Dear kirill,

I just performed a test calculation and compared it with an older version of Fleur. The output in the out file is in agreement with both outputs in the older version. The output in the out.xml file seems to have changed so the assumption is that it is wrong. There is a bug.

Here is the output for the older calculation:

gregor@analyticalEngine:~/work/Tutorial/Co/SOC/forceTheo$ grep -i "ev\-sum" out.xml
            <Angle theta="   0.0000000" phi="   0.0000000" ev-sum=" -33.4628490"/>
            <Angle theta="   1.5707963" phi="   0.0000000" ev-sum=" -33.4628356"/>
            <Angle theta="   0.7853982" phi="   0.0000000" ev-sum=" -33.4628423"/>
            <Angle theta="   1.5707963" phi="   1.5707963" ev-sum=" -33.4628356"/>
gregor@analyticalEngine:~/work/Tutorial/Co/SOC/forceTheo$ grep -i "single particle" out
          sum of val. single particle energies:       -33.4628489893 htr
          sum of val. single particle energies:       -33.4628356448 htr
          sum of val. single particle energies:       -33.4628423210 htr
          sum of val. single particle energies:       -33.4628356350 htr
 
 

And this is the output of the current development version:

gregor@analyticalEngine:~/work/Tutorial/Co/SOC/forceTheo/test$ grep -i "ev\-sum" out.xml
            <Angle theta="0.0000000" phi="0.0000000" ev-sum="-16.7320991"/>
            <Angle theta="1.5707963" phi="0.0000000" ev-sum="-16.7320924"/>
            <Angle theta="0.7853982" phi="0.0000000" ev-sum="-16.7320957"/>
            <Angle theta="1.5707963" phi="1.5707963" ev-sum="-16.7320924"/>
gregor@analyticalEngine:~/work/Tutorial/Co/SOC/forceTheo/test$ grep -i "single particle" out
          sum of val. single particle energies:       -33.4641982711 htr
          sum of val. single particle energies:       -33.4641982550 htr
          sum of val. single particle energies:       -33.4641847884 htr
          sum of val. single particle energies:       -33.4641914680 htr
          sum of val. single particle energies:       -33.4641847801 htr
 
 

I did not adjust all parameters to match between the two calculations, so don't wonder that there are deviations on the finer scale.

I think the output in the out file also features the number of digits you want to have. Personally I wouldn't trust those additional digits. They may depend on details of the calculation that are not under your control, like the precision of the determination of energy parameters and similar things. Be careful to interpret them in terms of physics.

I will create an issue related to the deviating out.xml output. We will analyze and fix this problem.

For completeness: This is the issue I created: https://iffgit.fz-juelich.de/fleur/fleur/-/issues/674 Please follow it to stay up-to date with the progress of the respective bug fix.

Dear kirill,

Maybe one or two more digits are okay. I just had an experience in a high-throughput project in which I performed the same calculations twice, without any difference in the setup of the calculation. For some free energies I got slightly different values. These are examples for two numbers I got for the same calculation (in eV):

1st calculation: -6130.3246333056
2nd calculation: -6130.3246333083
 
 

So you see on what precision level even identical calculations might differ, e.g., due to the order of certain computations in a parallelized Fleur run. In your calculations you change the SQA, so there actually are significant differences, which may interact with some hardcoded numbers in the code. If you do such a calculation and really rely on these digits please perform the calculations twice with slighlty different numerical setups to see whether the results are stable on that level. Vary Kmax slightly or the MT radii or whatever. Just tiny differences to make sure that there are no numerical issues in your results. It may be that this is not a problem at all, but I personally have no experience with calculations on that precision level.

The example from my former post is related to the MAE from one of the tutorials. It is the calculation for hcp Co. I think that shouldn't have too much MAE, but it is about two orders of magnitude larger than what you want to see. In that example I always tell the students that the tutorial at that point doesn't investigate parameter convergence. If you are interested in such a fine resolution of these quantities, please have an eye on this.

Dear kirill,

In commit https://iffgit.fz-juelich.de/fleur/fleur...185ecdc2e827de0 I now added two more digits to the ev-sum output from MAE force theorem calculations. At the moment I start to believe that the new number for ev-sum in the MAE output in the out.xml file is correct. It may be that for the other places and in the old fleur version there is a double-counting of the valence states in this context. At least the new number seems to be more in agreement with calculations without SOC. I have to analyze this a little more.

I never cloned a certain commit. I suggest to clone the repository and then checkout the commit, i.e., with

git checkout develop
git checkout <commitHash>
 
 

where you replace <commitHash> with the actual commit hash. You probably don't even need the second line. At the moment the develop branch should be stable.

That is what I wrote about a few days ago. I have not yet fixed this, but what I can say is that the eigenvalue sum in the MAE output in the out.xml file is more consistent with a calculation without SOC than the other output. It was actually explicitly fixed some time ago to have only half that value. I still have to figure out where the factor 2 in the other output comes from.

I now fixed the problem with commit: https://iffgit.fz-juelich.de/fleur/fleur...92be8df89dd361c

This was only related to the output of the valence eigenvalue sum in the case of spin-orbit coupling calculations without activating noncollinear magnetism.