[CPMD-list] PROGRAM STOPS IN SUBROUTINE UINV| ILLEGA

Juerg Hutter hutter at pci.unizh.ch
Mon Apr 28 13:16:16 CEST 2003


Hi

your problem seems to be that the wavefunction optimization
didn't converge. If this happens with PCG MINIMIZE
you can see that the step size goes to zero (LAMBDA=10^-309
in your case).
Experience tells us that this usually happens with systems
that a far from equilibrium (e.g. at the beginning of a
geometry optimization). It is therefore a good
practise to start an optimisation with a relaxed
convergence criteria and tighten it when approaching
the minimum.
There is an automatic scheme to do this with the LBFGS
optimizer, have a look in the manual.

In general, I would advice everybody to treat optimizations
not as black box jobs. Have a look at the output and
stop runs when irregularities appear. Then restart
with an adapted set of options.
This way a lot of computer time can be saved.

regards

Juerg Hutter

----------------------------------------------------------
Juerg Hutter                   Phone : ++41 1 635 4491
Physical Chemistry Institute   FAX   : ++41 1 635 6838
University of Zurich           E-mail: hutter at pci.unizh.ch
Winterthurerstrasse 190
CH-8057 Zurich, Switzerland
----------------------------------------------------------


On Sun, 27 Apr 2003 cdy2000 at vip.sina.com wrote:

> Hello, every CPMDer!
>
>     when I optimize the GEOMETRY of a oxide, there is the problem after many steps:
>
> 3485  3.416E-04   2.018E-05   -1447.214130    0.000E+00    185.50
>  LINE SEARCH: WARNING! 2*E0-4*E1+2*E2 == 0.0
>  LINE SEARCH : LAMBDA=.399-309       ENERGY =          -1447.2141
> 3486  3.416E-04   2.018E-05   -1447.214130    0.000E+00    185.38
>  LINE SEARCH : LAMBDA=.797-308       ENERGY =          -1447.2141
> 3487  3.416E-04   2.018E-05   -1447.214130    0.000E+00    185.56
>  LINE SEARCH: WARNING! 2*E0-4*E1+2*E2 == 0.0
>  LINE SEARCH : LAMBDA=.747-310       ENERGY =          -1447.2141
> 3488  3.416E-04   2.018E-05   -1447.214130    2.274E-13    185.81
>  LINE SEARCH : LAMBDA=.747-309       ENERGY =          -1447.2141
> 3489  3.416E-04   2.018E-05   -1447.214130    0.000E+00    186.19
>  UNIV| THE LEADING MINOR OF ORDER    1 IS NOT POSITIVE DEFINITE,
>  UNIV| AND THE FACTORIZATION COULD NOT BE COMPLETED.
>
>
>  PROGRAM STOPS IN SUBROUTINE UINV| ILLEGAL RESULTS DPOTRF
>
> then the process ends, I want to know how to solve the problem in modifying the parameters of the input code?
>
>     Thanks for any help!
>
> Sincerely Cao Dayong
> ______________________________________
>
> ===================================================================
> _______________________________________________
> CPMD-list mailing list
> CPMD-list at cpmd.org
> http://www.cpmd.org/mailman/listinfo/cpmd-list
>



More information about the CPMD-list mailing list