VMD-L Mailing List

From: René Hafner TUK (hamburge_at_physik.uni-kl.de)
Date: Fri Jun 10 2022 - 01:47:58 CDT

Hi Max,

    for my use cases (N~10^13 frames) I did not see any need to
    calculate MSD on GPU yet but to use a more efficient algorithm
    rather than the standard of N^2 complexity:

         Here
    (https://urldefense.com/v3/__https://stackoverflow.com/questions/34222272/computing-mean-square-displacement-using-python-and-fft__;!!DZ3fjg!-0W3GELLJpchZLlMeBuZJ92gbd1UobVsvAi9JslSBFtmqruSxGcuRWlHZewQWsRkIZmv4DzzIIW18nbHjRKRJilzwtM$ )

         you find a N*LOG(N) implementation in python numpy which could
    be translated into tcl or

    just get your particle trajectory to csv or any other format and use
    the python code right away.

    This can easily be parallelized e.g. by using pandas+pandarallel if
    you want average over multiple molecules.

    One could do the FFT on gpu but that may only be worth it for really
    huge trajectory sizes (but I did not test this).

Kind regards

René

On 6/9/2022 7:20 PM, Max Win wrote:
> Dear users,
> Is there a way to calculate MSD with GPU acceleration on TCL scripts?
> Max 

-- 
--
Dipl.-Phys. René Hafner
TU Kaiserslautern
Germany