[CPMD-list] Davidson diago of TDDFT
Juerg Hutter
hutter at pci.unizh.ch
Fri Nov 5 13:34:22 CET 2004
Hi
> Dear List Subscriber,
>
> I am trying to calculate electronic spectra using TDDFT (v3.9.1). When some of
> the roots are not converged after the Davidson diago step of the TDDFT matrix, I
> understand RDIIS kicks in. What puzzles me in my calculation is the point where
> this transition takes place. It seems to me that RDIIS is switched on even
> though either MAX ITER or MAX SUBSPACE of DAVIDSON DIAGO has not been exceeded
> (see portions of my output below). Is there a third criterion to stop DAVIDSON?
> Thanks very much for reading.
If progress in the Davidson diagonalisation is very slow and the gradient
is small enough, the program switches to RDIIS. This can be controled by
DAVIDSON RDIIS
nrdiismax,nrestdmax,rdiistin
where nrdiismax is max number of rdiis steps
nrestdmax maximum number of restarts
rdiistin minimum gradient to start RDIIS
regards
Juerg Hutter
>
> Best regards,
>
> EG Kim
>
> #############################################################################
>
> *********************** LINEAR RESPONSE ************************
> Step size for numeric dmu/dn : .500E-02
> Number of calculations for dmu/dn : 2
> Maximum number of optimisation steps: 1000
> Threshold for Hessian (Preconditioner) .5000
> Optimizer for LR equations AUTOMATIC
> Size of ODIIS buffer 5
> Size of ZDIIS buffer 4
> Switch from PCG to ODIIS at .1000E+00
> Switch to full preconditioning at .1000E-02
> Step length .1000
> Convergence criteria .1000E-04
> ****************************************************************
>
> *************************** TDDFT ****************************
> Step size for numeric dmu/dn : .500E-02
> Number of calculations for dmu/dn : 2
> Tamm-Dancoff Approximation
> Diagonalization Method DAVIDSON
> Max. number of iterations 200
> Convergence criteria .100E-06
> Max. size of Davidson matrix 50
> RDIIS Parameters
> NTDIISMAX= 20 NRESTDMAX= 3 RDIISTIN= .100E-02
> Number of Singlet States 3
> Forces calculated for state 1
> ****************************************************************
> .
> .
> .
> .
> .
> *** SPECTRA| THE NEW SIZE OF THE PROGRAM IS 286956 KBytes ***
> == SINGLET STATES ==
> NUMBER OF STATES TO BE INITIALIZED 3
> TOTAL NUMBER OF TEST VECTORS 6
> *** SPECTRA| THE NEW SIZE OF THE PROGRAM IS 319776 KBytes ***
> ================================================================
> == DAVIDSON DIAGONALISATION OF TDDFT MATRIX ==
> ================================================================
> ITER STATES SUBSPACE RESIDUAL TCPU
> 1 0 3 .13533773E-01 87.40
> 2 0 6 .46776986E-02 88.93
> 3 0 9 .30463019E-02 90.23
> 4 0 12 .33203299E-02 91.35
> 5 0 15 .22075106E-02 93.19
> 6 0 18 .11336981E-02 96.40
> 7 0 21 .67511167E-03 98.32
> 8 0 24 .64601509E-03 100.01
> 9 0 27 .57523001E-03 105.70
> 10 0 30 .47208419E-03 373.56
> 11 0 33 .36949736E-03 253.12
> 12 0 36 .40133983E-03 752.00
> 13 0 39 .59391461E-03 2563.71
> >>> RESIDUAL DIIS OPTIMIZATION OF ROOT 1
> ITER ENERGY SUBSPACE RESIDUAL TCPU
> 1 2.88328 1 .30627204E-04 108.58
> 2 2.88311 2 .82109007E-05 155.98
> 3 2.88299 3 .66425066E-05 61.41
> 4 2.88291 4 .43031780E-05 53.93
> .
> .
> .
> .
> .
>
> _______________________________________________
> CPMD-list mailing list
> CPMD-list at cpmd.org
> http://cpmd.org/mailman/listinfo/cpmd-list
>
More information about the CPMD-list
mailing list