TURBOMOLE Users Forum

TURBOMOLE Modules => Relax => Topic started by: ghill on May 22, 2007, 01:29:47 PM

Title: Fitting functions coalesce in cbasopt
Post by: ghill on May 22, 2007, 01:29:47 PM

I'm trying to optimize an RI-MP2 fitting basis with cbasopt and I'm running into problems with one or more exponents of the same l quantum number becoming very similar.

Is this just a sign that I need to create larger spaces between consecutive exponents or is it more a suggestion that I should be considering more or less functions in my basis set?

One thing that I have noticed is that the HF energy for the system is always reported as zero. Might this be an indication of a problem in my input or the way I'm attempting to run the optimization?

Any tips or advice on using cbasopt would be greatly appreciated.

Title: Re: Fitting functions coalesce in cbasopt
Post by: christof.haettig on May 23, 2007, 08:39:42 AM

Hi,

1) the HF energy is not used in the optimization of the auxiliary basis functions
     for RI-MP2/RI-CC2. At some point of the basis set optimization it is overwritten
     by something else. Relax needs the functional which should be minimized
     and its gradient.
2) coalescence of parameters is a very common problem in non-linear
     optimization/fitting... the reasons can be very different:
             * redundant exponents
             * missing tighter or diffuser exponents
             * missing exponents of higher or lower angular momentum

Christof

Title: Re: Fitting functions coalesce in cbasopt
Post by: ghill on May 23, 2007, 12:00:03 PM

Thanks for the advice.