1. Calculation:
Correct solution: see e.g. Tommi Järvis file
Calculating correction factor: script calccorr:
3 -2.69845 4 -1.25671 5 -0.486338 6 -0.173105 7 -0.058639 8 -0.0191903 9 -0.00611745 10 -0.00190965 11 -0.000585955 12 -0.0001772212.
Generate large enough cell to handle 12 Å cutoff:
~/md/structure/makeFCC 3.62 8 8 8 > atoms.in
The run with different cutoffs and add correction:
Epot Vcorr 3.0 -1.50932 -2.69845 -4.20777 4.0 -2.11530 -1.25671 -3.37201 5.0 -3.03515 -0.486338 -3.52149 6.0 -3.39088 -0.173105 -3.56399 7.0 -3.49893 -0.058639 -3.55757 8.0 -3.52213 -0.0191903 -3.54132 9.0 -3.53521 -0.00611745 -3.54133 10.0 -3.54069 -0.00190965 -3.5426 11.0 -3.54154 -0.000585955 -3.54213 12.0 -3.54203 -0.000177221 -3.54221Tolerance of 0.1 eV reached by 7.0 Å without correction, and at 5.0 with correction.
Problems: none really in bulk, but at surface would predict atoms have Epot even in vacuum. Cutoff of course causes its own problems: jumps in Epot if atoms cross the cutoff distance. And using Vcorr does not help with this at all, so that should not be an excuse for shortening the cutoff in finite T simulations.
If density not correct (e.g. nonhomogeneous liquid) correction factor also misleading
Plot: epot.ps
3. About stable up to deltat 10 fs, blows up for deltat >= 30 fs.
./mdmorse | & tee out.10fs
xgr -pick 1 ec -c 2 6 out.*
At 10 fs still about stable, at 30 fs blows up like mad.