Two Integral Sliding-Window Inequalities for Maximal Convolutions

Abstract

We prove two sliding-window inequalities for maximal convolutions. The first concerns the multiplicative maximal convolution. If $f$ and $g$ are nonnegative continuous functions on $[0,A]$ and $[0,B]$, respectively, define \[ h(x)=\max_{\substack{0\le u\le A\\0\le x-u\le B}} f(u)g(x-u),\qquad 0\le x\le A+B. \] Then there exists a window $[a,a+B]$ of length $B$ such that \[ \frac1B\int_a^{a+B}h(x)\,dx\ge \left(\frac1A\int_0^A f(x)\,dx\right) \left(\frac1B\int_0^B g(x)\,dx\right). \] The second concerns the additive maximal convolution. Let $f$ and $g$ be nonnegative continuous functions on $[0,C]$, and define \[ H(x)=\max_{\substack{0\le u\le C\\0\le x-u\le C}}\{f(u)+g(x-u)\},\qquad 0\le x\le 2C. \] Then, for every $p\ge1$, there exists a window $[a,a+C]$ of length $C$ such that \[ \left(\int_a^{a+C}H(x)^p\,dx\right)^{1/p} \ge \left(\int_0^C f(x)^p\,dx\right)^{1/p} + \left(\int_0^C g(x)^p\,dx\right)^{1/p}. \] We also record discrete analogues. The main point is that, in one-dimensional maximal-convolution settings, certain global Brunn--Minkowski or Prékopa--Leindler type phenomena admit natural sliding-window localizations.