Whole system hyperdynamics (or similar acceleration)?

LAMMPS LAMMPS Mailing List Mirror
1

Hello LAMMPS,

Thanks so much for creating this great program, and sharing with us.

I have a system, a crystalline solid in vacuum, and I need to study
the nature of its thermal decomposition i.e., study how atoms desorb,
by increasing the temperature. I'm counting desorption events by
making atoms leave the system in z-direction (I use 'thermo lost
ignore' and z-boundary is non-periodic).

I'm having issues of low desorption event counts, unless I run it for
prohibitively long time, or run it at an unrealistically high
temperature. Which means I could use some kind of accelerated MD
method.

I looked at hyper command, and it comes in two flavors:

- global hyper, where only 1 bond in the system can be given a bias potential.

- local hyper, where multiple bonds (separated by sufficient distance)
can be biased.

However, I'm thinking is it possible to bias all the bonds in the
whole system, so the whole system runs at accelerated time?

I was able to do this manually by re-scaling interatomic potentials,
e.g., by halving all the well-depths of pair-potentials. And it gave
statistically significant results at a much lower temperature. But my
guess is, my approach is not systematic, and I'm not sure how this
halved-well-depth result can be mapped back to a full-well-depth in
terms of temperature, as well as desorption event counts.

I have also looked at tad command, however I'm not sure if tad is a
whole system method. Another issue is that for these methods, the
event has to be detected by the event/displace command or something
similar, but I'm detecting my events by atoms leaving the z-boundary.
(Note: the desorption is due to multiple types of events. And I have
done NEB calcs for some of those types, but I don't know all the types
of events taking part in the desorption).

I highly appreciate any advice, tips, or pointers to relevant
literature, or lammps commands,

Thank you very much,
F

2

Although I have no real experience with this, my hunch is that there should be something possible with parallel tempering and some smart recalling or something.

If you can easily define/identify states that you’re interested in (e.g.,at least one atom this much away from the rest in z-direction) you might also be able to apply weighted ensemble dynamics. Unfortunately that is not possible with just lammps, so you’ll have to do some scripting yourself.

3

Both flavors of hyperdynamics (global and local) accelerate
time for the entire system. A key property your system
needs to have (to be eligible for significant time
acceleration) is a well-defined lowest barrier. The higher
that barrier is and the lower the temp you want to run
at, the higher the boost (time acceleration). See the
papers referenced on the hyperdynamic doc pages.

Steve

4

Thanks so much for the responses. As advised by Steve, I will go ahead
and look at hyperdynamic papers and doc in more detail.

Shafat, thanks for sharing your website and writing great tutorials.
It looks like your desorption/bond-boost work is very closely related
to what I'm trying to do.

Stefan, thanks for the pointers regarding parallel tempering, and
weighted ensemble dynamics. I will look into those methods as well.

Best regards,
F