On Vu's thin basis theorem in Waring's problem (Q1420208): Difference between revisions

Write \({\mathbb N}_ 0^ k\) for the set of \(k\)th powers of non-negative integers, and say that a set~\(\mathfrak B\) of integers is an asymptotic basis of order~\(h\) if every sufficiently large natural number is a sum of \(h\)~elements of~\(\mathfrak B\). Then the familiar number~\(G(k)\) in Waring's problem is the least order for which \({\mathbb N}_ 0^ k\) is an asymptotic basis, and a result of the author [J.~Lond. Math. Soc. (2) 51, 1--13 (1995; Zbl 0833.11041)] implies that for the larger values of~\(k\) the number~\(G(k)\) has an upper bound asymptotic to~\(k\log k\). Say that an asymptotic basis~\(\mathfrak X\) of order~\(s\) is thin when \(\bigl| \bigl\{{\mathfrak X} \cap [1,t]\bigr\}\bigr| \ll \smash{{(t\log t)}^ {1/s}}\) for sufficiently large~\(t\). \textit{V.~H.~Vu} [Duke Math. J.~105, 107--134 (2000; Zbl 1013.11063)] showed for each~\(k\) that when~\(s\) is sufficiently large there exists such~\(\mathfrak X \in {\mathbb N}_ 0^ k\). Vu's method involved probabilistic ideas and did not yield a sharp bound for an admissible value of~\(s\) for given~\(k\). The author reports that he finds that Vu's argument requires~\(s\) to be somewhat larger than \(2k^ 3\log k\). The author attacks the problem using the methods of analytic number theory in their modern form. He finds that there is a thin asymptotic basis of order~\(H(k)\) for which, again, \({H}(k) \sim k\log k\). In fact he is ``essentially'' able to show that the best value for \({H}(k)\) does not exceed~\(G(k)+2\), so the fact that \(\mathfrak X\) is thin makes little difference. Vu's paper used the circle method as well as probabilistic arguments. The paper under review uses ideas from Vu's paper as well as the circle method, estimates for exponential sums over smooth numbers, and a number of technical devices from the circle method's repertoire. The author remarks that the methods used are of more general applicability, and should be sufficient, for example, to show that there exist thin sets of \(k\)th powers of prime numbers that provide an asymptotic basis.