34.2 Options for EOM

see also section 25.1.

In the case of LT-DF-LCC2/ADC(2) multistate calculations are possible, and it is recommended to calculate more states as needed.

The parameters on the EOM card:

EOM,-n.1, key1=value1, key2=value2,...

where n.1 is the last state of interest, e.g., with EOM,-5.1 the four lowest excited states will be calculated. The following keywords key are possible:

If set to 1, singlet states will be calculated (default).
If set to 1, triplet states will be calculated (not implemented for ADC(2)).
Record for converged CIS eigenvectors (default 6100.2).
Record for save of restart information.
Record for restart information for triplet states (if calculated together with singlet states).
Record for restart of previous calculation.
Record for restart of previous calculation (if calculated together with singlet states).
In the first nst iterations in the Davidson diagonalisation the excited state domains are determined for each basis vector in the Davidson subspace ("search for states") (default 7).
threshold for the Davidson procedure. If smaller than zero, the Davidson procedure is skipped and DIIS is started directly instead (possible only for restart, SAVE and START have to be identical).
threshold for DIIS.
if $=1$, do ADC(2) calculation instead of CC2.
if $=1$, use LT-DF-LMP2 for the ground state.

Default local approximations are defined according to procedure described in Ref. [3] (Laplace domains).

Rough criterion for specifying eom-domains from laplace-transformed integrals (default 0.8).
Criterion for specifying important orbitals from laplace-transformed integrals (default 0.999).
Exact criterion for specifying eom-domains from laplace-transformed integrals (default 0.98).
Check for all orbital domains (complete sum over all orbitals) (default 0.95).

To switch to local aproximations calculated according to Ref. [9] (Boughton-Pulay procedure for excited states), set INTFRAC to zero.

Occupied orbital pair lists are calculated from important orbitals (Refs. [1,5]).

Distance criterion for excited state orbital pairs definition (default 5).
Different possibilities for excited state orbital pairs:
0: $[ij]\leq$ ewpair, $[im]\leq$ ewpair, $[mn]\leq$ ewpair;
1: $\forall [ij]$, $\forall [im]$, $[mn]\leq$ ewpair;
2: $\forall [ij]$, $[im]\leq$ ewpair, $[mn]\leq$ ewpair (default);
where $i$, $j$ are important orbitals and $m$, $n$ non-important. More detailed see Ref. [1].

One can try to improve the convergence of iterative procedures by changing following parameters

Type of preconditioner in LT-DF-LCC2/ADC(2) and DF-CIS:
0: canonical orbital energies;
3: solving linear equations including diagonal part of $(H-F)^{\rm CIS}$ with MINRES (default).
Scaling factor for diagonal part of $(H-F)^{\rm CIS}$ matrix in the case of linear equations preconditioner (default 0.7).

Properties are also activated on the EOM card:

States for which properties (dipole moments) should be calculated, e.g., PROPES=-3.1+5.1-8.1+15.1 $\Longrightarrow$ 2.1 3.1 5.1 6.1 7.1 8.1 15.1
States for which transition moments should be calculated, syntax like for properties. Is not implemented for ADC(2) and for triplet states.
Record, where densities calculated for states prop can be saved (for printing, see 34.4)
Note that ADC(2) transition moments require second-order ground-state singles and doubles amplitudes [see e.g. G. Wälz, D. Kats, D. Usvyat, T. Korona and M. Schütz, Phys. Rev. A 86, 052519 (2012); ADC(2) is equivalent to TD-UCC[2]-H]. This is done by carrying out a partial local MP4 calculation, with second-order doubles restricted to strong pairs. For ADC(2) excitation energies or first-order properties just the first-order LMP2 amplitudes are required. Strong, close, and weak pairs actually all enter the subsequent excited state calculations.

By default, orbital un-relaxed properties are computed. In order to calculate also orbital relaxed properties the option ORBREL on the LOCAL card should be set to 1, i.e.,

This also applies to the calculation of nuclear gradients. Options concerning nuclear gradients, which are also activated on the EOM card, are
GSONLY=0 or 1
1: only ground state gradient is calculated
Excited state for which the gradient is to be calculated, e.g., GRADSTATE=2.1; must be combined with GSONLY=0
0: DIIS after Davidson diagonalization is switched off for all states except gradstate
0: tries to follow a state in a geometry optimization after crossing on the basis of the overlap with eigenvector of previuous geometry; 1: calculate gradient of state number gradstate regardless of crossing
1: only total, rather than ground- and excited state Lagrange multipliers for orbital variations are computed. Hence, only one set of CPL and CPHF equations need to be solved rather than two. However, orbital relaxed properties are then not available.

molpro@molpro.net 2019-09-18