2014-01-26 Jakub Krajniak <[email protected]>: > On Sunday, 26 January 2014 21:02:58 UTC+1, Christoph Junghans wrote: >> >> 2014-01-26 Jakub Krajniak <[email protected]>: >> >> (...) >> >> >> >> > And the convergence, I have add <post_add>convergence</post_add> so >> >> > that >> >> > I >> >> > can measure the convergence. >> >> > 4. What should be the optimal case for the convergence? >> >> > What I am getting: (step_001: ~300, step_002: ~232, step_003: ~133) >> >> > and >> >> > then >> >> > suddenly for step_004 it is ~8000. >> >> > There was one step where the convergence was around 90.. but I run it >> >> > for >> >> > 100 steps and I don't see convergence at all. >> >> > (I used the <initial_configuration>laststep option). >> >> It seems something went wrong here, have a look at the different >> >> distributions to see if one of them is completely off. >> >> If so, you might need to scale the update and/or introduce an update >> >> cycle in do_potential. >> >> >> > >> > I have to ask about one more thing. I have noticed that when I am doing >> > the >> > cg simulation >> > in NVE then the differences between target and new rdf are very small >> > (well >> > question is >> > how to judge it, but it is around ~0.1). On the other hand the >> > temperature >> > in such >> > simulation is very huge (I am running the atomistic one in in 298K with >> > 1.0 >> > pressure). >> > If I try to turn on the thermostat to keep the temperature in average >> > constant then the differences >> > in rdf are not stable and well again does not lead to some convergences. >> > What should be a correct approach? I saw in the tutorials that the cg >> > simulation are done >> > by using the stochastic dynamics in the desire temperature. I don't >> > understand then why I got so strange results. >> The theory behind Boltzmann inversion assumes a Boltzmann distribution >> (NVT). >> Which thermostat are you using? For Espresso++ Langevin thermostat >> should to the same as stochastic dynamics. > > > Actually I have moved to Gromacs to not complicate too much. When I look on > the > temperature, it is kept correctly during the simulation (with sd > integrator). > > The strange thing is with rdfs. In the attachment there is a tgt and .new > A-A.dist > from step_001. The shapes seem similar but the y-values are different. In > the next > steps the rdfs look totally different (and the .conv value rise enormous). 1.) Your target distribution is really noisy, which will lead to a very noisy initial potential, maybe a bigger spacing would help. 2.) It seems the first update makes the A-A potential too attractive, hence you get the really high peak at x=0.42, maybe you will need to scale the update with a factor like 0.1 (postupd scale).
Also it is not uncommon that the convergence take a couple of iterations until it starts to decrease continuously. Christoph > > Jakub > > -- > You received this message because you are subscribed to the Google Groups > "votca" group. > To unsubscribe from this group and stop receiving emails from it, send an > email to [email protected]. > To post to this group, send email to [email protected]. > Visit this group at http://groups.google.com/group/votca. > For more options, visit https://groups.google.com/groups/opt_out. -- Christoph Junghans Web: http://www.compphys.de -- You received this message because you are subscribed to the Google Groups "votca" group. To unsubscribe from this group and stop receiving emails from it, send an email to [email protected]. To post to this group, send email to [email protected]. Visit this group at http://groups.google.com/group/votca. For more options, visit https://groups.google.com/groups/opt_out.
