Curie–Weiss Cell Model Overview

• Curie–Weiss Cell Model is a mean-field framework where microscopic variables in cells interact through a global average rather than local pairwise interactions.
• It encompasses diverse formulations—spin models, conditional variants, and cell-fluid systems—that reduce complex interactions to a small set of collective variables.
• The model reveals rich phase behavior, including first-order transitions, non-Gaussian critical fluctuations, and symmetry breaking in different occupancy regimes.

Searching arXiv for recent and foundational papers on Curie–Weiss cell models and related multi-group/cell-fluid formulations. A Curie–Weiss cell model is a mean-field interacting system in which the relevant microscopic variables are attached to “cells,” groups, or coarse-grained occupancy sites, while the dominant collective interaction has Curie–Weiss form: each degree of freedom couples to a global average rather than to a finite-range neighborhood. In the literature, this designation covers at least three closely related constructions: multi-group spin models with group magnetizations as order parameters, conditional or partially frozen Curie–Weiss populations, and cell-fluid models in which the volume is partitioned into cells with multiple occupancy, global attraction, and local repulsion (Kirsch et al., 2018, Opoku et al., 2016, Kozlovskii et al., 2021). Across these variants, the common mathematical structure is a reduction of the thermodynamic or dynamical problem to a small set of collective variables—typically magnetizations or mean occupancies—together with phase transitions characterized by law-of-large-numbers limits, Gaussian or non-Gaussian fluctuation regimes, and coexistence between symmetry-broken or density-distinct phases (Kirsch et al., 2017, Romanik et al., 1 Jul 2026).

1. Mean-field definition and principal variants

In spin formulations, the Curie–Weiss cell idea appears when the full system is decomposed into interacting populations or “cells,” each represented by its own empirical magnetization. A basic heterogeneous two-cell construction has spins $X_1,\dots,X_{N_1}$ in group 1 and $Y_1,\dots,Y_{N_2}$ in group 2, with $X_i,Y_j\in\{-1,+1\}$, proportions $\alpha_1=\lim N_1/N$, $\alpha_2=\lim N_2/N$, and a symmetric coupling matrix

$$J=\begin{pmatrix}J_1 & J\ J & J_2\end{pmatrix}, \qquad A:=J_1J_2-J^2>0,$$

where $J_1>0$ and $J_2>0$ are intra-group couplings and $J>0$ is the inter-group coupling (Kirsch et al., 2018). The associated Hamiltonian is quadratic in the group magnetizations $S_1=\sum_{i=1}^{N_1}X_i$ and $Y_1,\dots,Y_{N_2}$0, which is the defining Curie–Weiss feature: macroscopic observables enter through a mean-field quadratic form rather than through geometry-dependent pair distances (Kirsch et al., 2018).

A simpler two-group model keeps the classical homogeneous Curie–Weiss coupling $Y_1,\dots,Y_{N_2}$1 for all spins and only distinguishes two subsets at the level of observation. In that case, the two group magnetizations

$Y_1,\dots,Y_{N_2}$2

are monitored inside a single homogeneous mean-field environment (Kirsch et al., 2017). This produces a “two-cell” interpretation without changing the microscopic interaction law.

A different construction is the conditional Curie–Weiss model, where one part of the population is fixed and only the remaining spins fluctuate. The population is partitioned into a group fixed to $Y_1,\dots,Y_{N_2}$3, a group fixed to $Y_1,\dots,Y_{N_2}$4, and an undecided group of proportion $Y_1,\dots,Y_{N_2}$5, with the conditional Gibbs measure obtained by restricting the standard Curie–Weiss model to configurations respecting the fixed spins (Opoku et al., 2016). In this setting, the dynamic part is “a Curie-Weiss model on the sites in $Y_1,\dots,Y_{N_2}$6 with $Y_1,\dots,Y_{N_2}$7-dependent coupling strength $Y_1,\dots,Y_{N_2}$8 and $Y_1,\dots,Y_{N_2}$9-dependent external field $X_i,Y_j\in\{-1,+1\}$0” (Opoku et al., 2016).

In fluid and lattice-gas formulations, the cells are actual spatial boxes. The volume $X_i,Y_j\in\{-1,+1\}$1 is partitioned into $X_i,Y_j\in\{-1,+1\}$2 congruent cubic cells $X_i,Y_j\in\{-1,+1\}$3 of volume $X_i,Y_j\in\{-1,+1\}$4, each cell carries an occupation number, and the interaction combines a global Curie–Weiss attraction with a cell-local repulsion (Kozitsky et al., 2016, Kozlovskii et al., 2021, Dobush et al., 20 May 2025). One version allows unbounded occupancy $X_i,Y_j\in\{-1,+1\}$5 (Kozlovskii et al., 2021, Dobush et al., 20 May 2025), while another imposes double occupancy $X_i,Y_j\in\{-1,+1\}$6 and is isomorphic to the Blume–Capel model on a complete graph (Romanik et al., 1 Jul 2026).

2. Spin-cell formulations: group magnetizations and coupling structure

For the two-group heterogeneous spin model, the empirical magnetizations are

$X_i,Y_j\in\{-1,+1\}$7

The Gibbs measure is

$X_i,Y_j\in\{-1,+1\}$8

with coupling constants absorbing inverse temperature via $X_i,Y_j\in\{-1,+1\}$9, $\alpha_1=\lim N_1/N$0, $\alpha_1=\lim N_1/N$1 (Kirsch et al., 2018). The resulting phase structure is governed by the matrix

$\alpha_1=\lim N_1/N$2

so the distinction between high temperature, criticality, and low temperature is encoded by the spectral properties of $\alpha_1=\lim N_1/N$3 rather than by a single scalar threshold (Kirsch et al., 2018).

The high-temperature regime is characterized by $\alpha_1=\lim N_1/N$4 positive definite, equivalently by

$\alpha_1=\lim N_1/N$5

Under these conditions, $\alpha_1=\lim N_1/N$6 in distribution and the scaled magnetizations satisfy a bivariate central limit theorem with covariance matrix

$\alpha_1=\lim N_1/N$7

so cross-group couplings induce correlated Gaussian fluctuations even when the macroscopic magnetization vanishes (Kirsch et al., 2018).

At criticality,

$\alpha_1=\lim N_1/N$8

the law of large numbers still yields $\alpha_1=\lim N_1/N$9, but the Gaussian central limit theorem fails and the correct scaling is $\alpha_2=\lim N_2/N$0, with a non-Gaussian limit determined by quartic-order fluctuations (Kirsch et al., 2018). In the low-temperature regime, the magnetization vector converges in distribution to

$\alpha_2=\lim N_2/N$1

where $\alpha_2=\lim N_2/N$2 is the positive solution of the mean-field equations

$\alpha_2=\lim N_2/N$3

so the two cells exhibit spontaneous symmetry breaking into two symmetry-related ferromagnetic phases (Kirsch et al., 2018).

The homogeneous two-group model yields a more restrictive outcome. Since all spins share the same global Curie–Weiss interaction, the two group magnetizations converge jointly to

$\alpha_2=\lim N_2/N$4

where $\alpha_2=\lim N_2/N$5 solves $\alpha_2=\lim N_2/N$6 (Kirsch et al., 2017). This excludes distinct-sign limiting phases for the two groups and shows that any macroscopic subsets are locked to the same global phase in the homogeneous model (Kirsch et al., 2017).

3. Conditional and exchangeable formulations

The conditional Curie–Weiss model introduces a frozen-core architecture. The population is partitioned into $\alpha_2=\lim N_2/N$7 fixed to $\alpha_2=\lim N_2/N$8, $\alpha_2=\lim N_2/N$9 fixed to $$J=\begin{pmatrix}J_1 & J\ J & J_2\end{pmatrix}, \qquad A:=J_1J_2-J^2>0,$$0, and $$J=\begin{pmatrix}J_1 & J\ J & J_2\end{pmatrix}, \qquad A:=J_1J_2-J^2>0,$$1 of undecided spins, with $$J=\begin{pmatrix}J_1 & J\ J & J_2\end{pmatrix}, \qquad A:=J_1J_2-J^2>0,$$2, $$J=\begin{pmatrix}J_1 & J\ J & J_2\end{pmatrix}, \qquad A:=J_1J_2-J^2>0,$$3, and $$J=\begin{pmatrix}J_1 & J\ J & J_2\end{pmatrix}, \qquad A:=J_1J_2-J^2>0,$$4 (Opoku et al., 2016). The specific magnetization satisfies

$$J=\begin{pmatrix}J_1 & J\ J & J_2\end{pmatrix}, \qquad A:=J_1J_2-J^2>0,$$5

where $$J=\begin{pmatrix}J_1 & J\ J & J_2\end{pmatrix}, \qquad A:=J_1J_2-J^2>0,$$6 minimizes

$$J=\begin{pmatrix}J_1 & J\ J & J_2\end{pmatrix}, \qquad A:=J_1J_2-J^2>0,$$7

on $$J=\begin{pmatrix}J_1 & J\ J & J_2\end{pmatrix}, \qquad A:=J_1J_2-J^2>0,$$8 (Opoku et al., 2016). The threshold

$$J=\begin{pmatrix}J_1 & J\ J & J_2\end{pmatrix}, \qquad A:=J_1J_2-J^2>0,$$9

marks a first-order phase transition in magnetization at $J_1>0$0: for $J_1>0$1, the magnetization is continuous at that field, whereas for $J_1>0$2 it jumps discontinuously (Opoku et al., 2016). The paper explicitly distinguishes a discontinuous jump in the order parameter from a genuine change of phase, because the sign of the total magnetization can remain fixed when one frozen opinion has a sufficiently large majority (Opoku et al., 2016).

A different probabilistic viewpoint treats the Curie–Weiss model through exchangeability. The Gibbs law is represented as a De Finetti mixture

$J_1>0$3

so conditional on the latent parameter $J_1>0$4, the spins are i.i.d. Bernoulli and the dependence is carried by the random mixing measure $J_1>0$5 (Barhoumi-Andréani et al., 2023). In that framework, the magnetization admits the surrogate

$J_1>0$6

with $J_1>0$7 independent of the De Finetti random variable $J_1>0$8 (Barhoumi-Andréani et al., 2023). This suggests that the phase transition can be interpreted as a competition between “the randomness of the i.i.d. spin variables given the latent environment” and “the randomness of the latent environment $J_1>0$9 itself” (Barhoumi-Andréani et al., 2023).

That decoupling was sharpened through almost-sure Laplace inversion. Using the De Finetti representation, the uniform coupling of Bernoulli variables, and the inverse Laplace transform on a complex line, the magnetization is expressed as a functional of an i.i.d. field and an independent randomisation field; the limiting Gaussian objects are identified as a Gaussian analytic function, a Brownian bridge, and a Brownian sheet (Barhoumi-Andréani et al., 23 Jul 2025). The same analysis reproduces the subcritical Gaussian regime, the $J_2>0$0 critical scaling with non-Gaussian limit, and supercritical fluctuations around $J_2>0$1, where $J_2>0$2 solves

$J_2>0$3

(Barhoumi-Andréani et al., 23 Jul 2025). A plausible implication is that the “cell” interpretation of Curie–Weiss models extends naturally to a hierarchy of probabilistic fields: microscopic i.i.d. cells, a global randomisation field, and Gaussian functional limits governing fluctuations.

4. Cell-fluid models with multiple occupancy

In continuum cell-fluid formulations, the volume $J_2>0$4 is partitioned into $J_2>0$5 cells $J_2>0$6 of equal volume $J_2>0$7, and the occupation number of cell $J_2>0$8 is

$J_2>0$9

The pair potential is

$J>0$0

with $J>0$1 or equivalently $J>0$2 for stability (Dobush et al., 20 May 2025, Kozlovskii et al., 2021). The inter-cell attraction is uniform and of Curie–Weiss type, while the intra-cell repulsion penalizes multiple occupancy (Kozlovskii et al., 2021).

In the exactly solved grand-canonical reduction, the partition function takes the one-dimensional form

$J>0$3

where

$J>0$4

$J>0$5

(Dobush et al., 20 May 2025). The saddle-point equation gives

$J>0$6

so the reduced density is the average number of particles per cell (Dobush et al., 20 May 2025). The pressure is

$J>0$7

in the notation of that work (Dobush et al., 20 May 2025).

An earlier formulation on $J>0$8 writes the occupation-space partition function as

$J>0$9

with $S_1=\sum_{i=1}^{N_1}X_i$0, $S_1=\sum_{i=1}^{N_1}X_i$1, and $S_1=\sum_{i=1}^{N_1}X_i$2 (Kozitsky et al., 2016). There the thermodynamic phase is a product measure over cell occupation numbers determined by a Curie–Weiss self-consistency equation for the scalar parameter $S_1=\sum_{i=1}^{N_1}X_i$3,

$S_1=\sum_{i=1}^{N_1}X_i$4

and the pressure in the thermodynamic limit is $S_1=\sum_{i=1}^{N_1}X_i$5, where $S_1=\sum_{i=1}^{N_1}X_i$6 (Kozitsky et al., 2016). For sufficiently small attraction there is a single phase, while for large enough $S_1=\sum_{i=1}^{N_1}X_i$7 the model exhibits phase-coexistence points with two thermodynamic phases of distinct densities (Kozitsky et al., 2016).

A double-occupancy version restricts $S_1=\sum_{i=1}^{N_1}X_i$8 and isomorphicly maps to the Blume–Capel model on a complete graph via $S_1=\sum_{i=1}^{N_1}X_i$9, $Y_1,\dots,Y_{N_2}$00 (Romanik et al., 1 Jul 2026). Its Hamiltonian contains a Curie–Weiss attraction term $Y_1,\dots,Y_{N_2}$01, a local double-occupancy repulsion $Y_1,\dots,Y_{N_2}$02, and an effective linear term, with the grand partition function reduced by a Hubbard–Stratonovich transformation to a saddle-point problem over a single auxiliary variable (Romanik et al., 1 Jul 2026).

5. Phase structure, criticality, and coexistence

A characteristic feature of the unbounded-occupancy Curie–Weiss cell fluid is a sequence of first-order phase transitions below the first critical temperature. As the chemical potential or density increases, the system passes through phases I, II, III, and IV with density ranges approximately $Y_1,\dots,Y_{N_2}$03, $Y_1,\dots,Y_{N_2}$04, $Y_1,\dots,Y_{N_2}$05, and $Y_1,\dots,Y_{N_2}$06, and the transitions I–II, II–III, and III–IV occur near $Y_1,\dots,Y_{N_2}$07, $Y_1,\dots,Y_{N_2}$08, and $Y_1,\dots,Y_{N_2}$09, respectively (Dobush et al., 20 May 2025). The coexistence points are obtained microscopically from

$Y_1,\dots,Y_{N_2}$10

which is the mean-field analogue of Maxwell construction but derived directly from the saddle-point grand potential (Dobush et al., 20 May 2025).

The explicit equation of state derived for the same class of models is

$Y_1,\dots,Y_{N_2}$11

with

$Y_1,\dots,Y_{N_2}$12

where $Y_1,\dots,Y_{N_2}$13 is the saddle variable and $Y_1,\dots,Y_{N_2}$14 is the density (Kozlovskii et al., 2021). In that analysis, critical points are defined microscopically by

$Y_1,\dots,Y_{N_2}$15

which yields a sequence of critical temperatures

$Y_1,\dots,Y_{N_2}$16

(Kozlovskii et al., 2021). For $Y_1,\dots,Y_{N_2}$17 and $Y_1,\dots,Y_{N_2}$18, the first three critical densities are approximately $Y_1,\dots,Y_{N_2}$19, $Y_1,\dots,Y_{N_2}$20, and $Y_1,\dots,Y_{N_2}$21, demonstrating multiple critical points associated with successive density transitions (Kozlovskii et al., 2021).

The double-occupancy model displays a different but equally rich phase diagram. With $Y_1,\dots,Y_{N_2}$22, it shows one critical point for $Y_1,\dots,Y_{N_2}$23, two critical points for $Y_1,\dots,Y_{N_2}$24, a tricritical point at

$Y_1,\dots,Y_{N_2}$25

and a line of triple points for $Y_1,\dots,Y_{N_2}$26 in the distinguishable-particle formulation (Romanik et al., 1 Jul 2026). The model supports up to three density-distinct phases—gas-like, intermediate-density, and high-density—and thereby realizes both gas–liquid and liquid–liquid coexistence (Romanik et al., 1 Jul 2026).

By contrast, the earlier multiple-occupancy Curie–Weiss cell fluid with fixed interaction parameters has multiple critical points and a sequence of first-order transitions but no triple point (Kozlovskii et al., 21 Nov 2025). A triple point appears only after introducing an effective temperature-dependent attraction $Y_1,\dots,Y_{N_2}$27 that differs between phases, with

$Y_1,\dots,Y_{N_2}$28

for $Y_1,\dots,Y_{N_2}$29, and increasing exponents $Y_1,\dots,Y_{N_2}$30 (Kozlovskii et al., 21 Nov 2025). The three-phase coexistence condition then becomes

$Y_1,\dots,Y_{N_2}$31

and for $Y_1,\dots,Y_{N_2}$32, $Y_1,\dots,Y_{N_2}$33 the triple point is reported at

$Y_1,\dots,Y_{N_2}$34

with coexisting densities $Y_1,\dots,Y_{N_2}$35, $Y_1,\dots,Y_{N_2}$36, and $Y_1,\dots,Y_{N_2}$37 (Kozlovskii et al., 21 Nov 2025).

6. Response functions, entropy, and nonequilibrium extensions

Because the cell-fluid models are explicitly solvable, they permit closed-form thermodynamic response functions. In the supercritical region, the isothermal compressibility, thermal pressure coefficient, thermal expansion coefficient, and isochoric and isobaric heat capacities can be written explicitly in terms of the special functions $Y_1,\dots,Y_{N_2}$38 and the moments

$Y_1,\dots,Y_{N_2}$39

(Kozlovskii et al., 26 Nov 2025). For example, the reduced compressibility is

$Y_1,\dots,Y_{N_2}$40

while the reduced thermal expansion obeys $Y_1,\dots,Y_{N_2}$41, and the isobaric heat capacity satisfies

$Y_1,\dots,Y_{N_2}$42

(Kozlovskii et al., 26 Nov 2025). In the supercritical region these functions are smooth and single-valued, but they develop pronounced maxima near the critical region, consistent with a mean-field crossover structure (Kozlovskii et al., 26 Nov 2025).

The entropy of the cell-fluid model also admits explicit analytic form. With $Y_1,\dots,Y_{N_2}$43, the reduced entropy per particle is

$Y_1,\dots,Y_{N_2}$44

and the entropy per cell is

$Y_1,\dots,Y_{N_2}$45

(Romanik et al., 26 Oct 2025). In the multiple-occupancy regime, the entropy develops pronounced minima around integer-valued densities, which the authors suggest may be a generic feature of multiple-occupancy models (Romanik et al., 26 Oct 2025).

Nonequilibrium generalizations replace static Gibbs equilibrium by driven or dissipative dynamics. In the dissipative Curie–Weiss model, the interaction potential itself evolves stochastically with dissipation, and in the thermodynamic limit the magnetization can exhibit a stable limit cycle rather than convergence to a fixed point (Pra et al., 2013). In the symmetric noiseless reduction, the macroscopic variables satisfy

$Y_1,\dots,Y_{N_2}$46

and the threshold $Y_1,\dots,Y_{N_2}$47 separates global attraction to the origin from a regime with a unique globally attracting periodic orbit (Pra et al., 2013).

A periodically driven Curie–Weiss model with field $Y_1,\dots,Y_{N_2}$48 instead produces hysteresis and a dynamical critical temperature depending on amplitude and frequency (Fiori et al., 2024). In the thermodynamic limit, the magnetization obeys

$Y_1,\dots,Y_{N_2}$49

and the dissipated power is

$Y_1,\dots,Y_{N_2}$50

(Fiori et al., 2024). A nonequilibrium specific heat is defined from the excess heat under a slow temperature modulation,

$Y_1,\dots,Y_{N_2}$51

and it diverges at the dynamical critical temperature, unlike the equilibrium Curie–Weiss specific heat, which has only a finite discontinuity at $Y_1,\dots,Y_{N_2}$52 (Fiori et al., 2024).

The term “Curie–Weiss cell model” therefore does not designate a single canonical object. It denotes a family of mean-field constructions in which cells, populations, or occupancy boxes interact through a Curie–Weiss-type global coupling, and in which the model becomes tractable through collective variables, exact or asymptotic saddle-point reductions, and phase-diagram analysis. In spin systems this yields vector magnetizations, exchangeable mixture representations, and anomalous critical fluctuations (Kirsch et al., 2018, Barhoumi-Andréani et al., 2023, Barhoumi-Andréani et al., 23 Jul 2025). In fluid models it yields exactly solvable equations of state, sequences of first-order density transitions, multiple critical points, and, after suitable modification of the effective attraction, even triple-point formation (Kozlovskii et al., 2021, Romanik et al., 1 Jul 2026, Kozlovskii et al., 21 Nov 2025).