[lammps-users] Fix box/relax and minimization

Dear users:

Does fix box/relax combined with a minimization move the atoms during the minimization on top of rescaling them to the new volume? The documentation says that while the volume changes the atoms are just rescaled. When running a minimization with fix box/relax, I noticed that the thermo data from the outer minimization iterations shows the volume changing with each minimization step. However, I am unsure if the atoms are moving during any inner force iteration. In addition, I noticed if I run a second minimization without fix box/relax the potential energy will decrease even more.

Like it has been mentioned in previous postings, would alternating between fix box/relax on and off be the best way to find a minimum energy structure? If so, would it be better to use a small vmax and a more minimizations rather than a larger vmax? I am trying to evaluate a new potential for my system and previous runs with the fix box/relax command shows a tendency for the lattice to expand.

Thanks for your help,

Shawn Coleman
University of Arkansas

yes, the atoms move while the box volume changes. As the fix box/relax
doc page explains, doing a min while changing the volume
is an imprecisely defined mathematical task. LAMMPS tries to
get close to the right answer. But doing a subsequent min may move
the atoms some more, b/c you are starting from a new reference state.


Thanks for your response. Could you help me understand an interesting phenomenon that I’m observing during multiple minimizations of a solid alternating fix box/relax on and off.

I am finding that the minimized structure depends on what I choose to be vmax. I am varying vmax from 1e-4 to 1e-7. I run 400 minimizations alternating fix box/relax on and off with a target P as 0 bar. The stopping criteria for the last 10+ minimizations is the line search equals 0 so I’m assuming that I have found a minimum structure. However, none of the structures have reached 0 bar. I have tried running the simulations using a larger vmax, but the lattice expands greater than 50% of the original structure.

I am still evaluating the potentials for my system so I would appreciate any advice finding the true minimum structure so that I can compare the MD results to literature. If I have a valid potential should the pressure reach the desired 0 bar without expanding unreasonably?

Thanks for your help and advice.

Shawn Coleman
University of Arkansas

I'll let Aidan answer this.


alpha equal zero indicates that the linesearch failed. Sometimes this is due
to discontinuities in the potential energy function. Try using min_modify
line quadratic, which can handle discontinuities better. Ultimately, you
only care about one thing, whether the pressure has relaxed to the target
value or not. Use that to decide what the best choice of vmax is. Let us
know how it works.


Thanks. I forgot to mention that I already was using min_modify line quadratic. I set the target pressure for each minimization to 0 bar; however the minimization stops with the pressure for my system at 395347.83 bar (w/ vmax 1e-6) and 31124.507 bar (w/ vmax 1e-4).

I am researching several different polymorphs of alumina so verifying my potential’s ability to reproduce each minimum structure is important to me. Not only do I want the potential to accurately reproduce each geometry, but I also would like accurate energy and pressure data so I can compare the difference in enthalpy between each phase.

Thank you for your help.

Shawn P Coleman

University of Arkansas
Mechanical Engineering

There might be discontinuities in the energy that are causing this. You
could also experiment with dmax. Finally, you should try adjusting the
initial volume manually to get closer to the final state.