Hi,
I'm trying to relaxing the Ce2O3 structure, however I find various problems such as the symmetry broken of the unit cell and the difficult to converge the RMS change in Kohn-Sham potential.
I'd like any advice on possible improvement to my input.
I am doing a related calculation right now, on Lu2O3.
The problem with convergence is the first thing to solve. Having a stable SCF (task 0/1) is the basis for further calculations.
I recommend trying the following:
This setting generally helps in geometry-related jobs, just make sure that there is no atom at fractional coordinates 0.25 0.125 0.625.
vkloff0.250.1250.625
Use high quality:
highqt
I prefer using manual MT radii. Check different values for Ce in the 2-2.5 range. Use the largest at which the calculation converges. My usual value for O is 1.45. Small O RMT decreases chances of RMTs being changed during the optimization. Changing RMTs mean changing basis, which renders the whole optimization unstable.
Try larger k-point grid. Your 3 3 3 is way too small for this tiny cell. I bet it should be 7 7 7 or something like that.
radkpt50autokptt
See what kind of grid this will create, and then specify it with ngridk.
Also, best way is to do a series of SCFs with different grids - total energy should converge with the increasing grid.
Try using momfix and fsmtype=2 to specify target moments on atoms, it is generally better than reducebf:
That, however, might not work due to the fact that some portion of the moment might end up in the interstitial region, resulting in MT moments smaller than 1. If that happens, replace the target moment value (1) with the "converging" value.
Try optimizing your +U parameters, there are approaches to do that self-consistently. However, I'd made sure the calculation runs smoothly without +U, and then apply +U.
Try using SCAN meta-GGA (with libxc), without +U. The system is small, it might work in reasonable time. With SCAN, however, set ptnucl to false, nuclei will not be point charges (they will be spheres of the experimental nucleus radii), resulting in smaller jumps in potential at the nuclei.
ptnuclf
Hope that helps.
Good luck!
Andrew
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
The first problem is that Elk was not recognising the crystal structure as trigonal. This is because you have not specified the cell parameters precisely enough.
Elk uses the parameter 'epslat' to determine if two lattice vectors or atomic coordinates are the same. By default, epslat is set to 10⁻⁶. Your lattice parameters are not within this tolerance and hence Elk found only one crystal symmetry. See the files SYMLAT.OUT and SYMCRYS.OUT for the list of lattice and crystal symmetries.
Second, the structural opimisation code has been improved since version 6.3.2. One of the problems with performing lattice optimisation with most codes is that the number of plane waves (or augmented plane waves in the case of Elk) change discontinuously as the lattice vectors are changed. This is hard to mitigate for APW codes. Consequently, the energy can change discontinuously with lattice parameter.
Anyway, I've managed to optimise your structure with Elk version 6.8.4. Here is the updated input file:
Hi,
I'm trying to relaxing the Ce2O3 structure, however I find various problems such as the symmetry broken of the unit cell and the difficult to converge the RMS change in Kohn-Sham potential.
I'd like any advice on possible improvement to my input.
Best,
Andrea
In particular the cell is strongly deformed, I attach the initial and final structure.
Hi Andrea,
Is it your intention that the unit cell be hexagonal?
Regards,
Kay.
Hi John,
yes, the cell should be hexagonal (trigonal).
The unit cell is that attached.
https://materialsproject.org/materials/mp-2721/#
Best,
Andrea
Dear Andrea,
I am doing a related calculation right now, on Lu2O3.
The problem with convergence is the first thing to solve. Having a stable SCF (task 0/1) is the basis for further calculations.
I recommend trying the following:
I prefer using manual MT radii. Check different values for Ce in the 2-2.5 range. Use the largest at which the calculation converges. My usual value for O is 1.45. Small O RMT decreases chances of RMTs being changed during the optimization. Changing RMTs mean changing basis, which renders the whole optimization unstable.
Try larger k-point grid. Your 3 3 3 is way too small for this tiny cell. I bet it should be 7 7 7 or something like that.
See what kind of grid this will create, and then specify it with ngridk.
Also, best way is to do a series of SCFs with different grids - total energy should converge with the increasing grid.
That, however, might not work due to the fact that some portion of the moment might end up in the interstitial region, resulting in MT moments smaller than 1. If that happens, replace the target moment value (1) with the "converging" value.
Try optimizing your +U parameters, there are approaches to do that self-consistently. However, I'd made sure the calculation runs smoothly without +U, and then apply +U.
Try using SCAN meta-GGA (with libxc), without +U. The system is small, it might work in reasonable time. With SCAN, however, set ptnucl to false, nuclei will not be point charges (they will be spheres of the experimental nucleus radii), resulting in smaller jumps in potential at the nuclei.
Hope that helps.
Good luck!
Andrew
Dear Andrea and Andrew,
The first problem is that Elk was not recognising the crystal structure as trigonal. This is because you have not specified the cell parameters precisely enough.
Elk uses the parameter 'epslat' to determine if two lattice vectors or atomic coordinates are the same. By default, epslat is set to 10⁻⁶. Your lattice parameters are not within this tolerance and hence Elk found only one crystal symmetry. See the files SYMLAT.OUT and SYMCRYS.OUT for the list of lattice and crystal symmetries.
Second, the structural opimisation code has been improved since version 6.3.2. One of the problems with performing lattice optimisation with most codes is that the number of plane waves (or augmented plane waves in the case of Elk) change discontinuously as the lattice vectors are changed. This is hard to mitigate for APW codes. Consequently, the energy can change discontinuously with lattice parameter.
Anyway, I've managed to optimise your structure with Elk version 6.8.4. Here is the updated input file:
I've attached various output files generated during the optimisation run
Regards,
Kay.