AMBER Archive (2007)

Subject: Re: AMBER: results from TI

From: David Mobley (dmobley_at_gmail.com)
Date: Tue Oct 23 2007 - 12:09:31 CDT

Holly,

> Thanks for your response. The value for the LJ component is -17kcal/mol (I had said -16 before but
> that was wrong). In fact I looked at this and I didn't equilibrate long enough, so this is one of my
> problems. I suspect the reason that it's so negative is because like I was saying the molecule has
> different charges when different charges have been removed (the magnitudes of solvation free
> energies of ions are enormous, eg sodium with one unit of charge is about -77 kcal/mol). So when I
> repeat the calcuations I think I should change the groups being permuted to make them more
> neutral; that way the system will be sure to equilibrate reasonable fast. I'll let you know how this
> turns out.

Oh, but when you say LJ component, you are still changing charges at
the same time, right? I guess this could be a large number.

I still think your problem is most likely to be not having enough
lambda values, but it's hard to be sure -- I haven't done that many
relative free energy calculations. However, my experience has been
that just having large charges is not necessarily a problem (although
I do prefer to keep the system net neutral).

Keep us posted on what you figure out. We have a free energy wiki of
sorts at http://www.alchemistry.org which is still rather preliminary,
but we're trying to collect some "best practices" information there.
If you're up for it, it would be great if you can post any new
insights to the wiki
(http://www.alchemistry.org/index.php/Best_Practices).

Thanks,
David

> Thanks again,
> Holly
>
> > (1) See about adding more lambda points. In my experience even with
> > five point TI for the charging energies one gets significant (i.e.
> > ~0.2-0.5 kcal/mol) integration error even when leaving the net charged
> > fixed (i.e. for the charging free energy of SPC water in SPC water),
> > say roughly a 5-10% error. I could easily imagine that using three
> > point integration for something where the magnitude is more than 20
> > kcal/mol could give you at least 3 kcal/mol of integration error. 9
> > points is also few for the LJ part -- but adding more points won't
> > make as big a difference in the total answer since the net LJ
> > component is usually fairly small and should be similar for these two
> > molecules.
> > (2) Maybe send along your values for the LJ component. One would
> > expect this to be small (~0.5 kcal/mol?) so if it's not, that would be
> > a sign of some sort of problem there.
> >
> > Thanks,
> > David
>
>
>
> --
> Department of Physics, University of Alberta
> Edmonton CANADA
>
> -----------------------------------------------------------------------
> The AMBER Mail Reflector
> To post, send mail to amber_at_scripps.edu
> To unsubscribe, send "unsubscribe amber" to majordomo_at_scripps.edu
>
-----------------------------------------------------------------------
The AMBER Mail Reflector
To post, send mail to amber_at_scripps.edu
To unsubscribe, send "unsubscribe amber" to majordomo_at_scripps.edu