[Wien] problem with energy convergence

[Marco Schowalter]

Dear Angelica


>>Hello Wien2k users!
>>
>> 
>>
>>I have problem with energy convergence, the parameters that I used is: Rmax=9, 
>>mix=0.4 and I took kmesh=50. The energy convergence =0.0001Ry. Here are the results:
>>
>> 
>>
>>--- ENE -----------
>>
>>:ENE  : ********** TOTAL ENERGY IN Ry =       -24021.848965
>>
>>:ENE  : ********** TOTAL ENERGY IN Ry =       -24021.860569
>>
>>:ENE  : ********** TOTAL ENERGY IN Ry =       -24021.879533
>>
>>:ENE  : ********** TOTAL ENERGY IN Ry =       -24021.886065
>>
>>:ENE  : ********** TOTAL ENERGY IN Ry =       -24021.892995
>>
>>:ENE  : ********** TOTAL ENERGY IN Ry =       -24021.916703
>>
>>:ENE  : ********** TOTAL ENERGY IN Ry =       -24021.957049
>>
>>:ENE  : ********** TOTAL ENERGY IN Ry =       -24022.042321
>>
>>:ENE  : ********** TOTAL ENERGY IN Ry =       -24022.052118
>>
>>:ENE  : *WARNING** TOTAL ENERGY IN Ry =       -24022.061428
>>

I think this warning is due to the limited size of NMATMAX.. You have to 
increase it and to recompile, But you can check this in the vicinity of 
this Warning message in case.dayfile. Otherwise wien will reduce 
artificially your RKMAX..

>>--- DIS -----------
>>
>>:DIS  :  CHARGE DISTANCE       0.3933245
>>
>>:DIS  :  CHARGE DISTANCE       0.3875391
>>
>>:DIS  :  CHARGE DISTANCE       0.3743100
>>
>>:DIS  :  CHARGE DISTANCE       0.3718883
>>
>>:DIS  :  CHARGE DISTANCE       0.3729222
>>
>>:DIS  :  CHARGE DISTANCE       0.3450249
>>
>>:DIS  :  CHARGE DISTANCE       0.3055749
>>
>>:DIS  :  CHARGE DISTANCE       0.2449718
>>
>>:DIS  :  CHARGE DISTANCE       0.2294169
>>
>>:DIS  :  CHARGE DISTANCE       0.2018121
>>
>> 
>>
>>I tried to change Rmax to 10 in one case and to reduce the mixing parameter to 
>>0.2 instead of 0.4, but the convergence get even worse in both cases.
>>
>>I know that I need to enlarge the number of k points, but the SCF calculation 
>>takes to much time.

I do not really get your point. If you increase rkmax calculations will 
also take more time. So why do you not stay with smaller
RKMAX at the beginning and converge the k mesh? However, I have the 
impression that you are calculating a large system. Can you not start 
with a small system and optimize  the k mesh and rkmax? Then you can 
transfer the k-mesh towards the larger system.


>>
>>My question is, is there an option to improve the convergence, without taking 
>>bigger number of k points.
>>
>>Also is it possible to continue the SCF calculations after 20 iterations.

You can call run_lapw again. You then have further 20 iterations, Or you 
can call e.g. run_lapw -i 30. The -i Option
allows you to enlarge the number of maximum iterations.

>>
>> 
>>
>>Thanks in advance,
>>
>>Angelica 
>>
>> 
>>
>> 
>>

Best regards

Marco