I'm working with a quaternary system, and when I try to do the variational cell relaxation (vc-relax) with nonzero starting_magnetization terms added in the input file, it looks like the total energy is oscillating between two minima after 23 hours (please see attached). But when I did the same vc-relax without any starting_magnetization terms, it converged just fine with a total final energy (minimum).
Also, as you can see in output file that the total magnetization here is zero, which shouldn't be the case since the experimental result for this cif shows otherwise with a relatively high magnetization. So any help on this would be much appreciated! Thanks in advance!
Note: I applied noncolin = .true. without any angle terms. Previously, I also tried with nspin = 2 (for default z-axis mag), and got the same zero magnetic moment for this material.
Following are the measure I've already taken in an effort to troubleshoot the situation without any success: I decreased the the mixing_beta to 0.1d0, increased electron_maxstep to 1000 (since it stops after 100 iterations by default if the solution doesn't converge). And I only added nonzero starting_magnetization for para and ferro magnetic elements (and excluded dia/non magnetic atoms). I double checked the trivial stuff as well, like the convergence testing for k mesh, ecutwfc, ecutrho against hydrostatic pressure (tstress=.true.), with degauss = 0005d0, occupations='smearing', and smearing='mv'. I used ibrav = 0, since I added the cell parameters "manually".
Thanks again!
Have you tried conjugate-gradient for mixing_mode? It will either work or you'll get an error, saying "S matrix not positive definite". If the latter is the case, then it means some of the bonds are too short for your pseudo-potentials (PPs). To fix this you have to make new (harder) PPs, possibly with some semi-core states and proper settings for the magnetic states you're interested in.
Thanks Dr. Mohammad Saeed Bahramy! So far I only used TF and for mixing_mode. So I'll try the conjugate-gradient and update it here. And I used SSSP Precision pseudo-potentials (PPs), which, as you know, a mixture of different PPs (i.e. pbe uspp, pbe kjpaw, and others) for different elements.
Apologies, I meant conjugate gradient option for diagonalization. It very often happens that for non-homogenous magnetic systems, the wave functions get too oscillatory at and in the vicinity of PP cutoff radii. In this situation "Davidson" digitalization enters a never-converging loop and "conjugate gradient" ends up with the error I mentioned above. The fix will then be to make your own PPs that better describe the semi-core regions.
No worries, I'll fix that. Also, I didn't know anything about generating a customized PPs! So please kindly share if there's any resources you know whenever possible. Thanks in advance!
QE has a utility program, ld1.x, that allows you to generate PPs. A full description of input parameters used to run this program can be found here:
https://www.quantum-espresso.org/Doc/INPUT_LD1.html
Also, you can find some example input files for both relativistic and non-relativistic PPs in directory (/atomic/pseudo_library/PBE/). With some changes, you can generalize them and create your own PPs for other elements.
Finally, you can also try the norm conserving Vanderbilt PPs, optimized for QE : https://github.com/pipidog/ONCVPSP
Hope this information helps and good luck!
- Badges
- Science topic
- Similar topics
- Physical Sciences
- Materials Science
- Materials Chemistry
- Polymer Chemistry