[gmx-users] g_wham PMF profile

[chris.neale at utoronto.ca]

There is no such thing as over-sampling. There is no problem induced  
by having more sampling in one region except in relation to the fact  
that you now have less sampling in some other region. That is to say,  
if you have a totally converged PMF and then sample the first half of  
the umbrellas for another 100x as long, the PMF will not become  
incorrect.

Please note, though, that there are some possible problems here, but  
they involve one's conclusions and not the sampling or the wham. If,  
for example, you concluded that the error was less or the PMF had more  
features in a region that you had sampled for much longer times, then  
you can see how this conclusion could be an artefact of your method.

Generally, you need to have stronger umbrellas and thus closer  
umbrellas to obtain sufficient overlap in places where your PMF is  
concave down. The more strongly concave down the PMF is, the more you  
require umbrellas with strong force constants in that locale.  
Otherwise, you can introduce massive global PMF errors because the two  
umbrellas on either side of the local maximum fall to either side and  
the overlap becomes very poor.

Chris.

-- original message --

Thank you!

and, is there a problem if I'm over-sampling some regions in the case
where in the other regions I have got a good (not a under-one)
sampling?
This over-sampled region can invalidate my DeltaG value?

best,

Anna