when I do structural optimization with high accuracy, I find the calculation is difficult to converg

Message

ljzhang123 · #1 Post by **ljzhang123** » Wed Oct 16, 2013 10:07 am

when I do structural optimization with high accuracy, I find the calculation is difficult to converge, there s a suddenly obvious change of E0, then the E0 seem to be the same in the following several ion steps. but can not converge .
following is the OSZICAR
18 F= -.74433177E+03 E0= -.74441706E+03 d E =0.545887E-04
19 F= -.74433445E+03 E0= -.74441941E+03 d E =-.267739E-02
20 F= -.74433433E+03 E0= -.74441931E+03 d E =-.255814E-02
21 F= -.74433355E+03 E0= -.74441867E+03 d E =-.177958E-02
22 F= -.74381037E+03 E0= -.74389543E+03 d E =0.521396E+00
23 F= -.74367000E+03 E0= -.74375463E+03 d E =0.140376E+00
24 F= -.74367707E+03 E0= -.74376188E+03 d E =0.133298E+00
25 F= -.74377250E+03 E0= -.74385732E+03 d E =-.954237E-01
26 F= -.74381037E+03 E0= -.74389531E+03 d E =-.133300E+00
27 F= -.74381132E+03 E0= -.74389612E+03 d E =-.944028E-03
28 F= -.74381084E+03 E0= -.74389573E+03 d E =-.471490E-03
29 F= -.74381088E+03 E0= -.74389575E+03 d E =-.392610E-04
30 F= -.74381083E+03 E0= -.74389567E+03 d E =0.185320E-04

here is the INCAR
Start parameter for the Run:
ISTART = 0 job : 0-new 1-cont 2-samecut
ICHARG = 2 charge: 1-file 2-atom 10-const
INIWAV = 1 electr: 0-lowe 1-rand
NPAR=4
Electronic Relaxation
PREC = Accurate
ENCUT = 550
ISPIN = 1 1:no split 2:with split
NBANDS = 540 total number of bands including empty bands
IALGO = 48 algorithm:38(more reliable) or 48
NELM = 60; NELMIN= 0; NELMDL= 3 Number of ELM steps (NELMDL:Number of random steps)
EDIFF = 1E-06 stopping criterion for ELM
BMIX = 1.0 mixing parameter
ISYM = 1 symmetry: 0-off 1-on (default for PP) 2-more efficient mem (default for PAW)
VOSKOWN= 1

Geometric Relaxation
ISIF = 3 relax cell shape, volume, or ions(2 or 3 often)
EDIFFG = -1E-03 stopping criterion for IOM (negative: use force as criteria. positive:use energy as criteria)
NSW = 70 number of steps for IOM
IBRION = 2 move atom: -1-not move 0-MD 1-quasi-Newton 2-conjugate-gradient 3-with damping factor specified by
SMASS
POTIM = 0.4 for IBRION=1,2,3: scaling constant for the forces; for IBRION=0: user must supply this value

Others
LREAL =.TRUE. real space projections, TRUE recomended if atoms are more than 20
LWAVE=.FALSE. write wave function (default) or not
LCHARG=.FALSE. write charge density (default) or not
LAECHG=.FALSE. write core charge density or not(default)
DOS related values:
ISMEAR = 1 for metal use 1, 0 for insulator
SIGMA = 0.2

I have thought the step length was too long . then I change the IBRION (use 1 replace 2). then the problem seems solved.
but the same problem appear when I do structural optimization on the similar structure. here is the OSZICAR

16 F= -.74433399E+03 E0= -.74441894E+03 d E =-.158429E-03
17 F= -.74433403E+03 E0= -.74441896E+03 d E =-.397506E-04
18 F= -.74433399E+03 E0= -.74441894E+03 d E =0.457535E-04
19 F= -.74433406E+03 E0= -.74441896E+03 d E =-.702266E-04
20 F= -.74328323E+03 E0= -.74336803E+03 d E =0.105082E+01
21 F= -.74328328E+03 E0= -.74336808E+03 d E =-.500865E-04
22 F= -.74328337E+03 E0= -.74336818E+03 d E =-.920086E-04
23 F= -.74328343E+03 E0= -.74336823E+03 d E =-.539963E-04
24 F= -.74328342E+03 E0= -.74336822E+03 d E =0.814987E-05

INCAR
Start parameter for the Run:
ISTART = 0 job : 0-new 1-cont 2-samecut
ICHARG = 2 charge: 1-file 2-atom 10-const
INIWAV = 1 electr: 0-lowe 1-rand
NPAR=4
Electronic Relaxation
PREC = Accurate
ENCUT = 550
ISPIN = 1 1:no split 2:with split
NBANDS = 540 total number of bands including empty bands
IALGO = 48 algorithm:38(more reliable) or 48
NELM = 80; NELMIN= 0; NELMDL= 3 Number of ELM steps (NELMDL:Number of random steps)
EDIFF = 1E-06 stopping criterion for ELM
BMIX = 1.0 mixing parameter
ISYM = 1 symmetry: 0-off 1-on (default for PP) 2-more efficient mem (default for PAW)
VOSKOWN= 1

Geometric Relaxation
ISIF = 3 relax cell shape, volume, or ions(2 or 3 often)
EDIFFG = -1E-03 stopping criterion for IOM (negative: use force as criteria. positive:use energy as criteria)
NSW = 70 number of steps for IOM
IBRION = 1 move atom: -1-not move 0-MD 1-quasi-Newton 2-conjugate-gradient 3-with damping factor specified by
SMASS
POTIM = 0.4 for IBRION=1,2,3: scaling constant for the forces; for IBRION=0: user must supply this value

Others
LREAL =.TRUE. real space projections, TRUE recomended if atoms are more than 20
LWAVE=.FALSE. write wave function (default) or not
LCHARG=.FALSE. write charge density (default) or not
LAECHG=.FALSE. write core charge density or not(default)
DOS related values:
ISMEAR = 1 for metal use 1, 0 for insulator
SIGMA = 0.2

the most surprise is that I have twice see the structure energy minimisation completed after E0's suddenly change.
OSZICAR

1 F= -.74427990E+03 E0= -.74436259E+03 d E =-.744280E+03
2 F= -.74427982E+03 E0= -.74436273E+03 d E =0.759253E-04
3 F= -.74323096E+03 E0= -.74331383E+03 d E =0.104894E+01
4 F= -.74323097E+03 E0= -.74331385E+03 d E =0.104892E+01
5 F= -.74323099E+03 E0= -.74331387E+03 d E =0.104891E+01
6 F= -.74323102E+03 E0= -.74331389E+03 d E =-.263748E-04
7 F= -.74323106E+03 E0= -.74331393E+03 d E =-.706322E-04
8 F= -.74323107E+03 E0= -.74331394E+03 d E =-.838868E-04
9 F= -.74323108E+03 E0= -.74331395E+03 d E =-.372497E-05
10 F= -.74323110E+03 E0= -.74331397E+03 d E =-.229883E-04
11 F= -.74323117E+03 E0= -.74331404E+03 d E =-.996460E-04
OUTCAR
reached required accuracy - stopping structural energy minimisation
LOOP+: cpu time12040.22: real time12071.56
4ORBIT: cpu time 0.00: real time 0.00

INCAR
Start parameter for the Run:
ISTART = 0 job : 0-new 1-cont 2-samecut
ICHARG = 2 charge: 1-file 2-atom 10-const
INIWAV = 1 electr: 0-lowe 1-rand
NPAR=4
Electronic Relaxation
PREC = Accurate
ENCUT = 550
ISPIN = 1 1:no split 2:with split
NBANDS = 540 total number of bands including empty bands
IALGO = 48 algorithm:38(more reliable) or 48
NELM = 60; NELMIN= 0; NELMDL= 3 Number of ELM steps (NELMDL:Number of random steps)
EDIFF = 1E-06 stopping criterion for ELM
BMIX = 1.0 mixing parameter
ISYM = 1 symmetry: 0-off 1-on (default for PP) 2-more efficient mem (default for PAW)
VOSKOWN= 1

Geometric Relaxation
ISIF = 3 relax cell shape, volume, or ions(2 or 3 often)
EDIFFG = -1E-03 stopping criterion for IOM (negative: use force as criteria. positive:use energy as criteria)
NSW = 70 number of steps for IOM
IBRION = 2 move atom: -1-not move 0-MD 1-quasi-Newton 2-conjugate-gradient 3-with damping factor specified by
SMASS
POTIM = 0.4 for IBRION=1,2,3: scaling constant for the forces; for IBRION=0: user must supply this value

Others
LREAL =.TRUE. real space projections, TRUE recomended if atoms are more than 20
LWAVE=.FALSE. write wave function (default) or not
LCHARG=.FALSE. write charge density (default) or not
LAECHG=.FALSE. write core charge density or not(default)
DOS related values:
ISMEAR = 1 for metal use 1, 0 for insulator
SIGMA = 0.2
but after I calculated again ,the E0 change
OSZICAR

1 F= -.74427980E+03 E0= -.74436248E+03 d E =-.744280E+03
2 F= -.74427956E+03 E0= -.74436249E+03 d E =0.245249E-03
OUTCAR
reached required accuracy - stopping structural energy minimisation
LOOP+: cpu time15667.43: real time15711.24
4ORBIT: cpu time 0.00: real time 0.00

INCAR
Start parameter for the Run:
ISTART = 0 job : 0-new 1-cont 2-samecut
ICHARG = 2 charge: 1-file 2-atom 10-const
INIWAV = 1 electr: 0-lowe 1-rand
NPAR=4
Electronic Relaxation
PREC = Accurate
ENCUT = 550
ISPIN = 1 1:no split 2:with split
NBANDS = 540 total number of bands including empty bands
IALGO = 48 algorithm:38(more reliable) or 48
NELM = 60; NELMIN= 0; NELMDL= 3 Number of ELM steps (NELMDL:Number of random steps)
EDIFF = 1E-06 stopping criterion for ELM
BMIX = 1.0 mixing parameter
ISYM = 1 symmetry: 0-off 1-on (default for PP) 2-more efficient mem (default for PAW)
VOSKOWN= 1

Geometric Relaxation
ISIF = 3 relax cell shape, volume, or ions(2 or 3 often)
EDIFFG = -1E-03 stopping criterion for IOM (negative: use force as criteria. positive:use energy as criteria)
NSW = 70 number of steps for IOM
IBRION = 2 move atom: -1-not move 0-MD 1-quasi-Newton 2-conjugate-gradient 3-with damping factor specified by
SMASS
POTIM = 0.4 for IBRION=1,2,3: scaling constant for the forces; for IBRION=0: user must supply this value

Others
LREAL =.TRUE. real space projections, TRUE recomended if atoms are more than 20
LWAVE=.FALSE. write wave function (default) or not
LCHARG=.FALSE. write charge density (default) or not
LAECHG=.FALSE. write core charge density or not(default)
DOS related values:
ISMEAR = 1 for metal use 1, 0 for insulator
SIGMA = 0.2

<span class='smallblacktext'>[ Edited ]</span>

alex · #2 Post by **alex** » Wed Oct 16, 2013 10:51 am

Hi,

EDIFFG = -1E-03 is surely not a 'good' criteria for precise structures.
And the total energy might change because of spin flips.
It also helps if you do not post the complete OSZICAR but e.g. the converged energies. :-)

Cheers,

alex

ljzhang123 · #3 Post by **ljzhang123** » Wed Oct 16, 2013 11:05 am

[quote author= -1E-03 is surely not a 'good' criteria for precise structures.
And the total energy might change because of spin flips.
It also helps if you do not post the complete OSZICAR but e.g. the converged energies. :-)

Cheers,

alex
[/quote]
Thank you for your attention. what do you mean by saying "EDIFFG = -1E-03 is surely not a 'good' criteria for precise structures." because I need to calculate the phonon frequency. so I have to ensure the force on the atom is small.

alex · #4 Post by **alex** » Wed Oct 16, 2013 1:43 pm

Hi again,

if I'm doing a phonon calculation on top of a structure I'll set the criteria to at least

EDIFF = 1.e-6
EDIFFG = -1.e-4

or better to

EDIFF = 1.e-8
EDIFFG = -1.e-6

in case I need a very good structure. With the latter INCAR the atomic positions are converged better than 0.1 pm (important for vibrational spectroscopy).

Cheers,

alex

ljzhang123 · #5 Post by **ljzhang123** » Wed Oct 16, 2013 2:17 pm

[quote author= 1.e-6
EDIFFG = -1.e-4

or better to

EDIFF = 1.e-8
EDIFFG = -1.e-6

in case I need a very good structure. With the latter INCAR the atomic positions are converged better than 0.1 pm (important for vibrational spectroscopy).

Cheers,

alex

[/quote]
That sounds great ! I'll have a try! Thank you very much!