[gmx-users] gmx bar histogram data

[Eric Smoll]

Hello GROMACS users,

I am using the "gmx bar" tool in GROMACS 5.1.2 and am interested in
understanding the relevance of the histogram data output by the -oh flag.
Dr. Lemkul provides a very helpful reference
<http://www.sciencedirect.com/science/article/pii/0021999176900784?via%3Dihub>
and a summary
<http://www.bevanlab.biochem.vt.edu/Pages/Personal/justin/gmx-tutorials/free_energy/07_analysis.html>
which states:

"in order to obtain a reliable estimate of ΔG, we need adequate sampling
within each window (based on ∂H/∂λ distributions) and between neighboring
windows (ΔH distributions). The significant overlap in the distributions of
the data shown here and the low error estimate printed by gmx bar indicate
that we have achieved this goal. If your own calculations give you
non-overlapping histograms, resulting in large error estimates, then you
likely either need longer simulations, more λ states, or better parameters
for whatever molecule it is you are trying to transform."

When using the (default) mdp setting calc_lambda_neighbours=1, I assume
that the only overlap that matters is immediate-neighbour histogram
overlap. For example, when examining deltaH histograms that track the
"forward difference" lambda[i+1]-lambda[i], it only needs to show good
overlap with the lambda[i+2]-lambda[i+1] and lambda[i]-lambda[i-1]
histograms. Am I correct?

Best,
Eric