I am calculating some EOS curves for the purpose of pseudopotential testing. At the same time, I test basis size for Elk. I use a series of different scales, on the same cell, energy is recalculated from scratch at each point. Muffin-tin radii do not get reduced in the series, i.e. are small enough. I also use isgkmax = -2.
EOS seems to be much more sensitive to rgkmax than to gmaxvr. Is that a correct conclusion?
Rough (noisy parabolic) EOS indicates a low basis, rgkmax must be increased. But, in one of the tests (Lu2O3) I've got increase of the rougness after a certain increase in basis.
My series goes like:
rgkmax gmaxvr
7.0 15
7.5 16
7.8 17
8.0 18 < rougness goes down until here
8.5 19 < rougness increases abruptly here
Noteworthy, vanadium oxide (VO) does not show such behaviour, its EOS got better and better in the same series.
Why is it happening? What am I doing wrong?
I thought of using msmooth. I've restarted caluclations that converged with msmooth = 0, using msmooth =1. Then, restarted again with msmooth = 2 and so on. Msmooth does not heal the roughness. It rather shifts the whole curve slightly up in energy (tested with msmooth 0-5). The larger the value, the larger the shift. Is that an expected behaviour?
Also, please delete anonymous post with the same text. That was me, not noticing that I was not logged in.
Thank you in advance.
Andrew
Last edit: Andrew Shyichuk 2019-06-14
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
Long time ago "Anton Kozhevnikov" explained to me a similiar problem. That ultimatly somehow increasing RGKMAX leads to unstable situation. The conclusion of our discussion was the following maybe helpful for your situation as well:
RGKmax control the plane wave coefficients outside of the MT in the interstitial, wheras
lmaxapw controls the l-cutoff inside the MT. By construction of the basis one has to match one with the other. If you now increase RGKmax you somehow end up with a situation, where you can not match anymore the MT basis if you keep lmaxapw=const.. Because of this the computations become ''shaky''. Our solution was to increase the MT as well as the spherical grids and than things became smoother again.
best regards
Michael
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
I've done some more tests, namely with cubic and hexagonal Lu2O3.
I used the same basis and MT radii. Hexagonal Lu2O3 exhibits smooth EOS plot, while the cubic one is the described above. Why can this be?
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
The vkloff solution worked, I've got an adequate EOS.
Such a displacement matched no atom in my structure. It thus effectively disabled k-point reduction with symmetry. With my 3x3x3 grid, there were 8 k-points after reduction. With the vkloff option there were 27.
Will check if reducek 0 also does the trick.
Noteworthy, switching from 3x3x3 grid to 7x7x7 (24 k-points after reduction) changed nothing. Increasing lmaxapw to 9 or both lmaxapw and lmaxo to 9 also changed nothing.
Best regards,
Andrew
Last edit: Andrew Shyichuk 2019-06-24
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
Dear Users,
I am calculating some EOS curves for the purpose of pseudopotential testing. At the same time, I test basis size for Elk. I use a series of different scales, on the same cell, energy is recalculated from scratch at each point. Muffin-tin radii do not get reduced in the series, i.e. are small enough. I also use isgkmax = -2.
EOS seems to be much more sensitive to rgkmax than to gmaxvr. Is that a correct conclusion?
Rough (noisy parabolic) EOS indicates a low basis, rgkmax must be increased. But, in one of the tests (Lu2O3) I've got increase of the rougness after a certain increase in basis.
My series goes like:
rgkmax gmaxvr
7.0 15
7.5 16
7.8 17
8.0 18 < rougness goes down until here
8.5 19 < rougness increases abruptly here
Noteworthy, vanadium oxide (VO) does not show such behaviour, its EOS got better and better in the same series.
Why is it happening? What am I doing wrong?
I thought of using msmooth. I've restarted caluclations that converged with msmooth = 0, using msmooth =1. Then, restarted again with msmooth = 2 and so on. Msmooth does not heal the roughness. It rather shifts the whole curve slightly up in energy (tested with msmooth 0-5). The larger the value, the larger the shift. Is that an expected behaviour?
Also, please delete anonymous post with the same text. That was me, not noticing that I was not logged in.
Thank you in advance.
Andrew
Last edit: Andrew Shyichuk 2019-06-14
Dear Andrew,
Long time ago "Anton Kozhevnikov" explained to me a similiar problem. That ultimatly somehow increasing RGKMAX leads to unstable situation. The conclusion of our discussion was the following maybe helpful for your situation as well:
RGKmax control the plane wave coefficients outside of the MT in the interstitial, wheras
lmaxapw controls the l-cutoff inside the MT. By construction of the basis one has to match one with the other. If you now increase RGKmax you somehow end up with a situation, where you can not match anymore the MT basis if you keep lmaxapw=const.. Because of this the computations become ''shaky''. Our solution was to increase the MT as well as the spherical grids and than things became smoother again.
best regards
Michael
Dear Michael,
I will check if lmaxo and lmaxapw would have any affect on the problem.
Thank you.
Andrew
Dear Michael,
I will check if lmaxo and lmaxapw would have any affect on the problem.
Thank you.
Andrew
I've done some more tests, namely with cubic and hexagonal Lu2O3.
I used the same basis and MT radii. Hexagonal Lu2O3 exhibits smooth EOS plot, while the cubic one is the described above. Why can this be?
Dear Andrew,
could you check a 2nd thing, could you offcenter the k-point grid by vkloff
so add
vkloff
0.25 0.125 0.625
and check again, in some cases that ''healed'' rough curves. Moreover, could you share your input thus I could have a test for my own .....
best
Michael
Dear Michael,
The vkloff solution worked, I've got an adequate EOS.
Such a displacement matched no atom in my structure. It thus effectively disabled k-point reduction with symmetry. With my 3x3x3 grid, there were 8 k-points after reduction. With the vkloff option there were 27.
Will check if reducek 0 also does the trick.
Noteworthy, switching from 3x3x3 grid to 7x7x7 (24 k-points after reduction) changed nothing. Increasing lmaxapw to 9 or both lmaxapw and lmaxo to 9 also changed nothing.
Best regards,
Andrew
Last edit: Andrew Shyichuk 2019-06-24