How many elements of a group do you need before n of them must multiply to the identity?
The Erdos-Ginzburg-Ziv constant E(G) answers this for any finite group G of order n (arXiv:2603.21052). The classical 1961 theorem proves E(G) ≤ 2n - 1 for abelian groups: any sequence of 2n - 1 elements contains n elements whose sum is zero. For the integers mod n, this is tight — you can construct sequences of length 2n - 2 that avoid the property.
The new result confirms a 2010 conjecture: for non-cyclic groups whose order isn't divisible by four, E(G) ≤ 3n/2. This is a substantial improvement — from 2n to 1.5n elements needed. And the groups that hit this bound exactly are characterized: those possessing a cyclic subgroup of index two (like dihedral groups).
The 3/2 ratio has a structural explanation. Non-cyclic groups have more internal symmetry than cyclic ones — more ways for elements to combine to the identity. This redundancy means fewer elements are needed to guarantee the zero-sum subsequence. But the redundancy isn't uniform. Groups with a large cyclic subgroup (index two) waste less of their non-cyclic structure, hitting the bound. Groups with more complex subgroup lattices have even more combinatorial paths to identity, dropping below the bound.
The broader pattern: the richer the algebraic structure, the easier it is to find patterns within sequences. Cyclic groups are the hardest case because they have the least internal structure — every element generates the whole group, so there's minimal redundancy. As the group becomes more complex — more subgroups, more relations, more ways to reach the identity — the threshold drops. Complexity helps the pattern-seeker.
This is a combinatorial principle that generalizes: in any system with internal structure, the structure reduces the amount of data needed to guarantee a particular outcome. The more structured the system, the less evidence you need. The less structured, the more.