QuantumATK Forum

QuantumATK => General Questions and Answers => Topic started by: sukhito teh on July 25, 2024, 06:42

Title: Parallelization of ProjectedDensityOfStates
Post by: sukhito teh on July 25, 2024, 06:42

Dear users and staffs,
May I know how parallelization works in ProjectedDensityOfStates ? Lets say I have 4 kpoints, does that mean I should use not more than 4 processes in calculation? And how can I know what is the reduced number of kpoints?
Thank you for your attention.

best regars,
sukhito

Title: Re: Parallelization of ProjectedDensityOfStates
Post by: Anders Blom on July 30, 2024, 00:23

Most analysis functions are indeed parallelized over k-points. I suppose you just used the number 4 as an example, which could correspond to 2x2x1 k-points, but it's a very low number for any accurate DOS results. So let's use a more realistic example of 5x5x5 points. We then have roughly 5*5*5/2=62 points, by simply halving the product due to time-reversal symmetry. So running this on 32 or 64 cores will work quite well. The actual number can be computed as
len(MonkhorstPackGrid(5,5,5).kpoints())
which comes out to 63.