In classical statistical mechanics, identifying a phase transition relies heavily on finding a local order parameter—a macroscopic observable like magnetization that drops to zero when symmetry is restored. However, as we venture into modern quantum many-body systems, gauge theories, and highly disordered systems, this paradigm frequently breaks down. In many exotic regimes, such as topological phases or non-local transitions, defining a local order parameter a priori requires deep physical intuition or may be fundamentally impossible.
To bypass this limitation, a powerful alternative framework has emerged: diagnosing phase transitions through the lens of information theory and quantum entanglement. Instead of tracking the physical state of individual particles, this approach monitors how information is distributed, shared, and scrambled across the system.
Entanglement Entropy as a Critical Diagnostic
The most direct way to observe information-theoretic criticality is through bipartite entanglement entropy, typically measured via the von Neumann or Rényi entropies. When a quantum system approaches a critical point, long-range correlations develop globally. This structural reorganization leaves a distinct footprint on the system's entanglement structure.
For instance, in gauge theories where traditional Monte Carlo simulations suffer from the numerical "sign problem," entanglement metrics offer a clean path forward. Stefan Kühn et al. and Karl Jansen et al. demonstrated this by using matrix product states to simulate the two-flavor Schwinger model. They showed that a distinct peak in the von Neumann entropy reliably signals the critical point of a CP-violating Dashen phase transition, bypassing the need for traditional fermion condensates.
Entanglement Entropy (S)
^
| /\ <-- Peak signals the critical point
| / \
| / \
|____/ \____
+-------------------> Control Parameter (e.g., Field Strength)
However, in finite-sized systems, these entropy peaks are often smoothed out by boundary effects, making the exact critical point difficult to isolate. To resolve this, researchers look at the rate of change of entanglement. Qing Xu et al. and Titas Chanda et al. established that the first derivative of entanglement measures (such as logarithmic negativity) with respect to a control parameter exhibits sharp, singular peaks at the exact phase boundary. Even far from the thermodynamic limit, the derivative amplifies the subtle acceleration of quantum fluctuations, pinpointing the transition.
At the critical point itself, the entanglement entropy obeys universal scaling laws governed by the underlying conformal field theory (CFT). Benjamin Hsu et al. mapped 2D quantum wavefunctions to classical statistical mechanics to show that two-dimensional quantum critical points possess a universal, finite contribution to entanglement entropy determined by boundary CFT properties. Similarly, Zi Hong Liu et al. utilized the subleading terms of the second Rényi entropy to extract universal properties of symmetric mass generation critical points. Conversely, when a system freezes into a highly ordered crystalline state, it sheds its informational complexity; David Linteau et al. captured this via a sharp "entanglement drop" in Rényi-2 entropy, providing a zero-temperature quantum analog to the loss of configurational entropy in classical freezing.
Mutual Information and the Discontinuity of Inference
Bipartite entanglement entropy is highly effective for pure ground states, but it struggles when dealing with mixed states or non-local transitions where the correlations are spread across multiple subsystems. In these scenarios, researchers turn to Quantum Conditional Mutual Information (QCMI) and the principle of maximum entropy inference.
Maximum entropy inference is a method that selects the global quantum state with the highest von Neumann entropy consistent with a set of observed local measurements. In classical systems, as local observations change smoothly, the inferred maximum-entropy state also changes smoothly. Quantum mechanics, however, allows for non-local constraints driven by entanglement.
Jianxin Chen et al. discovered that at non-local quantum phase transitions, this inference process exhibits a sharp discontinuity. As local measurement outcomes vary smoothly across a critical threshold, the globally inferred state undergoes an abrupt, discontinuous jump. QCMI acts as a highly sensitive probe for this phenomenon, successfully signaling non-trivial quantum orders even within small, computationally accessible system sizes.
Local Measurements (Smooth) ---> [ Maximum Entropy Inference ] ---> Inferred Global State (Discontinuous Jump)
This informational boundary detection is not unique to quantum mechanics. In classical probabilistic graphical models, the mutual information between subsystems serves as a precise diagnostic for structural shifts in solution spaces. Amit Nir et al. developed a scalable machine-learning framework (MICE) that iteratively halves a physical system to compute mutual information, successfully identifying phase transitions without prior knowledge of the order parameters. This matches the rigorous mathematical foundation established by Amin Coja-Oghlan et al., who proved that deviations in mutual information from high-temperature formulas precisely mark the "condensation threshold" in random graph models, defining the limits of information-theoretic tractability.
Geometric Sensitivity and Alternative Information Probes
Beyond entropy and mutual information, the field utilizes several alternative information-theoretic metrics that probe different facets of a state's internal structure.
Quantum Fidelity Susceptibility
Instead of measuring how information is distributed within a single state, quantum fidelity susceptibility measures how sensitive a state is to external changes. It calculates the rate of change in the overlap between two ground states separated by an infinitesimal change in a Hamiltonian parameter $\lambda$:
$$\chi_F \propto \frac{1 - |\langle \psi(\lambda) | \psi(\lambda + d\lambda) \rangle|}{(d\lambda)^2}$$
Think of the ground state as a vector in Hilbert space. Throughout a stable phase, tuning $\lambda$ causes the vector to rotate slowly. At a critical point, the system undergoes a radical structural reorganization, causing the vector to suddenly swing in a completely new direction. The overlap drops precipitously, creating a massive spike in fidelity susceptibility. Ji-Hong Hu et al. and Yan-Wei Dai et al. demonstrated that this metric acts as a precise diagnostic for mapping both sharp quantum phase transitions and smooth crossovers, capturing the exact moment the ground state transforms.
Quantum Discord
Entanglement is notoriously fragile and vanishes rapidly at finite temperatures. To detect criticality in thermal regimes, researchers leverage quantum discord, which measures the total non-classical correlations between subsystems (including those that do not qualify as pure entanglement). Steve Campbell et al. and Ben‐Qiong Liu et al. showed that quantum discord is a far more robust indicator of phase transitions than entanglement, persisting over long distances and finite temperatures where standard entanglement measures drop to zero.
Participation Entropy
For localization transitions—such as the shift from a metallic, extended state to an insulating, localized state—standard spatial entanglement can be less informative than basis-dependent metrics. Hao Li et al. utilized participation entropy, which measures the delocalization of a quantum state within a specific configuration space. By tracking its time evolution and late-time saturation values on a superconducting circuit, the authors successfully mapped the critical phase diagram of the generalized Aubry-André-Harper model, demonstrating intermediate saturation values unique to the critical phase.
Data-Driven and Circuit-Level Criticality
The ultimate power of the information-theoretic paradigm is its abstraction away from specific Hamiltonians, allowing it to be applied directly to raw experimental data or circuit architectures.
On the data-driven front, Rajat Kumar Panda et al. showed that if one performs Principal Component Analysis (PCA) on raw thermal configurations of an Ising model, the numerical derivative of the resulting "PCA entropy" exhibits a sharp peak at the critical temperature. This derivative acts as a purely data-derived analog to physical specific heat, identifying critical points with less than 1% error without any knowledge of the underlying energy function. This aligns with the concept of "eigen microstate condensation," where a phase transition is signaled by a single configuration pattern suddenly emerging to capture a finite portion of the system's total variance.
Finally, this perspective enables the study of transitions driven not by energy, but by observation itself. In hybrid quantum circuits, unitary gates act to scramble information and build up volume-law entanglement throughout the system. Conversely, local projective measurements act as an information drain, collapsing the quantum state and reducing entanglement.
Marius Henry explored this balance in dual-unitary circuits, demonstrating that varying the measurement rate tunes the system across a measurement-induced phase transition. When measurements are rare, the circuit maintains a highly entangled volume-law phase; when measurements are frequent, the system collapses into a localized area-law phase. By engineering specific circuit architectures, the author demonstrated that systems can even bypass this binary transition entirely to reside in a unique, intermediate "critical purification phase" where entanglement scales as the square root of the subsystem size ($\sqrt{\ell}$).
By shifting the diagnostic focus from what particles are doing to how information flows, information-theoretic metrics provide a universal, highly sensitive toolkit for uncovering the boundaries of quantum matter.
Key Foundational Work
Universal entanglement entropy in two-dimensional conformal quantum critical points (2009) — contributed 1 concept elements to this direction.
Winding number transitions at finite temperature in thed= 2 Abelian-Higgs model (2000) — contributed 1 concept elements to this direction.