Large values of the Gowers-Host-Kra seminorms

Abstract

The Gowers uniformity norms \|f\|Uk(G) of a function f: G on a finite additive group G, together with the slight variant \|f\|Uk([N]) defined for functions on a discrete interval [N] := \1,...,N\, are of importance in the modern theory of counting additive patterns (such as arithmetic progressions) inside large sets. Closely related to these norms are the Gowers-Host-Kra seminorms \|f\|Uk(X) of a measurable function f: X on a measure-preserving system X = (X, X, μ, T). Much recent effort has been devoted to the question of obtaining necessary and sufficient conditions for these Gowers norms to have non-trivial size (e.g. at least η for some small η > 0), leading in particular to the inverse conjecture for the Gowers norms, and to the Host-Kra classification of characteristic factors for the Gowers-Host-Kra seminorms. In this paper we investigate the near-extremal (or "property testing") version of this question, when the Gowers norm or Gowers-Host-Kra seminorm of a function is almost as large as it can be subject to an L∞ or Lp bound on its magnitude. Our main results assert, roughly speaking, that this occurs if and only if f behaves like a polynomial phase, possibly localised to a subgroup of the domain; this can be viewed as a higher-order analogue of classical results of Russo and Fournier, and are also related to the polynomiality testing results over finite fields of Blum-Luby-Rubinfeld and Alon-Kaufman-Krivelevich-Litsyn-Ron. We investigate the situation further for the U3 norms, which are associated to 2-step nilsequences, and find that there is a threshold behaviour, in that non-trivial 2-step nilsequences (not associated with linear or quadratic phases) only emerge once the U3 norm is at most 2-1/8 of the L∞ norm.

0

Turn this paper into a lesson

ArcXiv compiles a structured reading guide from this paper's metadata: plain-English importance, contributions, prerequisite concepts, which sections to read first, flashcards, and a quiz. Grounded in the abstract, never invented.

Discussion (0)

Sign in to join the discussion.

Loading comments…