Approximating Text-to-Pattern Hamming Distances

We revisit a fundamental problem in string matching: given a pattern of length m and a text of length n, both over an alphabet of size σ, compute the Hamming distance (i.e., the number of mismatches) between the pattern and the text at every location. Several randomized (1+ε)-approximation algorithms have been proposed in the literature (e.g., by Karloff (Inf. Proc. Lett., 1993), Indyk (FOCS 1998), and Kopelowitz and Porat (SOSA 2018)), with running time of the form O(ε−O(1)nlognlogm), all using fast Fourier transform (FFT).  We describe a simple randomized (1+ε)-approximation algorithm that is faster and does not need FFT. Combining our approach with additional ideas leads to numerous new results (all Monte-Carlo randomized) in different settings: (1) We design the first truly linear-time approximation algorithm for constant ; the running time is O(ε−2n).  In fact, the time bound can be made slightly sublinear in n if the alphabet size σ is small (by using bit packing tricks). (2) We apply our approximation algorithms to design a faster exact algorithm computing all Hamming distances up to a threshold k; its runtime of O(n + min(nk√logm/√m,nk2/m)) improves upon previous results by logarithmic factors and is linear for k≤ √m. (3) We alternatively design approximation algorithms with better ε-dependence, by using fast rectangular matrix multiplication. In fact, the time bound is O(npolylogn) when the pattern is sufficiently long, i.e., m≥ ε−c for a specific constant c.  Previous algorithms with the best ε-dependence require O(ε−1npolylogn) time. (4) When k is not too small, we design a truly sublinear-time algorithm to find all locations with Hamming distance approximately (up to a constant factor) less than k, in time O((n/kΩ(1)+occ)no(1)) time, where occ is the output size. The algorithm leads to a property tester for pattern matching that costs O((δ−1/3n2/3 + δ−1n/m) n) time and, with high probability, returns true if an exact match exists and false if the Hamming distance is more than δ m at every location. (5) We design a streaming algorithm that approximately computes the Hamming distance for all locations with the distance approximately less than k, using O(ε−2√kn) space.  Previously, streaming algorithms were known for the exact problem with O(kn) space (which is tight up to the polylogn factor) or for the approximate problem with  O(ε−O(1)√mpolylogn) space. For the special case of k=m, we improve the space usage to O(ε−1.5√mpolylogn).