On sums of squares of primes and a \(k\)th power of prime (Q422015)

Let us call, respectively, \({\mathcal A}_3\) and \({\mathcal A}_4\) the sets of positive integers satisfying all the necessary congruence conditions allowing them to be representable, respectively, as the sum of two squares of primes and a \(k\)-th power of a prime, or as the sum of three squares of primes and a \(k\)-th power of a prime. \textit{L.-K. Hua} in 1938 [Math. Z. 44, 335--346 (1938; Zbl 0020.10502, JFM 64.0131.01)] was the first to prove that almost all \(n\in {\mathcal A}_j\) are the sum of \(j-1\) squares of primes and a \(k\)-th power of a prime (from now on, \(k\geq 2\) fixed). In other words, with \(E_j(N)\) the cardinality of the corresponding \(n\leq N\) (in \({\mathcal A}_j\), of course) which are ``exceptional'' for this kind of Goldbach-Waring representation (a genuine one, for \(k=2\), otherwise with unequal powers), Hua got \(E_j(N)=O(N(\log N)^{-A})\), for some \(A>0\) (here, \(j=3,4\)). After many improvements on both problems (especially for \({\mathcal A}_3\) with \(k=2\), i.e. the genuine Goldbach-Waring) during the years, the present author gives new bounds on \(E_3(N)\) and \(E_4(N)\) both stemming from the same arguments; these, actually, enable him also to bound the cardinality of the exceptional set, say \(E(N)\), for the representation of the \(n\leq N\) (satisfying all the necessary congruence conditions) as \(n=p_1+p_2^2+p_3^k\) improving recent results, too. His arguments are applied in the familiar environment of the Hardy-Littlewood circle method (à la Vinogradov, of course). They rely upon a 2002 nice trick of \textit{T. D. Wooley} [Can. J. Math. 54, No. 2, 417--448 (2002; Zbl 1007.11058)] to bound \(E_j(N)\) in terms of both lower bounds on major arcs and upper bounds on minor arcs; these, in turn, are from a 2006 work of \textit{A. V. Kumchev} [Mich. Math. J. 54, 243--268 (2006; Zbl 1137.11054)]; and he gets an asymptotic formula on major arcs (in terms of, say, singular integrals and singular series for the problem at hand), building mainly on \(L^1\) and \(L^2\) bounds for Dirichlet polynomials of the von Mangoldt \(\Lambda(n)\), ``twisted'' by characters \(\chi(n)\) (they are ``hybrid bounds'', since averaged both on characters \(\chi\) and their moduli).