On generalized quadratic equations in three prime variables (Q1261148): Difference between revisions

Although numbers that are sums of three squares form a set with asymptotic density only 5/6, \textit{H. Davenport} and \textit{H. Heilbronn} [Proc. Lond. Math. Soc., II. Ser. 43, 142-151 (1937; Zbl 0016.34806)] proved that almost all numbers are sums of two squares and an odd power. In the present paper the authors make significant progress on similar problems of this type. Let \(k\geq 2\), and \(a_ 1\), \(a_ 2\), \(a_ 3\) be coprime integers, not all negative. The problem here is the representation of positive integers \(b\) in the form \(a_ 1 p^ 2_ 1+ a_ 2 p^ 2_ 2+ a_ 3 p^ k_ 3\), where \(p_ 1\), \(p_ 2\), \(p_ 3\) are primes, and it will be natural to assume that \(b\) already satisfies the condition that, for each \(q\geq 1\), the congruence \(a_ 1 n^ 2_ 1+ a_ 2 n^ 2_ 2+ a_ 3 n^ k_ 3\equiv b\pmod q\) is soluble in \(n_ 1\), \(n_ 2\), \(n_ 3\) coprime with \(q\). Let \(W(X)\) denote the set of numbers \(b\leq X\) satisfying this congruent solubility condition, and let \(E(X)\) denote the subset of \(b\in W(X)\) which has no representation as \(a_ 1 p^ 2_ 1+ a_ 2 p^ 2_ 2+ a_ 3 p^ k_ 3\). Write \(B=\max\{3,| a_ 1|,| a_ 2|, | a_ 3|\}\). The first theorem is that \[ \text{Card }W(X)\gg X\exp (-C(k)\log B/\log\log B), \] where \(C(k)\) and the implied constant are effective and depend only on \(k\). The second theorem states that there are effective positive constants \(A\) and \(\varepsilon<1\) depending only on \(k\) such that \(\text{Card }E(X)\leq X^{1- \varepsilon}\) when \(X\geq B^ A\). The authors also point out that the lower bound for \(X\) is sharp irrespective of the value for \(A\). Besides dealing with the more difficult case of representations using primes, the estimate for the cardinal of the ``exceptional'' set \(E(X)\) is a substantial improvement on the original Davenport-Heilbronn theorem. The method is essentially the same as that used earlier by the second author and \textit{K. M. Tsang} [J. Reine Angew. Math. 399, 109-136 (1988; Zbl 0667.10030) and Monatsh. Math. 111, 147-169 (1991; Zbl 0719.11064)], but there are new difficulties arising from the fact that the associated singular series in the Hardy-Littlewood method is not absolutely convergent.