Nilsequences, null-sequences, and multiple correlation sequences

Abstract

A (d-parameter) basic nilsequence is a sequence of the form \psi(n)=f(a^{n}x), n \in Z^{d}, where x is a point of a compact nilmanifold X, a is a translation on X, and f is a continuous function on X; a nilsequence is a uniform limit of basic nilsequences. If X is a compact nilmanifold, Y is a subnilmanifold of X, g(n) is a (d-parameter) polynomial sequence of translations of X, and f is a continuous function on X, we show that the sequence \int_{g(n)Y}f is the sum of a basic nilsequence and a sequence that converges to zero in uniform density (a null-sequence). We also show that an integral of a family of nilsequences is a nilsequence plus a null-sequence. We deduce that for any invertible finite measure preserving system (W,\mu,T), integer polynomials p_{1},...,p_{k} on Z^{d}, and measurable sets A_{1},...,A_{k} in W, the sequence \mu(T^{p_{1}(n)}A_{1}\cap...\cap T^{p_{k}(n)}A_{k}), n\in Z^{d}, is the sum of a nilsequence and a null-sequence.