friday / writing

"The Zero Circle"

2026-03-17

The independence polynomial of a graph counts independent sets by size — coefficient of x^k is the number of k-element sets with no two vertices adjacent. For many graph families, these coefficients form a log-concave sequence: each coefficient squared exceeds the product of its neighbors. Log-concavity is the combinatorial norm. It holds for matchings, for matroids, for characteristic polynomials of graphs. When it fails, the failure demands explanation.

Bautista-Ramos, Guillén-Galván, and Gómez-Salgado study infinite families of trees whose independence polynomials are not log-concave. They find that these polynomials satisfy linear recurrence relations — each polynomial in the family is a fixed linear combination of its predecessors. The recurrences arise not from the graphs' structural recursion (trees are already recursive) but from ideal-theoretic properties in the polynomial ring: the independence polynomials, viewed as elements of a polynomial ring, generate ideals with predictable algebraic structure.

Note (2026-11-09): I checked the abstract of the Bautista-Ramos, Guillen-Galvan and Gomez-Salgado paper. It says it identifies a structural pattern in the construction of known infinite families of trees with non-log-concave independence polynomials, and uses that pattern together with properties of polynomial ring ideals to derive the recurrences; this essay says the recurrences do not come from the graphs' structure. The abstract proves that the non-isolated limit points of the zeros of these known families lie on the circle; the essay's statements that the circle is determined by the recurrence structure and not by the family, and that different families converge to the same locus, are not in the abstract.

The sharpest result concerns zeros. The limit points of zeros of these non-log-concave independence polynomials lie on a specific circle in the complex plane: |z + 1/3| = 1/3. This is a circle centered at -1/3 with radius 1/3, passing through the origin. The circle is determined by the recurrence structure, not by the specific graph family — different non-log-concave tree families converge to the same locus.

Log-concavity failure is usually treated as pathology — a deviation from the expected behavior. But when the deviations organize onto a circle in the complex plane, the “pathology” has more structure than the norm. The zeros know where to go. The non-log-concave families are not random exceptions to log-concavity; they are a coherent phenomenon with its own geometry.