Halfspaces are hard to test with relative error

Xi Chen; Anindya De; Yizhi Huang; Shivam Nadimpalli; Rocco A. Servedio; Tianqi Yang

Paper

Halfspaces are hard to test with relative error

Abstract

Several recent works [DHLNSY25, CPPS25a, CPPS25b] have studied a model of property testing of Boolean functions under a \emph{relative-error} criterion. In this model, the distance from a target function

that is being tested to a function

is defined relative to the number of inputs

for which

; moreover, testing algorithms in this model have access both to a black-box oracle for

and to independent uniform satisfying assignments of

. The motivation for this model is that it provides a natural framework for testing \emph{sparse} Boolean functions that have few satisfying assignments, analogous to well-studied models for property testing of sparse graphs. The main result of this paper is a lower bound for testing \emph{halfspaces} (i.e., linear threshold functions) in the relative error model: we show that

oracle calls are required for any relative-error halfspace testing algorithm over the Boolean hypercube

. This stands in sharp contrast both with the constant-query testability (independent of

) of halfspaces in the standard model [MORS10], and with the positive results for relative-error testing of many other classes given in [DHLNSY25, CPPS25a, CPPS25b]. Our lower bound for halfspaces gives the first example of a well-studied class of functions for which relative-error testing is provably more difficult than standard-model testing.

Parse Issues: 3 warnings