Breaking the $O(n)$-Barrier in the Construction of Compressed Suffix Arrays

The suffix array, describing the lexicographic order of suffixes of a given text, is the central data structure in string algorithms. The suffix array of a length-$n$ text uses $\Theta(n \log n)$ bits, which is prohibitive in many applications. To address this, Grossi and Vitter [STOC 2000] and, independently, Ferragina and Manzini [FOCS 2000] introduced space-efficient versions of the suffix array, known as the compressed suffix array (CSA) and the FM-index. For a length-$n$ text over an alphabet of size $\sigma$, these data structures use only $O(n \log \sigma)$ bits. Immediately after their discovery, they almost completely replaced plain suffix arrays in practical applications, and a race started to develop efficient construction procedures. Yet, after more than 20 years, even for $\sigma=2$, the fastest algorithm remains stuck at $O(n)$ time [Hon et al., FOCS 2003], which is slower by a $\Theta(\log n)$ factor than the lower bound of $\Omega(n / \log n)$ (following simply from the necessity to read the entire input). We break this long-standing barrier with a new data structure that takes $O(n \log \sigma)$ bits, answers suffix array queries in $O(\log^{\epsilon} n)$ time, and can be constructed in $O(n\log \sigma / \sqrt{\log n})$ time using $O(n\log \sigma)$ bits of space. Our result is based on several new insights into the recently developed notion of string synchronizing sets [STOC 2019]. In particular, compared to their previous applications, we eliminate orthogonal range queries, replacing them with new queries that we dub prefix rank and prefix selection queries. As a further demonstration of our techniques, we present a new pattern-matching index that simultaneously minimizes the construction time and the query time among all known compact indexes (i.e., those using $O(n \log \sigma)$ bits).