QuantumATK Forum

QuantumATK => General Questions and Answers => Topic started by: AsifShah on October 20, 2023, 05:30

Title: Unusual computational issue with geometry optimization cycle
Post by: AsifShah on October 20, 2023, 05:30

Dear Admin/All,
I am performing spin calculation on a system however, I observed something unusual. There have so far been 7 converged electronic cycles but only 5 geometry optimization steps have been performed. If I am not wrong, shouldn't each converged electronic cycle update atomic positions and hence both should be equal. Why is there this issue, please clarify. I have attached screenshot below as well.
1. grep OPT is used to see completed geometry steps = 5
2. fermi level is used to see completed electronic cycles = 7


Thanks
Regards

Title: Re: Unusual computational issue with geometry optimization cycle
Post by: Anders Blom on October 24, 2023, 01:41

The LBFGS algorithm needs additional force evaluations for the line search part

Title: Re: Unusual computational issue with geometry optimization cycle
Post by: AsifShah on October 24, 2023, 05:50

Is that normal or how can I speed it up?

Title: Re: Unusual computational issue with geometry optimization cycle
Post by: Anders Blom on October 24, 2023, 19:23

No it's normal, it's how the algorithm works