Secretary problem and two almost the same consecutive applicants

TL;DR

This paper introduces a new variant of the secretary problem focusing on selecting applicants with consecutive relative ranks differing by one, analyzing the success probability of a specific stopping rule as the number of applicants grows large.

Contribution

It proposes a novel secretary problem variant based on relative ranks and provides asymptotic probability bounds for a specific stopping rule's success.

Findings

Asymptotic success probability approaches 1/2 for the proposed rule.

The stopping rule's success probability is maximized at a threshold of half the applicants.

The new variant extends classical secretary problem models with a focus on consecutive relative ranks.

Abstract

We present a new variant of the secretary problem. Let $A$ be a totally ordered set of $n$ \emph{applicants}. Given $P\subseteq A$ and $x\in A$, let $rr(P,x)=\vert\{z\in P \mid z\leq x\}\vert\mbox{ }$ be the \emph{relative rank of} $x$ \emph{with regard to} $P$, and let $rr_n(x)=rr(A,x)$. Let $x_1,x_2,\dots,x_n\in A$ be a random sequence of distinct applicants. The aim is to select $1<j\leq n$ such that $rr_n(x_{j-1})-rr_n(x_j)\in\{-1,1\}$. Let $\alpha$ be a real constant with $0<\alpha<1$. Suppose the following stopping rule $\tau_n(\alpha)$: reject first $\alpha n$ applicants and then select the first $x_j$ such that $rr(P_j,x_{j-1})-rr(P_j,x_j)\in\{-1,1\}$, where $P_j=\{x_i\mid 1\leq i\leq j\}$. Let $p_{n,\tau}(\alpha)$ be the probability that $\tau_n(\alpha)$ selects $x_j$ such that $rr_n(x_{j-1})-rr_n(x_j)\in\{-1,1\}$. We show that \[\lim_{n\rightarrow\infty}p_{n,\tau}(\alpha)\leq \lim_{n\rightarrow\infty}p_{n,\tau}\left(\frac{1}{2}\right)=\frac{1}{2}\mbox{.}\]