It is much better for working with (a) low-rank approximations and (b) optimizing the order of operations. While (b) is "merely" a question of computational per

It is much better for working with (a) low-rank approximations and (b) optimizing the order of operations. While (b) is "merely" a question of computational performance, (a) can have extremely important fundamental/theoretical consequences for the problem under study (e.g. finding efficient classical algorithms for modeling quantum mechanics).