ADMM with Mixing

[Matt W]

Hi Martin,

Unfortunately, you have to use purification_method none at the moment with 
non OT methods. Forces/stress should then be available. Forces with 
purification and variable occupation number need a number of extra terms 
that haven't been implemented (to my knowledge).

Potentially (probably, conceptually certainly) this ADMM method is less 
accurate than when the auxiliary density matrix is forced to be well 
behaved using MO-DIAG or other purification. All I can offer is anecdotal 
evidence that it is reasonable. You may need better quality auxiliary basis 
sets to get as good an approximation to the full hybrid calc.

Please do not compare energies between different ADMM approximations - they 
are effectively running different density functionals.

HTH,

Matt

p.s I'm in A22 three doors down from Jochen's office if you want to 
discuss...

On Monday, March 9, 2015 at 4:39:42 PM UTC, M. Brehm wrote:
>
> Hi,
>
> I would like to perform some calculations (CELL_OPT) on solid state 
> lattices (metal oxides) with hybrid functionals in CP2k with the efficient 
> ADMM framework. However, the "ADMM_PURIFICATION_METHOD MO_DIAG", which is 
> the default, only works together with OT, and for my systems it would be 
> better to use Broyden mixing with Fermi-Dirac smearing instead. How should 
> I setup my calculation? Should I use another ADMM purification scheme (or 
> maybe none at all)?
>
> Best regards,
> Martin
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.cp2k.org/archives/cp2k-user/attachments/20150309/96351fa2/attachment.htm>