AMBER Archive (2007)

Subject: Re: AMBER: results from TI

From: David Mobley (dmobley_at_gmail.com)
Date: Wed Oct 24 2007 - 09:18:50 CDT

Holly, Thomas,

> I would try to find out how long you need to simulate to get convergence
> with one window and then start more windows with the same simulation time
> instead of longer sampling with fewer windows, but I'm not sure if that
> qualifies as best practice ;-)

I agree with this, especially with TI. With TI, no matter how well you
converge values at any particular set of lambda values, your
integration error places a fundamental limit on accuracy. Obviously,
you DO need to converge the <dV/dl> values -- but adding more lambda
values can in some cases provide a lot more gain than simply running
longer at the ones you have.

The situation is somewhat different with, for example, a method like
Bennett acceptance ratio, where integration error is not a factor. But
that's irrelevant to this discussion, of course.

Thanks,
David

> Regards,
>
> Thomas
>
> Dr. Thomas Steinbrecher
> The Scripps Research Institute
> 10550 N. Torrey Pines Rd.
> San Diego CA 92037, USA
> -----------------------------------------------------------------------
> 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