"A considerable group of numerical analysts still believes in the folk “theorem” that fast MM is always numerical unstable, but in actual tests loss of accuracy

"A considerable group of numerical analysts still believes in the folk “theorem” that fast MM is always numerical unstable, but in actual tests loss of accuracy in fast MM algorithms was limited, and formal proofs of quite reasonable numerical stability of all known fast MM algorithms is available (see [23], [90], [91], [62], and [61])." https://arxiv.org/abs/1804.04102 https://arxiv.org/abs/1804.04102