Restarting/Continuing Tinker9 Run

[cdpoon]

The web seems to have very little information on how to restart/continue Tinker9 run. For example, after a 10ns run, I create a new directory, copy the last snapshot from the ARC file in the previous run, and copy the the DYN file from the previous run. Together with the PRM file and the KEY file, I should be able to continue the "dynamic" run for another 10ns for example. From my understanding, the XYZ file does not matter as long as the filename is the same as the DYN file with xyz as file extension.

Is the above reasonable? Is there anything else one should pay attention to restart/continue a Tinker9 run?

[jayponder]

All versions of Tinker, including Tinker9, work the same way. What you say is basically correct. To continue an MD run you need the DYN file, which is essentially a restart file containing the state of the MD trajectory- coordinates, velocities, accelerations, and periodic box size, if applicable. When an MD run starts, the software will always check for a DYN file with the same base name as the XYZ file. If it finds a DYN file, the MD run will begin from the state contained in the DYN file. The XYZ file will only be used to get force field atom types and atom connectivity. The coordinates and everything else will be taken from the DYN file, so you don't need to copy the last frame of the ARC file to an XYZ file. Again, MD will always restart from the DYN file if available. So if you want to reinitialize and use the XYZ file coordinates, you need to delete or rename an existing DYN file.

Let us know if you have any other questions.

[cdpoon]

Thanks so much for your response. In a calculation we are working on, we have 20ns to equilibrate the system, then, restart/continue the run in another directory for 10ns. Using the ARC file in that 10ns run, we compute the diffusion constant using the Tinker "diffuse" command. We see that the MSD is high to begin with. It levels off to a lower MSD value. When we continue to have another round of 20ns-10ns run by continue the previous 20ns-10ns run, it behaves the same. If we restart/continue the Tinker9 run correctly, the diffusion constant should be a constant during the 10ns run.

We have just ruled out the possibility of not continuing Tinker9 run properly with your help.

We have also written a little Python program to compute diffusion constant from that the ARC file in the 10ns run. It shows that MSD is 20 times higher to begin with. It levels off to a much lower value after 9.5ns or so.

We are trying to figure out if it is due to some not-so-good parameters in the calculation or something else.

We appreciate your help and insight to the problem we are having.

Thanks again.

[jayponder]

Your interpretation of the diffusion is not correct. The diffusion constant is the slope of the MSD with respect to simulation time. It will always be high at the start of an MD run. What you want to do is to use the slope over the just middle part of your MD run, where it should be stable and gives a good estimate of the diffusion constant. The slope computed from just the start of the run will always be high, and computed from the end of the run will always be more noisy due to lack of data.

What you are seeing is exactly what you will always see... Even if you restart the MD from an equilibrated DYN file, the MSD will still give much too high of a diffusion constant at the beginning of the restarted run.

[cdpoon]

Thanks so much for the explanation. From what you said, we can take the last column of the "diffuse" output labelled as "Diff Const (x 10^5), remove some numbers of entries at the beginning and the end. Take the remaining and calculate the mean value. That should be the best estimate of diffusion constant. And, I believe the unit of that column is cm^2/ps. Is that correct?

[jayponder]

Yes, what you say above is correct- use an average from the middle region of the last column of the "diffuse" output. If you entered the time between trajectory frames in ps in the input to "diffuse", then the units for the final column of the output should be cm^2/s.