LU decomposition python with vector pivot

Question

The LU decomposition function provided by scipy returns a permutation matrix P

P,L,U = scipy.linalg.lu(A)

Where A is a rectangular matrix. However the size of my problem do not allow to store P (even temporary) due to its size, I really need a function that computes a permutation vector (like [L,U,P] = lu(A,'vector') in Matlab). I found a lapack function

LU,p,info = scipy.linalg.lapack.dgetrf(A)

which seems to return a vector p but I learned this latter is not an actual permutation vector since it contains twice the same value ( https://software.intel.com/en-us/forums/intel-math-kernel-library/topic/780655 ). I am thus looking for another function (which may be from another library) to perform this LU decomposition with pivoting. Since the computation time is also very important, I don't think implementing the decomposition myself will be efficient.

Answer 1

Yes, it's a pivot vector, which is a standard LAPACK return. So you'll need to convert it into whatever other form you want yourself (and this is much easier then reimplementing the factorization).