# Is it possible to perform ROMP2 numerical optimizations in Gaussian09?

I wish to perform an restricted open-shell MP2 (ROMP2) geometry optimization.

Gaussian09 currently only provides energies at ROMP2 level of theory.

It should be possible to perform an optimization by displacement of the structure followed by single-point-energy evaluations at the displaced geometries to obtain a gradient. In other quantum-chemical software packages this (slower) numerical optimization technique is readily available. I cannot seem to get it to work in the case of ROMP2 in Gaussian09. I have already tried the usual options, such as #p opt roMP2/cc-pVDZ scf=tight and so on. All of my approaches failed and yielded the following error:

 NEF-NEF-NEF-NEF-NEF-NEF-NEF-NEF-NEF-NEF-NEF-NEF-NEF-NEF-NEF-NEF-NEF-NEF-
NUMERICAL EIGENVECTOR FOLLOWING MINIMUM SEARCH
INITIALIZATION PASS

************************************************
** ERROR IN INITNF. NUMBER OF VARIABLES (  0) **
**   INCORRECT (SHOULD BE BETWEEN 1 AND 50)   **
************************************************


I am aware of this website: http://www.somewhereville.com/?p=1571 and I have tried the solution outlined therein, but it has failed in the same manner as well.

Any help or pointers towards other software that is capable of ROMP2 optimizations would be appreciated.

• Could you post a sample geometry that this fails for? Jun 11 '17 at 15:26
• For numerical optimization you have to provide a limited number of parameters you want to optimize. Gaussian doesn't find those in your input.
– Greg
Jun 11 '17 at 23:08
• Is there a reason, apart from computational cost, that you are not interested in unrestricted MP?
– user41033
Jun 14 '17 at 9:21
• @Vic Lineal: Severe spin contamination and symmetry broken orbitals. The system has multi reference character, therefore all single determinant methods will yield more or less non sensical data. I still want to benchmark to which extend cheaper methods yield accurate results. Jun 14 '17 at 17:34