bug in cp_fm_invert ?

flo fsch... at pci.uzh.ch
Wed Nov 24 08:55:09 UTC 2010


Hi Hanning,

I just noticed, when I wrote the subroutine, I committed the wrong
description for the subroutine. I will update this soon. Indeed the
sign can be wrong, as it depends on the entries in ipivot. As far as I
know CP2K uses the determinant only for computing the electronic
coupling matrix elements and here it is done by calling the pure
serial routine in mathlib.f, where it is corrected. In all other cases
the sign of the determinant does not matter.
In principle it is possible to correct this bug according to the way
it is done in mathlib.f. The point why it hasn't been done yet is,
that it wasn't needed and involves quite a bit of thinking about the
transformation between local positions in ipivot and the global
positions.

Cheers
Flo


More information about the CP2K-user mailing list