Pinning Down the Strong Wilber 1 Bound for Binary Search Trees

TL;DR

This paper investigates the relationship between Wilber's first bound and the optimal cost in binary search trees, providing new bounds, approximation algorithms, and proposing a stronger bound for future algorithmic improvements.

Contribution

It demonstrates the potential gap between Wilber's first bound and optimal solutions, and introduces new approximation algorithms and a stronger bound for binary search trees.

Findings

The gap between WB-1 and optimal cost can be as large as (\, log \, log \,n / log \, log \, log \,n)

A simple algorithm achieves an O(D)-approximation in sub-exponential time for any integer D>0

A new Guillotine Bound is proposed, stronger than WB-1, with potential for better algorithms.

Abstract

The dynamic optimality conjecture, postulating the existence of an $O(1)$-competitive online algorithm for binary search trees (BSTs), is among the most fundamental open problems in dynamic data structures. Despite extensive work and some notable progress, including, for example, the Tango Trees (Demaine et al., FOCS 2004), that give the best currently known $O(\log \log n)$-competitive algorithm, the conjecture remains widely open. One of the main hurdles towards settling the conjecture is that we currently do not have approximation algorithms achieving better than an $O(\log \log n)$-approximation, even in the offline setting. All known non-trivial algorithms for BST's so far rely on comparing the algorithm's cost with the so-called Wilber's first bound (WB-1). Therefore, establishing the worst-case relationship between this bound and the optimal solution cost appears crucial for further progress, and it is an interesting open question in its own right. Our contribution is two-fold. First, we show that the gap between the WB-1 bound and the optimal solution value can be as large as $\Omega(\log \log n/ \log \log \log n)$; in fact, the gap holds even for several stronger variants of the bound. Second, we provide a simple algorithm, that, given an integer $D>0$, obtains an $O(D)$-approximation in time $\exp\left(O\left (n^{1/2^{\Omega(D)}}\log n\right )\right )$. In particular, this gives a constant-factor approximation sub-exponential time algorithm. Moreover, we obtain a simpler and cleaner efficient $O(\log \log n)$-approximation algorithm that can be used in an online setting. Finally, we suggest a new bound, that we call {\em Guillotine Bound}, that is stronger than WB, while maintaining its algorithm-friendly nature, that we hope will lead to better algorithms. All our results use the geometric interpretation of the problem, leading to cleaner and simpler analysis.