[Pw_forum] pw 5.1.1 odd results in vc-relax

Alberto Otero de la Roza alberto at carbono.quimica.uniovi.es
Mon Feb 9 21:07:05 CET 2015

Hi all,

This is a copy of an e-mail exchange between Carlo and I that
addresses some of the questions in this thread. Sorry I missed it the
first time!

> I'm trying to do periodic PW calculations on solid structures using Quantum
> Espresso 5.1.1 with pbesol - PAW pseudopotentials on some systems that can
> contain C H N I Br Cl F atoms characterized by the presence of halogen
> bonds (N ... I).
> Since it is reported that these parameters "[fix] and [old]", and "Can be
> re-parametrized on demand", I'm wondering whether this is necessary (should
> I use different a1/a2 parameters?), and/or you can give me an advice.
> The xdm + pbesol seems to overestimate the attractive forces, while
> pbesol underestimate them.

PBEsol is not a particularly good functional when combined with
dispersion corrections (or at least with XDM). I suspect the reason is
that PBEsol was designed with the purpose of nailing the equilibrium
cell parameters by sacrificing accuracy in other departments (like,
for instance, cohesive energies). This results in a functional that is
more binding than PBE, which is not good news when you consider
non-covalent interactions. Particularly, donor-acceptor interactions
like halogen bonds. You can get an idea of the accuracy of PBEsol by
checking the MAPD in the parametrization fit on gatsby (18.9%).

The "parametrized on demand" bit on the webpage is not very well
phrased. The parameters marked with the [old] tag were obtained with a
previous implementation of XDM in QE, in which the atomic volumes were
calculated in a different way. When I wrote the XDM bit that is
currently in the QE distribution, I reparametrized PW86PBE, B86BPBE,
PBE, and BLYP, but didn't bother with the other two because I figured
that they would not be very useful. The [fix] tag means that I had to
force the a1 to be zero because the a1/a2 parametrization would find a
minimum MAPD at negative a1, which is unphysical. This is a clear sign
that things are not going well with the base functional.

If you need those a1/a2, I could easily run the parametrization calcs
for you and find them, although I don't expect things will improve
much with the new parameters. My advice is, if possible, to use one of
the XDM-corrected functionals that work better for non-covalent
interactions, such as PW86PBE (11.7% MAPD) or B86BPBE (11.8%).

> My first structure optimized with a smaller volume (1685 a.u.^3) than the
> real one (1835 a.u.^3) (if you wish I can send the input file for pw.x).
> I've been used the a1 and a2 parameters shown in the web pages for pbesol
> (a1=0.0, a2= 4.1503).

In any case, halogen bonds are a tough cookie in a periodic
solid. This is because all dispersion-corrected GGAs overbind halogen
bonds--both in terms of intermolecular distance and interaction
strength. The reason is delocalization error, as explained here:


The article shows that all GGAs present delocalization error by about
the same amount, and that halogen bonds can only be correctly modeled
(or rather, modeled at the same accuracy level as other non-covalent
interactions) if you use a dispersion-corrected half-and-half
functional. This is naturally a difficult thing to do in QE.

So if N...I non-covalent contacts are part of your crystal framework,
I'm not surprised at all that PBEsol-XDM grossly underestimates the
volume. I would suggest to go with one of the functionals that work
better for NCIs and see how that goes.

> Besides the structures, I 'm interested in phonons and Solid State NMR
> (since our group is doing SSNMR). I know that I would need to take in
> considerations the relativistic effects (i.e. ZORA), but at the moment this
> kind of calculation in not available,

The reason I didn't implement the nspin > 2 case is that I'm not sure
how to do the bit about the kinetic energy density, and I didn't have
much use for that bit at the time. If this problem can be
circumvented, then the implementation should be straightforward, but I
have no idea how to go about it off the top of my head.



Dr. Alberto Otero de la Roza
National Institute for Nanotechnology,
National Research Council of Canada,
11421 Saskatchewan Drive, Edmonton,
Alberta T6G 2M9, Canada

More information about the users mailing list