Rigid bodies will slightly deform if a barostat is used

Dear All,

I have been using LAMMPS doing simulations for some systems consisting of rigid bodies. I recently found that using the “fix rigid” commands with a barostat (i.e., fix rigid/nve or rigid/nve/small combined with a Berdenson barostat, or fix rigid/npt, rigid/nph (with or without “/small”)), the rigid bodies will slightly distorted about 0.1% of their length scale. For instant, if I run a simulation for a system consisting of 2-nm-sized rigid dimers in a NPT ensemble (300K, 1atm). The distance between the two beads in a dimer will slightly change between 19.99 to 20.01 Angstrom. I found this happened regardless the shape of the rigid bodies (dimers, polyhedrons, small molecules). The deform of the rigid body does not happen if no barostat is used (e.g., NVE, NVT ensembles)

The distortion is tiny, however, this precision does not meet the requirement for my current study. Therefore, I post this to seek helps. Any comments, suggestions, or answers are welcome. I suspect that the slight shape change of the rigid body is caused by the remapping procedure when applying a barostat (either Berdenson or Nose-Hoover). Does anyone know if this is true? More specifically, is it due to the (low) precision of the FFT used in the remap? If so, is there any way to increase the accuracy? Many thanks for your helps!!!

Best Wishes,
Zhaochuan Fan

Department of Chemistry,
University of Utah

Dear All,

I have been using LAMMPS doing simulations for some systems consisting of
rigid bodies. I recently found that using the "fix rigid" commands with a
barostat (i.e., fix rigid/nve or rigid/nve/small combined with a Berdenson
barostat, or fix rigid/npt, rigid/nph (with or without "/small")), the
rigid bodies will slightly distorted about 0.1% of their length scale. For
instant, if I run a simulation for a system consisting of 2-nm-sized rigid
dimers in a NPT ensemble (300K, 1atm). The distance between the two beads
in a dimer will slightly change between 19.99 to 20.01 Angstrom. I found
this happened regardless the shape of the rigid bodies (dimers,
polyhedrons, small molecules). The deform of the rigid body does not happen
if no barostat is used (e.g., NVE, NVT ensembles)

The distortion is tiny, however, this precision does not meet the
requirement for my current study. Therefore, I post this to seek helps. Any
comments, suggestions, or answers are welcome. I suspect that the slight
shape change of the rigid body is caused by the remapping procedure when
applying a barostat (either Berdenson or Nose-Hoover). Does anyone know if
this is true? More specifically, is it due to the (low) precision of the
FFT used in the remap? If so, is there any way to increase the accuracy?
Many thanks for your helps!!!

​there are no FFTs used when changing the box size or switching from
fractional to regular coordinates and back. where did you get the idea that
FFTs are involved?

​what happens if you run fix rigid like this (or fix press/berendsen):

group none type 999
fix 1 all rigid/npt [...​] dilate none

outside of that, please see: ​http://lammps.sandia.gov/guidelines.html

axel.

Dear Axel,

Many thanks for your reply and helps. I test your way to run fix rigid and the rigid bodies do not deform.

About the FFTs, I must understood wrongly. I saw the barostat codes called “remap” so I looked the “remap.cpp”. I found the “remap_3d_create_plan” in remap.cpp is also used in fft3d.cpp, therefore I thought relocating atoms used FFTs.

Thanks again for your fast response and helps.

Best,
Zhaochuan