BKS potential and non periodic box.

Dear all,

I’m having some issues running a simulation of amorphous SiO2.

I would like to simulate tension and compression test of cylindrical nanopillars;

I built some test system composed of 50k-70k atoms using VMD inorganic builder.

The pillars are cylinders oriented along the z direction, so that there are no
periodic boundary conditions on x and y.

I am using the buck/coul/msm potential with standard BKS parameters, and I am
using kspace_style msm.

What I observe is that the system is initially in a rather high potential energy

configuration. This leads to a rapid heating, so that I need to use lots of minimize
steps or need a very strong damping during the first timesteps, in order to

keep the temperature low (I would want to work at 100-300k).

Furthermore, the resulting system appears to be very “plastic”, whereas some

other test I made using reaxff were much more brittle.

I wonder if I am doing something inherently wrong here. I am attaching the key
parts of the script. Please let me what you think of it, any comment is

appreciated.

Thank you very much,

Alessandro

If potential energy is high and atoms have high velocity (T is high) at the starting of a simulation, then it usually indicates bad dynamics, i.e., bad structure, potential, setup, etc. I suggest you check the atoms at the z boundary and see if they are especially “hot”. If they are, you likely have overlapping (or very close) atoms at the z boundary.

Ray

My 1 1/2 cents. You talk about an amorphous config but where did it come from? No barebones builder can get you such arrangement as the state will depend on the forcefield (FF) among other things. If by any change you are exporting an amorphous configuration from a different FF such as reaxff, be very careful. QEq methods are known for generating charges with low absolute values. I doubt QEq will get the 2.4 and -1.2 charges that BKS employs within its parametrization. Put these numbers on a reaxff-derived surface morphology (no bulk balancing electrostatic forces) and you’ll see reconstruction at least if not atoms flying all around.
Carlos

Dear Carlos, Ray, and all,

thank you again for your time and suggestions.

About your comments:

  1. I verified that there is no initial atom overlap in the periodic boundary, so apparently the configuration was built correctly.

  2. Indeed the charges produced by reaxff were not exactly those expected by BKS so I took care of this.

Unfortunately I made some further efforts but cannot get through this, because I still get “blowups” during both dynamics

and minimizations. I suspect I’m doing something wrong with the forcefields.

My last attempt was this:

  • create the periodic cylinder with VMD and export to lammps format

  • load the data and minimize the system using reaxff, export the relaxed system

  • reload the relaxed system, and reset the charges to the BKS expected values.

  • minimize with BKS

however, after some time steps the minimize fails, showing crazy drops in energy.

I’m attaching an example here:

At this point threads tend to die because your issues are model-related and pretty much all the highly committed people to the forum are developers and mainly focus their efforts on the bug-free character of the software.
Anyways, always hard to tell what could be happening from far thus you should try talking to local colleagues who care about the physics of the problem and consult the lit for more info/hints. I personally don’t know how BKS performs when modeling finite systems/surfaces, but you make sure first you know how to handle and reproduce some of results that BKS yields for bulk silica. Additionally, do not mix forcefields, just generate your configs with the FF or interest. Make sure as well to increase the complexity of your system gradually, by for example, considering smaller systems sizes within your testing procedure. Those are my general comments. If you have covered all these steps already then just go over the procedure once again to ensure you did not miss anything.
Good Luck,
Carlos

At this point threads tend to die because your issues are model-related and
pretty much all the highly committed people to the forum are developers and
mainly focus their efforts on the bug-free character of the software.

perhaps we should start recommending people to post questions about
science to places like http://physics.stackexchange.com/
which seems to be evolving into something that is more focused on
explaining the physics and also looks more beginner compatible?

has anybody on the list here tried a web forum like that or something
else? i also recall seeing somebody setting up a LAMMPS group on
facebook where the discussions were more science related and more
aimed at beginners. from looking at many e-mail discussions it is
evident that especially beginners are not so much capable in the kind
of "batch processing" that e-mail requires and are more comfortable
with small incremental communication as in such more "social" forums.

Anyways, always hard to tell what could be happening from far thus you
should try talking to local colleagues who care about the physics of the
problem and consult the lit for more info/hints. I personally don't know how
BKS performs when modeling finite systems/surfaces, but you make sure first
you know how to handle and reproduce some of results that BKS yields for
bulk silica. Additionally, do not mix forcefields, just generate your
configs with the FF or interest. Make sure as well to increase the
complexity of your system gradually, by for example, considering smaller
systems sizes within your testing procedure. Those are my general comments.
If you have covered all these steps already then just go over the procedure
once again to ensure you did not miss anything.

hmmm... sounds a lot like we need somebody to formulate the "ten
commandments for successful simulations". :wink:

axel.

> At this point threads tend to die because your issues are model-related
and
> pretty much all the highly committed people to the forum are developers
and
> mainly focus their efforts on the bug-free character of the software.

perhaps we should start recommending people to post questions about
science to places like http://physics.stackexchange.com/
which seems to be evolving into something that is more focused on
explaining the physics and also looks more beginner compatible?

has anybody on the list here tried a web forum like that or something
else? i also recall seeing somebody setting up a LAMMPS group on
facebook where the discussions were more science related and more
aimed at beginners. from looking at many e-mail discussions it is
evident that especially beginners are not so much capable in the kind
of "batch processing" that e-mail requires and are more comfortable
with small incremental communication as in such more "social" forums.

Not sure how effective will be to re-route people to another service when
looking for answers.
On the personal side I have no experience posting on such sites but there
is no guarantee that
they will be as responsive as this list tends to be (one big plus of this
forum).
There is also the fact that good quality answers are given here all the
time. No idea again about
the other forum details.

> Anyways, always hard to tell what could be happening from far thus you
> should try talking to local colleagues who care about the physics of the
> problem and consult the lit for more info/hints. I personally don't know
how
> BKS performs when modeling finite systems/surfaces, but you make sure
first
> you know how to handle and reproduce some of results that BKS yields for
> bulk silica. Additionally, do not mix forcefields, just generate your
> configs with the FF or interest. Make sure as well to increase the
> complexity of your system gradually, by for example, considering smaller
> systems sizes within your testing procedure. Those are my general
comments.
> If you have covered all these steps already then just go over the
procedure
> once again to ensure you did not miss anything.

hmmm... sounds a lot like we need somebody to formulate the "ten
commandments for successful simulations". :wink:

Don't know if the "ten commandments" are due but perhaps a section added
to the Lammps FAQ with answers specifically tailoring "most common
mistakes" to be made by beginners and not so newbies. By browsing the
archives one can find questions that repeat themselves over and over.
Yet, don't ask me if people will bother reading it in detail.
Carlos

Dear all,

thanks for the additional advice.

I was eventually able to solve the bigger issue; the problem was quite simple but tricky to catch. Apparently, in the larger samples generated with the VMD inorganic builder, only two atoms, or four, were created very close to each other. This did not happen in the smaller testing samples I used for further testing: it is probably an issue that becomes important when building large structures. I had printed out the RDF but couldn’t see the tiny jump (at r < 1.5 ang) corresponding to the two nearby atoms. Of course this would eventually trigger the BKS blowup or “fusion”. I also tried to relax these samples using different potentials, but this didn’t work at the beginning, simply because I wasn’t relaxing enough. Minimizations would stop due to energy converging, because just one particle pair was far from equilibrium and would not contribute a lot to the whole system.

I eventually checked step by step my system and was able to find the problematic atoms adding (hybrid/overlay) a strong hard-core repulsion to the BKS potential.

I would warn interested people to closely check the structure generated by the inorganic builder plugins. It is interesting to observe that it wasn’t a periodic boundary issue, if anybody wonders. However, once these atoms are dealt with, the structure is quite stable.

I am now trying to replicate some published results and checking whether the non periodic BC give results comparable with those systems. This will be definitely more interesting!

About the general aspect of the model-specific questions on the ML, I’m basically working alone on this. And sometimes it happensI find a problem or issue I cannot solve unless I get to talk to people who have lots of experience and can point out something that’s going wrong. I have read many threads in this list in the past, and learn a lot from the discussions… it’s just that sometimes I make mistakes nontheless :slight_smile:

Therefore, sorry for the noise, and thanks again for all your suggestions/advice/help,
Alessandro

Just for future reference, instead of calculating the RDF
(integral/cumulative quantities are bad descriptors to spot local issues)
you could have just dumped out the forces on the atoms for a quick
comparison. That should have been good enough to spot those few overlapping
guys.
Glad to know your problem is solved,
Carlos

… Or you could have used the “delete_atoms overlap” with an appropriate cutoff value, not that you actually want to delete atoms since it’s SiO2, but just to check if there are any overlapping. Probably the laziest solution! Arthur