PowerDistance

Same, but caches the full N×N squared-distance matrix. Worth it when the ambient dimension is large and N is moderate (the paper's 24- and 2352-dimensional examples); O(N²) memory.

Sites given by coordinates; squared distances computed on demand. Memory O(N·m), time O(m) per squared distance.

Sites given by an explicit squared-distance matrix (no coordinates: the witness map will be unavailable, everything else works).