Geometric group testing

TL;DR

This paper explores non-adaptive group testing in a geometric setting, providing bounds on the number of tests needed when items are points in Euclidean space and tests are axis-parallel boxes, revealing polynomial bounds unlike the combinatorial case.

Contribution

It introduces bounds on the number of tests required for geometric group testing, a setting with spatial constraints, which differs from classical combinatorial group testing.

Findings

Identifying a defective pair in the plane requires at least (m^{3/5}) tests.

Configurations exist where (m^{2/3}) tests suffice for pair detection.

Single defective point identification in the plane needs (m^{1/2}) tests, sometimes achievable.

Abstract

Group testing is concerned with identifying $t$ defective items in a set of $m$ items, where each test reports whether a specific subset of items contains at least one defective. In non-adaptive group testing, the subsets to be tested are fixed in advance. By testing multiple items at once, the required number of tests can be made much smaller than $m$. In fact, for $t \in \mathcal{O}(1)$, the optimal number of (non-adaptive) tests is known to be $\Theta(\log{m})$. In this paper, we consider the problem of non-adaptive group testing in a geometric setting, where the items are points in $d$-dimensional Euclidean space and the tests are axis-parallel boxes (hyperrectangles). We present upper and lower bounds on the required number of tests under this geometric constraint. In contrast to the general, combinatorial case, the bounds in our geometric setting are polynomial in $m$. For instance, our results imply that identifying a defective pair in a set of $m$ points in the plane always requires $\Omega(m^{3/5})$ tests, and there exist configurations of $m$ points for which $\mathcal{O}(m^{2/3})$ tests are sufficient, whereas to identify a single defective point in the plane, $\Theta(m^{1/2})$ tests are always necessary and sometimes sufficient.