(1) Variable for geometry convergence status? (2) optg failure with M06-2X functional

[Irikura, Karl K. Dr. (Fed)]

Dear Molpro experts,

 

(1) Is there an accessible variable to indicate that a geometry optimization (OPTG) has converged, and not merely reached the iteration limit? 

 

(2) I find that the following job fails to converge the geometry even with 100 iterations.  A few other functionals that I tried worked fine.   

 

Thanks, as usual, for any suggestions!

Karl

 

***,HOCl test lack of geometry convergence

 

memory,1,G;

 

angstrom;

geometry={

O

H 1 oh

Cl 1 ocl 2 a

}

 

oh=1.

ocl=1.7

a=103.

 

basis=def2-tzvp

 

df-ks,M06-2X

optg,maxit=20

 

-------------------------------------------------------------------

Dr. Karl K. Irikura

National Institute of Standards and Technology

100 Bureau Drive, mail stop 8320

Gaithersburg, MD 20899-8320  USA

voice:  301-975-2510   fax:  301-975-3670

email:  karl.i...@nist.gov

http://www.nist.gov/mml/csd/informatics_research/index.cfm

-------------------------------------------------------------------

 

[andreas...@gmail.com]

(1):

the variables

 OPTCONVMX
 OPTCONVRMS
 OPTGRAD
 OPTGRADMX
 OPTGRADRMS

contain the values of the respective columns from the optg output after each iteration step.

So when optg reached the maximum number of iterations without reaching the

convergence criteria one can check this with some conditional in the input file, like

if(OPTCONVMX.gt.1d-5) then

   text,not yet converged!

endif

depending on the convergence criteria which were used.

Concerning (2):

it could be that the convergence gets better if you increase the grid size, eg, add

     gthresh,grid=1d-10

or similar at the beginning of the input file. Another point is that Molpro by default does not

include the gradient contribution for the quadrature weights (which is why for small grids the

gradient is slighty inaccurate). This can be changed by using:

   ks,m06-2x,gridgrad=1

   optg

I hope this will help to achieve convergence of optg for the system you are studying.

Other than that, one can of course always try out other optimisation methods for the geometry

optimisation, see the descriptions in the manual:

https://www.molpro.net/manual/doku.php?id=geometry_optimization_optg#options_for_optimization_methods

Best wishes,

Andreas

[andreas...@gmail.com]

(1): and variable OPTSTEP will store the number of iterations done in optg. So you can also just check

      if it equals the maximum number specified in order to see if the optimisation converged or not.

      (gradient norm could be checked as well in the unlikely but possible case that OPTG just required

       as many steps to converge as were given as the maximum in the input file).

Andreas