We explore the $\textit{average-case deterministic query complexity}$ of boolean functions under the $\textit{uniform distribution}$, denoted by $\mathrm{D}_\mathrm{ave}(f)$, the minimum average depth of zero-error decision tree computing a boolean function $f$. This measure found several applications across diverse fields. We study $\mathrm{D}_\mathrm{ave}(f)$ of several common functions, including penalty shoot-out functions, symmetric functions, linear threshold functions and tribes functions. Let $\mathrm{wt}(f)$ denote the number of the inputs on which $f$ outputs $1$. We prove that $\mathrm{D}_\mathrm{ave}(f) \le \log \frac{\mathrm{wt}(f)}{\log n} + O\left(\log \log \frac{\mathrm{wt}(f)}{\log n}\right)$ when $\mathrm{wt}(f) \ge 4 \log n$ (otherwise, $\mathrm{D}_\mathrm{ave}(f) = O(1)$), and that for almost all fixed-weight functions, $\mathrm{D}_\mathrm{ave}(f) \geq \log \frac{\mathrm{wt}(f)}{\log n} - O\left( \log \log \frac{\mathrm{wt}(f)}{\log n}\right)$, which implies the tightness of the upper bound up to an additive logarithmic term. We also study $\mathrm{D}_\mathrm{ave}(f)$ of circuits. Using H\r{a}stad's switching lemma or Rossman's switching lemma [Comput. Complexity Conf. 137, 2019], one can derive upper bounds $\mathrm{D}_\mathrm{ave}(f) \leq n\left(1 - \frac{1}{O(k)}\right)$ for width-$k$ CNFs/DNFs and $\mathrm{D}_\mathrm{ave}(f) \leq n\left(1 - \frac{1}{O(\log s)}\right)$ for size-$s$ CNFs/DNFs, respectively. For any $w \ge 1.1 \log n$, we prove the existence of some width-$w$ size-$(2^w/w)$ DNF formula with $\mathrm{D}_\mathrm{ave} (f) = n \left(1 - \frac{\log n}{\Theta(w)}\right)$, providing evidence on the tightness of the switching lemmas.
翻译:暂无翻译