A script written in Julia to automate runs of Extended Thermodynamic Integration by repeated calls of the GROMOS simulation and analysis packages.

The script only loads one argument from the command line interface: the absolute path of the extended TI simulation directory. This is to make it callable from GROMOS scripts generated with mk_script. The library file for mk_script should include the following line in the MISCELLANEOUS block.

lastcommand echo $OUTPUTTRE.gz >> en_files.dat && echo $OUTPUTTRG.gz >> fr_files.dat && julia /path/to/autexti/autexti.jl $SIMULDIR

(assuming there’s a working Julia installation is in PATH) A sample library file can be found in examples/.

Additionally, the script will look for a configuration file ext_ti.conf in the simulation directory.

Based on a working mk_script argument file, autexti.jl will generate Extended TI simulations; either adding new $λ$ points or extending them until the specified convergence criteria are met.

To start the process, call the program passing the same directory specified with @dir in the mk_script argument file.

$ julia /path/to/autexti/autexti.jl /path/to/simuldir/

A sample configuration file can be found in examples/

header String, currently unused.

• prefix

String. Prefix for simulation folders. Following the GROMOS tutorial it would be “L_”

• queuefile

String. File where queued jobs will be kept track of. Relative to simuldir.

• logfile

String. File where lines of queuefile will be saved to after running lobs. Only relevant if logqueue=1. Relative to simuldir.

• statefile

String. File where progress is logged. Relative to simuldir.

• command

String. Command to call the GROMOS scripts with, such as slurm. If lent blank it will be called directly.

• mk_script_args

String. Argument file for mk_script. Relative to simuldir.

• ene_ana_lib

Library file for ene_ana and ext_ti_ana. Relative to simuldir.

• gromospath

Path to a gromos++ installation (in case it’s not in path).

• eq_scripts Int. Number of equilibration scripts.
• eq_steps Int. Number of steps (NSTLIM) for equilibration scripts.
• md_stels Int. Number of steps (NSTLIM) for non-equilibration scripts.
• min_md_scripts Int. Number of non-equilibration scripts to run before any analysis.
• max_md_scripts Int. Maximum number of non-equilibration scripts to run regardless of analysis results.
• max_total_scripts Int. Maximum number of non-equilibrium steps across all simulated $λ$s. Currently not implemented.
• stepsb4ext Float. For the initial run, points will be created between $λ=0$ and $λ=1$ in increases of stepsb4ext. For example, stepsb4ext=1 will have an initial run of only 0 and 1, while stepsb4ext=.2 would create an initial run of 0, .2, .4, .6, .8, and 1. 0 and 1 will always be generated.
• start_scripts Int. If >0, set the number of non-equilibration simulations for $λ=0$, ignoring min_md_scripts and max_md_scripts. Required for endpoints where $\frac{∂ \mathcal{V}}{∂ λ}$ is 0.
• end_scripts Int. If >0, set the number of non-equilibration simulations for $λ=1$, ignoring min_md_scripts and max_md_scripts. Required for endpoints where $\frac{∂ \mathcal{V}}{∂ λ}$ is 0.

• dvdl_tol Float. For a given $λ$ point, scripts will keep being generated until the error estimate for $\frac{∂ \mathcal{V}}{∂ λ}$ at that point falls below dvdl_tol.
• pred_tol Float. Depends on criteria. If it’s set to “pred”, new $λ$ points will keep being added intil the error estimate for all points falls below pred_tol. For all other criteria, instead, points will keep being added until the difference between the integral of simulated points and that of predicted points falls below pred_tol.
• skip4ana Int. Number of energy trajectories to skip for analysis, starting from the first equilibrium simulation. It is recommended to set equal to eq_scripts.

• logqueue Bool. If set to 1, log simulations to logfile as they are ran.
• logmerges Bool. If set to 1, log previous resulults of ext_ti_ana when a new $lambda$ point is added.
• criteria String. How the values of new points are decided. Currently accepts: pred, ydist, xydist, areamid.

When choosing a new point to simulate, the following criteria are currently implemented.

Optimize predicted points. Looks at the predicted PMF and chooses the point with the highest error estimate until all predicted points have an error estimate below pred_tol.

Choose the point for which the prediction is furthest in the y-axis (energy) to a (linear interpolation) of simulated points.

Choose the point for which the predictions highest minimum distance (in both energy and $λ$) to (a linear interpolation of) simulated points.

For each segment of the simulated PMF, compares the area under the curve to the integral of the corresponding segment of the predicted PMF. Adds the next simulation at the mid-point of the segment with the highest difference to the predicted PMF.