1). I'm dealing with a system where the default mixing scheme always failed to converge. I would like to use simple linear mixing (mixtype=0) with very small mixing parameter. However, I can't find any appropriate does the job. Could anyone tell me how to set the mixing parameter when mixtype=0 ?
2) When use Boryden mixing, the mixing parameters can be set using broydmp. The variable broydmp has two values which correspnd to \alpha and \omega. Can anyone explain the meaning of these two values? I didn't find any useful resource to suggest me how to tune these paremeters.
3) I'm using MPI to run elk. However, the output file INFO.OUT always shows the information of each LOOP repeatedly. Is it because that I'm using MPI, so each process will return the LOOP information once?
Thanks!
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
I never used the pure linear mixing an would rather suggest you to use
1)
mixtype
1
and reduce the maximal and minimal mixing as
betamax
0.35
beta0
0.01
which in most cases helped for me. But note that also other things can effect the non-convergece including applied magnetic field, gmaxvr ... so without detail knowledge this is the only thing I can suggest
2) Never changed the broyden stuff ...
3) This sounds your compilation cluster has a problem since every loop should only be stated once. Could you try to take out the openMP flag from the make.inc file, recompile and try again
best
Michael
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
1) I second Michael's approach: reduce the beta parameters, maybe even down to beta0 0.005 and betamax 0.05 or so. This will almost enforce convergence, however, you may end up with the wrong "ground" state.
2) The Broyden mixer is in general more prone to convergence issues, specifically in magnetic systems. The parameters are carefully optimized for a large set of systems for which they give good across-the-board performance. However, they are by no means optimal for all systems. You may try to reduce the first number (by default 0.4) to, say, 0.2, but you might have to adjust the second number, (by default 0.15) to a smaller value as well. Difficult to say without testing.
3) Sounds like you have an error in the MPI section of make.inc. Please read the comments carefully and check that the file is correct. Then recompile. ;)
Regards,
Markus
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
Dear all,
I'm new to elk. I've two basic questions:
1). I'm dealing with a system where the default mixing scheme always failed to converge. I would like to use simple linear mixing (mixtype=0) with very small mixing parameter. However, I can't find any appropriate does the job. Could anyone tell me how to set the mixing parameter when mixtype=0 ?
2) When use Boryden mixing, the mixing parameters can be set using broydmp. The variable broydmp has two values which correspnd to \alpha and \omega. Can anyone explain the meaning of these two values? I didn't find any useful resource to suggest me how to tune these paremeters.
3) I'm using MPI to run elk. However, the output file INFO.OUT always shows the information of each LOOP repeatedly. Is it because that I'm using MPI, so each process will return the LOOP information once?
Thanks!
Dear Pididog,
I never used the pure linear mixing an would rather suggest you to use
1)
mixtype
1
and reduce the maximal and minimal mixing as
betamax
0.35
beta0
0.01
which in most cases helped for me. But note that also other things can effect the non-convergece including applied magnetic field, gmaxvr ... so without detail knowledge this is the only thing I can suggest
2) Never changed the broyden stuff ...
3) This sounds your compilation cluster has a problem since every loop should only be stated once. Could you try to take out the openMP flag from the make.inc file, recompile and try again
best
Michael
Hi all,
1) I second Michael's approach: reduce the beta parameters, maybe even down to beta0 0.005 and betamax 0.05 or so. This will almost enforce convergence, however, you may end up with the wrong "ground" state.
2) The Broyden mixer is in general more prone to convergence issues, specifically in magnetic systems. The parameters are carefully optimized for a large set of systems for which they give good across-the-board performance. However, they are by no means optimal for all systems. You may try to reduce the first number (by default 0.4) to, say, 0.2, but you might have to adjust the second number, (by default 0.15) to a smaller value as well. Difficult to say without testing.
3) Sounds like you have an error in the MPI section of make.inc. Please read the comments carefully and check that the file is correct. Then recompile. ;)
Regards,
Markus