Linear Time Construction of Cover Suffix Tree and Applications

TL;DR

This paper presents an optimal linear-time algorithm for constructing the Cover Suffix Tree of a string over an integer alphabet, enabling efficient computation of various quasiperiodic structures and pattern occurrences.

Contribution

It introduces a novel linear-time algorithm for constructing the Cover Suffix Tree, improving previous methods and facilitating faster computation of seeds and partial covers.

Findings

CST can be built in O(n) time for integer alphabets.

New characterization of overlapping occurrences using runs.

Efficient indexing for pattern occurrence reporting.

Abstract

The Cover Suffix Tree (CST) of a string $T$ is the suffix tree of $T$ with additional explicit nodes corresponding to halves of square substrings of $T$. In the CST an explicit node corresponding to a substring $C$ of $T$ is annotated with two numbers: the number of non-overlapping consecutive occurrences of $C$ and the total number of positions in $T$ that are covered by occurrences of $C$ in $T$. Kociumaka et al. (Algorithmica, 2015) have shown how to compute the CST of a length-$n$ string in $O(n \log n)$ time. We show how to compute the CST in $O(n)$ time assuming that $T$ is over an integer alphabet. Kociumaka et al. (Algorithmica, 2015; Theor. Comput. Sci., 2018) have shown that knowing the CST of a length-$n$ string $T$, one can compute a linear-sized representation of all seeds of $T$ as well as all shortest $\alpha$-partial covers and seeds in $T$ for a given $\alpha$ in $O(n)$ time. Thus our result implies linear-time algorithms computing these notions of quasiperiodicity. The resulting algorithm computing seeds is substantially different from the previous one (Kociumaka et al., SODA 2012, ACM Trans. Algorithms, 2020). Kociumaka et al. (Algorithmica, 2015) proposed an $O(n \log n)$-time algorithm for computing a shortest $\alpha$-partial cover for each $\alpha=1,\ldots,n$; we improve this complexity to $O(n)$. Our results are based on a new characterization of consecutive overlapping occurrences of a substring $S$ of $T$ in terms of the set of runs (see Kolpakov and Kucherov, FOCS 1999) in $T$. This new insight also leads to an $O(n)$-sized index for reporting overlapping consecutive occurrences of a given pattern $P$ of length $m$ in $O(m+output)$ time, where $output$ is the number of occurrences reported. In comparison, a general index for reporting bounded-gap consecutive occurrences of Navarro and Thankachan (Theor. Comput. Sci., 2016) uses $O(n \log n)$ space.