the structure does not converge

[weiyidan]

Hi professor,
I prepare to do a calculation than a atom moves in the surface between the two bulk.

First, i use Ediff=1e-2, Ediffg=-0.3, IBRION=2, it works well.
Second, i use Ediff=1e-5, Ediffg =-0.05, IBRION=2, it works well too.
Third, i use Ediff=1e-5, Ediffg =-0.01, IBRION=1, the problem appears.

when I relax the heterojunction with parameters in the third step， it seems hard to converge. What is the problem? Can you give me some advice？
My calculation history is in the attachment.

Thanks for your attention.
Yidan WEi

[graeme]

Try lowering POTIM; a value of 0.3 is pretty large for IBRION=1.

[weiyidan]

how about POTIM=0.15?

[graeme]

I typically use 0.1, but it really depends upon the stiffest mode in your system. You can try a few values to see when the convergence stabilizes.

[weiyidan]

Thanks for your reply, as You help me a lot from last year to now.

Concerning to the problem,
1. I used POTIM=0.15, however , it did not work. I attach the calculation files in attachments and the picture shows the relationship between energy and steps. Should I use a smaller POTIM?
2.how to judge the stabilization of convergence ? is continuous decrements in energy？

Yidan Wei

[weiyidan]

here is the attachment

[graeme]

I was not able to use the built-in IBRION=1 vasp LBFGS to obtain good convergence. My best attempt is here:

cy1.tar.gz

(44.66 MiB) Downloaded 24892 times

Using quick-min (IBRION=3) did better, but did not converge within 200 iterations:

cy2.tar.gz

(27.24 MiB) Downloaded 24630 times

I expect that quick-min would converge, but using our LBFGS optimizer (IBRION=3, POTIM=0, IOPT=1, INVCURV = 0.002) does a good job:

cy5.tar.gz

(20.86 MiB) Downloaded 24762 times

[weiyidan]

Thanks，professor. You really help me a lot.