# [lammps-users] Max component of global force vector

hi all,

I am trying to relax a crystal of RDX with 3x3x3 unit cells, with an external pressure applied on the simulation box using the fix box/relax command. I am using CG minimisation with a quadratic line-search. The energy tolerance is set to be very small so that LAMMPS will exit only if it hits the force tolerance or max iterations. The statistics after 60,000 iterations is given below:

Minimization stats:
Stopping criterion = max iterations
Energy initial, next-to-last, final =
48521.3928335 46722.5471291 46942.755204
Force two-norm initial, final = 110187 1068.16
Force max component initial, final = 110158 1067.81
Final line search alpha, max atom move = 9.29433e-07 0.000992461
Iterations, force evaluations = 60000 120000

LAMMPS reports the max component of the force to be 1067.81, while when I tried dumping the forces on individual atoms, I found that all the components of forces on all the atoms are less than unity. I would really appreciate it, if someone can explain what I am missing here.

I am attaching the LAMMPS script and data file which were used for this run. Also attached are the log file and the dump at the final iteration which has the forces on individual atoms. I am using January 2010 version of LAMMPS.

in.rdx.test (2.39 KB)

3x3x3cell_hmxcharge.data.rdx (914 KB)

log (53.5 KB)

dumpuw.60000 (382 KB)

I'll let Aidan take a look at this.

Steve

ReaxFF uses an iterative scheme to solve for the charges at each timestep.
The tolerance that you specify (or don't) will limit how tightly lammps can
zero in on the true energy minimum. So as a practical matter, if you want to
converge the forces to very high accuracy, you will need to increase the
accuracy of the charge equilibration scheme.

check the max force again. The minimizer includes enthalpy gradients in its
definition of "forces"

Aidan

Thanks, Aidan for your suggestions. I am using a flexible - Grant
Smith force field (J Phys Chem B 1999, 103, 3570-3575) and not Reaxff.
Hence charge equilibration is not a concern. Sorry that I did not
mention it in the first email.

The explanation about forces makes sense, as the forces which are
dumped are pairwise forces and the "forces" reported by the minimizer
are gradients of the objective function -which includes a work term in
this case. The forces are consistent when i turn off fix box/relax.