[gmx-users] Re: gmx 4.6.1, Expanded ensemble: weird balancing factors

[Michael Shirts]

That's a good question.  My understanding and experience is the
difference in required overlap in replica exchange and expanded
ensemble methods is not significant.  There are probably some more
detailed studies out there. Expanded ensemble can be somewhat more
efficient (see a paper by Park and Pande) But you really do want to
have at least 30% of the states moving each time in any case.

However, you DO need less overlap to get fairly good free energy
estimates.  I.e. the overlap for BAR to work fairly well is lower than
the what you need for good mixing.  That's more anecdotal -- I'd have
to dig a bit to get you good results on that . . .

On Tue, Jul 16, 2013 at 3:19 PM, Dejun Lin <dejun.lin at gmail.com> wrote:
> Hi Michael,
>
> Thanks for the reply. Just a quick follow-up. Do you think the overlap of
> energy histogram between different lambdas matter for lambda-dynamics in
> general? (Maybe not in this particular case or in the case of the tutorial
> you just posted.) I wonder if we need as many intermediate lambdas as in
> the case of replica exchange since the weights compensate the difference in
> energy.
>
> Thanks,
> Dejun
>
>
> 2013/7/16 Michael Shirts <mrshirts at gmail.com>
>
>> Hi, all-
>>
>> This not a problem with W-L, but is instead something that is wrong
>> with a particular combination of mdp options that are not working for
>> expanded ensemble simulations.  W-L can equilibrate to incorrect
>> distributions because it decreases the weights too fast (more on that
>> later), but that is not the problem here. The option wl-oneovert
>> allows switching to a slower scaling that is more likely to converge.
>>
>> The reason that it is not a W-L problem is that after the WL weights
>> are equilibrated, it is equilibrium sampling in the expanded ensemble.
>>  If the W-L equilibration was a problem, then the distribution of
>> states would not be flat.  They are flat, so therefore the expanded
>> ensemble dynamics are wrong.
>>
>> I have example expanded ensemble setups.
>>
>>
>> http://www.alchemistry.org/wiki/GROMACS_4.6_example:_Ethanol_solvation_with_expanded_ensemble
>>
>> These mdp files should work. Note that you should be able to swap out
>> any top, and it will still work.  If you get CORRECT results (and it
>> should take just a few ns to see) with these tops, then I will go
>> through and try to figure out which differences are causing the
>> problems.
>>
>> Thanks for the patience while we test this new functionality.
>> Frequently when one puts new code in the hands of others it breaks in
>> ways that were not foreseen!
>>
>> On Tue, Jul 16, 2013 at 12:53 PM, Dejun Lin <dejun.lin at gmail.com> wrote:
>> > I have a similar question here. Can anyone tell if the Wang-Landau
>> algorithm
>> > in lambda space is robust in that it doesn't depend on the choice of the
>> > convergence factor (weight-equil-wl-delta), the flatness of the histogram
>> > (wl-ratio) and/or the frequency of trial move (nstexpanded), especially
>> in
>> > the case of barely overlapping energy distribution corresponding to
>> > different lambdas? I can imagine in the extreme case, with only 2
>> lambdas (
>> > l = 1 or 0), the gap between the 2 energy distributions might be so large
>> > such that for most of the time, trial moves from the "center" of
>> > distribution 1 would frequently end up in the "tail" of distribution 0.
>> In
>> > this case, the Wang-Landau weights would be biased if there's not enough
>> > time for the system to relax back to equilibrium.
>> >
>> > Thanks,
>> > Dejun
>> >
>> >
>> >
>> > --
>> > View this message in context:
>> http://gromacs.5086.x6.nabble.com/gmx-4-6-1-Expanded-ensemble-weird-balancing-factors-tp5007681p5009892.html
>> > Sent from the GROMACS Users Forum mailing list archive at Nabble.com.
>> > --
>> > gmx-users mailing list    gmx-users at gromacs.org
>> > http://lists.gromacs.org/mailman/listinfo/gmx-users
>> > * Please search the archive at
>> http://www.gromacs.org/Support/Mailing_Lists/Search before posting!
>> > * Please don't post (un)subscribe requests to the list. Use the
>> > www interface or send it to gmx-users-request at gromacs.org.
>> > * Can't post? Read http://www.gromacs.org/Support/Mailing_Lists
>> --
>> gmx-users mailing list    gmx-users at gromacs.org
>> http://lists.gromacs.org/mailman/listinfo/gmx-users
>> * Please search the archive at
>> http://www.gromacs.org/Support/Mailing_Lists/Search before posting!
>> * Please don't post (un)subscribe requests to the list. Use the
>> www interface or send it to gmx-users-request at gromacs.org.
>> * Can't post? Read http://www.gromacs.org/Support/Mailing_Lists
>>
> --
> gmx-users mailing list    gmx-users at gromacs.org
> http://lists.gromacs.org/mailman/listinfo/gmx-users
> * Please search the archive at http://www.gromacs.org/Support/Mailing_Lists/Search before posting!
> * Please don't post (un)subscribe requests to the list. Use the
> www interface or send it to gmx-users-request at gromacs.org.
> * Can't post? Read http://www.gromacs.org/Support/Mailing_Lists