Implement a linear scaling algorithm without H_0 in numerators

Our code misses a simplification of a redundancy. There are terms like H^{AA} Y(X) and Y(X) H^{BB}, but these terms are related because H^{AA} Y(X) - Y(X) H^{BB} = X. This redundancy generates an excess of terms in Y and \tilde{H}. They should cancel, but producing extra terms costs time and computations. We attempted to get rid of the redundancy by recursively getting rid of H_0 on Y and \tilde{H}, but did not manage to formulate a new algorithm of O(n) per order.

See thread