Diophantine approximation with prime variables and mixed powers (Q5896116)

The famous Oppenheim conjecture, proved in 1987 by Margulis using methods from Ergodic theory, asserts that if \(Q\) is an indefinite non-degenerate quadratic form in \(n\ge 3\) variables which is not proportional to a form with integer coefficients, then the set \(\{Q(x) : x\in \mathbb{Z}^n\}\) is dense in \(\mathbb{R}\). Later, [\textit{R. J. Cook} and \textit{A. Fox}, Mathematika 48, No. 1--2, 137--149 (2001; Zbl 1035.11010)] and [\textit{G. Harman}, Mathematika 51, No. 1--2, 83--96 (2004; Zbl 1107.11043)] investigated the distribution of values of ternary quadratic forms \(Q(x)\) with arguments restricted to primes. Quintuple cubic forms at prime arguments were considered by [\textit{R.J. Cook}, J. Théor. Nombres Bordx. 13, No. 1, 77--91 (2001, Zbl 1047.11095)]. Diophantine approximation with mixed powers at prime arguments by terms of the form \[ \lambda_1p_1^2+\lambda_2p_2^2+\lambda_3p_3^3+\lambda_4p_4^4 \] was investigated by [\textit{W. P. Li} and \textit{T. Z. Wang}, Chin. Ann. Math., Ser. A 31, No. 2, 247--256 (2010, Zbl 1224.11061)]. Their result was subsequently improved by [\textit{Q. W. Mu} and \textit{X. D. Lü}, Chin. Ann. Math., Ser. A 36, No. 3, 303--312 (2015, Zbl 1340.11054)] and [\textit{Q. W. Mu} and \textit{X. D. Lü}, Adv. Math., Beijing 46, No. 2, 190--202 (2017, Zbl 1372.11050)]). The following sharpening of these results is obtained in the paper under review. Theorem. Let \(k\) be a positive integer with \(k\ge 4\) and \[ s(k)=\left[\frac{k+1}{2}\right]. \] Suppose that \(\lambda_1,\lambda_2,\lambda_3,\lambda_4\) are positive real numbers, \(\lambda_1/\lambda_2\) is irrational and algebraic, \(\mathcal{V}\) is a well-spaced sequence of positive real numbers, \(\delta>0\). Then the number \(E_k(\mathcal{V}, X,\delta)\) of \(v\in \mathcal{V}\) with \(v\le X\) for which \[ |\lambda_1p_1^2+\lambda_2p_2^2+\lambda_3p_3^4+\lambda_4p_4^4-v|\le v^{-\delta} \] has no solution in primes \(p_1,p_2,p_3,p_4\) satisfies \[ E_k(\mathcal{V},X,\delta)\ll X^{\hat{\sigma}+2\delta+\varepsilon} \] for any \(\varepsilon>0\), where \(\hat{\sigma}=\frac{15}{16}-\frac{1}{8\sigma(k)}\) and \[ \sigma(k)=\min\left(2^{s(k)-1}, \frac{1}{2}s(k)(s(k)+1)\right). \] The proof relies on a version of the circle method.