On sums of two coprime k-th powers (Q1123925): Difference between revisions

For a fixed natural number \(k\geq 2\), let \(r_ k(n)\) denote the number of integer pairs (u,v) with \(| u|^ k+| v|^ k=n\) and \(\rho_ k(n)\) the number of \((u,v)\in {\mathbb{N}}^ 2\) with \(\gcd (u,v)=1\), \(u^ k+v^ k=n\). On the basis of known results on \(\sum_{n\leq x}r_ k(n)\) [cf. \textit{W. Müller} and \textit{W. G. Nowak}, Arch. Math. 51, 55-59 (1988; Zbl 0627.10032)] and the elementary convolution formula \[ \rho_ k(n)=(1/4)\sum_{m^ k_ 1m_ 2=n}\mu (m_ 1)r_ k(m_ 2)\quad (n\geq 2), \] an asymptotic formula for the summatory function \(\sum_{n\leq x}\rho_ k(n)\) is established, assuming the validity of the Riemann Hypothesis. Theorem. If the Riemann Hypothesis is true, \[ \sum_{n\leq x}\rho_ k(n)=B_ kx^{2/k}+O(x^{v(k)+\epsilon)} \] for \(k\geq 3\), \(\epsilon >0\), where \(B_ k\) is an effective constant, \(v(k)=127/140k\) for \(3\leq k\leq 7\) and \(v(k)=\lambda (k)\) for \(k\geq 8\), where \(\lambda (k)=1/k- ((9q+28)/(9q+46))(1/k^ 2)\) and \(q=q(k)\in {\mathbb{N}}\) is uniquely defined by the condition \[ \frac{2^{q+1}74-36}{9q+46}\quad <\quad k\quad \leq \frac{2^{q+2}74-36}{9(q+1)+46}. \] The key point of the proof is a lemma for the estimation of exponential sums involving the Möbius function which is a generalization of a result of \textit{H. L. Montgomery} and \textit{R. C. Vaughan} [Recent progress in analytic number theory, Proc. Symp. Durham 1979, Vol. I, 247-256 (1981; Zbl 0462.10029)] and reads as follows. Lemma. Let (a,b) be any exponent pair, Z, \(Z'\) and W positive real parameters with \(W\geq 1,\quad 1\leq Z<Z'\leq 2Z,\quad c_ 0Z^{(1+5a- 2b)/3a}\leq W\leq Z^{(1+2a-b)/a}\) \((c_ 0\) a numerical constant), then it follows for any \(\epsilon >0\) that \[ \sum_{Z<m\leq Z'}\mu (m)e(W/m)\quad <<\quad W^{\eta_ 1}Z^{\eta_ 2+\epsilon} + W^{\eta_ 3+\epsilon} Z^{\eta_ 4+\epsilon} \] with \[ \eta_ 1=\frac{a}{2(1+a)},\quad \eta_ 2=\frac{3+b}{4(1+a)},\quad \eta_ 3=a- \frac{(1+a-b)a}{2+a-b},\quad \eta_ 4=b-2a+(1+a-b)\frac{1+2a-b}{2+a-b}. \] Furthermore, the new exponent pairs due to \textit{M. N. Huxley} and \textit{N. Watt} [Proc. Lond. Math. Soc., III. Ser. 57, 1-24 (1988; Zbl 0644.10027)] are used. At the end of the paper, several remarks concerning possible slight improvements are made. (In a joint work with \textit{W. Müller} [Lattice points in large planar domains: Applications of Huxley's ``Discrete Hardy-Littlewood method''; to appear in a Springer Lecture Notes volume, edited by E. Hlawka, in 1990], the author has meanwhile carried out the task to generalize the above result to a large class of planar domains, even with a slightly sharper bound. For the special arithmetic function \(\rho_ k(n)\) one obtains an error term \(O(x^{37/41k+\epsilon})\) for \(3\leq k\leq 6\), \(O(x^{\lambda (7)+\epsilon})\) for \(k=7\), respectively.)