[gpaw-users] meta-gga question

[Christopher O'Grady]

>Maybe you can clarify -- which meta-GGA's are built directly into GPAW?

Hi Glen,

The MGGA's that live outside of libxc are listed in gpaw/xc/kernel.py:

    'TPSS': 20,
    'M06L': 21,
    'revTPSS': 22

>Right now, the only xc value I found to be accepted was M06LLXC, and
>looking at the source code that refers back to the LibXC M06L
>functional.

I believe you can select any libxc functional using the recipe here,
under the "Exchange-Correlation functional" section:

https://wiki.fysik.dtu.dk/gpaw/documentation/manual.html

By the way, Jess and I tried to run a libxc-TPSS calculation the other
day and it also "blew up" in a manner similar to what you describe.
One could imagine that this is a problem in the new-ish libxc
interface for MGGA, but (as of today) I can run an SCF calculation
with Jess' mBEEF MGGA with the new interface and get the same numbers
as the old interface (to many decimal places) but I've only checked
spin-paired.  This suggests the interface doesn't have a bug.

The only other difference between libxc-mgga and gpaw-mgga that
immediately comes to mind is the handling of various "small density"
cases.  For example, I can make H2 converge (or not) by commenting out
tests in the mBEEF code like this

  if(ABS(tau - tauw) < 1.0e-5)
    {
    dalphadsigma = 0.0;
    dalphadtau = 0.0;
    dalphadd = 0.0; 
    }
  else
    {
    dalphadtau = 1.0/tau_lsda;
    dalphadsigma = -1.0/(tau_lsda*8.0*rho[0]);
    dalphadd = (tauw/rho[0]*tau_lsda - (tau-tauw)*dtau_lsdadd) / POW(tau_lsda, 2.0);
    }

chris