And on top of all these valuable suggestions you may try the desperate
(but often wise in my experience) trick:
electron_maxstep=50~100
scf_must_converge=.false.,
This is very useful when you want to optimize a structure and your
atomic configuration is far from equilibrium (e.g. a surface starting
from a cleaved bulk). But you must carefully check that your system
does not go bananas...
HTH
Giuseppe
Quoting Cococcioni Matteo <[email protected]>:
Dear Roberto,
in addition to what Lorenzo has already suggested, one more thing to
try is to switch on the value of U in small steps. you start from a
small value, e.g. 0.01 eV which should go to convergence as smoothly
as a plain DFT run does, then you read potential and wavefunctions
and start another run at a bit larger value of U, converge it,
restart at a larger value of U and repeat until you reach the value
you want to use.
In many case where convergence is difficult this helps (even though
it may take some steps).
Hth.
Best,
Matteo
On Jul 1, 2018, at 1:45 PM, Lorenzo Paulatto
<[email protected]<mailto:[email protected]>> wrote:
On 06/30/2018 04:28 PM, Roberto Gomes de Aguiar Veiga wrote:
However, when I try to perform calculations for the corresponding
surfaces, they do not converge. I've been using
mixing_mode='local-TF' and descreasing mixing_beta to values as low
as 0.05 for the surface calculations, with no success. Can any of
you with experience in this kind of simulation (DFT+U and surfaces)
share what you usually do to have converged calculations?
Hello Roberto,
DFT+U and surface is a bit tricky, because in principle there is not
guarantee that the value of U used for the bulk is transferable to
the surface atoms. We all do it, but it is not justified.
I also find it very difficult to converge DFT+U for surfaces. One
approach I have used is to defined different species for atoms of
the uppermost layers and for deeper atoms, to be able to define
different starting magnetization. Also, I do not refrain to increase
electrons max steps to 1000 or more, sometimes it just take that
long to converge!
Finally, a trick which I have used is to use starting_ns_eigenvalue
and mixing_fixed_ns to some large value (e.g. 1000) to get a first
converged charge density with ns eigenvalues kept fixed, that do a
restart="from_scratch" with startingpot="file" (this time without
mixing_fixed_ns, or setting it to a small value).
If you notice that the ns eigenvalues printed in output are less
symmetric than the initial magnetization, it may be a good idea to
split the definition of the +U species until they do, this way you
can define different initial ns for them.
hth, it is a lot of cooking
--
Lorenzo Paulatto - Paris
_______________________________________________
users mailing list
[email protected]<mailto:[email protected]>
https://lists.quantum-espresso.org/mailman/listinfo/users
************************************************
Matteo Cococcioni
Theory and Simulation of Materials
École Polytechnique Fédérale de Lausanne
GIUSEPPE MATTIOLI
CNR - ISTITUTO DI STRUTTURA DELLA MATERIA
Via Salaria Km 29,300 - C.P. 10
I-00015 - Monterotondo Scalo (RM)
Mob (*preferred*) +39 373 7305625
Tel + 39 06 90672342 - Fax +39 06 90672316
E-mail: <[email protected]>
_______________________________________________
users mailing list
[email protected]
https://lists.quantum-espresso.org/mailman/listinfo/users
