[QE-users] On the use of the modified Becke-Johnson (TB09) functional

Tue Nov 12 15:20:58 CET 2019

Dear Fabio,
you are right, I forgot to switch to the f03 version the call to the mgga
functional routine.
I do not guarantee anything about the results, but now you can properly set
the c parameter of TB09 as I told in the previous email. At line 24 I have
put zero as default value for c, but with no particular reason.
I don't know how old is you version of QE, but this file has been taken and
modified from the latest version in the develop branch.
Hope it helps,
Fabrizio

On Fri, Nov 8, 2019 at 5:44 PM Fabio Costa <fabiocosta4 at hotmail.com> wrote:

> Dear Fabrizio
>
> Thank you for your quick reply
>
> I followed your advice, and now I'm obtaining strange results. I'm running
> some tests with silicon, and to begin with, the scf run with mBJ reports a
> negative gap. In the pw.x output, the highest occupied level is higher in
> energy than the lowest unoccupied level, resulting on a "gap" of -0.972 eV.
> Just to compare, the same calculation done with PBE results on an expected
> result of ~0.8 eV gap. Going further, I plotted the density of states for
> both cases, and for my surprise, the result obtained with mBJ has no gap at
> all, as can be seen from the attached file.
>
> Also, I'm not sure if I proceeded in the right way, but I tried to change
> the value of the parameter "c" by changing it at line 24 from
> xc_mgga_drivers.f90, and recompiling pw.x . After doing this, the result
> was exactly the same, so my guess is that maybe I misunderstood how to
> change it.
>
> Sorry for the trouble, and thank you for your patience
> Fábio Costa
> ------------------------------
> *De:* users <users-bounces at lists.quantum-espresso.org> em nome de
> Fabrizio Ferrari <ferrariruffino.fz at gmail.com>
> *Enviado:* sexta-feira, 8 de novembro de 2019 11:51
> *Para:* Quantum ESPRESSO users Forum <users at lists.quantum-espresso.org>
> *Assunto:* Re: [QE-users] On the use of the modified Becke-Johnson (TB09)
> functional
>
> Dear Fabio,
> I added some lines in the attached file 'Modules/xc_mgga_drivers.f90' so
> that, by replacing it with the original one, you should be able to set the
> c parameter in tb09. You can do it directly into that module by setting the
> initial value of cc_param or in any part of the program by calling the
> routine 'set_cc_for_tb09(xx)' (but in case you may need to adjust the
> module dependencies).
> Before the compilation you need to add the '-lxcf03' in LD_LIBS in the
> make.inc file
> Hope it helps,
> Fabrizio
>
> On Wed, Nov 6, 2019 at 8:36 PM Fabio Costa <fabiocosta4 at hotmail.com>
> wrote:
>
> Dear users and developers
>
> After some more attempts to use the TB09 functional, I ended up finding
> this previous post:
> https://lists.quantum-espresso.org/pipermail/users/2019-April/042619.html ,
> and after changing v_of_rho.f90, the code is now able to end without any
> error.
>
> In my attempts, I'm performing the calculations as an initial scf run with
> a PBE pseudopotencial, and then another scf run, now with restart_mode =
> 'restart' and input_dft = 'MGGA_X_TB09'. By following this procedure, as
> described in E. Germaneau's paper, the calculations do converge.
>
> The issue now is that the gap energies I'm finding are very large. For
> example, for silicon, the calculations result in a 2.8 eV gap, larger even
> than the results presented in the paper, where the values ranged between
> 1.9 to 2.3. Despite the large gaps, the band structure and DOS agree well
> with the ones obtained with PBE.
>
> Is there any way to improve these results. My first guess is to adjust the
> "c" parameter in the mBJ functional. After some searching through the
> routines and Libxc files, I'm still clueless on how to proceed on this
> matter.
>
> Any further suggestions will be much appreciated
>
> Thank you
> Fábio Costa
>
>
>
> ------------------------------
> *De:* users <users-bounces at lists.quantum-espresso.org> em nome de Fabio
> Costa <fabiocosta4 at hotmail.com>
> *Enviado:* segunda-feira, 14 de outubro de 2019 13:10
> *Para:* users at lists.quantum-espresso.org <users at lists.quantum-espresso.org
> >
> *Assunto:* [QE-users] On the use of the modified Becke-Johnson (TB09)
> functional
>
> Dear all
>
> I'm having some difficulties to work with the TB09 functional. After some
> reading, I found in  some old posts that this can be done by compiling QE
> with Libxc, and by adding the line 'input_dft='tb09'' in the pw.x input.
>
> When I try to use this functional, the scf calculation always ends like
> this:
>
>      total energy              =              NaN Ry     Harris-Foulkes
> estimate   =     -11.24358313 Ry     estimated scf accuracy    <
> 0.01858366 Ry     iteration # 18     ecut=   120.00 Ry     beta= 0.10
>  Davidson diagonalization with
> overlap %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
>    Error in routine cdiaghg (7):     eigenvectors failed to
> converge %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
>
> What draws my attention is that always in the last cycle before the error
> the total energy is printed as NaN. The parameters to the calculation seem
> to be fine, as it ends smoothly without the input_dft line.
>
> By searching in the QE user guide, there is a reference to the work of E.
> Germaneau, which lead me to his paper (DOI: 10.1016/j.cpc.2013.02.020
> <http://dx.doi.org/10.1016/j.cpc.2013.02.020>), where the results of his
> implementation are presented. Even though, it is unclear to me how to use
> the functional correctly.
>
> Thanks for any assistance
> Fábio Costa
> _______________________________________________
> Quantum ESPRESSO is supported by MaX (www.max-centre.eu/quantum-espresso)
> users mailing list users at lists.quantum-espresso.org
> https://lists.quantum-espresso.org/mailman/listinfo/users
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.quantum-espresso.org/pipermail/users/attachments/20191112/503dcd6e/attachment.html>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: xc_mgga_drivers.f90
Type: text/x-fortran
Size: 14187 bytes
Desc: not available
URL: <http://lists.quantum-espresso.org/pipermail/users/attachments/20191112/503dcd6e/attachment.bin>