Hadwiger Models: Low-Temperature Behavior in a Natural Extension of the Ising Model

All isometrically invariant Markov (strictly local) fields on binary assignments are induced by energy functions that can be represented as linear combinations of area, perimeter, and Euler characteristic. This class of model includes the Ising model, both ferro- and antiferro-magnetic, with and without a field, as well as the “triplet” Ising model We determine the low-temperature behavior for this class of model, and construct a phase diagram of that behavior. In particular, we identify regions with three geometric phases, regions with a single unique phase, and coexistence lines between them.

💡 Research Summary

The paper introduces and thoroughly investigates a broad class of two‑dimensional binary spin models, called “Hadwiger models,” which are defined on the hexagonal lattice. The starting point is Hadwiger’s theorem: any isometry‑invariant, local (Markov) functional on subsets of the plane can be written as a linear combination of three geometric quantities – area (A), perimeter (P), and Euler characteristic (χ). Translating this into statistical mechanics, the Hamiltonian of any such model can be expressed as
H(σ)= x·χ(σ)+p·P(σ)+a·A(σ) ,
where σ assigns spins ±1 to the faces of the dual lattice. By mapping spin configurations to poly‑convex subsets of the plane, the three terms acquire a clear geometric meaning: the perimeter is unchanged under a global spin flip, while area and Euler characteristic change sign.

On the hexagonal lattice each vertex is incident to three faces, so the contribution of a vertex depends only on how many of those faces carry spin +1. The authors classify the four possible local environments as E (0 occupied faces), C (1 occupied), H (2 occupied), and F (3 occupied). Using the geometric definitions they compute the vertex energies in terms of the parameters (x,p,a):
e_C = x/6 + p + a/6,
e_H = –x/6 + p + a/3,
e_F = a/2,
while the empty state (E) has zero energy. Conversely, any choice of the three vertex energies uniquely determines (x,p,a). This re‑parameterisation is crucial because at low temperature the global ground state is dictated solely by which vertex type has the smallest energy.

The space of all models is three‑dimensional; scaling all three parameters corresponds to a temperature change, so the set of distinct models can be represented as a sphere. By stereographically projecting this sphere onto a plane the authors obtain a two‑dimensional phase diagram (Fig. 4). The diagram is divided into four large regions (E, C, H, F) where a single vertex type is energetically optimal, and into a network of “degeneracy lines” where two vertex types share the minimal energy. In the H‑region there are three symmetry‑related ground states (one for each sub‑lattice), giving rise to three distinct pure phases; the C‑region is the spin‑flipped counterpart. The E‑region corresponds to the all‑negative (ferromagnetic) ground state, while the F‑region corresponds to the all‑positive (anti‑ferromagnetic) ground state.

To analyze the Gibbs measures at low temperature the authors combine several rigorous tools:

1. Peierls condition – In any region where a single vertex type is strictly minimal, the energy cost of flipping a finite cluster of spins grows linearly with its perimeter. This guarantees exponential suppression of excitations.

2. Dobrushin–Shlosman theorem – In two dimensions, any translation‑invariant (or periodic) Gibbs state of a Peierls model is itself translation‑invariant (or periodic). Thus the possible Gibbs states are highly constrained.

3. Pirogov–Sinai theory – Given the Peierls condition, one can prove that for sufficiently low temperature each translation‑invariant Gibbs state is a convex combination of a finite set of extremal “pure phases,” each dominated by a particular ground configuration. Zahradník’s extension of this theory allows the authors to locate precisely the coexistence curves where two distinct pure phases have comparable free energy.

4. Disagreement percolation and reflection positivity – On the degeneracy lines the Peierls condition fails, so Pirogov–Sinai cannot be applied directly. The authors use disagreement percolation to show that, for example along the E–C line with e_F ≥ e_H, there is a unique Gibbs state and no configuration dominates at any temperature. Reflection positivity yields exponential decay estimates for the probability of seeing the non‑dominant vertex type.

The main rigorous results are:

• Theorem 3 – When H or C is the unique lowest‑energy vertex, there exists a temperature T₀ such that for all T < T₀ the model possesses exactly three extremal Gibbs states (the three sub‑lattice ground states). Every Gibbs state is a convex combination of these.

• Theorems 4 and 5 – Along closed segments of the E–H and C–F degeneracy lines (excluding the endpoints) the dominant phase is determined by the relative size of e_F versus e_C (or e_H versus e_C). The coexistence curves intersect these lines at explicitly computed points (e.g. e_F = e_C = √2/2, e_E = e_H = 0).

• Theorem 6 – Along the E–C and H–F lines, if e_F ≥ e_H (or e_E ≥ e_C) the Gibbs measure is unique and no configuration dominates at any temperature; the system exhibits a non‑Peierls behavior with infinitely many ground configurations.

• Theorem 7 – In the same regimes, the probability of observing the non‑dominant vertex type decays exponentially in the inverse temperature, a consequence of reflection positivity.

Overall, the paper provides a complete low‑temperature classification of a geometrically natural extension of the Ising model. By grounding the Hamiltonian in area, perimeter, and Euler characteristic, the authors bridge statistical mechanics with integral geometry, revealing new types of phase behavior (e.g., non‑Peierls degeneracy lines) that do not appear in the classical Ising model. The work not only generalizes known results for the ferromagnetic, antiferromagnetic, and Baxter‑Wu models but also establishes a framework that could be extended to higher dimensions (via intrinsic volumes) and to other lattices.