Re: [sparta-users] [EXTERNAL] Ensemble averaging in SPARTA

[Gallis, M. A. <ma...@sa...>]

Rommil

Ensemble averaging is the best option for this case. You can launch multiple jobs under the same allocation of nodes. Make the seed and the output filenames  a function of the job number, so that you end up with statistically independent results and different output files.

After that you will have to average the results in a post processing fashion. The code cannot average results from different runs.

Michael

From: Rommil Emperado <rem...@ni...>
Date: Wednesday, September 11, 2024 at 9:18 PM
To: spa...@li... <spa...@li...>
Subject: [EXTERNAL] [sparta-users] Ensemble averaging in SPARTA
Hi everyone!

I am working on a DSMC modelling of laser-produced plasma expansion dynamics, and the unsteady nature of the flow seems to require repeating my simulations several times using different seed values and averaging the dumped data over all trials to get precise time-resolved data.

Is there a way to implement this kind of averaging using the input script? Can I assign my runs to different groups of processors so I can automate the multiple runs?

For those who have previous experience on unsteady DSMC, is there a better way to get precise simulations aside from ensemble averaging?

Thank you so much for your help.

Kind regards,

Rommil Emperado
PhD Candidate
National Institute of Physics
University of the Philippines Diliman