--- title: Fredrickson–Andersen j-Spin Model url: https://www.emergentmind.com/topics/fredrickson-andersen-j-spin-facilitated-model type: topic --- # Fredrickson–Andersen j-Spin Model The Fredrickson–Andersen \(j\)-spin facilitated model is a family of kinetically constrained spin models in which the equilibrium measure is trivial or non-interacting, while the dynamics is restricted by a facilitation rule: a site can change state only if at least \(j\) neighboring sites are in a facilitating state. In the literature the same parameter is often denoted \(f\), so the “\(j\)-spin” and “\(f\)-spin” formulations are equivalent at the level of the constraint \(j\leftrightarrow f\). Depending on convention, the microscopic variables are either Ising spins \(\sigma_i\in\{-1,+1\}\) or occupation variables \(\eta_x\in\{0,1\}\); in both representations, glassy behavior is generated by dynamic facilitation rather than by nontrivial static interactions [1009.6221]. ## 1. Microscopic definition and notational conventions In the occupancy formulation on \(\mathbb Z^d\), the configuration space is \(\Omega=\{0,1\}^{\mathbb Z^d}\), with \(\eta_x=0\) interpreted as an empty, infected, or excited site and \(\eta_x=1\) as a filled, healthy, or inactive site. A standard FA-\(j\)f dynamics is a continuous-time Markov process in which each site attempts to resample from a Bernoulli\((1-q)\) law when its Poisson clock rings, but the update is accepted only if at least \(j\) nearest neighbors are empty. The corresponding constraint indicator is \[ c_x(\eta)= \begin{cases} 1 & \text{if } \sum_{y\sim x}(1-\eta_y)\ge j,\\ 0 & \text{otherwise.} \end{cases} \] The Dirichlet form can be written as \[ \mathcal D(f)=\sum_{x\in\mathbb Z^2}\mu_q\!\left(\mathbf 1_{\{|\{y\in N_x:\eta_y=0\}|\ge j\}}\mathrm{Var}_x(f(\eta))\right), \] and the generator as \[ \mathcal L f(\eta)=\sum_{x} c_x(\eta)\bigl(\mathbb E_{\eta_x}[f(\eta)]-f(\eta)\bigr) \] in the two-dimensional presentation [2205.02297]. In the Ising-spin formulation often used on Bethe lattices and random regular graphs, one takes \(\sigma_i\in\{-1,+1\}\) with a non-interacting Hamiltonian \[ H=-\frac{1}{2}\sum_{i=1}^N \sigma_i, \] so the equilibrium up-spin probability is \[ p=\frac{1}{1+\exp(-1/T)}. \] The single-spin flip rate is Glauber- or Metropolis-like, \[ w(\sigma_i\to-\sigma_i)=\min\{1,e^{-\sigma_i/T}\}, \] but only if the facilitation constraint is satisfied. In the FA model on a Bethe lattice studied with random pinning, the rule is that the number of nearest neighbors in state \(-1\) must be greater than or equal to \(f\), with the concrete choice \(z=4\) and \(f=2\) corresponding to a two-spin facilitated model in the usual FA classification [1505.02912]. These two descriptions are equivalent up to convention. A common source of confusion is the identity of the facilitating state: some papers use empty sites or zeros, others down spins or defects. The invariant content is the same. The Hamiltonian is deliberately simple, and the slow dynamics arises because mobility can propagate only through neighborhoods that already contain enough facilitating sites. This separation between trivial thermodynamics and nontrivial kinetics is one of the defining features of the FA family [2205.02297]. ## 2. Bethe-lattice formulation and bootstrap-percolation structure On locally tree-like graphs, especially random regular graphs and Bethe lattices, the FA \(j\)-spin model admits an exact description in terms of bootstrap percolation or \(k\)-core–type recursion relations. This is the canonical mean-field setting for the model. On a 4-regular random graph, for example, the glass order parameter can be taken as the fraction \(\Phi\) of frozen spins or, dynamically, as the long-time limit of the persistence \(\phi(t)\), where \(\phi(t)\) is the probability that a spin has never flipped up to time \(t\) [1405.0195]. The bootstrap-percolation interpretation is central. For uniform facilitation on a Bethe lattice, dynamical arrest corresponds to the appearance of an infinite cluster of frozen spins. For \(z-1>f>1\), the transition is of bootstrap-percolation type and is hybrid: the frozen fraction jumps discontinuously, but critical fluctuations diverge. For \(f=z-1\) or \(f=z\), the transition is continuous and equivalent to standard percolation. In facilitated mixtures on a \(z=4\) Bethe lattice, the distribution \[ P(f_i)=(1-q)\delta_{f_i,2}+(q-r)\delta_{f_i,3}+r\delta_{f_i,4} \] interpolates between these regimes. The resulting frozen fraction \(\Phi\) is obtained exactly from a recursive solution involving rooted-tree probabilities \(x\) and \(y\), and the critical line is \[ T_c(q)= \begin{cases} \frac{1}{\log(8-12q)}, & 0\le q\le \frac12,\\[1ex] -\frac{1}{\log(3q-1)}, & \frac120\) and \(\beta=2\) when \(r=0\) [1009.6221]. A closely related dynamical viewpoint is obtained from exact equations for occupation correlations on the Bethe lattice. In the oriented FA model on a \(k\)-ary rooted tree with facilitation \(f\), the long-time blocked fraction \(q\) satisfies \[ q=pQ(q),\qquad Q(q)=\sum_{i=0}^{f-1}\binom{k}{i}q^{k-i}(1-q)^i. \] This is simultaneously the bootstrap-percolation fixed-point equation and the equation that determines the plateau of the dynamical occupation correlation. The same structure governs the asymptotic memory kernel of the model, which makes the bootstrap-percolation mapping not merely geometric but dynamically exact in the scaling regime near the transition [1902.06537]. The Bethe-lattice formulation therefore supplies more than a mean-field caricature. It identifies the precise blocked structures, yields exact recursion relations for the nonergodicity parameter, and separates two distinct geometries of frozen clusters: compact bootstrap-percolation clusters in the discontinuous regime and fractal standard-percolation clusters in the continuous regime. This distinction reappears in the critical exponents and in the character of relaxation. ## 3. Mean-field criticality, memory kernels, and higher-order singularities Near the Bethe-lattice dynamical transition, the FA \(j\)-spin model displays the full mode-coupling-theory scaling scenario. For the occupation correlation \[ \phi(t)=\langle n_s(0)n_s(t)\rangle, \] the Mori–Zwanzig equation takes the form \[ \tau_0\dot\phi(t)+\phi(t)+\int_0^t \mathrm dt'\, m(t-t')\dot\phi(t')=0. \] In the scaling regime near the critical point, the memory kernel becomes asymptotically exact, and the deviation \(G_\sigma(t)=\phi(t)-q_c\) from the plateau satisfies the standard MCT scaling equation. The exponents \(a\) and \(b\) obey \[ \lambda=\frac{\Gamma^2(1-a)}{\Gamma(1-2a)}=\frac{\Gamma^2(1+b)}{\Gamma(1+2b)}, \] and the characteristic times diverge as \[ t_\sigma\sim \sigma^{-1/2a},\qquad t_\sigma'\sim \sigma^{-\gamma},\qquad \gamma=\frac{1}{2a}+\frac{1}{2b}. \] For the case \(k=3\), \(f=2\), the exponent parameter is \(\lambda=839/864\), with \(a\approx0.121\), \(b\approx0.147\), and \(\gamma\approx7.53\) [1902.06537]. Random pinning reveals a higher-order version of the same structure. On a 4-regular Bethe lattice with facilitation \(f=2\), pinning a fraction \(c\) of spins from an equilibrium configuration generates a line of dynamical transitions \(T_d(c)\) in the \(T\)–\(c\) plane. For \(ct)\) has exponential tails with the corresponding characteristic scales: \[ F_0(t)\le \begin{cases} e^{-Cq^3 t}, & d=1,\\ e^{-Cq^2\log(1/q)t}, & d=2,\\ e^{-Cq^2 t}, & d\ge3, \end{cases} \] while \(E_\mu(\tau_0)\) satisfies the matching lower bounds \(Cq^{-3}\), \(Cq^{-2}\log(1/q)\), and \(Cq^{-2}\) in dimensions \(1\), \(2\), and \(3\), respectively [2005.13327]. For FA-2f, the asymptotics are much more singular. In \(\mathbb Z^d\), the infection time of the origin in the stationary process obeys \[ \tau_0=\exp\!\Big(\frac{d\,\lambda(d,2)+o(1)}{q^{1/(d-1)}}\Big) \qquad \text{w.h.p. as } q\downarrow0, \] and the same leading asymptotics hold for \(\mathbb E_{\mu_q}[\tau_0]\). In \(d=2\), since \(\lambda(2,2)=\pi^2/18\), this becomes \[ \mathbb E_{\mu_q}[\tau_0] = \exp\!\Big(\frac{\pi^2}{9q}\bigl(1+o(1)\bigr)\Big), \] with corresponding high-probability bounds for \(\tau_0\). This was the first sharp threshold result for a critical KCM and explicitly tied the FA-2f timescale to the sharp threshold constant for 2-neighbor bootstrap percolation [2012.02557]. The two-dimensional overview reformulates this result as \[ \log T_{\mathrm{rel}}(q)=\frac{\pi^2}{9q}\bigl(1+o(1)\bigr) \] for stationary FA-2f on \(\mathbb Z^2\), and interprets the mechanism in terms of rare mobile droplets whose probability is \(\exp(-\pi^2/(9q))\). This settled several long-standing controversies in the physics literature over the correct FA-2f scaling regime [2205.02297]. A persistent misconception is that all FA models should share a single low-temperature universality class. The rigorous results show otherwise. FA-1f has polynomial scales and mobile single-defect dynamics; FA-2f has stretched-exponential scales governed by critical droplets and bootstrap-percolation constants; higher facilitation in low dimension can produce genuine frozen configurations with positive probability. The facilitation threshold is therefore not a perturbative parameter but a structural classifier of the dynamics. ## 5. Front propagation, transport, and trajectory-space phase behavior A particularly detailed picture is available for the nonequilibrium front in one-dimensional FA-1f. Starting from a configuration that is fully occupied on the left half-line and has a single zero at the origin, one tracks the leftmost zero \(X(\sigma_t)\). For \(q>\bar q\), where \[ \bar q=\frac{2\lambda_c(\mathbb Z)}{1+2\lambda_c(\mathbb Z)}, \] the process seen from the front has a unique invariant measure \(\nu\), and there exist \(s=s(q)\) and \(v=v(q)\) such that \[ \frac{X(\sigma_t)}{t}\xrightarrow[t\to\infty]{} v \quad\text{a.s.}, \qquad \frac{X(\sigma_t)-vt}{\sqrt t}\xrightarrow[t\to\infty]{d}\mathcal N(0,s^2), \] with \[ v=p\cdot \nu[\tilde\sigma(1)=0]-q<0. \] Thus the front moves ballistically and has Gaussian fluctuations at scale \(\sqrt t\) [1803.08761]. This front motion controls finite-volume relaxation. On an interval \([1,L]\) with empty boundary conditions at 0 and \(L+1\), FA-1f exhibits cutoff at time \[ \frac{L}{2v} \] with window \(O(\sqrt L)\) for a positive constant \(v\). More precisely, for \(q>\bar q\), the worst-case total variation distance satisfies \[ d\Bigl(\frac{L}{2v}-\alpha(\varepsilon)\sqrt L\Bigr)\ge 1-\varepsilon, \qquad d\Bigl(\frac{L}{2v}+\beta(\varepsilon)\sqrt L\Bigr)\le \varepsilon \] for suitable \(\alpha(\varepsilon,q),\beta(\varepsilon,q)\) and all large \(L\) [2103.00019]. Transport by a tagged particle in FA environments has a similarly sharp characterization. For noncooperative KCM in the \(k\)-zeros class, the tracer diffusion coefficient satisfies \[ c\,q^{k+1}\|u\|_2^2\le u\cdot D u\le C\,q^{k+1}\|u\|_2^2. \] For FA-1f, which corresponds to \(k=1\) in this class, this gives \[ D(q)\asymp q^2. \] This rigorously confirmed the \(D\sim q^2\) prediction made earlier for the one-spin facilitated FA model [1306.6500]. The FA family also supports trajectory-space phase transitions. In the one-dimensional FA model, the large deviations of the dynamical activity \(K(\tau)\) define a scaled cumulant generating function \({\cal G}(s)\) with a first-order singularity at \(s=0\), corresponding to coexistence between active and inactive dynamical phases. In a two-dimensional spin-model representation of trajectories, the susceptibility at coexistence diverges as \[ \chi_{L,\infty}(s^*)\sim e^{\Gamma_{\rm s}L} \] in the long-time dynamical limit \(Y\gg L\), while for isotropic aspect ratio \(Y\sim L\) it follows standard first-order finite-size scaling \(\chi^*\approx LY(\Delta k)^2/4\) [1912.02465]. These results indicate that FA dynamics supports several distinct macroscopic structures: moving fronts in real space, tracer subdynamics in a fluctuating constrained environment, and active–inactive coexistence in trajectory space. They are different manifestations of the same facilitation principle. ## 6. Random environments, pinned variants, and broader connections Quenched heterogeneity modifies FA \(j\)-spin dynamics in a way that sharply separates local and global timescales. In the mixed-threshold FA model on \(\mathbb Z^2\), each site is assigned a random threshold \(\omega_x\in\{1,2\}\) with \(\nu(\omega_x=1)=\pi\) and \(\nu(\omega_x=2)=1-\pi\), so the system is a quenched mixture of FA-1f and FA-2f constraints. For the associated bootstrap percolation, the emptying time of the origin has deterministic scaling \[ \tau_0\sim q^{-1/2} \] in two dimensions, and more generally \(q^{-1/d}\) in \(\mathbb Z^d\). For the stochastic KCM, by contrast, the local emptying time has random exponents: \[ P_\mu[\tau_0\ge q^{-\overline\alpha}]\to0,\qquad P_\mu[\tau_0\le q^{-\underline\alpha}]\to0, \] with \(\underline\alpha,\overline\alpha\) depending on the disorder realization. At the same time, the global relaxation time is dominated by rare FA-2f-like islands and satisfies \[ T_{\mathrm{rel}}\ge \exp(c/q) \] almost surely [1812.00774]. A different cross-disciplinary extension arises in noisy quantum many-body dynamics. In the strong-noise limit of an exchange-coupled qubit chain without conserved spin component, the second moments of operators map onto an FA-1f process with binary variables \(n_j\in\{0,1\}\). The effective local rates are \[ \Gamma_0=\frac{4J^2}{3g},\qquad \Gamma_1=\frac{4J^2}{g}, \] and the stationary excitation probability is \(p_1=3/4\). In this representation, operator spreading becomes front propagation in FA-1f. In one dimension, the asymptotic front fluctuations are Gaussian, and numerical evidence in two dimensions supports the conjecture that the long-time fluctuations belong to the Kardar–Parisi–Zhang universality class [1806.01723]. Random pinning, mixed facilitation, and quantum mappings all reinforce the same conclusion: the FA \(j\)-spin model is not a narrowly defined toy process but a flexible dynamical framework. It supports mean-field glass transitions with \(A_2\) and \(A_3\) structure, sharp finite-dimensional threshold laws, trajectory-space first-order transitions, and mappings to problems outside classical glass theory. A plausible implication is that dynamic facilitation organizes a broader universality structure than any single microscopic realization, provided the local rule still enforces mobility through neighboring mobile sites. Source: https://www.emergentmind.com/topics/fredrickson-andersen-j-spin-facilitated-model