'CrashRestart' taking long time

[Vardhan]

Dear Professor,

I am using RASPA to study the adsorption of a gas on a MOF. I have transferred my jobs from one cluster to another cluster, (after killing the jobs). In the new cluster, I have submitted the jobs, using 'CrashRestart    yes' command. I haven't changed anything else in the input file. I have 11 different files (each for a particular pressure). The simulations are taking almost 24 hours to pick up from where they have left. Also, the ones with high pressure (above 1 bar) are not running (I presume that it is still reading the Crash restart file). Is there a way I can speed up my calculations during using the 'Crash Restart'?

Thanks,

Regards,
Vardhan.

[David Dubbeldam]

What is "CrashRestart yes"? Please check the manual "Restart and crash-recovery".
There is an option to restart from a restart-file containing positions, velocities, forces, and one option to continue from a binary file.
The binary-file format is not transferable between different versions of the code, so make sure you are using the latest version of the code (2.0.35).

[Vardhan]

Thank you very much for the reply.

1. I have installed RASPA 2.0 in both the clusters (gcc compiler). My earlier calculations were done using RASPA 2.0, so I am continuing with that.

2. Sorry, by "CrashRestart  yes", I meant the 'Binary Restart File'. In the input file, I have added "ContinueAfterCrash   yes" command, to continue using the binary file in the folder "Crash Restart".

3. I have tried the two options given in the manual as follows, (following are the input file commands I have given for a GCMC simulation, with 200000 initialization and production cycles, each)
Approach 1: (using restart file, where the positions are taken)

SimulationType            MonteCarlo
NumberOfCycles                188800
PrintEvery                    200

PrintPropertiesEvery          200

RestartFile                   yes
WriteBinaryRestartFileEvery   10000

Approach 2: (using binary restart file (crash restart))
SimulationType            MonteCarlo
NumberOfCycles                200000
NumberOfInitializationCycles  200000
PrintEvery                    200

PrintPropertiesEvery          200

ContinueAfterCrash            yes
WriteBinaryRestartFileEvery   10000



So, in the first approach I have to change, in the input file, the number of production and initialization cycles accordingly. But, the problem here is, the earlier averages of the properties are ignored and being calculated from the current value.

The second approach, where I include a command in the input file ("ContinueAfterCrash   yes"), is working properly. It is exactly picking up from where it has left without changing anything else in the input file. But, when I start the simulations using this command, it is taking a long time to restart. Every time my wall time is reached, they stop and take a long time to pick up from where it has left.
Is there anything I can do to overcome this and speed up the calculations?


Thank you,

Best regards,
Suvardhan J.

[David Dubbeldam]

The binary-restart should immediately pick up where it left. So at most 10000 cycles before catching up again with the output. Make sure it really reads the binary restart and does not restart from scratch (for example, by testing it using WriteBinaryRestartFileEvery with a smaller number and run it interactively in a console). Also, you have the source code, and you can use a debugger to see what is happening).

[Vardhan]

Thank you very much for the reply, Professor.

I was running the simulations at 11 different pressures (0.1,0.3,0.5,0.7,0.9,1,2,4,6,8,10). The simulations at low pressure (below 1bar) are starting after a gap of around 5-8 hours. But the ones at high pressures are not even starting. When I do any changes in the input file (I have changed the 'WriteBinaryRestartFileEvery' to 5000 from 10000), before re-starting them using the binary file, then it starts fresh from zero (as can be seen in the output file).
I configured RASPA using gcc compiler in the first cluster. Later, I moved the data to another cluster to run the simulations there. In the new cluster, RASPA was configured using gcc compiler, but a different version. I am guessing this might be the reason.
Also, is there anything I can do while configuring RASPA, so that it speeds the simulations?

Thank you again.

[David Dubbeldam]

If you know that there is no hysteresis in the system, you could simulate all of these pressure in parallel. That really speeds up computations of isotherms.