrametrically.
However, this structure is fragile. Helmut G. Katzgraber and Florent Krzakala found that while mean-field models maintain their hierarchy in an external magnetic field, realistic short-range systems might lose it (Ultrametric probe of the spin-glass state in a field). Similarly, Marcus K. Benna noted that certain "baby" de Sitter models lack ultrametricity, suggesting that not all "glassy" behavior requires a tree-like organization.
Why It Matters
Understanding this hierarchy is the key to navigating complexity. As Ada Altieri notes in An introduction to the theory of spin glasses, this tree-like geometry explains why complex systems are so slow to relax: to move from one state to another, the system often has to "climb" all the way up the hierarchy to a common ancestor before descending into a new valley. By mapping these "nested cycles," as Muhittin Mungan does for amorphous materials in Networks and Hierarchies, we can finally predict how complex materials learn, remember, and respond to the world.