Asymptotics Related to a Binary Search Scheme

TL;DR

This paper analyzes the asymptotic behavior of the number of tests needed in a binary search group testing scheme for contaminated specimens, especially when contamination probability scales with the number of sources.

Contribution

It derives asymptotic formulas for the expectation and variance of tests required, focusing on the case where contamination probability decreases as a power law with the number of sources.

Findings

Asymptotic expectation of tests as N grows large

Variance behavior under different contamination probabilities

Special emphasis on the case p ~ a/N when N is large

Abstract

Specimens are collected from $N$ different sources. Each specimen has probability $p$ of being contaminated, independently of the other specimens. We assume group testing is applicable, namely one can take small portions from several specimens, mix them together, and test the mixture for contamination, so that if the test turns positive, then at least one of the samples in the mixture is contaminated. In this paper we derive asymptotics, as $N$ gets large, of the expectation and the variance of the number $T(N)$ of tests required in order to find all contaminated specimens, under the binary search scheme we introduced in \cite{P} (see, also, arXiv:2007.11910). In \cite{P} the probability $p$ was fixed, whereas in the present work we consider the case where $p \sim a / N^{-\beta}$, $a, \beta > 0$ with emphasis on the case $\beta = 1$, which turns out to be the most interesting.