The paper develops a new mathematical tool called a "robust OR polynomial framework" and applies it to two classic problems in combinatorics. The core idea is a method for combining certain mathematical certificates, called positive semidefinite certificates, when you have a problem that involves many conditions linked by OR logic (meaning at least one of several conditions must hold). This kind of composition is technically difficult because standard tools either lose accuracy when combined or lack the algebraic structure needed to combine them at all.
The first application involves counting how large a collection of vectors can be while satisfying a certain geometric condition called "r-way acute-free," meaning that for any two vectors in the collection, they must form an obtuse or right angle in at least one coordinate direction. A classical graph theory tool called the Lovasz theta number is easy to multiply across repeated graph products but gives bounds that are far too weak, while a sharper tool called the Schrijver number gives tight bounds but resists multiplication. The paper proves a "quasi-tensorization" result showing that the Schrijver number is almost multiplicative, up to some extra logarithmic factors, which gives new upper bounds on how large these acute-free families can be.
The second application improves bounds on multicolor Ramsey numbers, a central question in combinatorics asking: how large must a complete graph be before any coloring of its edges with r colors must contain a single-color complete subgraph of size k? A recent breakthrough by Balister and collaborators gave the first improvement in decades on the dependence of this number on r. By using the robust OR polynomial framework to sharpen a geometric step in their argument, this paper reduces the dependence on r in the exponent from roughly the 12th power to roughly the 9th power, giving a quantitatively better bound.