[DFTB-Plus-User] MD runs terminate without warning

Bálint Aradi balint.aradi at bccms.uni-bremen.de
Tue Nov 5 09:17:00 CET 2013

Dear Vanessa,

> I'm attempting to perform DFTB-MD to simulate the bombardment of a quartz
> surface by an atom using dftb+ 1.2.2. My simulations have exhibited some
> strange behaviour. When I run my input (see attached), which consists of a
> small alpha-quartz substrate and an impinging atom, the simulation ends
> (almost always) without giving any error before the MD simulation finishes.
> Some simulations progress much farther than others, but usually the
> simulations terminate around 10 time steps- sometimes before and sometimes
> much later (well after the impinging atom contacts the substrate). Also, I
> notice that if you run identical input again, the simulation terminates at
> a time step from a previous run. I've tried replacing the
> "DivideAndConquer" option for the "Eigensolver" with "RelativelyRobust",
> but that didn't really change the outcome.

Your system simple does not converge in terms of SCC-iterations. Having
spatially separated unbounded O atom and a covalently bounded system
(SiO2), is tricky, as you have two non-interacting systems with
different Fermi-levels. However, DFTB+ (similar to other QM-simulation
softwares) fills up the system with electrons according one common
Fermi-level. This results in charges going back and forth between the
unbonded O atoms and the covalently bonded SiO2 system. Some of the
steps need more than 1000 SCC-iterations (if they converge as all), and
it is merely luck and numerical inaccuracy, which leads to a sort of

You could try to raise the *electronic* filling temperature by a factor
of 10. I've tried your system with 3000 Kelvin, and obtained very fast
convergence. If it is too high for your system, then start the
bombardement from very close to the surface, so that you have a sort of
covalent bond between bomarding atom and substrate right from the
beginning. Alternatively, separate your MD in two stages: 1) Bombarding
atom reaching the surface (high electronic temperature or alternatiely
non-scc DFTB) 2) Bombarding atom interacting with the surface (SCC-DFTB
with low filling temperature).

>  Also, I'm curious about the fact that
> identical input does not result in the identical output (i.e., failure at
> the same time step and # of iSCC). The velocities for each atom are
> specified in the input, so these don't vary between runs. I'm assuming that
> the initial forces between atoms, which are not specified in the input, are
> changing between runs and that results in the differences in output. Is
> that correct?

No, that should not be the case. Since positions and velocities are
fixed in your input and you are driving an NVE simulation (no
thermostat), the trajectories for the same input should be the same. The
only thing I can think about right now is, that your system is
incredible unstable in terms of SCC convergence, so it was practically
driven by numerical noise to convergence.

  Best regards,


Dr. Bálint Aradi
Bremen Center for Computational Materials Science, University of Bremen

-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 263 bytes
Desc: OpenPGP digital signature
URL: <http://www.dftb-plus.info/pipermail/dftb-plus-user/attachments/20131105/4c4d95fe/attachment.pgp>

More information about the DFTB-Plus-User mailing list