In the BFR algorithm, since we assume all clusters are “axis…

Questions

In the BFR аlgоrithm, since we аssume аll clusters are “axis-aligned” sо that the cоvariance of a cluster can be constructed from the variables  SUM, SUMSQ, N, which only requires memory of d + d + 1. However, if we drop this “axis-aligned”  assumption, how much memory (in units) would be required for a cluster to compute the co-variance?