[CPMD-list] BO-MD starting problem after successful WFO

Wed Jun 6 16:43:44 CEST 2007

Dear Matus,

thank you for the comments!

> Why You set velocities in WF optimization run?
> Why You restart velocities in MD run and also set them? I think it can cause errors.
Velocities are ignored during WF optimization. That's why I do not remove this piece. 

> You have to set much higher cutoff with psps You are using, at least 15.00 maybe 20.00 - low cutoff can cause unstable MD.
In fact at some parameters (cutoff, electron temperature) BO-MD works fine.
Perhaps initially I have written not clear enough. But my question was: why it can be so that BO-MD initial diagonalization *after* WF optimization can fail (sometimes it *is* successful but takes too much time). 
All the relevant parameters in BO-MD are just the same as in WF opt.

Of course I have noticed that generally all things go much better at higher cut-offs. However in order to increase the speed at acceptable precision I try to choose the lowest cut-off possible (e.g. the same cut-off 8Ry with the same potential AL_SGS was used for BO-MD in Stumpf and Scheffler, Phys. Rev. B 53 (1996) 4958, however not for MD).

As to the time step, the 100 a.u. seems to be normal. When MD happened to do well I checked that the total energy drift is not big. Besides in the textbook of Marx&Hutter (page 30) it is shown to be acceptable. At least BOMD can work.
And again, as I understand, the timestep is irrelevant to the initial diagonalization step in BOMD where the program fails in my case.

Best regards,
Vladimir

P.S. K-points *is* implemented for BO-MD.

P.P.S. I think the copy to the list could be useful.

  ----- Original Message ----- 
  From: Matus Dubecky 
  To: Vladimir Stegailov 
  Sent: Wednesday, June 06, 2007 2:18 PM
  Subject: Re: [CPMD-list] BO-MD starting problem after successful WFO

  Dear Vladimir,

  I am not an cpmd expert, but..I see some inconsistencies.
  Why You set velocities in WF optimization run?
  Why You restart velocities in MD run and also set them? I think it can cause errors. Not good idea.
  You have to set much higher cutoff with psps You are using, at least 15.00 maybe 20.00 - low cutoff can cause unstable MD. You have very large timestep, try to set it to 20. This timestep will probably never converge. Firstly try Your MD run with lower number of atoms and without k-points. (I am not sure that BOMD is implemented with K-points, you have to dig into manual).
  Hope this helps.

  Best regards, 
  Matus Dubecky.

  PS. If You have more questions to me, please, do not reply to cpmd-mail list, write me directly.

    &CPMD
    OPTIMIZE WAVEFUNCTION 
    FREE ENERGY FUNCTIONAL
    LANCZOS DIAGONALISATION 
    LANCZOS PARAMETERS
    50 8 15 1.D-18
    TROTTER FACTOR
    0.001
    BOGOLIUBOV CORRECTION OFF 
    GRAM-SCHMIDT ORTHOGONALISATION
    CONVERGENCE ORBITALS
    1.D-6
    MAXSTEP
    5000
    BROYDEN MIXING
    0.15 200 0.01 0 8
    ALEXANDER MIXING
    1.1 
    ELECTRON TEMPERATURE
    1000.
    COMPRESS WRITE32
    RHOOUT
    &END
    &SYSTEM
    POINT GROUP
    AUTO
    SYMMETRY
    1
    CELL
    8.064 1.0 1.0 0.0 0.0 0.0 (8.064=2*4.032, 4.032A is the eq lattice const of Al)
    CUTOFF
    8.000
    ANGSTROMS
    STATES
    90
    SCALE
    TESR
    1
    KPOINTS MONKHORST-PACK FULL
    2 2 2 
    &END
    &ATOMS
    *AL_SGS KLEINMAN-BYLANDER
    LMAX=D
    32
    0.0178998 -0.0197471 0.0145112
    ....
    0.500011 0.750491 0.756745
    VELOCITIES
    32 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32
    -5.10982 0.101812 -1.43574
    ....
    7.04835 2.96887 2.84766
    END VELOCITIES
    &END
    &BASIS
    PSEUDO AO 2
    0 1 
    &END
    &DFT
    NEWCODE
    &END

    &INFO
    BO-MD
    &END
    &CPMD
    FILEPATH
    /home/stegailov/work/cpmd/nose/
    RESTART WAVEFUNCTION OCCUPATION COORDINATES VELOCITIES KPOINTS
    MOLECULAR DYNAMICS BO
    FREE ENERGY FUNCTIONAL
    LANCZOS DIAGONALISATION 
    LANCZOS PARAMETERS N=5
    50 8 15 1.D-9
    0.05 1.D-9
    0.01 1.D-13
    0.0025 1.D-16
    0.001 1.D-18
    TROTTER FACTOR
    0.001
    BOGOLIUBOV CORRECTION OFF 
    GRAM-SCHMIDT ORTHOGONALISATION
    CONVERGENCE ORBITALS
    1.D-4
    MAXSTEP
    100
    BROYDEN MIXING
    0.15 200 0.01 0 8
    ALEXANDER MIXING
    1.1 
    TEMPERATURE
    1000.
    ELECTRON TEMPERATURE
    1000.
    TIMESTEP
    100
    STORE
    100
    TRAJECTORY XYZ
    COMPRESS WRITE32
    ELECTROSTATIC POTENTIAL
    RHOOUT
    &END
    &SYSTEM
    POINT GROUP
    AUTO
    SYMMETRY
    1
    CELL
    8.064 1.0 1.0 0.0 0.0 0.0 (8.064=2*4.032, 4.032A is the eq lattice const of Al)
    CUTOFF
    8.000
    ANGSTROMS
    STATES
    90
    SCALE
    TESR
    1
    KPOINTS MONKHORST-PACK FULL
    2 2 2 
    &END
    &ATOMS
    *AL_SGS KLEINMAN-BYLANDER
    LMAX=D
    32
    0.0178998 -0.0197471 0.0145112
    ....
    0.500011 0.750491 0.756745
    VELOCITIES
    32 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32
    -5.10982 0.101812 -1.43574
    ....
    7.04835 2.96887 2.84766
    END VELOCITIES
    &END
    &BASIS
    PSEUDO AO 2
    0 1 
    &END
    &DFT
    NEWCODE
    &END
    _______________________________________________
    CPMD-list mailing list
    CPMD-list at cpmd.org
    http://cpmd.org/mailman/listinfo/cpmd-list

------------------------------------------------------------------------------
  Looking for a deal? Find great prices on flights and hotels with Yahoo! FareChase.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://cpmd.org/pipermail/cpmd-list/attachments/20070606/f5fdb335/attachment.html 