Title: Entropy-Optimal Algorithms for Multiple-Choice Secretary

Abstract: In this paper, we study the problem for the entropy of both admissible and optimal distributions of permutations to the multiple-choice secretary problem. In particular, we construct a distribution with entropy $\Theta(\log\log n)$ such that a deterministic threshold-based algorithm gives a nearly-optimal competitive ratio $1-O(\log{k}/k^{1/3})$ for $k=O((\log n)^{3/14})$. Our error is simultaneously nearly-optimal and with optimal entropy $\Theta(\log\log n)$. Our result improves in two ways the previous best construction by Kesselheim, Kleinberg and Niazadeh (KKN) [STOC'15] whose competitive ratio is $1 - O(1/k^{1/3}) - o(1)$. First, our solution works for an exponentially larger range of parameters $k$, as in the work of KKN $k=O((\log\log\log n)^{\epsilon})$. Second, our algorithm is a simple deterministic single-threshold algorithm (only drawing a permutation from a stochastic uniform distribution), while the algorithm in KKN uses additional randomness. We also prove a new corresponding lower bound for entropy of optimal solutions to the $k$-secretary problem, matching the entropy of our algorithm. We further show the strength of our techniques by obtaining fine-grained results for optimal distributions of permutations for the classic secretary problem. For optimal entropy $\Theta(\log\log n)$, we precisely characterize the success probability of uniform distributions that is below, and close to, $1/e$, and construct such distributions in polynomial time. We prove even higher entropy $\Theta(\log(n))$ suffices for a success probability above $1/e$, but, no uniform probability distribution with small support and entropy strictly less than $\log (n)$ can have success probability above $1/e$. For maximum entropy $\Theta(n \log(n))$, improving upon a result of Samuels from 1981, we find the precise formula for the optimal success probability of any secretary algorithm.