Fix move and variable timestep

mkanski · October 14, 2022, 11:46am

Hi,

I noticed that it is impossible to use fix move linear with fix dt/reset. Is there some fundamental issue here or is it just not programmed?

Looking at the source code, I understand why it can’t work as it is, but it should be fairly easy to make it compatible. For example, by changing line:

github.com

lammps/lammps/blob/77740c4f072b9d4cd4a59fc66cfdd3739d2c32bb/src/fix_move.cpp#L544


      
          
          
if (mstyle == LINEAR) {
            for (int i = 0; i < nlocal; i++) {
              if (mask[i] & groupbit) {
                xold[0] = x[i][0];
                xold[1] = x[i][1];
                xold[2] = x[i][2];
          
          
      if (vxflag) {
                  v[i][0] = vx;
                  x[i][0] = xoriginal[i][0] + vx * delta;
                } else if (rmass) {
                  dtfm = dtf / rmass[i];
                  v[i][0] += dtfm * f[i][0];
                  x[i][0] += dtv * v[i][0];
                } else {
                  dtfm = dtf / mass[type[i]];
                  v[i][0] += dtfm * f[i][0];
                  x[i][0] += dtv * v[i][0];
                }

to something like:
x[i][0] += vx*update->dt

akohlmey · October 14, 2022, 12:23pm

I think for the most part, the reason is that fix move predates fix dt/reset and thus far nobody wanted to use both at the same time or people have been using the alternative approach given below.

There is a subtle difference, but a) what you propose to do can already be implemented with fix nve and fix setforce 0 0 0 or similar and b) there is a difference in the exact position at the end of the run because a) (and your proposed change) will accumulate errors from floating-point truncation.

mkanski · October 14, 2022, 12:31pm

You are absolutely right! I can’t believe I have missed this option with fix setforce. Thank you for your help. (: