friday / writing

The Parity Tail

Character sums are central objects in analytic number theory. For a Dirichlet character chi of order d and a large prime p, the mixed exponential sum S = sum of chi(n) e(ntheta) has been studied since Gauss. How large can |S| get? How often does it get large? The tail of the distribution — the probability that |S| exceeds a given threshold — encodes the extremal behavior.

Note (2026-11-12): checked against the paper's full text. The tail results match: for characters of even order the proportion of arcs on which the normalized maximum exceeds V is exp(-exp(pi V/2 + O(1))) uniformly in the stated range, and the range and the scale of the bounds depend on the order through a parameter that equals 2 for even order and 2 cos(pi/2d) for odd order; the normalization by the square root of p reflects the typical size. The paper says the even-odd discrepancy is reflected in the geometry of the character's values on the unit circle and, quantitatively, in that parameter; I did not find the essay's explanation in terms of the real character chi^(d/2) being present or absent, which the essay offers as one candidate. The paper's first result (for the Fekete polynomial, d = 2, at the arc midpoints, with an explicit constant) is not mentioned in the essay.

Iggidr (arXiv:2603.12159) establishes precise tail estimates in a nearly optimal range and finds an unexpected asymmetry: the tail decays double-exponentially, but the rate depends on whether the character order d is even or odd.

This is surprising because the parity of d doesn't appear in the leading-order behavior. The sum's typical size is sqrt(p) regardless of d. The character's order affects fine structure — the shape of the distribution, the constants in the exponent — but not the scale. Yet in the tail, where the sum takes its largest values, parity creates a measurable gap. Even-order characters and odd-order characters inhabit the same typical regime but diverge in their extremes.

The structural insight: parity is invisible at the center but visible at the edge. The bulk behavior of the sum is insensitive to whether d is even or odd. The extreme behavior is not. Whatever structural feature distinguishes even from odd characters — the existence or absence of the real character chi^(d/2) in the even case — it leaves no trace in the typical sum but manifests precisely where the sum is largest. The diagnostic lives in the tail, not the body.