Hi all,
I have difficulties to line up results from pardiso forward substitution with results from explicit trials (eg. in R or pyhton).
Let "C" be a positive definite symmetric squared matrix and "v" be a vector of 1 with length equal to the rows of "C".
When solving "Cb=v" pardiso results looking good with phase 33, where "b" contains the sum over rows of "C^{-1}".
Now "C" maybe decomposed into "LL'" (cholesky factor). A forward substitution would solve "Lb=v", thus the result must be the sum over the rows of "L^{-1}".
However, when feeding the same matrix "C" into pardiso, but setting phase to 331 the results doesn't seem to reflect the above. I guessed that it might be due to permutation. Retrieving the permutation by setting iparm(5) to reshuffle the result vector did not do anything. Moreover, since the range fo the results from pardiso where different to an explicit operation (taking the row sums of "L^{-1}"), the cause cannot the permutation.
Did I miss anything?? For instance is pardiso's LL' not the cholesky of C??
Any advice appreciated.
Thanks
NB: iparm(36) is zero