Hello, all users,

hello chol-jun,

I think it would be necessary that MD code has the automatic checking

function for the convergence (enough equilibration) as like ab initio

convergence checking. Rather one has to set the simulation time

while this would be a nice to have, this is practically not possible.

in quantum chemistry calculations you can generate an estimate

for how far you are away from the ground state, the equivalent in

classical modelling would be a geometry optimization, _not_ an MD.

to use the term 'convergence' for 'being in equilibrium' is something

very different. since the dimensionality of your phase space is so enormous,

you cannot check everything and one just has to use statistical

estimates and ultimately some sort of 'leap of faith' to decide whether

equilibrium has been reached or not. in principle, you system can look

as if you are equilibrated and still 'find' a different local minimum in

phase space (and thus fall out of equilibrium) at _any_ time.

sufficient long (run 5000000) to reach the equilibration based on the

experience. Is it possible to do this in Lammps?

how long is enough depends a lot on what kind of property you want

to determine. some properties (e.g. radial distribution functions) converge

quickly and are also less dependend on the system size, while others

(e.g. self-diffusion) can seem to be converged after a several nanoseconds

but if you look at different chunks of the same size from a longer trajectory,

they may 'converge' to rather different results. so how would you define

convergence here? in my personal experience it proved to be the best

way to assume equilibrium rather quickly, i.e. when there is no significant

drift in the total energy and then collect data for analysis and then decide

what part to cut off from the beginning of the trajectory as not completely

equilibrated after the fact (by looking at different size parts of the

trajectory

independently).

it might help to have a look or two into your favorite text book on

statistical mechanics for confirmation.

cheers,

axel.