sample-skpcnf

Sampling interval of MD

The erdst program is a post-process of an MD run and reads an MD trajectory; the trajectory fed to erdst is linked in the name of HISTORY. The trajectory file takes less disk space when the interval to save the MD snapshot (configuration) is longer, while some amount of trajectory data may be necessary for good statistics in obtaining distribution functions with erdst.

To determine the sampling interval (interval to save the snapshot) in MD, the parameter skpcnf may be used in parameters_er. As described in Parameter files for erdst, skpcnf specifies the interval to "use" the configuration in HISTORY.
When skpcnf = 1 (default), all the configurations in HISTORY are used to compute the distribution functions.
When skpcnf = 2, the configuration saved in HISTORY is used every other step.
When skpcnf = 10, the configuration is used every 10 steps.

If the MD trajectory is saved every 10 fs, the configuration is actually analyzed by erdst at every 100 fs when skpcnf is set to 10. This situation is the same as saving the MD trajectory every 100 fs and using all the trajectory data with skpcnf = 1. Accordingly, when the result obtained with an MD sampling interval of 10 fs and skpcnf = 10 is unchanged from the one with a sampling interval of 100 fs and skpcnf = 1, the sampling interval of MD should be set to 100 fs (with running erdst at skpcnf = 1). It is indeed desirable from the viewpoint of disk storage that the MD trajectory is saved with a long interval.

Some guidelines for the sampling interval of MD for soln can be as follows.
For a small molecule in water or homogeneous solvent, the sampling interval of MD is initially set to 20 fs and a larger interval is sought with changing skpcnf.
For a large molecule such as protein in water, an initial choice is 100 fs for the sampling interval of MD.
For a small molecule in micelle or lipid membrane, 100 fs can also be employed as the initial value for the sampling interval of MD.
The sampling interval can be initially set to 100 fs for the MD for a refs calculation.

In addition to the sampling interval of MD, the total time of MD simulation is a factor governing the computation time of erdst. The total time of MD can be determined from the convergence of the solvation free energy. For the MD used for soln run, the necessary time of simulation can be assessed from the output of the slvfe run. The slvfe program outputs the free-energy value in each of the trajectory blocks of the solution system and its cumulative average, as described in the section of Running slvfe: Getting Final Output of Quick Start Guide. The cumulative average is then shown as the last part of the output, and provides the convergence behavior of the solvation free energy with respect to the MD length of the solution system.

The convergence with respect to the refs run can be assessed with the scheme written in Assessing the convergence with respect to the reference-solvent calculation. The necessary time of simulation can be determined from the convergence behavior with respect to the simulation time of MD for refs.

The computation time of erdst for refs is also affected by the number of insertions per reference-solvent configuration sampled. The total number of solute-solvent configurations subject to erdst calculation is equal to the product of the number of reference-solvent configurations and the number of insertions per reference-solvent configuration sampled. The number of insertions per reference-solvent configuration can be changed by the maxins parameter within parameters_er; its default value is 1000 as described in Parameter files for erdst. The computation time of erdst for refs is essentially determined by maxins multiplied by the number of reference-solvent configurations. It is usually better to have a larger number of reference-solvent configurations (smaller maxins) when the computation time of erdst for refs is kept.

The above does not apply to an erdst calculation for soln. In a soln run, the computation corresponds effectively to maxins = 1. The computation time of erdst is determined by the number of configurations saved in HISTORY divided by skpcnf.