While CCSD excitation energies and the treatment of excited state potential energy surfaces with CCSD are reasonably good, vast improvement is obtained at the CCSDT and CCSDTQ levels. However, these latter methods are very expensive, and it is desirable to have a compromise between the accuracy of the latter approaches and the simplicity and cost of the former. For ground-state calculations, of course, the CCSD(T) method has enjoyed great success for a quarter-century, but this method does not formally lend itself to an excited state extension.
CFOUR has a few options for treating the effects of triple excitations noniteratively in excited state calculations. These are:
1. The "EOM-CCSD(T)" method of Bartlett and coworkers.
2. The "EOM-CCSD*" method of Stanton and Gauss.
3. The "EOM-CCSD(T)(a)*" method of Matthews and Stanton.
To run option #1 above, specify EOM_NONIT=QTP, together with CALC=CCSD,EXCITE=EOMEE.
To run option #2 above, specify EOM_NONIT=STAR together with CALC=CCSD,EXCITE=EOMEE.
To run option #3 above, specify EOM_NONIT=STAR, together with CALC=CCSD(T)(a),EXCITE=EOMEE.
Analytic gradients do not work for any of these methods, so any geometry optimizations and frequency calculations must be done with the corresponding energy-only options.
These methods are implemented only for closed-shell reference functions (REFERENCE=RHF).
The quality of results obtained with these methods increases monotonically as one goes from option #1 to option #3 above. The EOM-CCSD(T)(a) method is overwhelmingly the most accurate for excitation energies, but also the most expensive method since it calls for triple excitation corrections to be computed for both the reference state and the final state. Potential surfaces for EOM-CCSD* and EOM-CCSD(T)(a) are both better than those from EOM-CCSD or EOM-CCSD(T), although this is only an anecdotal observation that has not yet been published in the open literature.
EOM-CCSD(T)(a) must be run with CC_PROG=ECC.