Hi, I don't know how to do it with "linear algebra" but you can still do it in a different way. For instance, you can use CD-Hit algorithm, the one used for reducing the NCBI Non-Redudant database redundancy, or if you have your sequences aligned, you could try a trimAl [http://trimal.cgenomics.org] where you set the redundancy threshold and the program applies a clustering algorithm for removing redundant sequences.