committed: better handling of wave functions in Slat_det

[Lucas Wagner]

Hi all, I went ahead and fixed the normalization issues we were having
with very large cells. The determinant is now never explicitly
evaluated in Slat_det, only the log and the sign. Lubos may recall
that he suggested that long time ago. It was mostly implemented, but
I removed the last bottleneck, which seems to help a log on big
systems. You can look how I did it to update Fslat_wf, BCS_wf,
Backflow_wf, and CSlat_wf, should you feel the urge. Otherwise, I
will likely do it at least for CSlat_wf in the near future.

Cheers,

Lucas

[Paul]

I will investigate doing this for FSlat_wf. This has the
implementation of iterative updates for multideterminants, so is
rather critical. However the current implementation is entirely in
terms of the determinant values. -- Paul Kent

[Lucas Wagner]

Great, it turns out that the number of changes necessary is pretty
small. I just don't understand FSlat_wf well enough to make them. By
the way, would you say that FSlat_wf is always better that Slat_wf?
If so, we can replace Slat_wf with the newer version.

Lucas

[Paul]

(i) I didn't test all the paramter optimization variants, so Slat_wf
is safer, at least until we've done more testing

(ii) I think that it would be a good idea to keep the capability of
having all the cofactor matrices present.

On Sep 22, 1:40 pm, Lucas Wagner <lucas.wag...@gmail.com> wrote:
> Great, it turns out that the number of changes necessary is pretty
> small.  I just don't understand FSlat_wf well enough to make them.  By
> the way, would you say that FSlat_wf is always better that Slat_wf?
> If so, we can replace Slat_wf with the newer version.
>
> Lucas
>