Question about running eOn with external potential

Current eOn code and boinc distributed computing

Moderator: moderators

Post Reply
qian
Posts: 1
Joined: Tue Nov 10, 2020 9:11 pm

Question about running eOn with external potential

Post by qian »

Hi,

I am trying to run eOn with ext_pot as the option for potential. l used the binary ‘ext_pot’ to call cp2k to calculate the energy and forces. But it seems to be extremely slow. I tested on the pt system using the example in akmc-pt. With the build-in morse_pt potential, it only took ~5 seconds for one process search. However, when I tried to use the same Morse model, but with ext_pot calculating energy and forces using cp2k, it took ~500 seconds, which is ~100 times slower.

I’m wondering if the ext_pot is supposed to be so slow or is there any way to improve the computation speed when I have to use an external package to get energy and forces? Thank you!

Best,
Qian
graeme
Site Admin
Posts: 1946
Joined: Tue Apr 26, 2005 4:25 am
Contact:

Re: Question about running eOn with external potential

Post by graeme »

Hi Qian,

I'm not really sure what cp2k does when it evaluates an empirical potential, but I could imagine that there is a significant overhead in running cp2k to evaluate a simple Morse interaction. To check this, you might just check the time to calculate a single point energy and force with cp2k and see if that is consistent with the eon run time and the number of force evaluations that it did.

The idea behind the ext_pot is to allow coupling of EON to other codes. Our thought was that these codes would be based upon DFT or some approximate electronic structure. In that case, the overhead from running another program to get the forces and energy could be modest as compared to the cost of calculating that energy and force. With a Morse potential, however, there should be closer integration (unless the system is huge). We do support integration with the LAMMPS library with allows for relatively efficient calling of a wide variety of empirical potentials.

Graeme
Post Reply