--- title: Translation-Invariant Splitting Gibbs Measures url: https://www.emergentmind.com/topics/translation-invariant-splitting-gibbs-measures-tisgms type: topic --- # Translation-Invariant Splitting Gibbs Measures Searching arXiv for the primary paper and closely related TISGM literature on Cayley trees. Search query: 1706.06130 Translation-invariant Gibbs measures Blum-Kapel Cayley tree Translation-invariant splitting Gibbs measures (TISGMs) are Gibbs measures on trees that are simultaneously homogeneous and Markovian in the tree sense: they arise from constant boundary laws, equivalently from translation-invariant tree-indexed Markov chains. On Cayley trees, they serve as the canonical homogeneous phases, because the DLR consistency problem reduces to nonlinear recursion on descendants, and translation invariance turns this recursion into a finite-dimensional fixed-point problem. In the literature, TISGMs are therefore the natural interface between phase transition theory, boundary-law methods, and reconstruction on trees; the Blum–Kapel model on a Cayley tree is a particularly explicit instance of this framework [1706.06130]. ## 1. Formal framework on Cayley trees A Cayley tree of order \(k\) is an infinite tree in which each vertex has degree \(k+1\). After fixing a root \(x^0\), one uses the spheres \(W_n=\{x:d(x,x^0)=n\}\), balls \(V_n=\{x:d(x,x^0)\le n\}\), and successor sets \(S(x)\) to express Gibbs specifications recursively along generations. This rooted description is not merely notational: it is the mechanism through which tree models admit boundary-law equations and splitting constructions. For nearest-neighbor models on a Cayley tree, finite-volume Gibbs measures are defined on \(V_n\) with boundary fields placed on the outer layer \(W_n\). Consistency of these finite-volume measures under marginalization is equivalent to the Dobrushin–Lanford–Ruelle condition and yields an infinite-volume Gibbs measure by Kolmogorov’s extension theorem. In the tree literature, a splitting Gibbs measure is exactly a Gibbs measure generated by such a compatible boundary-law recursion; equivalently, it is a tree-indexed Markov chain whose subtrees are conditionally independent given the parent spin [1310.6220]. A TISGM is obtained when the boundary law is constant across the tree. In practice, one starts with a site-dependent recursion \(h_x=\sum_{y\in S(x)}F(h_y)\), or an equivalent recursion for positive ratios \(z_{i,x}\), and imposes \(h_x\equiv h\) or \(z_{i,x}\equiv z_i\). The infinite-dimensional consistency problem then collapses to a fixed-point equation in finitely many variables. This reduction is the central structural feature of TISGM theory. ## 2. Boundary laws, fixed points, and the Markov-chain viewpoint The modern tree formulation of TISGMs is boundary-law based. For the ferromagnetic \(q\)-state Potts model, compatibility of the finite-volume Gibbs distributions is equivalent to a recursion of the form \[ h_x=\sum_{y\in S(x)}F(h_y,\theta), \] where \(\theta=e^{\beta J}\), \(h_x\in\mathbb{R}^{q-1}\), and \(F\) is an explicit nonlinear map. Under translation invariance this becomes \[ h=k\,F(h,\theta), \] or, after exponentiating coordinates, \[ z_i=\left(\frac{(\theta-1)z_i+\sum_{j=1}^{q-1}z_j+1}{\theta+\sum_{j=1}^{q-1}z_j}\right)^k,\qquad i=1,\dots,q-1. \] A central structural theorem states that every translation-invariant solution has a block form in which some subset of coordinates is equal to a common \(z\) and the rest are equal to \(1\); this reduces the classification to scalar equations \(z=f_m(z)\), indexed by subset size \(m\) [1310.6220]. That reduction has several consequences. First, the number of TISGMs can be counted exactly in many Potts regimes; in particular, at sufficiently low temperatures the number is \(2^q-1\), and there are \([q/2]\) critical temperatures at which the count changes [1310.6220]. Second, every TISGM comes with an explicit transition matrix \[ P_{ij}\propto z_j e^{\beta J\delta_{ij}}, \] so extremality becomes a question about reconstruction for a concrete tree-indexed Markov chain. Third, boundary conditions can be tied directly to TISGMs: for the Potts model on the binary tree, explicit classes of boundary configurations were constructed that converge to each prescribed TISGM [1504.01265]. This boundary-law interpretation also clarifies terminology. Translation invariance concerns homogeneity of the boundary law or transition kernel; splitting refers to the Markov decomposition along forward subtrees; extremality concerns whether the resulting Gibbs measure is a pure state. These notions are related but not identical. ## 3. The Blum–Kapel model as a worked example In the Blum–Kapel model studied on the Cayley tree, the spin space is \[ \Phi=\{-1,0,+1\}, \] and the Hamiltonian is \[ H(\sigma)=-J\sum_{\langle x,y\rangle\in L}\sigma(x)\sigma(y),\qquad J>0. \] Finite-volume Gibbs measures on \(V_n\) are defined with boundary fields \(h_{i,x}\) on \(W_n\), and one introduces \[ \lambda=e^{J\beta},\qquad z_{i,x}=\exp(h_{i,x}-h_{0,x}),\quad i=\pm1. \] The consistency theorem gives the boundary-law recursion \[ z_{+1,x}=\prod_{y\in S(x)}\frac{\lambda z_{+1,y}+\lambda^{-1}z_{-1,y}+1}{z_{+1,y}+z_{-1,y}+1}, \] \[ z_{-1,x}=\prod_{y\in S(x)}\frac{\lambda^{-1} z_{+1,y}+\lambda z_{-1,y}+1}{z_{+1,y}+z_{-1,y}+1}. \] A TISGM is therefore a positive constant solution \((z_1,z_2)\) of \[ z_1=\left(\frac{\lambda z_1+\lambda^{-1}z_2+1}{z_1+z_2+1}\right)^k,\qquad z_2=\left(\frac{\lambda^{-1}z_1+\lambda z_2+1}{z_1+z_2+1}\right)^k. \] Each such solution defines a translation-invariant splitting Gibbs measure, and the measure is represented explicitly by a homogeneous transition matrix of a tree-indexed Markov chain [1706.06130]. The symmetric branch \(z_1=z_2=z\) satisfies \[ z=\left(\frac{(\lambda+\lambda^{-1})z+1}{2z+1}\right)^k. \] For every \(k\ge2\) and every \(\lambda>0\), this scalar equation has a unique positive solution. Hence there is always at least one symmetric TISGM. For \(k=2\), the symmetric equation becomes a cubic and admits an explicit Cardano-type solution \(z^*(\lambda)\) [1706.06130]. ## 4. Multiplicity, phase transition, and symmetry breaking The full classification in the Blum–Kapel paper is carried out explicitly for the binary tree \(k=2\). In that case there exists \[ \lambda_{cr}\approx 2.1132163 \] such that the translation-invariant phase diagram is exactly: \[ 0<\lambda\le \lambda_{cr}\quad\Longrightarrow\quad \text{one translation-invariant Gibbs measure } \mu_0, \] \[ \lambda>\lambda_{cr}\quad\Longrightarrow\quad \text{three translation-invariant Gibbs measures } \mu_0,\mu_1,\mu_2. \] Here \(\mu_0\) is the symmetric TISGM associated with \((z^*,z^*)\), while \(\mu_1\) and \(\mu_2\) correspond to the asymmetric solutions \((z_1^{(1)},z_1^{(2)})\) and \((z_1^{(2)},z_1^{(1)})\) [1706.06130]. Using \(\lambda=e^{J/T}\), the corresponding critical temperature is \[ T_{cr}=\frac{J}{\ln\lambda_{cr}}. \] Thus the model has a unique TISGM for \(T\ge T_{cr}\) and exactly three TISGMs for \(01. \] Numerically, this holds for \[ \lambda\in(0,\lambda_1)\cup(\lambda_2,\infty), \qquad \lambda_1\approx0.336135,\quad \lambda_2\approx2.975. \] Hence \(\mu_0\) is non-extreme in those regions [1706.06130]. A sufficient condition for extremality comes from the Martinelli–Sinclair–Weitz criterion \(k\kappa\gamma<1\). In this model it reduces to the same threshold interval, yielding \[ \mu_0 \text{ is extreme for } \lambda_1<\lambda<\lambda_2. \] For the physically relevant ferromagnetic parametrization \(\lambda=e^{J/T}>1\), this means that the symmetric TISGM is extreme at intermediate and high temperatures but becomes non-extreme at sufficiently low temperature, namely for \(\lambda>\lambda_2\) [1706.06130]. Two points follow. First, multiplicity of TISGMs and extremality are distinct issues: counting translation-invariant fixed points does not by itself identify pure phases. Second, even when a TISGM is uniquely specified inside the translation-invariant class, that does not automatically settle the structure of the full Gibbs simplex. In the Blum–Kapel paper, it is further shown that for \(\lambda_{cr}<\lambda<\lambda_2\) there are at least two extreme Gibbs measures, while the individual extremality of the asymmetric TISGMs \(\mu_1,\mu_2\) remains open [1706.06130]. ## 6. Position within the broader TISGM literature The Potts model is the benchmark case. For the ferromagnetic Potts model on a Cayley tree, all TISGMs are obtained from the scalar equations \(z=f_m(z)\), and at sufficiently low temperature their number is \(2^q-1\) [1310.6220]. Their extremality can be studied by coarse-graining: fuzzy transformations map Potts TISGMs to effective two-state models, and in a temperature interval there are at least \(2^{q-1}+q\) extremal TISGMs on the binary tree [1403.5775]. The problem of which boundary conditions select which TISGM was solved explicitly for the order-two tree, again in terms of the tree recursion [1504.01265]. The relation between translation invariance and periodicity is model dependent. In the zero-field Potts model, some periodic Gibbs measures collapse to translation-invariant ones: for the \(q\)-state antiferromagnetic Potts model on the Cayley tree of order two and for the \(q\)-state ferromagnetic Potts model on the Cayley tree of order \(k\), all periodic Gibbs measures are translation-invariant [1804.08708]. This shows that TISGMs may exhaust the periodic sector in certain regimes, but not universally. Mixed-spin and coupled models broaden the TISGM landscape substantially. The mixed spin-\((1,1/2)\) Ising model on the second-order Cayley tree has three TISGMs in both the ferromagnetic and anti-ferromagnetic regimes, and the disordered TISGM has an exact extremality characterization via tree-indexed Markov chains [2201.12615]. A periodic triple mixed-spin Ising model with spins \((\tfrac12,1,\tfrac32)\) reduces to a scalar fixed-point equation and has at least three TISGMs when a derivative instability criterion is positive; on the binary tree, the plus and minus limits are themselves TISGMs [2602.12369]. Coupled Ising–Potts models can be much richer still: one paper proves at least three TISGMs in general and at least eight for the \((2,3)\)-model on the binary tree [2511.01507], while another finds at least \(2^q+1\) TISGMs at sufficiently low temperature and an exact count of \(335\) TISGMs for \(k=2\), \(q=5\) [2502.12014]. In the HC–Blume–Capel model with wand graph, there is again a sharp one-versus-three TISGM dichotomy, together with a complete extremality analysis for one symmetric measure [2603.28830]. Taken together, these results establish the encyclopedic significance of TISGMs on trees. They are the principal homogeneous Gibbs phases, they are classified by boundary-law fixed points, they encode phase transitions through multiplicity changes, and their extremality is governed by reconstruction criteria for the associated tree-indexed Markov chains. The Blum–Kapel model provides a particularly transparent example, but the same boundary-law architecture persists across Potts, mixed-spin, hard-core, and coupled-spin systems [1706.06130]. Source: https://www.emergentmind.com/topics/translation-invariant-splitting-gibbs-measures-tisgms