Optimal bounds for monotonicity and Lipschitz testing over hypercubes and hypergrids

Abstract

The problem of monotonicity testing over the hypergrid and its special case, the hypercube, is a classic, well-studied, yet unsolved question in property testing. We are given query access to $f:[k]^n \mapsto \R$ (for some ordered range $\R$). The hypergrid/cube has a natural partial order given by coordinate-wise ordering, denoted by $\prec$. A function is \emph{monotone} if for all pairs $x \prec y$, $f(x) \leq f(y)$. The distance to monotonicity, $\eps_f$, is the minimum fraction of values of $f$ that need to be changed to make $f$ monotone. For $k=2$ (the boolean hypercube), the usual tester is the \emph{edge tester}, which checks monotonicity on adjacent pairs of domain points. It is known that the edge tester using $O(\eps^{-1}n\log|\R|)$ samples can distinguish a monotone function from one where $\eps_f > \eps$. On the other hand, the best lower bound for monotonicity testing over the hypercube is $\min(|\R|^2,n)$. This leaves a quadratic gap in our knowledge, since $|\R|$ can be $2^n$. We resolve this long standing open problem and prove that $O(n/\eps)$ samples suffice for the edge tester. For hypergrids, known testers require $O(\eps^{-1}n\log k\log |\R|)$ samples, while the best known (non-adaptive) lower bound is $\Omega(\eps^{-1} n\log k)$. We give a (non-adaptive) monotonicity tester for hypergrids running in $O(\eps^{-1} n\log k)$ time. Our techniques lead to optimal property testers (with the same running time) for the natural \emph{Lipschitz property} on hypercubes and hypergrids. (A $c$-Lipschitz function is one where $|f(x) - f(y)| \leq c\|x-y\|_1$.) In fact, we give a general unified proof for $O(\eps^{-1}n\log k)$-query testers for a class of "bounded-derivative" properties, a class containing both monotonicity and Lipschitz.