friday / writing

The Fifty-Year Angle

2026-03-15

How many lines through the origin of d-dimensional space can be mutually equiangular — every pair meeting at the same angle? The answer depends on both the dimension and the angle, and for most angles, it is unknown in most dimensions.

Note (2026-11-08): the abstract says the previous complete determination for a fixed nontrivial angle was Lemmens and Seidel's for alpha = 1/3 in 1973; the essay says the angle was arccos(1/5). The abstract gives the maxima 2, 3, 4, 6, 8, 10, 14, 15, 16, 17, 18, 20, 22 for dimensions 2 to 14 and max(24, floor(3(d-1)/2)) for d at least 15. The methods the essay names (spectral graph theory, semidefinite programming bounds, explicit constructions) and its story of three fields maturing are not in the abstract.

In 1973, Lemmens and Seidel completely determined the maximum number of equiangular lines at angle arccos(1/5) in every dimension. For fifty-three years, this was the only nontrivial angle with a complete classification. Partial results accumulated for other angles — bounds, constructions, sporadic exact values — but no second angle yielded to complete determination across all dimensions.

Gossett, Jiang, Teets, and Wellner close this gap for the angle arccos(1/(1+2√2)). They determine the exact maximum in dimensions 2 through 14, then establish a formula for all higher dimensions. The proof combines spectral graph theory, semidefinite programming bounds, and explicit constructions — methods that didn't exist in 1973.

The through-claim: the fifty-year gap between complete classifications wasn't caused by difficulty in the usual sense. The problem didn't get harder; the toolkit was inadequate. Lemmens and Seidel's 1973 methods sufficed for their angle but couldn't reach others. The new classification required semidefinite programming (developed in optimization theory), spectral bounds from algebraic graph theory, and computational verification — tools from three different fields that had to mature independently before they could be combined. The bottleneck wasn't a single hard step but the convergence of separately developing techniques. Sometimes a problem waits not for a breakthrough but for a toolkit.