This thread is locked.Only browsing is available.
Top Page > Browsing
Cell optimization with Krylov subspace method
Date: 2019/05/28 20:18
Name: Pavel Ondra&#269;ka   <>

Dear all,

in short: is it possible to use the Krylov subspace method for cell optimization?

Long version:
I've been experimenting with the Krylov subspace method for some small examples ATM, hoping to eventually scale it to larger system. What I have seen though is that when I run relaxation of positions and cell with "MD.Type OptC5" the cell vectors are few percent off and more importantly, the a/b/c ratio is vary bad...

This is with quite simple structure 64 atoms TiON in fcc (basically just TiN with few O atoms at N sites and a single vacancy, I can provide the input if needed). I was basically testing the precision with respect to standard band method (again with "MD.Type OptC5" which worked well here), where the structure relaxes a little, but the cell vectors stay very similar and the atoms also don't move much as expected).

I did test with different runs with orderN.KrylovH.order from 800 to 1600 and orderN.HoppingRanges from 6.0 to 10.0, but to no avail, there doesn't even seem to be some trend suggesting I need to go higher, the values just jump around randomly.

The article PHYSICAL REVIEW B 74, 245101 (2006) states that the forces (position optimization) should be reasonable with sufficiently converged parameters, but there is no info about the lattice parameters...
Any comments would be appreciated?

Best regards
Page: [1]