Problem convergence SCF

Total energy, geometry optimization, DFT+U, spin....

Moderator: bguster

Locked
Samuel
Posts: 3
Joined: Mon Mar 22, 2010 7:16 pm
Location: LLN

Problem convergence SCF

Post by Samuel » Mon Mar 22, 2010 10:08 pm

Hi,

I'm currently trying to calculate the optimal height of a fcc FeNi L10 ferromagnetic supercell containing 10 atoms (it is an hexagonal structure in the (111) plane).


My SCF cycle does not converge:

Code: Select all

BROYDEN STEP NUMBER   0  ------------------------------------------------------

     iter   Etot(hartree)     deltaE(h) residm    nres2   diffor   maxfor   magn
 ETOT  1  -312.60618381500    -3.13E+02 2.09E+00 1.81E+03 5.3E-02 5.29E-02 9.199
 ETOT  2  -313.24670613267    -6.41E-01 4.52E-01 2.35E+03 4.9E-01 4.33E-01 11.84
 ETOT  3  -310.71193697027     2.53E+00 1.84E+00 2.49E+03 9.4E-01 7.09E-01 5.860
 ETOT  4  -309.43344505848     1.28E+00 1.92E+00 6.60E+02 1.2E+00 4.53E-01 6.908
 ETOT  5  -310.37168186074    -9.38E-01 1.32E+00 1.41E+03 1.7E+00 1.22E+00 5.316
 ETOT  6  -309.43477850700     9.37E-01 5.34E-01 6.82E+02 1.6E+00 6.85E-01 5.260
 ETOT  7  -309.95216905246    -5.17E-01 1.26E+00 1.88E+03 1.7E+00 1.65E+00 3.747
 ETOT  8  -309.40324893839     5.49E-01 9.23E-01 5.01E+02 1.6E+00 3.54E-01 5.816
 ETOT  9  -309.22283938569     1.80E-01 3.23E-01 1.04E+02 3.9E-01 1.25E-01 6.370
 ETOT 10  -309.34652237466    -1.24E-01 3.99E-01 3.34E+02 4.1E-01 3.39E-01 6.352
 ETOT 11  -309.22248899386     1.24E-01 1.04E-01 7.92E+01 5.6E-01 2.25E-01 7.310
 ETOT 12  -309.24050243038    -1.80E-02 1.12E-01 1.09E+02 2.1E-01 1.58E-01 7.399
ETOT 13  -309.22064803040     1.99E-02 3.18E-01 3.55E+01 1.8E-01 6.29E-02 7.718
 ETOT 14  -309.29479133657    -7.41E-02 5.13E-02 2.44E+02 4.0E-01 3.98E-01 6.885
 ETOT 15  -309.26522510592     2.96E-02 5.10E-02 1.08E+02 4.4E-01 1.64E-01 7.429
 ETOT 16  -309.26166770342     3.56E-03 1.12E-01 9.31E+01 3.6E-01 2.08E-01 8.079
 ETOT 17  -309.25967886189     1.99E-03 8.70E-02 6.36E+01 1.6E-01 9.57E-02 7.918
 ETOT 18  -309.24742252023     1.23E-02 5.69E-02 3.27E+01 1.3E-01 9.51E-02 7.917
 ETOT 19  -309.27230810643    -2.49E-02 2.66E-01 9.54E+01 2.1E-01 2.54E-01 8.150
 ETOT 20  -309.24597053519     2.63E-02 6.48E-02 2.98E+01 2.9E-01 1.66E-01 8.464
 ETOT 21  -309.24414502700     1.83E-03 1.46E-02 1.88E+01 1.7E-01 5.22E-02 8.504
 ETOT 22  -309.29235083028    -4.82E-02 6.20E-02 1.57E+02 1.4E-01 1.45E-01 8.148
 ETOT 23  -309.41836779211    -1.26E-01 2.48E-01 4.67E+02 7.3E-01 6.38E-01 6.611
 ETOT 24  -309.46204537162    -4.37E-02 2.07E-01 5.53E+02 9.2E-01 3.13E-01 6.778
 ETOT 25  -309.24637216961     2.16E-01 2.00E-01 2.08E+01 3.5E-01 9.52E-02 8.662
 ETOT 26  -309.25293656823    -6.56E-03 1.73E-02 2.70E+01 5.2E-02 7.24E-02 8.527
 ETOT 27  -309.27467387807    -2.17E-02 3.87E-02 6.90E+01 1.6E-01 1.83E-01 8.073
 ETOT 28  -309.32746201720    -5.28E-02 8.62E-02 2.07E+02 2.8E-01 3.73E-01 7.813
 ETOT 29  -309.25436609148     7.31E-02 6.83E-02 2.75E+01 4.2E-01 1.65E-01 8.693
 ETOT 30  -309.26130623710    -6.94E-03 4.11E-02 3.82E+01 2.2E-01 1.03E-01 8.608
 ETOT 31  -309.33341869759    -7.21E-02 1.08E-01 2.67E+02 1.6E-01 2.06E-01 8.205
 ETOT 32  -309.44296810488    -1.10E-01 3.65E-01 5.06E+02 8.2E-01 6.68E-01 6.759
 ETOT 33  -309.42917655047     1.38E-02 2.20E-01 5.20E+02 9.7E-01 3.01E-01 7.469
 ETOT 34  -309.25382772422     1.75E-01 2.02E-01 2.78E+01 3.6E-01 1.13E-01 8.981
 ETOT 35  -309.27961735513    -2.58E-02 6.26E-02 7.65E+01 9.4E-02 1.17E-01 8.546
 ETOT 36  -309.34997923735    -7.04E-02 1.22E-01 2.54E+02 4.7E-01 5.07E-01 7.314
 ETOT 37  -309.31302404989     3.70E-02 1.18E-01 1.74E+02 5.0E-01 3.65E-01 8.104
 ETOT 38  -309.35876277181    -4.57E-02 1.03E-01 3.00E+02 6.2E-01 6.74E-01 7.667
 ETOT 39  -309.31328334626     4.55E-02 3.06E-01 1.52E+02 7.9E-01 2.52E-01 8.503
 ETOT 40  -309.32274257938    -9.46E-03 2.01E-01 2.47E+02 3.0E-01 2.03E-01 8.328
 ETOT 41  -309.39059258932    -6.79E-02 2.80E-01 3.93E+02 7.3E-01 5.62E-01 7.162
 ETOT 42  -309.34860082067     4.20E-02 1.52E-01 3.15E+02 7.8E-01 2.24E-01 8.173
 ETOT 43  -309.44698355550    -9.84E-02 3.90E-01 7.58E+02 9.2E-01 7.74E-01 7.880
 ETOT 44  -309.50325836390    -5.63E-02 5.55E-01 6.68E+02 7.9E-01 5.13E-01 6.525
 ETOT 45  -309.61951086313    -1.16E-01 4.99E-01 7.63E+02 1.1E+00 1.16E+00 6.005
 ETOT 46  -309.69608081524    -7.66E-02 4.12E-01 9.03E+02 1.1E+00 1.17E+00 5.768
 ETOT 47  -309.53003125265     1.66E-01 2.95E-01 6.42E+02 1.3E+00 1.05E+00 6.485
 ETOT 48  -309.75580495089    -2.26E-01 1.34E+00 1.43E+03 1.6E+00 1.07E+00 5.996
 ETOT 49  -309.94254293567    -1.87E-01 1.58E+00 1.80E+03 1.0E+00 9.12E-01 6.036
 ETOT 50  -309.98266200099    -4.01E-02 1.30E+00 1.07E+03 1.9E+00 9.78E-01 5.246

 Cartesian components of stress tensor (hartree/bohr^3)
  sigma(1 1)= -1.31424401E-02  sigma(3 2)= -7.62161247E-04
  sigma(2 2)= -6.74268992E-03  sigma(3 1)= -4.82769636E-07
  sigma(3 3)=  1.61621942E-03  sigma(2 1)= -4.09508824E-07


 scprqt:  WARNING -
  nstep=   50 was not enough SCF cycles to converge;
  density residual=  1.074E+03 exceeds tolvrs=  1.000E-12


I already have try with:

1) iprcel 45
2) iprcel 145
3) iprcel 145 and nnsclo à 3.
4) iprcel 145 and nnsclo à 4.
5) iprcel 145, nnsclo à 4 and nline 5
6) iprcel 0 and diemixmag -4

None of this converge.

I've attached the input file.

Best regards.

Ps: I cannot put the log file because .tar and .gz files are not allowed. In addition the max size is 2MB.
Attachments
Position0TEMP1.in
(2.09 KiB) Downloaded 414 times
Last edited by Samuel on Tue Mar 23, 2010 11:02 am, edited 1 time in total.

Samuel
Posts: 3
Joined: Mon Mar 22, 2010 7:16 pm
Location: LLN

Re: Problem convergence SCF

Post by Samuel » Tue Mar 23, 2010 11:04 am

Advertisement spam?

7) I've also try with only one long SCF cycle:

Code: Select all

 scprqt:  WARNING -
  nstep=  500 was not enough SCF cycles to converge;
  density residual=  2.619E+03 exceeds tolvrs=  1.000E-12

mverstra
Posts: 655
Joined: Wed Aug 19, 2009 12:01 pm

Re: Problem convergence SCF

Post by mverstra » Tue Mar 23, 2010 11:52 am

magnetic systems are often hard to converge. Try with different initial spinat values, try increasing nline, and you can try preconditionning with a small diemix and some value for diemac (but I doubt this will be sufficient).

I have heard of a very clever combination of scf algorithms which does an additional rotation of the states to make the convergence more monotonic (and much quicker) but this is not implemented in abinit, alas.

good luck

Matthieu
Matthieu Verstraete
University of Liege, Belgium

Locked