Majority Vote Model & Phase Transitions

• Majority Vote Model is a nonequilibrium spin system where binary spins align with the local majority, with a noise parameter allowing dissent.
• It employs a stochastic local-majority rule and finite-size scaling to analyze order–disorder phase transitions, drawing parallels to Ising-like behavior.
• Extensions include heterogeneous agents, inertial dynamics, memory effects, and multilayer networks, which enrich studies on consensus and critical phenomena.

The majority-vote model is a nonequilibrium spin system in which binary variables $\sigma_i=\pm 1$ evolve by a stochastic local-majority rule: a site tends to align with the majority state of its neighbors, but with a nonzero probability it does the opposite. In its standard form, the model has up–down symmetry, short-range interactions on a chosen substrate, and a control parameter interpreted as noise rather than equilibrium temperature. Although it does not, in general, satisfy detailed balance with respect to a Boltzmann–Gibbs distribution, it displays order–disorder phase transitions, finite-size scaling, and a broad family of extensions on lattices, complex networks, multiplex structures, reaction–diffusion systems, and non-Markovian settings (Lima, 2013, Santos et al., 2010, Chen et al., 2020).

1. Microscopic rule and nonequilibrium character

In the formulation introduced by M. J. Oliveira (1992), the model is defined on a lattice or graph whose sites carry Ising-like spins $\sigma_i=\pm1$. For a given site $i$, one computes the local majority through the sign of the neighborhood sum,

$$S\!\left(\sum_j \sigma_j\right)= \begin{cases} +1,& \sum_j \sigma_j>0,\ 0,& \sum_j \sigma_j=0,\ -1,& \sum_j \sigma_j<0. \end{cases}$$

The single-spin-flip probability is

$$w_i=\frac{1}{2}\left[1-(1-2q)\,\sigma_i\,S\!\left(\sum_j \sigma_j\right)\right].$$

When the local majority is nonzero, this implies alignment with probability $1-q$ and anti-alignment with probability $q$; in a tie, the flip probability is $1/2$ (Lima, 2013, Lima, 2011).

The parameter $q$ is the standard noise or misalignment probability. Low $q$ favors ordered, consensus-like states, whereas large $\sigma_i=\pm1$0 favors disorder. A recurring observation across the literature is that $\sigma_i=\pm1$1 plays a role analogous to temperature at the level of phase organization, but the analogy is limited: the dynamics is nonequilibrium, there is generally no Hamiltonian generating the transition probabilities, and detailed balance is not satisfied in the usual thermodynamic sense (Santos et al., 2010).

This nonequilibrium but up–down symmetric structure made the model a canonical test case for the Grinstein–Jayaprakash–He conjecture that non-equilibrium spin systems with up–down symmetry and short-range interactions share the universality class of the equilibrium Ising model. Subsequent work showed that this expectation is robust on some substrates and for some variants, but not uniformly across all geometries, update rules, or disorder patterns (Lima, 2011, Santos et al., 2010).

2. Observables, scaling relations, and critical diagnostics

The standard order parameter is the magnetization per site,

$\sigma_i=\pm1$2

Two further observables dominate the numerical analysis of the model: $\sigma_i=\pm1$3 Here $\sigma_i=\pm1$4 measures macroscopic ordering, $\sigma_i=\pm1$5 captures magnetization fluctuations, and the Binder cumulant $\sigma_i=\pm1$6 is used to locate critical points through size-dependent crossing behavior (Santos et al., 2010, Lima, 2013).

Near a continuous transition, the model is typically analyzed through finite-size scaling. For linear size $\sigma_i=\pm1$7,

$\sigma_i=\pm1$8

$\sigma_i=\pm1$9

$i$0

and the pseudocritical position of the susceptibility peak obeys

$i$1

These relations underpin the extraction of $i$2, $i$3, and $i$4 from Monte Carlo data (Santos et al., 2010, Lima, 2013).

A further diagnostic is the hyperscaling-like combination

$i$5

used extensively in network settings to define an effective dimensionality. On regular two-dimensional lattices this quantity is expected to be compatible with the embedding dimension two, whereas on several complex networks it is reported to be close to one (Santos et al., 2010, Lima, 2013).

Short-time methods provide a complementary route to criticality. On cubic lattices, the auxiliary function

$i$6

was introduced to locate $i$7 from crossings at fixed times and to obtain $i$8; finite-time scaling then yields $i$9, the dynamical exponent $$S\!\left(\sum_j \sigma_j\right)= \begin{cases} +1,& \sum_j \sigma_j>0,\ 0,& \sum_j \sigma_j=0,\ -1,& \sum_j \sigma_j<0. \end{cases}$$0, and the initial-slip exponent $$S\!\left(\sum_j \sigma_j\right)= \begin{cases} +1,& \sum_j \sigma_j>0,\ 0,& \sum_j \sigma_j=0,\ -1,& \sum_j \sigma_j<0. \end{cases}$$1 (Nascimento et al., 2020).

3. Regular lattices, Archimedean lattices, and the Ising question

On the square lattice, the two-state majority-vote model was long regarded as an Ising-like nonequilibrium counterpart. A three-state generalization with $$S\!\left(\sum_j \sigma_j\right)= \begin{cases} +1,& \sum_j \sigma_j>0,\ 0,& \sum_j \sigma_j=0,\ -1,& \sum_j \sigma_j<0. \end{cases}$$2 retained this behavior: Monte Carlo simulations gave $$S\!\left(\sum_j \sigma_j\right)= \begin{cases} +1,& \sum_j \sigma_j>0,\ 0,& \sum_j \sigma_j=0,\ -1,& \sum_j \sigma_j<0. \end{cases}$$3, $$S\!\left(\sum_j \sigma_j\right)= \begin{cases} +1,& \sum_j \sigma_j>0,\ 0,& \sum_j \sigma_j=0,\ -1,& \sum_j \sigma_j<0. \end{cases}$$4, $$S\!\left(\sum_j \sigma_j\right)= \begin{cases} +1,& \sum_j \sigma_j>0,\ 0,& \sum_j \sigma_j=0,\ -1,& \sum_j \sigma_j<0. \end{cases}$$5, $$S\!\left(\sum_j \sigma_j\right)= \begin{cases} +1,& \sum_j \sigma_j>0,\ 0,& \sum_j \sigma_j=0,\ -1,& \sum_j \sigma_j<0. \end{cases}$$6, and $$S\!\left(\sum_j \sigma_j\right)= \begin{cases} +1,& \sum_j \sigma_j>0,\ 0,& \sum_j \sigma_j=0,\ -1,& \sum_j \sigma_j<0. \end{cases}$$7, in agreement with the spin-$$S\!\left(\sum_j \sigma_j\right)= \begin{cases} +1,& \sum_j \sigma_j>0,\ 0,& \sum_j \sigma_j=0,\ -1,& \sum_j \sigma_j<0. \end{cases}$$8 and spin-1 Ising universality class (Lima, 2011).

Archimedean lattices became the main arena for testing whether the Ising correspondence survives nontrivial planar geometry. An early study of honeycomb $$S\!\left(\sum_j \sigma_j\right)= \begin{cases} +1,& \sum_j \sigma_j>0,\ 0,& \sum_j \sigma_j=0,\ -1,& \sum_j \sigma_j<0. \end{cases}$$9, Kagome $$w_i=\frac{1}{2}\left[1-(1-2q)\,\sigma_i\,S\!\left(\sum_j \sigma_j\right)\right].$$0, and triangular $$w_i=\frac{1}{2}\left[1-(1-2q)\,\sigma_i\,S\!\left(\sum_j \sigma_j\right)\right].$$1 lattices reported critical noises $$w_i=\frac{1}{2}\left[1-(1-2q)\,\sigma_i\,S\!\left(\sum_j \sigma_j\right)\right].$$2, $$w_i=\frac{1}{2}\left[1-(1-2q)\,\sigma_i\,S\!\left(\sum_j \sigma_j\right)\right].$$3, and $$w_i=\frac{1}{2}\left[1-(1-2q)\,\sigma_i\,S\!\left(\sum_j \sigma_j\right)\right].$$4, together with exponent ratios $$w_i=\frac{1}{2}\left[1-(1-2q)\,\sigma_i\,S\!\left(\sum_j \sigma_j\right)\right].$$5, $$w_i=\frac{1}{2}\left[1-(1-2q)\,\sigma_i\,S\!\left(\sum_j \sigma_j\right)\right].$$6, and $$w_i=\frac{1}{2}\left[1-(1-2q)\,\sigma_i\,S\!\left(\sum_j \sigma_j\right)\right].$$7, respectively, and concluded that these results differ from the usual Ising model results (Santos et al., 2010).

That conclusion was later challenged by a high-precision study of all eleven Archimedean lattices, which reported that, contrary to some previous reports, the majority-vote model with noise belongs to the two-dimensional Ising universality class on the full Archimedean family. A central methodological point was that very precise determination of the critical noise is required to obtain proper values of the critical exponents (Yu, 2016). This established an important controversy in the literature: some deviations previously attributed to novel universality were reinterpreted as consequences of finite-size and critical-point uncertainties.

A distinct but related branch replaced the standard noise parameter by a Glauber-type transition rate,

$$w_i=\frac{1}{2}\left[1-(1-2q)\,\sigma_i\,S\!\left(\sum_j \sigma_j\right)\right].$$8

with the mapping $$w_i=\frac{1}{2}\left[1-(1-2q)\,\sigma_i\,S\!\left(\sum_j \sigma_j\right)\right].$$9. On the Archimedean lattices $1-q$0, $1-q$1, and $1-q$2, this formulation yielded $1-q$3, $1-q$4, and $1-q$5, and exponent ratios that were reported to differ from Ising values and from one another (Lima, 2013). A plausible implication is that universality issues in majority-vote systems depend not only on substrate geometry but also on the precise dynamical implementation.

Setting	Representative results	Reported interpretation
Square-lattice three-state model	$1-q$6, $1-q$7, $1-q$8, $1-q$9	Same universality class as spin-1 and spin-$q$0 Ising (Lima, 2011)
Honeycomb, Kagome, triangular	$q$1	Non-Ising exponents reported in an early study (Santos et al., 2010)
Eleven Archimedean lattices	high-precision $q$2 and exponent estimates	Two-dimensional Ising universality, contrary to some previous reports (Yu, 2016)
Glauber-type MV on $q$3, $q$4, $q$5	$q$6	Critical behavior reported as non-Ising (Lima, 2013)

4. Complex networks, heterogeneity, and finite-size controversies

On complex and adaptive networks, the model departs more sharply from the square-lattice picture. On Stauffer–Hohnisch–Pittnauer opinion-dependent networks, the reported values $q$7, $q$8, $q$9, $1/2$0, and $1/2$1 satisfy $1/2$2, indicating $1/2$3 and a universality class distinct from the equilibrium Ising model on regular lattices (Lima, 2013).

Agent heterogeneity can change the critical behavior even on a square lattice. In the heterogeneous-agent model, each site receives an individual noise $1/2$4 drawn uniformly from $1/2$5, so that

$1/2$6

This quenched disorder produced $1/2$7, $1/2$8, $1/2$9, $q$0, and $q$1, and was interpreted as a different universality class from the homogeneous square-lattice majority-vote model (Lima, 2013).

A further extension adds ancillary independent noise. In the noisy majority-vote model,

$q$2

so that with probability $q$3 the usual majority-vote rule is applied and with probability $q$4 a random flip occurs independently of the neighborhood. On random regular, Erdős–Rényi, and Barabási–Albert networks, both mean-field calculations and simulations indicate that the additional noise does not alter the continuous character of the phase transition, although the critical point and critical exponents depend on the network class (Encinas et al., 2018).

Scale-free networks have generated a second, still-active controversy. A droplet finite-size scaling theory for uncorrelated scale-free networks reports non-universal critical behavior for $q$5, with exponent ratios depending on the degree-distribution exponent and with $q$6; for $q$7, the model is placed in the same universality class as the majority-vote model on Erdős–Rényi random graphs, while $q$8 carries logarithmic corrections (Alencar et al., 2023). Short-time heterogeneous mean-field theory reaches a parallel conclusion for dynamical scaling, yielding a $q$9-dependent dynamical exponent $q$0 for $q$1, $q$2, and mean-field Ising criticality with logarithmic corrections at $q$3 and above (Alencar et al., 2024). By contrast, a scaling-corrections analysis on Barabási–Albert networks argues that the leading critical exponents are in fact universal and connectivity-independent, with the apparent non-universality arising from strong logarithmic corrections to finite-size scaling (Alves et al., 2019). This suggests that the asymptotic status of scale-free majority-vote criticality depends sensitively on the finite-size framework used to interpret heterogeneous-network data.

5. Dynamical exponents, inertia, memory, and thermodynamic consistency

In three dimensions, short-time Monte Carlo simulations on cubic lattices of volume $q$4 yielded a high-precision critical point $q$5, together with $q$6, $q$7, $q$8, and $q$9. These values were interpreted as evidence that the isotropic two-state majority-vote model on cubic lattices belongs to the same universality class as the three-dimensional Ising model (Nascimento et al., 2020).

The addition of inertia produces a qualitatively different regime. In the inertial majority-vote model, the local field becomes

$\sigma_i=\pm1$00

and the spin-flip probability retains the majority-vote form with $\sigma_i=\pm1$01. On Erdős–Rényi, random degree-regular, and Barabási–Albert networks, increasing $\sigma_i=\pm1$02 above a threshold converts the usual continuous transition into a discontinuous or explosive one, with strong hysteresis, coexistence of a disordered phase and two symmetric ordered phases, and transition rates that decay exponentially with system size in the hysteresis region (Chen et al., 2016).

Memory effects can be introduced without changing the binary state space. In the non-Markovian majority-vote model, each agent carries an age $\sigma_i=\pm1$03, and the flip rate is modulated as

$\sigma_i=\pm1$04

with $\sigma_i=\pm1$05 decreasing in the aging regime and increasing in the anti-aging regime. For exponential, linear, rational, and power-law kernels without characteristic age scales, the critical noise becomes a non-monotonic function of the rate $\sigma_i=\pm1$06: in the aging regime it displays a maximum, and in the anti-aging regime it displays a minimum. This was verified within heterogeneous mean-field theory and by simulations on random regular, Erdős–Rényi, scale-free, two-dimensional, and three-dimensional substrates (Chen et al., 2020).

A separate development embeds the model into stochastic thermodynamics. The original majority-vote model is not, in general, thermodynamically consistent, so the paper on nonequilibrium thermodynamics introduces a distinct heat bath for each local neighborhood value $\sigma_i=\pm1$07. For the inertial model,

$\sigma_i=\pm1$08

and the multibath construction yields neighborhood-dependent inverse temperatures

$\sigma_i=\pm1$09

with the plateau value

$\sigma_i=\pm1$10

at $\sigma_i=\pm1$11. This permits the definition of heat fluxes, entropy production, and the decomposition of dissipation by local configuration, for both continuous and discontinuous transitions on regular and complex substrates (Hawthorne et al., 2023).

6. Mobile populations, multiplex rules, and feedback-coupled noise

The majority-vote rule has also been generalized to reaction–diffusion and multilayer settings. In the diffusive majority-vote model, each time step comprises diffusion and reaction. Walkers carrying spins $\sigma_i=\pm1$12 hop between neighboring nodes with probabilities $\sigma_i=\pm1$13 and $\sigma_i=\pm1$14, and the spins at each node then update according to the local majority among co-located walkers. For equal diffusion probabilities $\sigma_i=\pm1$15, the model has a density-controlled transition between a paramagnetic phase and a consensus phase; at $\sigma_i=\pm1$16 and $\sigma_i=\pm1$17, the threshold densities are $\sigma_i=\pm1$18 in two dimensions and $\sigma_i=\pm1$19 in three dimensions. For unequal diffusion probabilities, the threshold vanishes and the stationary collective opinion is dominated by the faster-diffusing individuals (Lima et al., 2021).

On multiplex networks, the central issue is how different social layers are combined at the update stage. In the AND-model, a voter updates only if all layers share the same local majority; in the OR-model, a layer is chosen at random and its local majority is followed. The AND-model reaches the largest consensus below the critical noise parameter $\sigma_i=\pm1$20, but needs much longer time to reach consensus than the OR-rule or single-network models; near the transition, its consensus can collapse abruptly. The OR-model attains smaller consensus, reaches it more quickly, and displays a continuous transition (Choi et al., 2018).

A further multilayer extension makes the noise itself dynamical. In the feedback-coupled model, the vote layer follows a site-dependent rule,

$\sigma_i=\pm1$21

while the noise layer evolves through

$\sigma_i=\pm1$22

Here $\sigma_i=\pm1$23 is the size of the vote-layer cluster containing node $\sigma_i=\pm1$24. Across square lattices, random-regular networks, and their combinations, the study reports independent third-order transitions in all cases and dependent third-order transitions when critical transitions occur, suggesting that dependent third-order transitions may serve as precursors of critical transitions in nonequilibrium systems. It also reports that when the structure of vote layers is local, coupling to the noise layer leads to the absence of critical phenomena (Liu et al., 2023).

Related majority-based imitation processes show that the boundary of the majority-vote family is not purely taxonomic. In a discrete-time imitation model based on the majority, an agent imitates a sampled neighbor only if that neighbor’s opinion has strictly larger support in the focal neighborhood. On finite connected graphs, consensus is guaranteed on complete graphs but can fail on general graphs because locally stable mixed configurations persist (Li, 2023). This suggests that the defining feature of the majority-vote model is not merely majority preference, but the specific stochastic majority-rule flip mechanism and its associated nonequilibrium scaling structure.

Across these formulations, the majority-vote model serves as a compact framework for studying how local conformity, dissent, inertia, heterogeneity, mobility, memory, and multilayer coupling reorganize collective behavior. The resulting literature does not support a single universal conclusion beyond the standard square-lattice case; rather, it shows that the model’s critical behavior is highly sensitive to substrate, update rule, and auxiliary degrees of freedom, while remaining one of the most extensively analyzed nonequilibrium laboratories for opinion dynamics and symmetry-driven phase transitions (Santos et al., 2010, Choi et al., 2018).