For some $s, k$, let $J_{s, k}(X; \mathbf{n})$ be the number of solutions to the system $$\sum_{i\le s} (x_i^j - y_i^j) = n_j$$ for $j\le k$ with $x_1, \dots, x_s, y_1, \dots, y_s\in [1, X]\cap\mathbb{Z}$. It is not hard to show that $$J_{s, k}(X; \mathbf{n}) = \int_{[0, 1)^k} |f(\mathbf{\alpha}; X)|^{2s}e(-n_1\alpha_1-\dots-n_k\alpha_k)d\mathbf{\alpha}$$ where $$f(\alpha; X) = \sum_{n\le X} e(\alpha_1 n^1 + \dots + \alpha_k n^k)$$ from which it easily follows that for all $\mathbf{n}\in\mathbb{Z}^k$, $J_{s, k}(X;\mathbf{n})\le J_{s, k}(X; \mathbf{0})$. Is there any way to show that this is the case only by considering the underlying Diophantine equations which doesn't use the fact that $J_{s, k}(X;\mathbf{n})$ is equal to the above integral?


Let $N_{\bf n}$ be the number of solutions to $\sum x_i^s=n_i$. Them you only need to show $\sum N_{\bf m}N_{\bf m+n}\le\sum N_{\bf n}^2$, which is just Cauchy-Schwarz.

  • 1
    $\begingroup$ $N_mN_{n+m}\leqslant \frac12(N_m^2+N_{n+m}^2)$ is not even Cauchy-Schwarz $\endgroup$ – Fedor Petrov Apr 5 '17 at 5:52

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.