[Pw_forum] bfgs optimization not going anywhere

[Matej Huš]

Thanks for very useful comments.

Yes, vacuum region is indeed too low in this example (whoops, seems like
I've attached an old input file). I also ran a calculation with 10 A of
vacuum without any improvement. Open shell is unnecessary as it quickly
converges to a closed shell solution. I've just wasted some CPU time with
that.

Ectuwfc, ecutrhc are large enough (converged). I believe K points should
also not be the source of a problem.

As for DFT-D2, I admit I am a bit in the dark here. How 'wrong' is it? Or to
put differently, is it more wrong that not using dispersion correction at
all? How about vdw-DF, how do I turn it on and does it perform better? Where
can I find more important about that?

Also, if I change/switch off the dispersion correction, then I will have to
re-optimize all already converged structures (metal, isolated molecules,
adsorbed intermediates), right? Looks rather tedious


Matej



-----Original Message-----
From: pw_forum-bounces at pwscf.org [mailto:pw_forum-bounces at pwscf.org] On
Behalf Of Ari P Seitsonen
Sent: Tuesday, November 24, 2015 11:22 AM
To: PWSCF Forum <pw_forum at pwscf.org>
Subject: Re: [Pw_forum] bfgs optimization not going anywhere


Dear Matej Hus,

   Adding to the previous comments (I think that it is ok to use the k
points, even though you might gain some CPU time by first indeed using
Gamma-only and then increasing, as your cell is quite large), have you
checked the length of your vacuum/lattice vector along the surface normal? 
If I see correctly, you have only about six Ångströms, which is already very
(too) little, plus the DFT-D2 (Grimme-D2) adds quite strong an interaction
through the vacuum at such distances I believe, plus the dipole
correction...

     Greetings from Peaceful Montrouge/France,

        apsi

PS My personal opinion of the vdW-DF* is, yes, they might be better, but
first check the literature about a "suitable empirical choice of the
particular mixture" of the functional, as some of the choices give quite
large lattice constants already (4-5 % larger than experimental, which, of
course, is not necessarily the target, rather a good adsorption geometry and
energy). And often the best agreement is found by tuning the exchange
functional, which naturally should have not nothing to do with the
vdW/London dispersion, which is pure correlation... So much about "ab
initio". ;)  The good news of the day is that it seems that seldomly things
seem to go completely wrong, no matter which choice of the vdW-DF* you take,
but still, I would study a bit the literature first. If any one has a
reference for a good review on the vdW-DF* et co applied to surfaces and
adsorption, I would also be very thankful! :)

-=*=-=*=-=*=-=*=-=*=-=*=-=*=-=*=-=*=-=*=-=*=-=*=-=*=-=*=-=*=-=*=-=*=-=*=-=*=
-
   Ari Paavo Seitsonen / Ari.P.Seitsonen at iki.fi / http://www.iki.fi/~apsi/
   Ecole Normale Supérieure (ENS), Département de Chimie, Paris
   Mobile (F) : +33 789 37 24 25    (CH) : +41 79 71 90 935


On Tue, 24 Nov 2015, Matej Huš wrote:

> 
> Dear PW community
> 
> I have a problem when trying to find optimized geometry of an adsorbed 
> intermediate in "benzene" hydrogenation on Ru surface (see input below).
> 
> I believe I am fairly close to optimised structure since I'm only 
> adding one hydrogen  atom to an already optimised reactant and only 
> slight displacing the adjacent atoms.
> 
>  
> 
> The problem is that BFGS doesn't seem to go anywhere. Energy only 
> fluctuates, which to my understanding is a sign of being close to a
minimum.
> However, forces remain large.
> 
>  
> 
> I've switched to the damp algorithm and the problem persists. I've 
> searched through the forum and the usual approaches of tackling this 
> issue do not seem to work.
> 
>  
> 
> Any ideas? Thank you!
> 
> Matej Hus
> 
> Laboratory of Catalysis and Chemcial Reaction Engineering
> 
> National Institute of Chemistry
> 
> Ljubljana, Slovenia
>