gmx wham [-ix [<.dat>]] [-if [<.dat>]] [-it [<.dat>]] [-is [<.dat>]] [-iiact [<.dat>]] [-tab [<.dat>]] [-o [<.xvg>]] [-hist [<.xvg>]] [-oiact [<.xvg>]] [-bsres [<.xvg>]] [-bsprof [<.xvg>]] [-xvg <enum>] [-min <real>] [-max <real>] [-[no]auto] [-bins <int>] [-temp <real>] [-tol <real>] [-[no]v] [-b <real>] [-e <real>] [-dt <real>] [-[no]histonly] [-[no]boundsonly] [-[no]log] [-unit <enum>] [-zprof0 <real>] [-[no]cycl] [-[no]sym] [-[no]ac] [-acsig <real>] [-ac-trestart <real>] [-nBootstrap <int>] [-bs-method <enum>] [-bs-tau <real>] [-bs-seed <int>] [-histbs-block <int>] [-[no]vbs]
gmx wham is an analysis program that implements the Weighted
Histogram Analysis Method (WHAM). It is intended to analyze
output files generated by umbrella sampling simulations to
compute a potential of mean force (PMF).
gmx wham is currently not fully up to date. It only supports pull setups
where the first pull coordinate(s) is/are umbrella pull coordinates
and, if multiple coordinates need to be analyzed, all used the same
geometry and dimensions. In most cases this is not an issue.
At present, three input modes are supported.
-it, the user provides a file which contains the file names of the umbrella simulation run-input files (.tpr files), AND, with option
-ix, a file which contains file names of the pullx
mdrunoutput files. The .tpr and pullx files must be in corresponding order, i.e. the first .tpr created the first pullx, etc.
Same as the previous input mode, except that the user provides the pull force output file names (
pullf.xvg) with option
-if. From the pull force the position in the umbrella potential is computed. This does not work with tabulated umbrella potentials.
By default, all pull coordinates found in all pullx/pullf files are used in WHAM. If
some of the pull coordinates should be used, a pull coordinate selection file (option
be provided. The selection file must contain one line for each tpr file in tpr-files.dat.
Each of these lines must contain one digit (0 or 1) for each pull coordinate in the tpr
Here, 1 indicates that the pull coordinate is used in WHAM, and 0 means it is omitted.
If you have three tpr files, each containing 4 pull coordinates, but only pull
coordinates 1 and 2 should be
used, coordsel.dat looks like this:
1 1 0 0 1 1 0 0 1 1 0 0
By default, the output files are:
``-o`` PMF output file ``-hist`` Histograms output file
Always check whether the histograms sufficiently overlap.
The umbrella potential is assumed to be harmonic and the force constants are
read from the .tpr files. If a non-harmonic umbrella force
a tabulated potential can be provided with
-binsNumber of bins used in analysis
-tempTemperature in the simulations
-tolStop iteration if profile (probability) changed less than tolerance
-autoAutomatic determination of boundaries
-min,-maxBoundaries of the profile
The data points that are used to compute the profile
can be restricted with options
-b to ensure sufficient equilibration in each
-log (default) the profile is written in energy units, otherwise
-nolog) as probability. The unit can be specified with
With energy output, the energy in the first bin is defined to be zero.
If you want the free energy at a different
position to be zero, set
-zprof0 (useful with bootstrapping, see below).
For cyclic or periodic reaction coordinates (dihedral angle, channel PMF
without osmotic gradient), the option
-cycl is useful.
gmx wham will make use of the
periodicity of the system and generate a periodic PMF. The first and the last bin of the
reaction coordinate will assumed be be neighbors.
-sym symmetrizes the profile around z=0 before output,
which may be useful for, e.g. membranes.
If available, the number of OpenMP threads used by gmx wham can be controlled by setting
OMP_NUM_THREADS environment variable.
gmx wham estimates the integrated autocorrelation
time (IACT) tau for each umbrella window and weights the respective
window with 1/[1+2*tau/dt]. The IACTs are written
to the file defined with
-oiact. In verbose mode, all
autocorrelation functions (ACFs) are written to
Because the IACTs can be severely underestimated in case of limited
-acsig allows one to smooth the IACTs along the
reaction coordinate with a Gaussian (sigma provided with
see output in
iact.xvg). Note that the IACTs are estimated by simple
integration of the ACFs while the ACFs are larger 0.05.
If you prefer to compute the IACTs by a more sophisticated (but possibly
less robust) method such as fitting to a double exponential, you can
compute the IACTs with gmx analyze and provide them to
gmx wham with the file
-iiact), which should contain one line per
input file (pullx/pullf file) and one column per pull coordinate in the
Statistical errors may be estimated with bootstrap analysis. Use it with care,
otherwise the statistical error may be substantially underestimated.
More background and examples for the bootstrap technique can be found in
Hub, de Groot and Van der Spoel, JCTC (2010) 6: 3713-3720.
-nBootstrap defines the number of bootstraps (use, e.g., 100).
Four bootstrapping methods are supported and
b-histDefault: complete histograms are considered as independent data points, and the bootstrap is carried out by assigning random weights to the histograms (“Bayesian bootstrap”). Note that each point along the reaction coordinate must be covered by multiple independent histograms (e.g. 10 histograms), otherwise the statistical error is underestimated.
histComplete histograms are considered as independent data points. For each bootstrap, N histograms are randomly chosen from the N given histograms (allowing duplication, i.e. sampling with replacement). To avoid gaps without data along the reaction coordinate blocks of histograms (
-histbs-block) may be defined. In that case, the given histograms are divided into blocks and only histograms within each block are mixed. Note that the histograms within each block must be representative for all possible histograms, otherwise the statistical error is underestimated.
trajThe given histograms are used to generate new random trajectories, such that the generated data points are distributed according the given histograms and properly autocorrelated. The autocorrelation time (ACT) for each window must be known, so use
-acor provide the ACT with
-iiact. If the ACT of all windows are identical (and known), you can also provide them with
-bs-tau. Note that this method may severely underestimate the error in case of limited sampling, that is if individual histograms do not represent the complete phase space at the respective positions.
traj-gaussThe same as method
traj, but the trajectories are not bootstrapped from the umbrella histograms but from Gaussians with the average and width of the umbrella histograms. That method yields similar error estimates like method
-bsresAverage profile and standard deviations
-bsprofAll bootstrapping profiles
-vbs (verbose bootstrapping), the histograms of each bootstrap are written,
and, with bootstrap method
traj, the cumulative distribution functions of
Options to specify input files:
-ix[<.dat>] (pullx-files.dat) (Optional)
Generic data file
-if[<.dat>] (pullf-files.dat) (Optional)
Generic data file
-it[<.dat>] (tpr-files.dat) (Optional)
Generic data file
-is[<.dat>] (coordsel.dat) (Optional)
Generic data file
-iiact[<.dat>] (iact-in.dat) (Optional)
Generic data file
-tab[<.dat>] (umb-pot.dat) (Optional)
Generic data file
Options to specify output files:
-oiact[<.xvg>] (iact.xvg) (Optional)
-bsres[<.xvg>] (bsResult.xvg) (Optional)
-bsprof[<.xvg>] (bsProfs.xvg) (Optional)
xvg plot formatting: xmgrace, xmgr, none
Minimum coordinate in profile
Maximum coordinate in profile
Determine min and max automatically
Number of bins in profile
First time to analyse (ps)
Last time to analyse (ps)
Analyse only every dt ps
Write histograms and exit
Determine min and max and exit (with
Calculate the log of the profile before printing
Energy unit in case of log output: kJ, kCal, kT
Define profile to 0.0 at this position (with
Create cyclic/periodic profile. Assumes min and max are the same point.
Symmetrize profile around z=0
Calculate integrated autocorrelation times and use in wham
Smooth autocorrelation times along reaction coordinate with Gaussian of this sigma
When computing autocorrelation functions, restart computing every .. (ps)
nr of bootstraps to estimate statistical uncertainty (e.g., 200)
Bootstrap method: b-hist, hist, traj, traj-gauss
Autocorrelation time (ACT) assumed for all histograms. Use option
-acif ACT is unknown.
Seed for bootstrapping. (-1 = use time)
When mixing histograms only mix within blocks of
Verbose bootstrapping. Print the CDFs and a histogram file for each bootstrap.