When a magnetic material is cooled, its spins typically align in a uniform direction, creating a ferromagnet. However, when the interactions between spins are both random and conflicting—a property known as frustration—the system cannot satisfy all bonds simultaneously. Instead, it freezes into a highly disordered yet rigid configuration known as a spin glass.
For decades, statistical physicists have grappled with a fundamental question: What happens to a spin glass when you turn on an external magnetic field?
Because an external field explicitly breaks time-reversal symmetry by forcing spins to favor a specific direction, it threatens the fragile equilibrium of the glassy state. Understanding whether the spin glass phase genuinely survives under an external field, or if it is immediately destroyed, lies at the heart of understanding complexity in disordered systems.
The Topology of Frustration
To understand why an external field poses such a challenge, we must look at the energy landscape of a spin glass. In a standard ferromagnet, there are only two ground states (all spins up or all spins down). In a spin glass, quenched randomness creates a rugged landscape with an astronomical number of stable valleys, known as "pure states" (Multicritical point relations in three dual pairs of hierarchical-lattice Ising spin glasses).
As shown by M. A. Virasoro in The microstructure of ultrametricity, these pure states are not randomly distributed; they are organized hierarchically in a taxonomic tree, a geometric property called ultrametricity. If you choose any three equilibrium states at random, the "distances" (or degrees of similarity) between them form either an equilateral or an isosceles triangle. This hierarchical organization is the hallmark of Replica Symmetry Breaking (RSB), the theoretical framework describing how a system spontaneously fractures into many distinct thermodynamic phases (Mode-coupling approximations, glass theory and disordered systems).
[Root: High Temperature Paramagnet]
/ \
/ \
[Sub-valley A] [Sub-valley B]
/ \ / \
[State1] [State2] [State3] [State4]
When an external magnetic field is applied, it introduces a uniform organizing force that competes directly with this delicate, disordered hierarchy.
Two Competing Paradigms: Mean-Field vs. Droplets
The physics community split into two camps regarding how a spin glass responds to an external field, centering on the nature of its low-energy excitations.
1. The Mean-Field View (The AT Line)
In infinite dimensions—where every spin interacts with every other spin—exact solutions show that the spin glass phase survives weak external fields. The boundary between the spin glass and the conventional paramagnet is known as the de Almeida-Thouless (AT) line. According to this view, replica symmetry breaking and its ultrametric tree remain intact until the field exceeds a critical threshold $B_c$.
The reason this phase survives is rooted in the geometry of its excitations. As Giorgio Parisi demonstrated in Local excitations in mean-field spin glasses, low-energy excitations in mean-field models are not compact clusters. Instead, they are sprawling, fractal structures resembling "lattice animals." Flipping these massive, stringy structures costs a surprisingly small, finite amount of energy regardless of their size, allowing the system to rearrange itself into alternative, equally stable valleys without collapsing into a paramagnet (Non-compact local excitations in spin-glasses).
2. The Droplet Model View
Conversely, the droplet model argues that in realistic, three-dimensional systems with short-range interactions, the AT line is an illusion. This model posits that low-energy excitations are compact "blobs." The energy cost to flip a droplet of size $L$ grows with its boundary area.
When you apply an external field, aligning a large droplet with the field yields an energy benefit proportional to its volume ($L^d$). Because volume growth ($L^3$) outpaces surface growth ($L^\theta$), the magnetic field will always win at large enough length scales. Therefore, the droplet model predicts that any non-zero external field immediately destroys the spin glass phase in three dimensions, reducing the system to a disordered paramagnet.
The Numerical Impasse and the "Silent Majority"
Resolving this debate requires simulating finite-dimensional spin glasses, a task severely bottlenecked by the astronomical relaxation times of glassy dynamics (Critical parameters of the three-dimensional Ising spin glass).
Early absolute-zero simulations by Houdayer et al. in Zero-Temperature Responses of a 3D Spin Glass in a Magnetic Field found evidence of a sharp phase transition at a critical field $B_c \approx 0.65$, suggesting that the spin glass phase could withstand external field pressure. However, as algorithms advanced, the evidence began to shift. Using massive parallelization, Jörg et al. demonstrated in Behavior of Ising Spin Glasses in a Magnetic Field that the purported AT line seemed to vanish as the system size increased, dealing a heavy blow to the mean-field prediction in 3D.
To diagnose why the phase disappears, Katzgraber et al. devised a structural litmus test in Ultrametric probe of the spin-glass state in a field. They directly measured the ultrametric tree structure of the states. They discovered that while infinite-range models preserve their tree-like topology under a field, short-range, realistic models see their ultrametric structure completely break down, implying that the field fundamentally disrupts the multi-valley landscape.
The Statistical Subtlety
If the phase is destroyed, why did so many simulations detect a transition? The answer lies in a statistical trap uncovered by the Janus Collaboration in The three-dimensional Ising spin glass in an external magnetic field: the role of the silent majority.
In highly disordered systems, standard mathematical averages (means) are heavily skewed by rare, extreme sample fluctuations. By shifting to a quantile analysis—looking at the median behavior of the system, or the "silent majority"—the authors revealed that the system exhibits scale invariance where standard averages show nothing. This subtle distinction explains why decades of numerical studies produced conflicting results: the signature of the transition was hidden in the tail of the probability distributions.
Standard Average (Mean): [Skewed by rare, highly stable samples] -> Suggests no transition
Quantile (Median): [Reflects the typical bulk behavior] -> Uncovers scale invariance
Boundaries of Glassiness: Dimensions and Random Fields
The survival of the glassy phase depends heavily on spatial dimensions and the exact nature of the disorder.
While the existence of the AT line in 3D remains highly contentious, moving to four dimensions ($4\text{D}$) changes the calculus. In Thermodynamic glass transition in a spin glass without time-reversal symmetry, Baños et al. proved conclusively that a thermodynamic glass transition survives an external field in $4\text{D}$, establishing that the mean-field scenario becomes viable once the system has enough spatial degrees of freedom.
Furthermore, the spin glass phase is highly sensitive to whether the external field is uniform or random. In models where the external field itself is disordered—such as the Random Field Ising Model (RFIM) or the random-field Ginzburg-Landau model—the random fields can completely suppress glassy freezing. As shown by Krzakala et al. in No spin glass phase in the ferromagnetic random-field random-temperature scalar Ginzburg–Landau model, the spin glass susceptibility in these systems is strictly capped by the ferromagnetic susceptibility. This mathematical bound prevents the system from undergoing a pure glassy transition without simultaneously ordering ferromagnetically, ruling out an independent spin glass phase under those specific random conditions (Elusive Spin-Glass Phase in the Random Field Ising Model).
When continuous spin orientations are allowed, such as in the random-field Heisenberg model, the interplay between spatial dimensions and disorder becomes even richer, giving rise to phenomena like reentrant ferromagnetic ordering, where the system transitions from ordered to disordered and back to ordered as temperature drops (Reentrant ferromagnetic ordering of the random-field Heisenberg model in d > 2 dimensions: Fourier-Legendre renormalization-group theory). Similar complex landscapes with multiple low-energy states also emerge when studying related disordered systems, such as directed polymers in random potentials (On the glassy nature of random directed polymers in two dimensions).
From Theory to the Lab: Quantum Optical Spin Glasses
For nearly half a century, testing the microscopic reality of replica symmetry breaking and ultrametricity under external perturbations was confined to pen-and-paper theory and supercomputer simulations. Physical spin glasses in the lab only allowed for macroscopic measurements, like magnetic susceptibility.
This gap was bridged in a breakthrough study by Kroeze et al. in Directly observing replica symmetry breaking in a vector quantum-optical spin glass. Instead of relying on traditional magnetic alloys, the authors engineered a spin glass using a cold atomic gas inside an optical cavity, where atoms communicate via photons.
Using a quantum gas microscope, the authors managed to prepare identical copies ("replicas") of the disordered system and directly measure the spin configurations at individual atomic sites. By analyzing the overlaps between these replicas, they provided the first direct, experimental visualization of replica symmetry breaking and its emergent ultrametric structure in a physical system. This programmable platform opens the door to mapping out exactly how external fields reshape the complex energy landscapes of disordered matter in the real world.
Key Foundational Work
The microstructure of ultrametricity (1985) — contributed 4 concept elements to this direction.
Self-averaging correlation functions in the mean field theory of spin glasses (1984) — contributed 3 concept elements to this direction.
Replica symmetry breaking and the nature of the spin glass phase (1984) — contributed 1 concept elements to this direction.