Alphabet-Dependent String Searching with Wexponential Search Trees

TL;DR

This paper introduces new deterministic worst-case bounds for static and dynamic string searching in tries, achieving faster query times by combining known ideas and employing weighted exponential search trees, with applications to suffix trees.

Contribution

It presents improved deterministic worst-case search times for static and dynamic tries using weighted exponential search trees, surpassing previous bounds.

Findings

Static tries: $O(m+ ext{loglog} \sigma)$ search time.

Dynamic tries: $O(m+rac{ ext{loglog}^2 \sigma}{ ext{logloglog}\sigma})$ search/update time.

Suffix tree updates: $O( ext{loglog} n + rac{ ext{loglog}^2 \sigma}{ ext{logloglog}\sigma})$ time.

Abstract

It is widely assumed that $O(m+\lg \sigma)$ is the best one can do for finding a pattern of length $m$ in a compacted trie storing strings over an alphabet of size $\sigma$, if one insists on linear-size data structures and deterministic worst-case running times [Cole et al., ICALP'06]. In this article, we first show that a rather straightforward combination of well-known ideas yields $O(m+\lg\lg \sigma)$ deterministic worst-case searching time for static tries. Then we move on to dynamic tries, where we achieve a worst-case bound of $O(m+\frac{\lg^{2}\lg\sigma}{\lg\lg\lg\sigma})$ per query or update, which should again be compared to the previously known $O(m+\lg\sigma)$ deterministic worst-case bounds [Cole et al., ICALP'06], and to the alphabet \emph{in}dependent $O(m+\sqrt{\lg n/\lg\lg n})$ deterministic worst-case bounds [Andersson and Thorup, SODA'01], where $n$ is the number of nodes in the trie. The basis of our update procedure is a weighted variant of exponential search trees which, while simple, might be of independent interest. As one particular application, the above bounds (static and dynamic) apply to suffix trees. There, an update corresponds to pre- or appending a letter to the text, and an additional goal is to do the updates quicker than rematching entire suffixes. We show how to do this in $O(\lg\lg n + \frac{\lg^{2}\lg\sigma}{\lg\lg\lg\sigma})$ time, which improves the previously known $O(\lg n)$ bound [Amir et al., SPIRE'05].