Excited states with equation-of-motion CCSD (EOM-CCSD)

Excitation energies for singlet states can be computed using equation-of-motion (EOM) approach. For the excitation energies the EOM-CCSD method gives the same results as linear response CCSD (LR-CCSD) theory. Accurate results can only be expected for excited states dominated by single excitations. The states to be computed are specified on an EOM input card, which is a subcommand of CCSD. The following input forms are possible

EOM, state1, state2, state3, …

Computes the given states. Each state is specified in the form number.sym, e.g., 5.3 means the fifth state in symmetry 3. Note that state 1.1 corresponds to the ground state CCSD wavefunction and is ignored if given.

EOM, $-n1.sym1$, $-n2,sym2$, …

computes the first $n1$ states in symmetry sym1, $n2$ in sym2 etc.

EOM, $n1.sym1$, $-n2,sym1$, …

computes states $n1$ through $n2$ in symmetry sym1.

The different forms can be combined, e.g.,

EOM, $-3.1$, $2.2$, $2.3$, $-5.3$

computes states 1-3 in symmetry 1, the second excited state in symmetry 2, and the second through fifth excited states in symmetry 3. Note that state 1.1 is the ground-state CCSD wavefunction.

By default, an error exit will result if the CCSD did not converge and a subsequent EOM calculation is attempted. The error exit can be avoided using the NOCHECK option on the CCSD command (see also CCSD(T)).

Normally, no further input is needed for the calculation of excitation energies.

EOM-CCSD amplitudes can be saved using SAVE=record.ifil. The vectors will be saved after every refreshing of the iteration space and at the end of the calculation. The calculation can be restarted from the saved vectors, if START=record.ifil is specified. The set of vectors to be computed can be different in old and restarted calculations. However, if both cards (SAVE and START) are specified and the records for saving and restarting are identical, the sets of vectors should be also identical, otherwise chaos. The identical SAVE and START records can be useful for potential energy surfaces calculations, see section PES for lowest excited states for hydrogen fluoride.

By default, only excitation energies are calculated, since the calculation of properties is about two times as expensive, as the calculation of energies only. The one-electron properties and transition moments (expectation type, as defined in: J.F. Stanton and R.J. Bartlett, J. Chem. Phys. 98, 7029 (1993)) can be calculated by adding TRANS=1 to EOM card. The CCSD ground state is treated as a special case. If RELAX option is specified on the EXPEC card, also the relaxed one-electron density matrix is calculated for the ground state. (Currently, the relaxed CCSD density matrix through the EOM program is available for all-electron calculations only.) By default, dipole moments are calculated. Other required properties can be specified using EXPEC card. Properties are saved in MOLPRO variables, e.g. the $x$-component of the dipole moment is saved in DMX, its pure electron part in DMXE, transition moment – in TRDMX (left and right transition moments are stored separately). If DENSAVE=record.ifil is specified, excited-state densities are saved to record.ifil, otherwise they are saved to the record given in DM card. If TRANS=2, transition density matrices from/to the ground state will be saved provided that DENSAVE=record.ifil or DM card are specified and nonzero. If TRANS=3, transition moments among excited states are also calculated, and finally if TRANS=4, all transition densities will be saved (note that the last option should be used with caution because the number of densities to be stored will quickly exceed the allowed maximum). For an example see section EOM-CCSD transition moments for hydrogen fluoride.

When properties are needed, the left EOM-CCSD wave functions are calculated first. It is possible to use them as starting guesses for the right EOM-CCSD wave functions. This option is controlled by STARTLE (default 0). If STARTLE=1, left vectors are just used as a start for right vectors; if STARTLE=2, starting vectors, obtained from the left vectors are additionally biorthogonalized to the left vectors; finally, if STARTLE=3, also the final right vectors are biorthogonalized to the left vectors. The last possibility is of particular importance for degenerate states.

It is possible to make the program to converge to a vector, which resembles a specified singles vector. This option is switched on by FOLLOW=$n$ card (usually $n$=2 should be set). FOLLOW card should be always accompanied with EXFILE=record.ifil card, where record.ifil contains singles vectors from a previous calculation, see section calculate an EOM-CCSD state most similar to a given CIS state.

Normally, no further input is needed. However, some defaults can be changed using the EOMPAR directive:

EOMPAR, key1=value1, key2=value2,…

where the following keywords key are possible:

  • MAXDAV=nv Maximum value of expansion vectors per state in Davidson procedure (default 20).
  • INISINGL=ns Number of singly excited configurations to be included in initial Hamiltonian (default 20; the configurations are ordered according to their energy). Sometimes INISINGL should be put to zero in order to catch states dominated by double excitations.
  • INIDOUBL=nd Number of doubly excited configurations to be included in initial Hamiltonian (default 10).
  • INIMAX=nmax Maximum number of excited configurations to be included in initial Hamiltonian. By default, $nmax=ns+nd$.
  • MAXITER=itmax Maximum number of iterations in EOM-CCSD (default 50).
  • MAXEXTRA=maxex Maximum number of extra configurations allowed to be included in initial Hamiltonian (default 0). In the case of near degeneracy it is better to include a few extra configurations to avoid a slow convergence.
  • EOMLOCAL=eoml If set to 0, non-local calculation (default). EOMLOCAL=1 switches on the local module (experimental!).
  • INIMAX=ini Number of CSFs included in initial Hamiltonian, used only if INISINGL and INIDOUBL are both zero.

All keywords can be abbreviated by at least four characters.

The following print options are mostly for testing purposes and for looking for the convergence problems.

EOMPRINT, key1=value1, key2=value2,…

where the following keywords key are possible:

  • DAVIDSON=ipr Information about Davidson procedure:

ipr=1 print results of each “small diagonalization”
ipr=2 also print warning information about complex eigenvalues
ipr=3 also print hamiltonian and overlap matrix in trial space.

  • DIAGONAL=ipr Information about configurations:

ipr=1 print the lowest approximate diagonal elements of the transformed hamiltonian
ipr=2 print orbital labels of important configurations
ipr=3 print all approximate diagonal elements
ipr=4 also print the long form of above.

  • PSPACE=ipr Print information about the initial approximate hamiltonian:

ipr=2 print the approximate hamiltonian used to find the first approximation.

  • HEFF=ipr Print information about effective Hamiltonian:

ipr=2 print columns of effective hamiltonian and overlap matrix in each iteration

  • RESIDUUM=ipr Print information about residual vectors:

ipr=-1 no print in iteration
ipr=0 print energy values + residuum norm (squared) for each iteration (default)
ipr=1 also print warning about complex eigenvalue, and a warning when no new vectors is added to the trial space due to the too small norm of the residuum vector.
ipr=2 also print how many vectors are left

  • LOCEOM=ipr ipr=1 prints overlaps of sample and tested vectors in each iteration, if FOLLOW card is present. Increasing ipr switches on more and more printing, mostly related to the local EOM-CCSD method.
  • POPUL=ipr if ipr=1, do a population analysis of the singles part of the rhs EOM-CCSD wave function. By default the Löwdin method is used. The Mulliken analysis can be forced by adding MULLPRINT=1 to EOM card. Note that a more correct (but more expensive) approach is to calculate and analyse the EOM-CCSD density matrix, see section options for EOM.
  • INTERMED=ipr Print intermediates dependent on ground state CCSD amplitudes:

ipr=0 no print (default)
ipr=1 print newly created intermediates
ipr=2 also print more intermediates-related information

This example shows how to calculate potential energy surfaces for several excited states using restart from a previous calculation.

examples/hf_eom_pes.inp
***, PES for several lowest states of hydrogen fluoride
basis=avdz                                       ! define basis set
geometry={h;f,h,r}                               ! z-matrix
r=0.8 Ang                                        ! start from this distance
do n=1,100                                       ! loop over distances
rr(n)=r                                          ! save distance for table
hf                                               ! do SCF calculation
ccsd                                             ! do CCSD calculation, try to restart
start,4000.2,save,4000.2                         ! and save final T amplitudes
eom,-2.1,-1.2,-1.4,start=6000.2,save=6000.2      ! do EOM-CCSD calculation, try to restart
                                                 ! and save final excited states' amplitudes



ebase(n)=energy(1)                               ! save ground state energy for this geometry
e2(n)=energy(2)-energy(1)                        ! save excitation energies for this geometry
e3(n)=energy(3)-energy(1)
e4(n)=energy(4)-energy(1)
r=r+0.01                                         ! increment distance
enddo                                            ! end of do loop
table,rr,ebase,e2,e3,e4                          ! make table with results
digits,2,8,5,5,5,5,5,5,5,5                       ! modify number of digits
head,R(Ang),EGRST,E_EXC(2.1),E_EXC(1.2),E_EXC(1.4)! modify headers of table
! title of table
title,EOM-CCSD excitation energies for hydrogen fluoride (in hartree), basis $basis
save,hf_eom_ccsd.tab                             ! save table in file

This calculation produces the following table:

 EOM-CCSD excitation energies for hydrogen fluoride (in hartree), basis AVDZ

   R(ANG)        EGRST     E_EXC(2.1)  E_EXC(1.2)  E_EXC(1.4)
     0.80   -100.23687380     0.56664     0.41204     0.56934
     0.81   -100.24094256     0.56543     0.40952     0.56812
     0.82   -100.24451598     0.56422     0.40695     0.56690

etc.

This example shows how to calculate and store CCSD and EOM-CCSD density matrices, calculate dipole and quadrupole moments (transition moments from the ground to excited states are calculated), and how to use the EOM-CCSD excited state density for Mulliken population analysis.

examples/hf_eom_prop.inp
***, Properties and transition moments for several lowest states of hydrogen fluoride
basis=avdz                                       ! define basis set
geometry={h;f,h,r}                               ! z-matrix
r=0.92 Ang                                       ! define distance

hf                                               ! do SCF calculation
{ccsd                                            ! do CCSD calculation
dm,5600.2                                        ! density matrices will be stored here
expec,response,qm                                ! require quadrupole moments
eom,-3.1,-2.2,-2.3,-2.4,trans=1}                 ! do EOM-CCSD calculation + properties

pop;density,5600.2,state=2.4                     ! make population analysis for state 2.4

This calculation produces the following table:

--------------------------------------------------------------------------------
                        Final Results for EOM-CCSD
                             (moments in a.u.)
--------------------------------------------------------------------------------

  State    Exc. Energy (eV)            X              Y              Z
   2.1         14.436

 Right transition moment           0.00000000     0.00000000     0.65349466
 Left  transition moment           0.00000000     0.00000000     0.68871635
 Dipole strength                   0.45007246
 Oscillator strength               0.15917669
 Dipole moment                     0.00000000     0.00000000     0.88758090

etc.

This example shows how to force the convergence of the EOM-CCSD program to a state, which resembles at most a given CIS state.

examples/hf_eom_conv.inp
***, EOM-CCSD, vector following procedure
basis=avdz                                     ! define basis set
geometry={h;f,h,r}                             ! z-matrix
r=0.92 Ang                                     ! define distance
hf;save,2100.2                                 ! do SCF calculation, save orbitals
cis,-4.4,exfile=6000.2                         ! do CIS calculation, save amplitudes
ccsd;save,4000.2                               ! do CCSD calculation, save amplitudes
eom,-4.4,checkovlp=1,exfile=6000.2             ! do EOM-CCSD calculation,
                                               ! check overlap of singles with CIS vectors
                                               ! stored in record given in exfile
eompar,inisingl=200,inidoubl=0                 ! for first approximation take 200 single CSF
                                               ! of approximate hamiltonian
ccsd;start,4000.2                              ! do CCSD calculation, try to restart
eom,2.4,follow=2,exfile=6000.2,checkovlp=1     ! do EOM-CCSD calculation for state closest
                                               ! to 2.4 CIS state, check overlap of singles
                                               ! with CIS vectors stored in exfile
eompar,inisingl=200,inidoubl=0
eomprint,loce=1                                ! print overlaps of sample and EOM vectors in
                                               ! each iteration

Excitation energies can also be calculated using the Configuration-Interaction Singles (CIS) method. By default, singlet excited states are calculated. Triplet excited states can be obtained by setting triplet=1 in EOM card. This method cannot be expected to give accurate results, but can be used for quite large molecules. The states to be computed are specified as in EOM. Setting trans=1 switches on the calculation of dipole transition moments (length gauge), while trans=2 allows to obtain additionally one-electron properties of excited states. By default, dipole moments are calculated. Other required properties can be specified using EXPEC card. trans=0 and 1 work in direct mode.

hf
cis,-3.1,1.2,trans=2

[cprop_ccsd] First-order and frequency-dependent second-order properties, derived from the expressions based on the expectation value of a one-electron operator, can be obtained with the CPROP directive for the closed-shell CCSD method. For first-order properties obtained from the energy-derivative approach see section expectation values. The expectation-value methods utilized in the program are described in the following papers:

$[1]$ B. Jeziorski and R. Moszynski, Int. J. Quantum Chem., 48, 161 (1993);
$[2]$ T. Korona and B. Jeziorski, J. Chem. Phys. 125, 184109 (2006);
$[3]$ R. Moszynski, P. S. Żuchowski and B. Jeziorski, Coll. Czech. Chem. Commun., 70, 1109 (2005);
$[4]$ T. Korona, M. Przybytek and B. Jeziorski, Mol. Phys. 104, 2303 (2006),
$[5]$ T. Korona, Theor. Chem. Acc., 129, 15 (2011).

Note that properties obtained from the expectation-value expression with the coupled cluster wave function are not equivalent to these derived from gradient or linear-response methods, although the results obtained with both methods are quite similar. In XCC the exponential ansatz for the wave function is utilized for the bra side, too.

For the first-order properties the one-electron operators should be specified in the EXPEC card, while for the second-order properties – in the POLARI card. A density can be saved by specifying the DM card.

For the first-order properties the option XDEN=1 should be usually given. Other options specify a type of the one-electron density, which can be either the density directly derived from the expectation-value expression, see Eq. (8) of Paper 2, or the modified formula, rigorously correct through the ${\cal O}(W^3)$ Møller-Plesset (MP) order, denoted as $\bar X(3)_{\rm resp}$ in Papers 1 and 2. In the first case the option PROP_ORDER=n can be used to specify the approximation level for single and double excitation parts of the so-called $S$ operator (see [2], Eq. (9)); $n=\pm 2,\pm 3,\pm 4$, where for a positive $n$: all approximations to $S$ up to $n$ are used, and for a negative $n$ only a density with $S$ obtained on the $|n|$ level will be calculated. Another option related to the $S$ operator is HIGHW=n, where $n=0,1$; if $n$=0, some parts of $S_1$ and $S_2$ operators of a high MP order are neglected. Below an example of a standard use of this method is given:

CPROP,XDEN=1,PROP_ORDER=-3,HIGHW=0

The combination above is also available by writing EXPEC,XCCSD after the CCSD card. A cheap method denoted as XCCSD(3), obtained by a simplification of the original XCCSD formula, is available by setting

CPROP,XDEN=-21

or by writing EXPEC,XCCSD(3) after the CCSD card.

In the second case the options X3RESP=1 and the CPHF,1 card (or alternatively the EXPEC card) should be specified,

CPROP,XDEN=1,X3RESP=1;CPHF,1

For the second-order properties always the following options should be given:

CPROP,PROPAGATOR=1,EOMPROP=1

The recommended CCSD(3) model from Paper 4 requires that additionally the PROP_ORDER=3 and HIGHW=0 options are specified. Frequencies for dynamic properties (in atomic units) should be given in variables OMEGA_RE (real parts) and OMEGA_IM (imaginary parts). If one of these arrays is not given, it is filled with zeros. Other options for the second-order properties involve

  • OMEGAG (default 0.3). There are two linear-equation solvers, OMEGAG is a minimum frequency, for which the second solver (working for large frequencies) is used.
  • DISPCOEF=n if $n>0$, calculate dispersion integrals for the van der Waals coefficients with operators given in the POLARI card, using $n$ as a number of frequencies for the numerical integration. In this case the frequency values given in OMEGA_RE and OMEGA_IM are ignored. If two molecules are calculated in the same script one after another, also the mixed dispersion integrals are calculated. The isotropic $C_6$ coefficient is stored in a variable DISPC6, the isotropic $C_9$ nonadditive coefficient – in a variable DISPC9. All necessary informations for the calculation of dispersion integrals are written to the ascii file name.dispinfo, where name is the name of the MOLPRO script.
  • THRPROPAG if given, use this threshold as a convergence criterion for the linear-equation solver for the first-order perturbed CCSD amplitudes.
  • STARTT1=n various start options for the iterative linear-equation solver for the first-order perturbed CCSD amplitudes, the most useful is $n=0$ (zero start) and $n=7$ (start from the negative of the r.h.s. vector rescaled by some energetic factors dependent on the diagonal of the Fock matrix and the specified frequency).