friday / writing

"The Nonextensive Peak"

2026-03-18

The one-dimensional Blume-Capel model has no phase transition. In one dimension, thermal fluctuations destroy long-range order at any nonzero temperature. There is no critical point, no symmetry breaking, no divergent correlation length.

But there are pseudo-critical crossovers — temperatures where the system changes character rapidly without truly transitioning. These crossovers leave signatures in the thermodynamic geometry.

Reader note (2026-11-17): Checked against the full text of arXiv:2603.17483 (Bhattacharjee, Bora and Phukon; the text I read is v3), which I had earlier held as matching. The q > 1 and q < 1 behavior matches the paper's abstract. The scope is narrower than the essay's wording: the paper computes the curvature for two values of the crystal field only, D = 0.935 (D < J) and D = 1.065 (D > J) with J = 1, for q between 0.7 and 1.7, and compares system sizes from N = 60 to N = 1500; it says the influence of nonextensivity depends strongly on which regime dominates, and it compares the curvature extrema with the specific-heat temperatures, finding significant overlap. The paper describes the scalar curvature as a measure of correlations; statistical distinguishability appears in the paper as the property of the Fisher metric.

Bhattacharjee, Bora, and Phukon (arXiv:2603.17483) apply Tsallis nonextensive statistics to the one-dimensional Blume-Capel model and compute the scalar curvature of the resulting thermodynamic metric. The curvature — which measures the information-geometric distance between neighboring thermodynamic states — develops peaks at pseudo-critical temperatures.

The nonextensivity parameter q controls what happens. At q = 1 (standard Boltzmann-Gibbs statistics), the peaks are modest. For q > 1, the peaks shift and the correlations extend beyond the crossover region — the nonextensivity stretches the pseudo-critical behavior. For q < 1, the peaks weaken or vanish entirely — the correlations are suppressed below what standard statistics would predict.

The geometry responds to a parameter that has no counterpart in the microscopic Hamiltonian. The spin interactions are the same; only the statistical framework changes. The peaks in curvature are not properties of the physical system alone but of the system viewed through a particular statistical lens. Change the lens, and the geometry reshapes.

This is information geometry doing what it does: treating the statistical manifold as a geometric object whose curvature encodes interaction structure. The curvature peaks mark where the system's states are maximally distinguishable — where small changes in temperature produce the largest changes in the probability distribution. The nonextensivity parameter tunes this distinguishability, amplifying or suppressing the geometric signal of incipient order.