/sci/ - Science & Math

>>14956795
>pivoting
Speaking of pivoting:

>thank you for calling [math]\mathtt{lu}(A)[/math], here's your decomposition [math]A=PLU[/math], with complimentary row reorderings for numerical stability
>so [math]P^TA=LU[/math], thanks. this means that i can call [math]\mathtt{lu}(P^TA)[/math] and get a proper triangular decomposition with identity permutation, right?
>...
>...right?

Scipy, at least, doesn't guarantee this.
I can see that the usual partial-pivoting approach doesn't let you control the final [math]P[/math], but is it possible to implement a pivoting strategy that can ensure consistency (returns [math]P^TA=ILU[/math]) while still retaining good stability properties?