Troubles converging low scaling RPA for metallic systems with band crossings

[kdoblhoff]

Dear VASP community,

I am having troubles in converging the low scaling ACFDT-RPA method for bulk Al. What happens is the following: As soon as I increase my K-point mesh to 10x10x10, I see the minimum excitation energy getting very small. As soon as OMEGAMIN (as detected automatically) gets smaller than 0.05 eV, the RPA correlation energy starts to give incorrect values. Increasing NOMEGA from 16 to 20 to 24 does not solve the problem:

Code: Select all

NOMEGA = 16
      cutoff energy     smooth cutoff   RPA   correlation   Hartree contr. to MP2
---------------------------------------------------------------------------------
            166.667           133.333    -1780.9308643871********************
            158.730           126.984    -1781.0094794909********************
            151.172           120.937    -1781.2266521184********************
            143.973           115.178    -1781.2923937959********************
            137.117           109.694    -1781.2340806750********************
            130.588           104.470    -1781.2611010667********************
            124.369            99.495    -1781.3385357701********************
            118.447            94.758    -1781.3666223710********************
  linear regression
  converged value                        -1780.4551769844********************
  
  
NOMEGA = 20
      cutoff energy     smooth cutoff   RPA   correlation   Hartree contr. to MP2
---------------------------------------------------------------------------------
            166.667           133.333 -1841750.2556570359********************
            158.730           126.984 -1841758.2503074536********************
            151.172           120.937 -1841757.7727216494********************
            143.973           115.178 -1841724.0818368408********************
            137.117           109.694 -1841697.6935746986********************
            130.588           104.470 -1841649.1188613100********************
            124.369            99.495 -1841595.1484981293********************
            118.447            94.758 -1841550.6474406589********************
  linear regression
  converged value                     -1842113.0359588007********************

Restricting OMEGAMIN to 0.05 eV in the INCAR does allow me to get reasonable values,

Code: Select all

NOMEGA = 16
      cutoff energy     smooth cutoff   RPA   correlation   Hartree contr. to MP2
---------------------------------------------------------------------------------
            166.667           133.333       -4.9300282298      -15.7623601002
            158.730           126.984       -4.9228244905      -15.7544975990
            151.172           120.937       -4.9150881364      -15.7461159963
            143.973           115.178       -4.9064852493      -15.7371977502
            137.117           109.694       -4.8976110150      -15.7276748460
            130.588           104.470       -4.8878102895      -15.7174740140
            124.369            99.495       -4.8775908195      -15.7065760995
            118.447            94.758       -4.8665886704      -15.6948510066
  linear regression
  converged value                           -5.0249631892      -15.8629128458

but the energy fluctuates on the order of several tens of meV. Consequently, it is not possible any more to compute a reasonable equation of state (energy vs. volume/lattice constant).
It is obvious that there is still a problem, as the results get worse when increasing NOMEGA. However, I do not quite understand this behavior and I would be happy if someone could explain it to me (sorry, might be a very naive question...)

Is such a system "convergable" using the minimax method? If so, at which knobs do I need to turn?
(I should probably mention that the system converges nicely for NOMEGA>=16) when using the 4-step routine with the high scaling RPA method as was already implemented in VASP 5.)

Thank you for your help,
Best regards,
Katharina Doblhoff-Dier

[merzuk.kaltak]

Dear Katharina,

Please use the finite-temperture implementation of the ACFDT/RPA code in vasp as suggested in forum/viewtopic.php?t=18039
This boils down to add (or replace) following entries in the INCAR file

Code: Select all

ISMEAR=-1 ; LFINITE_TEMPERATURE = .TRUE. 
NOMEGA=12
ALGO = ACFDTR # or ALGO=RPAR

Best,
Merzuk