33.2.10 Conical Intersection optimization (CONICAL)

To optimize a Conical Intersection (CI) between two electronic states having the same spin, three vectors must be evaluated at SA-CPMCSCF level:

1) Gradient Difference between the states involved (GD).

2) Non-Adiabatic Derivative Coupling (DC).

3) Gradient of the upper state (USG).

This can be done by adding three different CPMCSCF cards in the MULTI input

CPMCSCF, DGRAD, $S_{i}$, $S_j$, ACCU=1.0d-7, record=record1.file

CPMCSCF, NACM, $S_i$, $S_j$, ACCU=1.0d-7, record=record2.file

CPMCSCF, GRAD, $S_j$, ACCU=1.0d-7, record=record3.file

where $S_i$,$S_j$ are the electronic states in the usual format istate.istsym and record[n].file specifies the name and the file number where CPMCSCF infos should be stored. One must remember to:

i) specify always three different record.file in the CPMCSCF cards.

ii) evaluate the CPMCSCF for USG always as last.

iii) skip the DC evaluation if the CI involves states with different spin (eg

a Singlet/Triplet crossing) because the vector would be identically zero.

Three sets of FORCE cards (only two for Singlet/Triplet CI) follow the MULTI input. All the sets are like

FORCE

SAMC,record[n].file

CONICAL,record4.file[,NODC]

where record.file is one of the records containing CPMCSCF infos (the order here is not important) and record4.file points to a free record used for internal storage by the CONICAL code.

record4.file must be the same in all the CONICAL cards.

WARNING: The present implementation works properly only if file=1 in the CONICAL cards.

The optional keyword NODC is used in case of S/T crossing when DC is not evaluated and therefore not used.

The actual optimization is performed by repeatedly calling OPT inside a DO-LOOP cycle until the variable OPTCONV is below some predefined threshold. The example below is a job which optimizes the CI S0/S1 in $LiH_2$ .

Input: lih2_conical.com

Output: lih2_conical.out