Many SCF cycles in each geometry optimization step

[ma455...@gmail.com]

Hi,

I'm a new user of cp2k 8.1. I'm calculating a P doped Si system (output file attached). The system contains 1000 atoms. The calculation goes well and the SCF converges fairly smoothly. However, different from my previous calculations, I found in this calculation there are many SCF cycles (SCF converged again and again) for each geometry optimization step. In my previous calculations, the basic steps are: SCF ->  OPTIMIZATION STEP 1 -> SCF -> OPTIMIZATION STEP 2 -> SCF -> OPTIMIZATION STEP 3. But in this calculation, the steps are: SCF -> OPTIMIZATION STEP 1 -> SCF (converged) -> SCF (converged) -> SCF (converged) -> SCF (converged) ...... -> OPTIMIZATION STEP 2. I suspect this maybe caused by the different OPTIMISER. In my previous calculations, I used BFGS. But in this calculation I used LBFGS. However, based on the tutorial, LBFGS should be faster in large system calculation but the results are not what I expected. Could someone please provide some suggestions on improving this issue? Thanks

Regards,

Hongyang

[Thomas Kühne]

Dear Hongyang, 

you have given the answer to your own question, the observed behavior is a consequence 

of the employed geometry optimizer. LBFGS, is typically faster anyhow, which is not only 

due to the algorithm, but also due to the particular implementation. 

Best, 

Thomas

--
You received this message because you are subscribed to the Google Groups "cp2k" group.
To unsubscribe from this group and stop receiving emails from it, send an email to cp2k+uns...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/cp2k/e713484b-b0f4-4886-b689-6459ab67cd5dn%40googlegroups.com.
<s5-hse-01-out.txt>

==============================

Thomas D. Kühne

Dynamics of Condensed Matter

Chair of Theoretical Chemistry

University of Paderborn

Warburger Str. 100

D-33098 Paderborn

Germany

thomas...@upb.de

+49/(0)5251/60-5726

[ma455...@gmail.com]

Dear Thomas,

Thanks for this information!

Regards,

Hongyang