Griffiths Phases in Disordered Systems

• Griffiths Phases are regions in disordered systems where rare, locally ordered patches cause non-universal power-law scaling of observables.
• They arise in various settings, including magnetic systems, quantum magnets, ecological models, and complex networks, demonstrating slow, algebraic relaxation.
• Observable signatures such as divergent susceptibilities and continuously varying exponents provide insights into dynamical phase transitions and error resilience.

Griffiths phases constitute a major paradigm in disordered statistical physics and dynamical systems, marking extended regions of phase space where observables exhibit non-universal, power-law scalings governed by rare-region effects. These phases arise generically when spatial (or, analogously, temporal) disorder permits rare patches that locally reside in an ordered, active, or supercritical regime—even as the bulk system remains globally disordered or inactive. Characterized by slow, algebraic relaxation, divergent susceptibilities, and continuously varying exponents, Griffiths phases have been observed across quantum magnets, ecological models, brain networks, evolutionary games, complex networks, topologically disordered quantum error-correcting codes, and more. In each context, the crucial mechanism is the dominance of rare, exponentially long-lived regions in the dynamics or thermodynamics of the system.

1. Mechanism and General Theory

The Griffiths phase emerges when disorder (in coupling strengths, network topology, or local rules) produces rare regions ("RRs") of atypical local properties. These RRs remain locally supercritical or ordered, sustaining activity or order for exponentially long times relative to their size. The probability $P(s)$ to find a region of size $s$ is typically $P(s) \sim e^{-c s}$, while the relaxation (lifetime) for such a region scales as $\tau(s) \sim e^{\beta s}$ (Arrhenius form). In the time domain, the global observable—such as magnetization, activity density, or order parameter—relaxes as a convolution over clusters or patches

$$\rho(t) \sim \int ds\, P(s) e^{-t/\tau(s)} \sim t^{-\alpha}$$

yielding a continuously variable power-law exponent $\alpha = c/\beta$. This slow decay persists throughout the Griffiths region, which lies between the global critical point and a lower threshold set by the disorder realization.

Griffiths phases are distinct from conventional second-order transitions (which have a single critical point, diverging correlation length, and universal exponents) and from first-order transitions (which feature discontinuities). Instead, they show a continuum of critical exponents and a stretched ("smeared") critical regime, in both equilibrium (e.g., classical/quantum magnets) and nonequilibrium models (e.g., contact processes, neural networks, evolutionary games) (Muñoz et al., 2010, Nozadze et al., 2011, Chatelain, 2013, Chatelain, 2014, Bhoyar et al., 2020, Moretti et al., 2013).

2. Griffiths Phases in Disordered Classical and Quantum Systems

2.1 Magnetic Systems

In the prototype scenario introduced by R.B. Griffiths, dilute magnets above but near $T_c$ contain rare, large clusters capable of local order. The resulting non-analytical free energy and non-universal divergence in susceptibility over $T_c < T < T_G$ are hallmarks of the classical Griffiths phase (Ślebarski et al., 4 Jun 2025, Bhoi et al., 2011, Chatelain, 2014). The susceptibility behaves as

$$\chi(T) \propto (T-T^R_C)^{-(1-\lambda)}, \qquad 0 < \lambda < 1$$

where $\lambda$ is the Griffiths exponent encoding the (disorder-dependent) weight of rare regions (Bhoi et al., 2011).

In quantum magnets, and Kondo-lattice compounds near quantum phase transitions, rare regions persist to $s$0, producing a quantum Griffiths phase. Observables follow non-Fermi-liquid power laws: $s$1 with $s$2 continuously varying over the quantum Griffiths region (Ślebarski et al., 4 Jun 2025). Experimental realizations include disordered CeRhSn, which shows both classical and quantum Griffiths regimes with $s$3, power-law susceptibility, and specific heat (Ślebarski et al., 4 Jun 2025).

2.2 Transport and Thermodynamics

In quantum Griffiths phases, rare-region physics controls transport, leading to a single, non-universal exponent $s$4 for all low-temperature power laws: $s$5 and violating "universal" Fermi liquid ratios, such as the Wiedemann–Franz and Wilson ratios (Nozadze et al., 2011, Nozadze et al., 2011). The exponent $s$6 vanishes at the quantum critical point and increases monotonically into the Griffiths region.

3. Griffiths Phases on Complex and Modular Networks

3.1 Dynamical Phase Transitions on Graphs

On random, modular, or hierarchical networks, Griffiths phases arise in spreading models—such as the contact process (CP), SIS, and threshold models—when topological or coupling disorder creates rare, locally supercritical clusters (Muñoz et al., 2010, Ódor et al., 2010, Ódor et al., 2015, Cota et al., 2018, Li, 2016).

The presence of a finite topological dimension $s$7 is crucial: in finite-$s$8 (large-world) networks, rare subgraphs remain locally active for exponentially long times. The global density then decays algebraically in time: $s$9 for an extended $P(s) \sim e^{-c s}$0-interval, the Griffiths phase (Ódor et al., 2010, Ódor et al., 2015). In contrast, small-world networks ($P(s) \sim e^{-c s}$1) lack such regions, and transitions are sharp.

3.2 Hierarchical Modularity, Localization, and Brain Networks

Hierarchical modular network models, inspired by connectomic data, exhibit Griffiths phases when the inter-module link probability decays exponentially with module size, yielding rare, weakly coupled modules that act as long-lived rare regions (Moretti et al., 2013, Ódor et al., 2015, Li, 2016). Activity avalanches in such networks produce power-law distributions of size and duration with exponents varying across the Griffiths region. Localization analysis (e.g., via the inverse participation ratio of network eigenmodes) confirms the confinement of activity to rare modules in the GP (Ódor et al., 2015, Cota et al., 2018).

Empirical connectomes (human, C. elegans) show evidence for stretched criticality and scale-free avalanching compatible with Griffiths-phase dynamics (Moretti et al., 2013, Pretel et al., 2023).

3.3 Threshold and Hybrid Transitions

Models with higher activation thresholds ($P(s) \sim e^{-c s}$2), even in high-dimensional or densely connected graphs, may exhibit GPs below (or alongside) hybrid transitions—a combination of discontinuity and critical scaling rounded by disorder, with the GP characterized by non-universal avalanche scaling (Ódor et al., 2020).

4. Extensions: Evolutionary Games, Temporal Disorder, and Error Correction

4.1 Evolutionary Game Theory

Quenched heterogeneity in payoff matrices in spatial evolutionary games (e.g., the Prisoner's Dilemma) produces extended intervals (upon varying temptation or payoff parameters) of slow, power-law decay for both cooperative and defector densities, i.e., dual Griffiths phases symmetrically placed around the clean critical points (Amaral et al., 2021). Here, rare favorable regions for cooperation or defection act analogously to rare regions in magnets or spreading models.

4.2 Temporal Griffiths Phases

Temporal (rather than spatial) disorder induces "Temporal Griffiths Phases" (TGPs): rare, long intervals of globally unfavorable or favorable conditions dominate the scaling of system lifetimes and observables (Vazquez et al., 2011). In $P(s) \sim e^{-c s}$3 dimensions, TGPs show algebraically scaling mean lifetimes $P(s) \sim e^{-c s}$4 and diverging response as a function of frequency, with exponents parameterized by statistics of the temporal disorder.

4.3 Topological Quantum Error Correction

In topological quantum codes, non-uniform, correlated error rates map the decoding problem to random-bond statistical mechanics. In 1D repetition codes, extended rare regions of high error rates yield a Griffiths phase where logical failure probabilities decay only as stretched exponentials (or even power laws), dominated by the probability distribution of rare regions (Sriram et al., 2024). For the 2D toric code, rare planar regions are exponentially suppressed in area, precluding a true decodable GP (Sriram et al., 2024).

5. Mathematical Structure and Observable Signatures

5.1 Universal Griffiths Scaling

A defining mathematical hallmark is a distribution of rare-region sizes $P(s) \sim e^{-c s}$5 and lifetimes $P(s) \sim e^{-c s}$6 that leads to integrals of the form

$P(s) \sim e^{-c s}$7

for global observables $P(s) \sim e^{-c s}$8 such as the density, magnetization, or survival probability. The exponent $P(s) \sim e^{-c s}$9 varies with system parameters and with disorder statistics (Ódor et al., 2010, Leonardi et al., 29 Sep 2025).

5.2 Continuously Varying Exponents

In the GP, exponents for power-law decay (e.g., $\tau(s) \sim e^{\beta s}$0 or $\tau(s) \sim e^{\beta s}$1) and finite-size scaling (e.g., $\tau(s) \sim e^{\beta s}$2) are non-universal, drifting with control parameters and the statistical properties of disorder (Chatelain, 2014, Chatelain, 2013, Li, 2016). This is reflected in, e.g.,

$\tau(s) \sim e^{\beta s}$3

for abundance distributions in Lotka–Volterra systems, with

$\tau(s) \sim e^{\beta s}$4

as in mutualistic ER-networks under strong interaction disorder (Leonardi et al., 29 Sep 2025).

5.3 Crossover to Activated or Conventional Scaling

At the boundaries of the Griffiths phase, exponents vanish and scaling may become activated (logarithmic) at infinite-randomness fixed points, or crossover to pure exponential relaxation in the bulk-inactive phase (Ódor et al., 2010, Bhoyar et al., 2020).

6. Robustness, Topology, and Universality

The existence and width of the Griffiths phase are controlled by both disorder amplitude and the correlation structure (e.g., spatial or modular correlations in couplings, network topology). Griffiths phases persist for broad disorder distributions, in non-hierarchical infinite-dimensional modular structures, and even for strong sign-biased interactions (Cota et al., 2018, Leonardi et al., 29 Sep 2025). Hierarchy is not required; what is essential is the presence of slowly mixing, weakly coupled large modules or domains (Cota et al., 2018).

Competing weak uniform effects (e.g., in Lotka–Volterra theory, global competition) may shift critical boundaries, but generally do not destroy the Griffiths regime (Leonardi et al., 29 Sep 2025).

7. Functional and Practical Implications

Griffiths phases are central to understanding:

• Empirical power-law abundance distributions in ecology without fine-tuning, as rare mutualistic modules generate robust power-law SADs (Leonardi et al., 29 Sep 2025).
• Neural criticality and information processing in hierarchical, modular brain networks by enabling self-organized critical-like dynamics over a wide parameter regime (Moretti et al., 2013, Pretel et al., 2023, Li, 2016, Ódor et al., 2020).
• Non-universal exponents in avalanching activity and stretched memory decay, supporting functional adaptability and resilience.
• Enhanced robustness of complex systems by eliminating the need for fine-tuning to a unique critical point.

Griffiths-phase behavior has practical consequences for the design and fault tolerance of quantum codes, the interpretation of evolutionary and ecological data, and the analysis of dynamical patterns in biological and computational networks (Sriram et al., 2024, Amaral et al., 2021, Leonardi et al., 29 Sep 2025).

---

References:

• "Griffiths phase emerging from strong mutualistic disorder in high-dimensional interacting systems" (Leonardi et al., 29 Sep 2025)
• "Griffiths phases in structurally disordered CeRhSn: Experimental evidence and theoretical modeling" (Ślebarski et al., 4 Jun 2025)
• "Transport properties in antiferromagnetic quantum Griffiths phases" (Nozadze et al., 2011)
• "Stability of the Griffiths phase in the 2D Potts model with correlated disorder" (Chatelain, 2014)
• "Griffiths phases on complex networks" (Muñoz et al., 2010)
• "Griffiths phases and localization in hierarchical modular networks" (Ódor et al., 2015)
• "Griffiths phases in infinite-dimensional, non-hierarchical modular networks" (Cota et al., 2018)
• "Griffiths phase and critical behavior of the 2D Potts models with long-range correlated disorder" (Chatelain, 2013)
• "Griffiths phases and the stretching of criticality in brain networks" (Moretti et al., 2013)
• "From asynchronous states to Griffiths phases and back: structural heterogeneity and homeostasis in excitatory-inhibitory networks" (Pretel et al., 2023)
• "Non-Uniform Noise Rates and Griffiths Phases in Topological Quantum Error Correction" (Sriram et al., 2024)
• "Criticality and Griffiths phases in random games with quenched disorder" (Amaral et al., 2021)
• "Dynamic Phase Transition in the Contact Process with Spatial Disorder: Griffths Phase and Complex Persistence Exponents" (Bhoyar et al., 2020)
• "Heterogeneous excitable systems exhibit Griffiths phases below hybrid phase transitions" (Ódor et al., 2020)
• "The Griffiths Phase on Hierarchical Modular Networks with Small-world Edges" (Li, 2016)
• "Temporal Griffiths Phases" (Vazquez et al., 2011)