# Re: [AMBER] Wham tolerance

Date: Tue, 19 Nov 2013 10:21:05 -0500

Just for comparison, the default tolerance used in the pymbar library for
MBAR calculations is 10^-11. One can of course define all kinds of
tolerances. Specifically, they check the tolerance on the "maximum relative
change in free energy" between iterations:

max{ [f_i - f_(i-1)] / f_i } < 10^-11

where f is the vector of all (dimensionless) free energies and i is the
iteration index. I don't recall what Grossfield uses as a tolerance, but I
expect it is something similar.

The other major factor here, and one you may or may not have already
considered, is that your answer will also depend on the size/location of
the histogram bins you specify. This is not true of all WHAM
implementations, but it is true of Grossfield's.

Regards,
Brian

On Mon, Nov 18, 2013 at 5:41 PM, Jason Swails <jason.swails.gmail.com>wrote:

> On Mon, Nov 18, 2013 at 5:17 PM, Fabrício Bracht <fabracht1.gmail.com
> >wrote:
>
> > Yes. They are overlapping. The force constant is the lowest that allowed
> me
> > to properly sample high energy regions. The differences in the free
> energy
> > values I get with the different tol values are proportional to the
> amounts
> > of zeros that I use for tol. For example:
> > Tol Value at 0.584444 Angstrom
> > 0.01 6.559799
> > 0.001 7.006187
> >
>
> These are very large tolerance values. I would never use values this
> large.
>
>
> > 0.0001 7.051056
> > 0.00001 7.055565
> > 0.000001 7.056019
> > 0.0000001 7.056064
> > 0.00000001 7.056069
> > 0.000000001 7.056069
> >
>
> This seems much more reasonable. I think my default tolerance is 10^-5 or
> 10^-6, or something to that extent. In any case, your free energies change
> less than 10^-3 kal/mol from tolerance = 10^-4 to 10^-9, which suggests
> that any value in that range is reasonable.
>
> I believe Alan Grossfield's program will also do bootstrapping to estimate
> uncertainties. This is worth doing in my opinion, although if you _really_
> want reliable uncertainty estimates you'll want to use a better estimator
> than WHAM. MBAR (the multistate Bennett acceptance ratio) has been shown
> to be a superior estimator to the free energy than WHAM (and equivalent in
> the limit of infinite sampling), and an advantage of MBAR is that you get
> the uncertainties directly.
>
> Is this normal?
> >
>
> It doesn't look unusual to me.
>
> Good luck,
> Jason
>
> --
> Jason M. Swails
> BioMaPS,
> Rutgers University
> Postdoctoral Researcher
> _______________________________________________
> AMBER mailing list
> AMBER.ambermd.org
> http://lists.ambermd.org/mailman/listinfo/amber
>

```--
Institute for Quantitative Biology
PhD candidate - York Research Group       :     Rutgers, The State
University of New Jersey
University of Minnesota - Twin Cities         :     Center for Integrative
Proteomics Room 308
Department of Chemistry                          :     Piscataway, NJ
08854-8066