Core Concepts
For every prime p, there exists a constant δ > 0 such that for every field F of characteristic p and for all integers k ≥ 2 and d ≥ k, there exists a k-nearly orthogonal set of at least dδ·k/ log k vectors in Fd. The size of this set is optimal up to the log k term in the exponent.
Abstract
The content discusses the problem of determining the largest possible size of nearly orthogonal sets over finite fields.
Key highlights:
For a field F and integers d and k, a set A ⊆ Fd is called k-nearly orthogonal if its members are non-self-orthogonal and every k + 1 vectors of A include an orthogonal pair.
Let α(d, k, F) denote the largest possible size of a k-nearly orthogonal subset of Fd.
A simple upper bound on α(d, k, F) stems from Ramsey theory, giving α(d, k, F) ≤ O(dk).
Over the real field R, the problem of determining α(d, k, R) was studied extensively, with lower and upper bounds being rather far apart.
The study of nearly orthogonal sets over finite fields was proposed by Codenotti, Pudlák, and Resta, motivated by questions in circuit complexity.
In contrast to the real field, it was shown that for the binary field F2, there exists a constant δ > 0 such that α(d, 2, F2) ≥ d1+δ for infinitely many integers d.
The main results of the paper are:
For every prime p, there exists a constant δ = δ(p) > 0 such that for every field F of characteristic p and for all integers k ≥ 2 and d ≥ k, it holds that β(d, k, F) ≥ dδ·k/ log k, where β(d, k, F) denotes the largest possible size of a set A ⊆ Fd of non-self-orthogonal vectors such that for every two (not necessarily disjoint) sets A1, A2 ⊆ A of size k + 1 each, there exist vectors v1 ∈ A1 and v2 ∈ A2 with ⟨v1, v2⟩ = 0.
For every prime p and every integer ℓ ≥ 1, there exists a constant δ = δ(p, ℓ) > 0 such that for every field F of characteristic p and for all integers k ≥ 2 and d ≥ k ≥ ℓ, it holds that α(d, k, ℓ, F) ≥ dδ·k/ log k, where α(d, k, ℓ, F) denotes the largest possible size of a (k, ℓ)-nearly orthogonal subset of Fd.
The proofs rely on probabilistic and spectral arguments, as well as the hypergraph container method.