I would like to optimize a structure given an initial guess structure. Is it a good idea to try to perform simulated annealing for this purpose ? Are there better approaches ?
I have made an attempt to perform simulated annealing. Basic steps I used are:
(i) I created a supercell, say 4x4x4 of the structure. Then I used the boundary condition on the supercell as “f f f”.
(ii) “minimize 1.0e-4 1.0e-6 100 1000” on the guess structure.
(iii) Then I used a series of the following steps:
(A) fix 1 all nve
fix 2 all langevin 0.0 50.0 0.1 48278
run for 2.5 ps
(B) unfix 1
fix 1 all nve
run for 100 ps
This way I would heat up from 0 - 500K and then cool down the system in steps. Is it a correct approach.
Should I do the above using NPT or NVT instead for heating, cooling the system ?
I have another problem: even if I change mass of hydrogen to 3 and use a timestep of 0.1 fs, I have often ended up loosing atoms during the course of simulation after some heating.
I would appreciate some suggestion/discussion and/or some suggestion of good literature (papers, tutorials, web pages, books, etc) to help me learn how other people have dealt this issue ?