<div><BR> Hi David,<BR> First, thank you very much for your elaborate answer to my question.<BR> Your fast response has given me a little time to think about the info<BR> you supplied.</div> <div><BR>Quoting David Mobley <<A href="mailto:dmobley@gmail.com">dmobley@gmail.com</A>>:</div> <div>> Soren,<BR>> <BR>> > Given a certain amount of time and computer power available, is it better<BR>> > in general to do short simulations for many lambda values or to do long<BR>> > simulations for fewer lambda values?<BR>> <BR>> This is a really good question, and one that is hard to answer<BR>> generally (that is, the answer is probably somewhat system-dependent).<BR>> Here are a couple considerations:<BR>> <BR>> 1) You need enough lambda values to ensure good phase space overlap<BR>> from one lambda value to the next, otherwise the results you get will<BR>> be meaningless (all of the FEP/TI type
expressions require sufficient<BR>> overlap between the Hamiltonians of interest).</div> <div> </div> <div> So if I understand correctly, there is no way, apart from from the indirect <BR> method you describe below, that can say how big a part of the sampled <BR> configurations that overlap?</div> <div> </div> <div>> Bennett Acceptance<BR>> Ratio (a type of FEP) is somewhat more efficient and may require<BR>> slightly fewer lambda values. It is probably nontrivial to figure out<BR>> how many lambda values is "enough", either. The approach I would<BR>> typically take for a particular problem is to start with lots of<BR>> lambda values for a "typical" problem of that type, and compute the<BR>> "correct" answer, and then reduce the number of lambda values until<BR>> the answer starts to deteriorate, then add a couple back in and use<BR>> that number.<BR>> 2) Regardless of whether you have enough lambda
values, if your<BR>> simulations are not long enough, your results will be meaningless.<BR>> Your simulations should probably be at least 10x the typical<BR>> correlation time for your system (for your observable of interest).<BR>> For example, if you are calculating binding free energies of a ligand,<BR>> and the ligand has two different sub-conformations it can occupy in<BR>> the binding site, and the timescale for switching between these is 100<BR>> ps, you would need to run at least 1 ns at each lambda value to ensure<BR>> you have time to sample each conformation a sufficient number of<BR>> times. In the extreme case, suppose you started with only one of these<BR>> conformations and ran 50 ps, observing no swaps -- instead of<BR>> computing the binding free energy of the ligand, you would compute<BR>> what the binding free energy would have been if the ligand could only<BR>> occupy one conformation in the binding site. This could
be not at all<BR>> related to the correct binding free energy.<BR>> <BR>> Anyway, that all basically boils down to this: If the transformation<BR>> you are doing is "hard" (like turning off LJ interactions for some<BR>> atoms in your system), (1) dictates that you will need a relatively<BR>> large number of lambda values, since the phase space changes a great<BR>> deal,</div> <div> </div> <div> What is it that makes the phase space change a great deal for a "hard"<BR> transformation? (And why do you consider it particularly "hard"?)<BR> In other words, why would turning of LJ interactions be more difficult than<BR> other transformations? And, forgive my ignorance which other kinds of<BR> transformations could such other transformations be?<BR> <BR> All the best regards, and thank you again for enlighting me on this,<BR> -Søren</div> <div> </div> <div>> and if the system you are looking at has
long correlation times,<BR>> (2) dictates that you will need long simulations. You need to do some<BR>> investigation to see what your problem is like. I would especially<BR>> recommend looking at correlation times. Maybe run out one or two<BR>> really long MD trajectories to get an idea of some of the relevant<BR>> timescales in your system.<BR>> <BR>> I would also recommend not thinking of it so much as an optimization<BR>> problem of, "How do I best use X amount of computer time?" but, "How<BR>> much computer time will I have to spend where to get my statistical<BR>> uncertainty down to X in the most efficient way?" Depending on your<BR>> system, (1) and (2) may dictate that the amount of computer time you<BR>> need to spend is more than you actually want to spend. Then you have<BR>> to decide, basically, how big of error bars you are willing to<BR>> tolerate.<BR>> <BR>> David Mobley<BR>> UCSF<BR>> ><BR>>
> Thanks for all the good advices on the list,<BR>> > Soren<BR>> ><BR>> > _______________________________________________<BR>> > gmx-users mailing list <A href="mailto:gmx-users@gromacs.org">gmx-users@gromacs.org</A><BR>> <A href="http://www.gromacs.org/mailman/listinfo/gmx-users">http://www.gromacs.org/mailman/listinfo/gmx-users</A><BR>> > Please don't post (un)subscribe requests to the list. Use the<BR>> > www interface or send it to <A href="mailto:gmx-users-request@gromacs.org">gmx-users-request@gromacs.org</A>.<BR>> > Can't post? Read<BR>> > <A href="http://www.gromacs.org/mailing_lists/users.php">http://www.gromacs.org/mailing_lists/users.php</A><BR>> ><BR>> ><BR>> _______________________________________________<BR>> gmx-users mailing list <A href="mailto:gmx-users@gromacs.org">gmx-users@gromacs.org</A><BR>> <A
href="http://www.gromacs.org/mailman/listinfo/gmx-users">http://www.gromacs.org/mailman/listinfo/gmx-users</A><BR>> Please don't post (un)subscribe requests to the list. Use the <BR>> www interface or send it to <A href="mailto:gmx-users-request@gromacs.org">gmx-users-request@gromacs.org</A>.<BR>> Can't post? Read <A href="http://www.gromacs.org/mailing_lists/users.php">http://www.gromacs.org/mailing_lists/users.php</A><BR>> </div>