---
title: Active Ising Model
url: https://www.emergentmind.com/topics/active-ising-model
type: topic
---

# Active Ising Model

Searching arXiv for recent and foundational Active Ising Model papers to ground the article.
The Active Ising Model (AIM) is a nonequilibrium lattice model for collective motion in which particles carrying Ising spins combine local ferromagnetic alignment with spin-dependent biased hopping. In its standard two-dimensional formulation, each particle has spin $\sigma\in\{+1,-1\}$, multiple occupancy is allowed on each site, density is conserved, and the local magnetization is non-conserved because spin flips change it. The AIM is widely used as a minimal flocking model with discrete rotational symmetry, and its central phenomenology is that collective motion corresponds to a liquid–gas phase transition between a disordered gas and an ordered polar liquid, often with a coexistence regime consisting of a traveling dense ordered band in a dilute disordered background [1506.05749]. Recent work has further clarified its ordering kinetics, showing that in $2d$ the coarsening length scale follows the Lifshitz–Cahn–Allen law $R(t)\sim t^{1/2}$ both in the coexistence region and in the ordered liquid, despite the coupling of a conserved density to a non-conserved magnetization [2401.13471].

## 1. Microscopic definition and dynamical rules

In the standard AIM, particles occupy a $2d$ square lattice with periodic boundary conditions and no on-site exclusion, so a site $i$ can contain arbitrary numbers $n_i^+$ and $n_i^-$ of spins $\sigma=+1$ and $\sigma=-1$ respectively. The local density and magnetization are
$$
\rho_i=n_i^+ + n_i^-,
\qquad
m_i=n_i^+ - n_i^-,
$$
with $-\rho_i \le m_i \le \rho_i$ [2401.13471]. The density field is conserved by the stochastic dynamics, whereas the magnetization is not, since spin flips locally change $m_i$ without conserving it [2401.13471].

Alignment is implemented through on-site spin-flip dynamics. In the formulation used for the kinetic study of ordering, the local Hamiltonian is
$$
H_i = -\frac{J}{2}\left(\frac{m_i^2}{\rho_i}-1\right),
$$
which yields an energy change
$$
\Delta H_i = \frac{2J}{\rho_i}(\sigma m_i - 1)
$$
upon flipping a spin $\sigma$ at site $i$ [2401.13471]. With $J=1$ and $\gamma=1$ in the simulations, the spin-flip rate is
$$
W_{\rm flip}(\sigma\to-\sigma)=\exp\!\left[-\frac{2\beta}{\rho_i}(\sigma m_i-1)\right],
$$
where $\beta=1/T$ controls the noise strength and $T$ is an athermal control parameter [2401.13471]. A closely related earlier formulation uses
$$
W(s\to -s)=\gamma \exp(-s\beta m_i/\rho_i),
$$
which satisfies detailed balance with respect to a sum of fully connected on-site Ising Hamiltonians [1506.05749].

Self-propulsion is encoded as biased hopping. In the standard $2d$ AIM, hopping is biased along the $x$ axis and unbiased along $y$:
$$
W_{\rm hop}(\sigma,p)=D[1+\sigma\epsilon (p\!\cdot\! e_x)],
$$
with $p\in\{\pm e_x,\pm e_y\}$, baseline hop rate $D$, and bias $\epsilon\in[0,1]$ [2401.13471]. Thus, along $\pm x$, the rates are $D(1\pm \sigma\epsilon)$, while along $\pm y$ they are simply $D$ [2401.13471, 1506.05749]. The mean drift is $v=2D\epsilon$ along the propulsion axis [2401.13471]. One Monte Carlo step is defined by choosing a random particle, attempting a spin flip with probability $W_{\rm flip}\Delta t$, a hop with probability $W_{\rm hop}\Delta t$, and doing nothing otherwise, with
$$
\Delta t = [4D+\exp(2\beta)]^{-1}
$$
to keep probabilities bounded by unity [2401.13471].

A key structural point is that AIM dynamics remain out of equilibrium even at $\epsilon=0$. In the original $2d$ analysis, a Kolmogorov loop of four configurations gives unequal products of transition rates in the two directions, showing that detailed balance is violated even without drift [1506.05749]. This nonequilibrium character becomes important in discussions of criticality and universality [2306.10791].

## 2. Phases, phase separation, and canonical interpretation

For fixed noise and self-propulsion, the AIM exhibits three nonequilibrium steady states: a disordered gas, an ordered polar liquid, and a phase-separated liquid–gas coexistence regime [1506.05749, 2401.13471]. In the disordered gas, the system is spatially homogeneous and $\langle m_i\rangle \approx 0$; in the polar liquid, it is homogeneous with $|\langle m_i\rangle|>0$ and all particles drift coherently; in coexistence, an ordered dense liquid band propagates through a dilute disordered gas [1506.05749].

A defining result of the original AIM study is that the transition to collective motion is a bona fide liquid–gas phase separation in the canonical ensemble [1506.05749]. At coexistence, the densities of the gas and liquid regions are fixed at $\rho_g(T,v)$ and $\rho_\ell(T,v)$, and only the liquid fraction changes with the mean density $\rho_0$. The lever rule holds:
$$
\Phi = \frac{\rho_0-\rho_g}{\rho_\ell-\rho_g}.
$$
This is precisely the canonical structure of an equilibrium liquid–gas transition, except that the gas has $m=0$ while the liquid has $m\neq 0$ [1506.05749].

The phase-diagram topology reflects the discrete $Z_2$ symmetry of the model. In the $(T,\rho_0)$ plane at fixed $v>0$, coexistence is bounded by two binodals separating gas, coexistence, and liquid. As $\rho_0\to\infty$, both binodals asymptote to $T_c=1$ in the 2015 parametrization [1506.05749]. Because the disordered gas and ordered liquid have different symmetries, there is no supercritical region: one cannot continuously connect them without crossing a transition line [1506.05749]. In the $(v,\rho_0)$ plane at fixed $T<1$, the coexistence lines merge at $v=0$, producing a finite-density critical point belonging to the $2d$ Ising universality class [1506.05749].

The later numerical coarsening study uses a slightly different flip-rate parametrization and reports a critical temperature $T_c\simeq 2$ for the chosen rates, explicitly noting that this value is about twice that of Solon and Tailleur 2015 because of the different parametrization [2401.13471]. This is not a contradiction; it reflects a model-definition difference.

## 3. Hydrodynamic descriptions and continuum theories

The coarse-grained AIM is typically formulated in terms of a conserved density $\rho(\mathbf{x},t)$ and a scalar magnetization $m(\mathbf{x},t)$. A refined mean-field hydrodynamic theory derived for the AIM has the form
$$
\partial_t \rho = D \nabla^2 \rho - v \partial_x m,
$$
$$
\partial_t m = D \nabla^2 m - v \partial_x \rho + 2(2\beta-1-r/\rho)m - \alpha m^3/\rho^2,
$$
with $\alpha = 4\beta^2(1-2\beta/3)$, $r>0$, and $v=2D\epsilon$ [2401.13471]. These equations incorporate both diffusion and activity: each field diffuses, while advection couples them only along the propulsion direction [2401.13471]. They reproduce phase separation and the observed coarsening kinetics [2401.13471].

An earlier refined mean-field model derived from the 2015 AIM gives
$$
\partial_t \rho = D\Delta \rho - v\partial_x m,
$$
$$
\partial_t m = D\Delta m - v\partial_x \rho + 2(\beta-1-r/\rho)m - \alpha m^3/\rho^2,
$$
with $\alpha=\beta^2(1-\beta/3)$ and $\beta_t(\rho)=1+r/\rho$ [1506.05749]. In that framework, linear stability analysis yields a gas spinodal
$$
\phi_g(\beta)=\frac{r}{\beta-1}
$$
and an ordered spinodal $\phi_\ell$ given by Eq. (B10) in the paper [1506.05749]. Between these spinodals, the homogeneous solutions are unstable and the PDEs develop traveling-band solutions [1506.05749].

Traveling-wave analysis captures the phase-separated profiles. Using the comoving coordinate $z=x-ct$, the hydrodynamic equations reduce to coupled ODEs for $\rho(z)$ and $m(z)$ [1506.05749]. Near $\beta\to 1$, the dense-band profiles are well approximated by a tanh form
$$
m(z)=\frac{m_\ell}{2}[1+\tanh(kz)],
$$
with asymmetry between the ascending and descending fronts at finite activity [1506.05749]. The same theory predicts coexistence densities near the critical regime:
$$
\rho_g=\phi_g-\frac{4r}{9\alpha},
\qquad
\rho_\ell=\phi_g+\frac{8r}{9\alpha},
\qquad
m_\ell=\frac{4r}{3\alpha},
$$
and a band speed
$$
c=v+\frac{2D(\beta-1)^2}{3v\alpha}
$$
close to $\beta\to 1$ [1506.05749]. Microscopic simulations confirm that $c\to v$ as $\beta\to 1$ and that front asymmetry grows away from the critical regime [1506.05749].

A more systematic route to continuum AIM theories uses Doi–Peliti field theory. This approach has been developed for several AIM variants and recovers deterministic hydrodynamics while also deriving fluctuating noise terms [2306.10791]. For AIM1, the hydrodynamic equations can be written as
$$
\partial_t m=D\nabla^2 m-v\partial_{\hat x}\rho-2F(m,\rho),
$$
$$
\partial_t \rho=D\nabla^2\rho-v\partial_{\hat x}m,
$$
with a nonpolynomial alignment term $F(m,\rho)$ inherited from the microscopic rates [2306.10791]. For AIM2, the same structure holds but with
$$
F(m,\rho)=m\left(\gamma + \tau \frac{m^2-\rho^2}{8}\right),
$$
which makes clear that purely pairwise local alignment does not contribute to the deterministic drift in the hydrodynamic limit [2306.10791].

## 4. Ordering kinetics and coarsening laws

A central recent result is that the $2d$ AIM exhibits Lifshitz–Cahn–Allen coarsening, with characteristic length scale
$$
R(t)\sim t^{1/2},
$$
for quenches both into the liquid–gas coexistence region and into the ordered liquid [2401.13471]. This was established through microscopic Monte Carlo simulations and independently by solving the refined hydrodynamic equations [2401.13471].

The ordering study considers quenches from a random disordered configuration at $T\to \infty$ into $T=0.9$ with $\epsilon=1$, using representative densities $\rho_0=3$ inside the coexistence region and $\rho_0=10$ deep in the ordered liquid [2401.13471]. Equal-time two-point correlations of density and magnetization are measured,
$$
C_{\rho\rho}(r,t)=\frac{1}{L^2}\sum_i \langle \Delta \rho_{i,t}\Delta \rho_{i+r,t}\rangle,
$$
$$
C_{mm}(r,t)=\frac{1}{L^2}\sum_i \langle \Delta m_{i,t}\Delta m_{i+r,t}\rangle,
$$
and the characteristic scale $R(t)$ is defined by the condition
$$
C(r=R,t)=0.2\,C(0,t),
$$
with results insensitive to the specific fraction [2401.13471]. Both density and magnetization yield the same asymptotic growth law [2401.13471].

The observed exponent is robust with respect to noise and self-propulsion. Varying $\beta=0.9,1.1,1.4$ or $\epsilon=0,0.5,1$ does not change the asymptotic exponent; only prefactors and the time needed to reach the nonequilibrium steady state are affected [2401.13471]. Quantitatively, the effective exponent
$$
Z_{\rm eff}=d\ln R/d\ln t
$$
plateaus at late times in the range $0.45\lesssim Z_{\rm eff}\lesssim 0.52$ across quench regimes and parameters [2401.13471].

The physical mechanism differs from passive scalar coarsening in detail but not in asymptotic scaling. For $\epsilon=0$, the magnetization field undergoes curvature-driven coarsening and the density follows diffusively, leading naturally to $R\sim t^{1/2}$ [2401.13471]. For $\epsilon>0$, self-propelled clusters move along $x$, but they can merge only if they spread transversely. The study therefore identifies transverse diffusion along $y$ as the dominant coarsening mechanism in the active case [2401.13471]. When the transverse diffusion rate $D_\perp$ is reduced to zero, growth arrests into narrow horizontal stripes; finite $D_\perp\ge 0.2$ restores diffusive coarsening with the same LCA exponent [2401.13471]. This establishes that the asymptotic AIM kinetics are diffusion-controlled even though the model couples conserved and non-conserved fields [2401.13471].

Hydrodynamic numerics corroborate this interpretation. In an $x$-independent limit corresponding to a thin horizontal stripe, the PDEs reduce to diffusion-reaction equations in $y$ only, and the stripe broadens with width $\sim \sqrt{t}$ [2401.13471]. This provides an explicit continuum mechanism for the observed growth law.

## 5. Morphology, scaling functions, and structure factors

The AIM exhibits dynamical scaling during coarsening. The equal-time correlation function satisfies
$$
C(r,t)=f(r/R(t)),
$$
which indicates statistically self-similar morphologies as domains grow [2401.13471]. The small-$r$ behavior has a cusp
$$
1-C(r,t)\sim r^\alpha,
$$
with $\alpha\approx 1$, consistent with sharp interfaces [2401.13471].

The structure factor,
$$
S(k,t)=\int d^d r\, e^{-i k\cdot r} C(r,t),
$$
scales as
$$
S(k,t)=R(t)^d g(kR(t))
$$
and obeys Porod’s law in $d=2$,
$$
S(k)\sim k^{-(d+1)}=k^{-3},
$$
at large $k$ [2401.13471]. Numerical data show excellent collapse of $S(k,t)R(t)^{-2}$ versus $kR$ and clear $k^{-3}$ tails, implying compact domains with smooth boundaries [2401.13471].

This morphology sharply distinguishes the AIM from Vicsek-type flocking models. In the comparison made in the coarsening study, the Vicsek model shows distinct growth scales for density and orientational order, with
$$
R_\rho\sim t^{0.25}, \qquad R_m\sim t^{0.83},
$$
and non-Porod behavior associated with fractal clusters and giant number fluctuations [2401.13471]. By contrast, the AIM yields compact domains, normal density fluctuations in homogeneous phases, and identical growth exponents for density and magnetization [1506.05749, 2401.13471].

The origin of this difference lies in the hydrodynamics. In the AIM, both $\rho$ and $m$ possess explicit diffusion terms, while advection couples them only through $-v\partial_x m$ and $-v\partial_x \rho$ [2401.13471]. In Vicsek/Toner–Tu hydrodynamics, the density lacks a diffusive $D\nabla^2 \rho$ term and the order field contains an advective nonlinearity $(m\cdot\nabla)m$, which qualitatively changes coarsening and domain geometry [2401.13471]. This suggests that diffusion dominance, rather than activity per se, governs the AIM morphology.

## 6. Critical behavior, universality, and theoretical extensions

At zero self-propulsion, the AIM reaches a continuous critical point. In the original $2d$ model, Binder-cumulant crossings and finite-size scaling at $\beta=1.9$, $D=1$ give a critical density
$$
\rho^*=2.798\pm 0.002
$$
and a Binder cumulant $G\approx 0.61$, consistent with the $2d$ Ising universality class [1506.05749]. The finite-size scaling collapses use the standard Ising exponents $\beta=1/8$, $\gamma=7/4$, and $\nu=1$ [1506.05749].

However, the field-theoretic status of the $\epsilon=0$ AIM is more subtle than the phrase “Ising universality” might suggest. A Doi–Peliti analysis shows that AIM variants without self-propulsion remain out of equilibrium and, for the models studied there, lie outside equilibrium Model C because the density current lacks the relevant $\nabla(m^2)$ coupling characteristic of Model C [2306.10791]. In the coarse-grained zero-propulsion theory,
$$
\partial_t m=D\nabla^2 m-a m-b m^3-g\,\delta\rho\,m+\sqrt{2}\eta,
$$
$$
\partial_t \delta\rho = -\nabla\cdot \mathbf{J}, \qquad \mathbf{J}=-D\nabla\delta\rho+\sqrt{2D}\,\boldsymbol\zeta,
$$
the density dynamics is independent of the spin state at $v=0$ [2306.10791]. This structural feature excludes the equilibrium Model C current and implies a distinct nonequilibrium critical theory for those AIM variants [2306.10791]. A plausible implication is that equilibrium-looking exponent estimates at accessible scales need not imply equilibrium universality at the field-theoretic level.

For $v\neq 0$, field theory confirms that AIMs generically exhibit first-order flocking transitions with coexistence and banding, except in the special case of purely local pairwise alignment [2306.10791]. In AIM2, the deterministic drift contains no dependence on the pairwise alignment rate $\lambda$, so pairwise local alignment alone cannot sustain flocking in the hydrodynamic limit; any ordering in finite systems is noise-driven and vanishes as system size grows [2306.10791]. This result clarifies which microscopic ingredients are relevant for the AIM phenomenology.

More recent renormalization-group work has explored active Ising criticality beyond the canonical AIM. A density-impeded active Ising theory with conserved density reveals six fixed points and three new universality classes, with a generic fixed point that supersedes Wilson–Fisher criticality once the nonlinear coupling between order parameter and density is retained [2507.06068]. In that theory, the coarse-grained equations are
$$
\partial_t \rho = -\gamma \partial_x \phi + K \nabla_\perp^2 \rho,
$$
$$
\partial_t \phi + \frac{\lambda}{2}\partial_x \phi^2
= -\kappa_1 \partial_x \rho - \kappa_2 \partial_x(\rho^2)
+ \mu_x \partial_x^2 \phi + \mu_\perp \nabla_\perp^2 \phi
+ (-\alpha_0-\alpha_1\rho-\alpha_2\rho^2-\beta \phi^2)\phi + f_x,
$$
and the one-loop flow yields a generic fixed point with coordinates
$$
g^*=\epsilon[2/9,\,0,\,0,\,2/3]
$$
and exponents
$$
z=2,\qquad \zeta=1,\qquad \chi_\phi=\chi_\rho=-1+\epsilon/2,\qquad y_{\alpha_0}=2-\epsilon/3
$$
[2507.06068]. Although this model is not identical to the canonical lattice AIM, it shows how coupling to a conserved density soft mode can qualitatively change universality.

A different extension, the Active Malthusian Ising Model, replaces the conserved density sector by birth–death dynamics and exhibits two distinct Lifshitz points, one longitudinal and one transverse [2511.02566]. This is no longer the standard AIM, but it highlights how modifying conservation laws changes the critical landscape. The contrast drawn there is explicit: the conserved AIM and the Malthusian active Ising theory belong to different large-scale critical classes [2511.02566].

Quantum generalizations have also appeared. A one-dimensional bosonic active Ising model formulated as a Lindblad open system retains biased hopping and local alignment in a quantum setting, with Bose enhancement factors strengthening flocking and aster formation [2606.18091]. This suggests that “active Ising model” has become a broader family of discrete-symmetry flocking models, while the classical conserved-density lattice AIM remains the canonical reference point.

## 7. Methods, limitations, and open problems

The modern numerical study of AIM ordering kinetics uses microscopic Monte Carlo simulations on systems of size $L=400$ up to $t=10^5$ Monte Carlo steps, with averages over at least $300$ independent realizations [2401.13471]. The hydrodynamic PDEs are solved with an FTCS finite-difference scheme using $\Delta x=1$, $\Delta t=10^{-3}$, $L=400$, $t_{\max}=5\times 10^6$, periodic boundary conditions, and averages over $25$ runs [2401.13471]. Both microscopic and continuum approaches recover the same growth exponent, strengthening the claim of asymptotic robustness [2401.13471].

Several caveats remain. First, finite-size and finite-time effects are significant: scaling collapse and effective-exponent plateaus emerge only after $t\gtrsim 10^2$, while at later times $R(t)$ saturates as the system approaches the final single-band or polar-liquid steady state [2401.13471]. Second, the coarsening is anisotropic at intermediate times because advection acts only along $x$ and transverse diffusion controls mergers across $y$ [2401.13471]. Third, the small-$D_\perp$ regime is not fully understood: growth arrests for $D_\perp\le 0.1$, producing effectively decoupled one-dimensional rings with alternating magnetization, and the crossover scaling as $D_\perp\to 0$ remains open [2401.13471].

The field-theoretic literature also emphasizes model dependence. The symmetry that excludes the Model C current at $v=0$ is specific to the AIM variants studied in the Doi–Peliti analysis and need not apply to all AIM generalizations [2306.10791]. Likewise, the new critical universality classes found in the density-impeded theory require a specific hydrodynamic mechanism in which increased density suppresses collective motion [2507.06068]. These developments suggest that the term “Active Ising Model” now names a broader theoretical family rather than a single unique universality class.

Open directions explicitly identified in the coarsening work include three-dimensional AIM ordering kinetics, stronger activity or nonlocal interactions, disorder effects such as pinning and logarithmic corrections, and analytical derivations of the growth law from interface equations [2401.13471]. A plausible broader implication is that the AIM has shifted from being only a minimal flocking model to becoming a testbed for nonequilibrium phase separation, anisotropic coarsening, and critical phenomena involving coupled conserved and non-conserved fields.

Source: https://www.emergentmind.com/topics/active-ising-model