Core–valence-separated coupled-cluster-singles-and-doubles complex-polarization-propagator approach to X-ray spectroscopies

Rasmus Faber* and Sonia Coriani*
DTU Chemistry - Department of Chemistry, Technical University of Denmark, Kemitorvet Building 207, DK-2800 Kongens Lyngby, Denmark. E-mail: rfaber@kemi.dtu.dk; soco@kemi.dtu.dk

Received 30th June 2019 , Accepted 8th August 2019

First published on 8th August 2019


The iterative subspace algorithm to solve the complex linear response equation of damped coupled cluster response theory presented, up to CCSD level, by Kauczor et al., J. Chem. Phys., 2013, 139, 211102, and recently extended to the solution of the complex left response multipliers by Faber and Coriani, J. Chem. Theory Comput., 2019, 15, 520, has been modified to include a core–valence separation projection step in the iterative procedure. This allows one to overcome serious convergence issues that specifically manifest themselves at the CCSD level when addressing core-related spectroscopic effects using large basis sets. The spectra, obtained adopting the new scheme for X-ray absorption and circular dichroism, as well as resonant inelastic-X-ray scattering, are presented and discussed. Core–valence separated results for non-resonant X-ray emission are also reported.


1 Introduction

The complex-polarization-propagator (CPP) approach,1–10 also known as damped response theory,11 is a valuable method to tackle the computational simulation of spectroscopic effects in cases where a large density of states and/or additional resonant conditions are involved, and traditional stick-spectra-based approaches may therefore fail.

The method relies on the ability to solve linear response equations for a complex, or damped, frequency,10 and it has been successfully implemented at various levels of electronic structure theory, including Hartree–Fock and time-dependent density functional theory,1–3,12 as well as multiconfigurational self-consistent field,1,2 algebraic diagrammatic construction (ADC)13 and coupled-cluster (CC) theories.7–9,14 Extensions to solvated environments15,16 and in the relativistic domain17 have also been presented. Applications encompass the calculation of linear absorption spectra in different frequency regions, including X-ray absorption spectra,4,5,8,14,17 electronic circular dichroism spectra,18 magnetic-field and nuclear-spin induced circular dichroism,19–21 magneto-chiral dichroism and birefringence dispersion,22 two-photon absorption in both UV-vis and X-ray regimes,6,23 Cauchy coefficients at imaginary frequency,7,13,24 and, more recently, resonant inelastic X-ray scattering.9,25

At the CC level, two different algorithms have been proposed to obtain complex response functions. In the first one,7,8 a diagonal basis of eigenvectors was generated by diagonalization of an approximate Lanczos-based tridiagonal representation of the CC Jacobian, and used to construct the imaginary or real components of the complex linear response function via a sum-over-state-like expression that includes the damping parameter γ. In the second one,9,14 the damped linear response equations yielding the complex (real and imaginary components thereof) amplitudes and multipliers were solved via a generalization of the reduced-space algorithm used in the conventional CC response.

The main drawbacks of the first approach are the need to pre-decide the size of the truncated Lanczos basis, which prevents the a priori control of the convergence thresholds, and the need to store a large number of Lanczos vectors on file, resulting in disk and I/O issues for larger systems. Moreover, if the target excited states lie in the X-ray region, large Lanczos chain lengths are required to obtain converged X-ray energies, unless specific core–valence separated (CVS) techniques are adopted.26–28

In the CPP-CC method of ref. 14, on the other hand, the damped response solver manifests severe convergence issues at the CC singles and doubles (CCSD) level in the high-energy frequency region and in particular when larger basis sets are used.

These issues can be rationalized as originating from the high density of doubly excited/ionized valence states that form the continuum in which the X-ray absorption bands are embedded. Thus, even though the essence of the CPP method is to introduce a damping parameter γ in the (linear) response functions to specifically remove the singularities when the external frequency approaches a resonant value, in the CCSD case, this damping is not sufficient to guarantee convergence as the basis set increases, due to the enormous number of closely lying double excited/ionized states that become accessible when a large basis set is used. Too many potential resonances can occur for just one parameter to take care of, and the less and less diagonal the CC Jacobian becomes, which compromises the efficiency of the diagonal preconditioner that is typically used to accelerate convergence. As further proof, in the case of the CC singles and approximate doubles method, CC2, where the double-double block of the Jacobian exactly corresponds to the orbital energy differences matrix, the damped response equations do converge.

To overcome these problems, we here propose to apply a core–valence separation projector27 during the solution of the CCSD complex response equations, to remove the continuum of valence ionized states. The performance of the resulting CVS-CPP algorithm is illustrated by calculations, within the CCSD linear response (LR-CCSD) framework, of near-edge-absorption fine structure (NEXAFS) and X-ray circular dichroism (XCD) and, within CCSD quadratic response (QR-CCSD), of resonant inelastic X-ray scattering (RIXS). CVS-CCSD results for non-resonant X-ray emission (XES), computed, according to our previously proposed recipe,9 as transitions between valence- and core-ionized states, are also reported for completeness. Equation-of-motion (CVS-CPP-EOM-CCSD) variants of the same properties could also be derived by modifications to the property Jacobian matrix and final property expressions.9,29,30 An alternative derivation of RIXS and XES exploiting a conceptually analogous CVS-DIIS damped solver within the frozen-core core–valence separated fc-CVS-EOM-CCSD framework31 is presented in ref. 32.

2 Theory

2.1 NEXAFS, XCD and RIXS within the CPP-(CVS)-CC formalism

NEXAFS cross sections σ(ω) can be computed from the imaginary electric dipole-electric dipole linear response function1,4,5,8
 
σXAS(ω) ∝ − ωℑ〈〈μα;μα〉〉γω (1)
where μα is the electric dipole moment, and the incident frequency is in the region of absorption of the relevant X-ray edge. XCD cross sections, on the other hand, are obtained from the real part of the electric dipole and magnetic dipole linear response function1,18,22
 
σXCD(ω) ∝ ωℜ〈〈μα;mα〉〉γω (2)
(since the magnetic dipole operator m is an imaginary operator). The RIXS cross section is obtained from the transition strengths σ0f averaged over all molecular orientations and over the polarization of the emitted radiation. The latter depends on the angle θ between the polarization vector of the incident photon, and the propagation vector of the scattered photon9,25
 
image file: c9cp03696b-t1.tif(3)
where [scr F, script letter F]0fXY(ω) and [scr F, script letter F]f0XY(ω) are the left and right Kramers–Heisenberg–Dirac (KHD) amplitudes,33,34 respectively. Sum-over-states expressions for the amplitudes assuming the same inverse lifetime for all states can be found in ref. 9 and 25.

In the CC damped linear response, the complex polarizability is computed according to:

 
image file: c9cp03696b-t2.tif(4)
The solution of the response equations yielding the amplitudes tY(ω + iγ) is discussed later in this section.

For the NEXAFS cross section, the imaginary part of the complex dipole–dipole polarizability in eqn (4) is needed, which is obtained according to

 
image file: c9cp03696b-t3.tif(5)
where we have explicitly split the complex response amplitudes into real and imaginary parts
 
tX(ω + iγ) = tX(ω + iγ) + itX(ω + iγ), (6)
and made use of the relation
 
tX(−ω + iγ) = tX(−ω − iγ), (7)
which is valid for all real operators (components) X. For an imaginary operator (component) χ, like the magnetic moment m, the latter relation reads
 
tχ(−ω + iγ) = tχ(−ω − iγ). (8)

An imaginary operator is needed to compute the XCD cross section in eqn (2), according to

 
image file: c9cp03696b-t4.tif(9)

In the RIXS amplitudes, the operators X and Y are always real, and both real and imaginary components are needed. The general complex expressions of the KHD amplitudes within damped CC response theory (as well as EOM-CCSD) were presented in ref. 9, and can be split into real and imaginary parts as given below for CC[scr F, script letter F]f0XY(ω):

 
image file: c9cp03696b-t5.tif(10)
A similar expression can be derived for CC[scr F, script letter F]0fXY(ω).

While referring to, e.g., ref. 9 and 35, for a definition of the remaining CC building blocks in the expressions above, we return to the solution of the complex response equations needed to obtain the real and imaginary components of the complex amplitudes tX(ω + iγ) and multipliers [t with combining macron]X(ω − iγ):

 
(A − (ω + iγ)I)tX(ω + iγ) = −ξX (11)
 
[t with combining macron]X(ω′ − iγ)(A + (ω′ − iγ)I) = −ηXFtX(ω′ − iγ) (12)

They can be recast in (pseudo-symmetric) matrix form as, e.g.,

 
image file: c9cp03696b-t6.tif(13)
and solved via an iterative subspace algorithm as discussed in ref. 9 and 14.

An important step to ensure convergence in the iterative algorithm is the generation of new trial vectors as

 
image file: c9cp03696b-t7.tif(14)
where image file: c9cp03696b-t8.tif is a general new trial vector, image file: c9cp03696b-t9.tif is a general residual in iteration n + 1 and [scr P, script letter P] is a preconditioner. In the iterative algorithm of ref. 9 and 14, eqn (14) is implemented as
 
image file: c9cp03696b-t10.tif(15)
where A0 ia a diagonal approximation of the full Jacobian A. A typical choice as A0 is to use the energy differences between virtual and occupied orbitals Δεμ, where μ refers to a given excitation level. In the CCSD case, (εaεi) is used for the corresponding singles, and (εa + εbεiεj) is used for the doubles.

Despite the preconditioner, and as anticipated in the introduction, the iterative subspace algorithm of ref. 9 and 14 does not converge at the CCSD level (in larger basis sets) when solving for the complex response amplitudes at positive values of ω falling in the X-region. The same occurs when solving for the response multipliers at negative frequencies in the X-region.

While a finite value of γ ensures that eqn (13) will not be exactly singular for any value of ω, the solution of this equation becomes nonetheless unfeasible if many excitation energies are very close to ω. The X-ray region is beyond the (valence)ionization limit of common molecules and there is therefore a continuum of ionized states near any given core-excitation energy. In calculations using finite basis sets, a discrete set of ionization energies is found and, for methods that only include single excitations, this set might be sparse enough not to prevent convergence of the damped response equations in the X-ray region. This explains why methods like CC2 and ADC(2) in general do not manifest (according to our experience) converge problem. For models, such as CCSD, that explicitly include double excitations, however, the set of ionized states is so dense that the normal damped response equations will not converge except for the smallest molecules and basis sets.

The problem of separating bound states from the continuum of ionized states can be solved using the CVS approximation,26 which has recently been introduced in the context of CC theory to describe core-excited and core-ionized states.27,31 In the present work, we propose to use a CVS projector in eqn (13), whenever the frequency ω is positive and falls in the X-ray region.

This projector allows only the part of the CC amplitudes in eqn (11) that involve core electrons to respond to the external electric field, effectively removing the valence ionization contribution. Similarly, a CVS projector is applied in the multiplier equation, eqn (12), for negative ω′. In the opposite case, negative ω in eqn (11) and positive ω′ in eqn (12), the equations are strongly diagonally dominant for absolute values of ω in the X-ray region and can easily be solved without applying projectors of any kind.

Comparing to the formulation of CVS-CCLR of Coriani and Koch,27 projecting out the valence excitation space from the right and left response equations only for the above-mentioned frequency values corresponds to projecting out exclusively from the eigenvalue equations. However, in ref. 27, it was suggested that one could apply CVS also to the left multiplier [M with combining macron]f equations, even though they are convergent. To obtain a damped-response CVS scheme equivalent to a CVS scheme where the [M with combining macron]f equations are also projected out, we would need to apply the CVS projector to all equations where the magnitude of the frequency is in the X-ray region, no matter its sign. Even though the effect of projecting out from the [M with combining macron]f vectors is negligible, our recommendation is however not to project them.

3 Results and discussion

3.1 Computational details

For the illustrative results on water, the same geometry and basis set we adopted previously9 were used. The geometry of acetone was optimized at the CCSD(T) level and the aug-cc-pVTZ basis set using CFOUR.36 The structure of methyloxirane was optimized at the MP2/aug-cc-pCVTZ level also using CFOUR. In determining the valence excited states for the RIXS calculations, the core orbitals have been kept frozen in the excited state calculation. The ground state parameters were optimized in full space. The same procedure was adopted for the valence ionized states required to obtain the XES strengths.

3.2 NEXAFS and XCD

As an illustration of cases that we could otherwise not tackle, due to convergence issues, with the full-space CPP solver in a basis set as large as 6-311++G**, we show, in Fig. 1 and 2, the CPP-CVS-XAS spectra of S-methyloxirane and acetone at the carbon K-edge, respectively, together with the XAS sticks obtained from a conventional linear response calculation using the CVS Davidson algorithm of ref. 27. The XAS spectrum of methyloxirane has been measured by Piancastelli et al.37 at both C and O K-edges, and by Turchini et al.38 at the C K-edge. The XAS spectra of acetone have been measured by Prince et al.39
image file: c9cp03696b-f1.tif
Fig. 1 CVS-CPP-CCSD results for the XAS spectrum of S-methyloxirane at the carbon K-edge in the 6-311++G** basis set, with underlying oscillator strengths obtained from standard CVS-CCSD calculations. The dashed red curve is the CVS-CPP result, whereas the black curve was obtained by applying a Lorentzian broadening to the CVS-CCSD sticks.

image file: c9cp03696b-f2.tif
Fig. 2 CVS-CPP-CCSD results for the XAS spectrum of acetone at the carbon K-edge in the 6-311++G** basis set, with underlying oscillator strengths obtained from standard CVS-CCSD calculations. The dashed red curve is the CVS-CPP result, whereas the black curve was obtained by applying a Lorentzian broadening to the CVS-CCSD sticks.

The XCD of methyloxirane has been experimentally measured by Turchini et al.38 at the carbon K-edge. The study of Turchini et al. is the first (and only) CD measurement on a randomly oriented system and it was performed in the vapour phase. Computational investigations have been presented at the random phase approximation level by Alagna et al.,40 and using STEX by Carravetta et al.41

In Fig. 3, we compare the XCD spectrum of methyloxirane calculated using CVS-CPP-CCSD with the one obtained from a CVS-CCSD-LR calculation. The CPP-CVS-CCSD calculations were performed on a grid with a spacing of 5 × 10−4 Hartree between points and with a broadening factor of 1000 cm−1. Our calculated spectra present bands and sticks somewhat consistent with the computed spectral sticks of ref. 41, whereas the agreement with the experimental spectrum is far from satisfactory. Further investigations are required to firmly assess the origin of the observed discrepancies.


image file: c9cp03696b-f3.tif
Fig. 3 CVS-CPP-CCSD results for the XCD spectrum of S-methyloxirane in the 6-311++G** basis set, with underlying rotatory strengths obtained from standard CVS-CCSD calculations. The dashed red curve is the CVS-CPP result whereas the black curve was obtained by applying a Lorentzian broadening to the CVS-CCSD sticks.

3.3 RIXS and XES

Turning our attention to RIXS and non-resonant XES, we report in Fig. 4 a comparison of RIXS and XES spectral slices obtained with the here-proposed CVS-CPP approach, and with the full-space approach of ref. 9 for water. As observed, the CVS projection introduces only a very modest shift of the peaks on the energy axis. The differences in intensities are also quite modest and the assignment of the origin of the spectral bands remains as previously done.9 XES and RIXS spectra for water generated using the CVS-CPP solver and 40 valence excited states are shown in Fig. 5. The third relatively intense band that is also experimentally observed emerges now at the resonant frequencies of the 1s → 4a1 and 1s → 2b2 core excitations. Even 40 valence excited states are, on the other hand, not sufficient to yield it when a resonant pump frequency close to the third XAS band is used. The third band is reproduced in the XES spectrum.
image file: c9cp03696b-f4.tif
Fig. 4 Full-space CPP-CCSD (black) versus CVS-CPP-CCSD (red) results for the RIXS (at the indicated incident frequency) and XES spectra of water in the 6-311++G**+Ryd basis set.

image file: c9cp03696b-f5.tif
Fig. 5 CVS-CPP-CCSD results for the RIXS (at the indicated incident frequency) and XES spectra of water in the 6-311++G**+Ryd basis set including the third peak, generated using 40 valence excited states.

As another example of the applicability of the CVS-CPP approach to RIXS, we considered the case of acetone at the oxygen K-edge. Despite the still moderate size of the molecule, its RIXS spectra could not be computed using the full-space CPP-CCSD approach, whereas they are easily obtained with the CVS-CPP solver. Table 1 collects the excitation energies and strengths of the valence excited states that have been considered in the spectral simulation. The resonant pump frequency was chosen at the value of the first core excitation at the oxygen K-edge, see also Table 2. The computed RIXS and XES spectral slices are shown in Fig. 6. Experimental resonant (RIXS) and non-resonant (XES) spectral slices at the oxygen K-edge in solution are available from ref. 42, see also ref. 43. The re-digitized experimental curves have been overlapped with the computed spectra without any shift on the energy axis. Despite the misalignment, the main spectral features at the lower core excitation in the resonant RIXS spectrum, as well as the XES spectrum, are reproduced. As can be appreciated from Fig. 6, the overall shift of the XES spectrum is slightly larger than for the RIXS one, possibly an indication that CCSD describes relaxation effects on XES (or, at least, on XES as computed here9 at the CCSD level) less accurately than those on RIXS.

Table 1 The valence excitation energy (ωn), and transition (Sn0) and oscillator (fn0) strengths of acetone as calculated at the CCSD/6-311++G** level
Symmetry ωn (a.u.) ωn (eV) Sn0 (a.u.) fn0
A2 0.166849 4.5402 0.0000000 0.00000
B2 0.238143 6.4802 0.1991480 0.03162
A1 0.278881 7.5887 0.0000052 0.00000
A2 0.279308 7.6003 0.0000000 0.00000
B2 0.286427 7.7941 0.0479960 0.00916
B2 0.300078 8.1655 0.1921846 0.03845
A1 0.309681 8.4269 0.3608999 0.07451
B1 0.321666 8.7530 0.0891817 0.01912
B2 0.325418 8.8551 0.0069247 0.00150
A2 0.329077 8.9546 0.0000000 0.00000
B1 0.345325 9.3968 0.0219041 0.00504
B1 0.346340 9.4244 0.0315136 0.00728
A1 0.346448 9.4273 1.3455875 0.31078
B2 0.346749 9.4355 0.0759524 0.01756
A2 0.354554 9.6479 0.0000000 0.00000
B1 0.364546 9.9198 0.0042734 0.00104
A1 0.365335 9.9413 0.0096702 0.00235
B2 0.375621 10.221 0.1865032 0.04670
B2 0.378524 10.300 0.1435405 0.03622
A2 0.381769 10.388 0.0000000 0.00000
A1 0.387199 10.536 0.0279431 0.00721
A2 0.388462 10.571 0.0000000 0.00000
A1 0.389988 10.612 0.1629636 0.04237
B1 0.397173 10.808 0.2212859 0.05859
A1 0.398122 10.833 0.4069430 0.10801
A2 0.404636 11.012 0.0000000 0.00000
B2 0.405736 11.041 0.3600519 0.09739
B1 0.408028 11.103 0.0198221 0.00539
A1 0.412197 11.216 0.1073709 0.02950
B2 0.415604 11.309 0.2804371 0.07770


Table 2 Oxygen K-edge excitation energy (ωn), and transition (Sn0) and oscillator (fn0) strengths of acetone as calculated at the CVS-CCSD/6-311++G** level
ωn (a.u.) ωn (eV) Sn0 (a.u.) fn0
19.5892 533.05 0.0033221 0.04338
19.7441 537.26 0.0000000 0.00000
19.7855 538.39 0.0001939 0.00256
19.7880 538.46 0.0000809 0.00107
19.7910 538.54 0.0001268 0.00167
19.8098 539.05 0.0000418 0.00055



image file: c9cp03696b-f6.tif
Fig. 6 CVS-CPP-CCSD results for the RIXS (bottom panel) and XES (upper panel, label “non-resonant”) spectra at the oxygen K-edge of acetone in the 6-311++G** basis set. The experimental spectral slices in solution from ref. 42 are also shown. Note that the computed spectra have not been shifted to align to the experimental ones.

4 Conclusions

We have presented a core–valence separated strategy to solve the complex response equations of coupled cluster damped response theory at the singles and doubles level. Introducing the CVS at specific values of the external frequency allows one to overcome convergence problems that appear in the CCSD case due to the continuum of valence ionized states. The modified solver has been used to compute XAS, XCD and RIXS spectra of selected molecules, showing moderate shifts of the peak positions and intensities compared to the cases where full-space calculations could be performed. Moderate differences are also observed between CVS-XES and full-space XES results.

Conflicts of interest

There are no conflicts of interest to declare.

Acknowledgements

We thank A. I. Krylov for useful discussion. We acknowledge support from DTU Chemistry and from the Independent Research Fund Denmark – DFF-Forskningsprojekt2 grant no. 7014-00258B.

References

  1. P. Norman, D. M. Bishop, H. J. A. Jensen and J. Oddershede, J. Chem. Phys., 2001, 115, 10323–10334 CrossRef CAS.
  2. P. Norman, D. Bishop, H. J. A. Jensen and J. Oddershede, J. Chem. Phys., 2005, 123, 194103 CrossRef PubMed.
  3. L. Jensen, J. Autschbach and G. C. Schatz, J. Chem. Phys., 2005, 122, 224115 CrossRef CAS PubMed.
  4. U. Ekström, P. Norman, V. Carravetta and H. Ågren, Phys. Rev. Lett., 2006, 97, 143001 CrossRef PubMed.
  5. U. Ekström and P. Norman, Phys. Rev. A: At., Mol., Opt. Phys., 2006, 74, 042722 CrossRef.
  6. T. Fahleson, H. Ågren and P. Norman, J. Phys. Chem. Lett., 2016, 7, 1991–1995 CrossRef CAS PubMed.
  7. S. Coriani, T. Fransson, O. Christiansen and P. Norman, J. Chem. Theory Comput., 2012, 8, 1616–1628 CrossRef CAS PubMed.
  8. S. Coriani, O. Christiansen, T. Fransson and P. Norman, Phys. Rev. A: At., Mol., Opt. Phys., 2012, 85, 022507 Search PubMed.
  9. R. Faber and S. Coriani, J. Chem. Theory Comput., 2019, 15, 520–528 CAS.
  10. P. Norman, Phys. Chem. Chem. Phys., 2011, 13, 20519–20535 CAS.
  11. K. Kristensen, J. Kauczor, T. Kjaergaard and P. Jørgensen, J. Chem. Phys., 2009, 131, 044112 CrossRef PubMed.
  12. J. Kauczor and P. Norman, J. Chem. Theory Comput., 2014, 10, 2449–2455 CrossRef CAS PubMed.
  13. T. Fransson, D. R. Rehn, A. Dreuw and P. Norman, J. Chem. Phys., 2017, 146, 094301 Search PubMed.
  14. J. Kauczor, P. Norman, O. Christiansen and S. Coriani, J. Chem. Phys., 2013, 139, 211102 Search PubMed.
  15. P. Reinholdt, M. S. Nørby and J. Kongsted, J. Chem. Theory Comput., 2018, 14, 6391–6404 CAS.
  16. M. Nørby, S. Coriani and J. Kongsted, Theor. Chem. Acc., 2018, 137, 49 Search PubMed.
  17. T. Fransson, D. Burdakova and P. Norman, Phys. Chem. Chem. Phys., 2016, 18, 13591–13603 RSC.
  18. A. Jiemchooroj and P. Norman, J. Chem. Phys., 2007, 126, 134102 CrossRef PubMed.
  19. H. Solheim, K. Ruud, S. Coriani and P. Norman, J. Chem. Phys., 2008, 128, 094103 CrossRef PubMed.
  20. T. Fahleson, J. Kauczor, P. Norman, F. Santoro, R. Improta and S. Coriani, J. Phys. Chem. A, 2015, 119, 5476–5489 CrossRef CAS PubMed.
  21. J. Vaara, A. Rizzo, J. Kauczor, P. Norman and S. Coriani, J. Chem. Phys., 2014, 140, 134103 CrossRef PubMed.
  22. J. Cukras, J. Kauczor, P. Norman, A. Rizzo, G. L. J. A. Rikken and S. Coriani, Phys. Chem. Chem. Phys., 2016, 18, 13267–13279 RSC.
  23. K. Kristensen, J. Kauczor, A. J. Thorvaldsen, P. Jørgensen, T. Kjærgaard and A. Rizzo, J. Chem. Phys., 2011, 134, 214104 CrossRef PubMed.
  24. A. Jiemchooroj, B. E. Sernelius and P. Norman, Phys. Rev. A: At., Mol., Opt. Phys., 2004, 69, 044701 CrossRef.
  25. D. R. Rehn, A. Dreuw and P. Norman, J. Chem. Theory Comput., 2017, 13, 5552–5559 CrossRef CAS PubMed.
  26. L. S. Cederbaum, W. Domcke and J. Schirmer, Phys. Rev. A: At., Mol., Opt. Phys., 1980, 22, 206–222 CrossRef CAS.
  27. S. Coriani and H. Koch, J. Chem. Phys., 2015, 143, 181103 CrossRef PubMed.
  28. B. N. C. Tenorio, T. Moitra, M. A. C. Nascimento, A. B. Rocha and S. Coriani, J. Chem. Phys., 2019, 150, 224104 CrossRef PubMed.
  29. F. Pawłowski, J. Olsen and P. Jørgensen, J. Chem. Phys., 2015, 142, 114109 CrossRef PubMed.
  30. S. Coriani, F. Pawłowski, J. Olsen and P. Jørgensen, J. Chem. Phys., 2016, 144, 024102 CrossRef PubMed.
  31. M. L. Vidal, X. Feng, E. Epifanovsky, A. I. Krylov and S. Coriani, J. Chem. Theory Comput., 2019, 15, 3117–3133 CrossRef CAS PubMed.
  32. K. Nanda, M. L. Vidal, R. Faber, S. Coriani and A. I. Krylov, Phys. Chem. Chem. Phys., 2019 DOI:10.26434/chemrxiv.9598715.v1.
  33. F. Gel'mukhanov and H. Ågren, Phys. Rep., 1999, 312, 87–330 Search PubMed.
  34. D. A. Long, The Raman Effect: A Unified Treatment of the Theory of Raman Scattering by Molecules, John Wiley & Sons Ltd., 2002 Search PubMed.
  35. O. Christiansen, P. Jørgensen and C. Hättig, Int. J. Quantum Chem., 1998, 68, 1–52 CrossRef CAS.
  36. J. F. Stanton, J. Gauss, L. Cheng, M. E. Harding, D. A. Matthews and P. G. Szalay, CFOUR, Coupled-Cluster techniques for Computational Chemistry, a quantum-chemical program package, With contributions from A. A. Auer, R. J. Bartlett, U. Benedikt, C. Berger, D. E. Bernholdt, Y. J. Bomble, O. Christiansen, F. Engel, R. Faber, M. Heckert, O. Heun, M. Hilgenberg, C. Huber, T.-C. Jagau, D. Jonsson, J. Jusélius, T. Kirsch, K. Klein, W. J. Lauderdale, F. Lipparini, T. Metzroth, L. A. Mück, D. P. O'Neill, D. R. Price, E. Prochnow, C. Puzzarini, K. Ruud, F. Schiffmann, W. Schwalbach, C. Simmons, S. Stopkowicz, A. Tajti, J. Vázquez, F. Wang, J. D. Watts and the integral packages MOLECULE (J. Almlöf and P.R. Taylor), PROPS (P.R. Taylor), ABACUS (T. Helgaker, H. J. Aa. Jensen, P. Jørgensen, and J. Olsen), and ECP routines by A. V. Mitin and C. van Wüllen, For the current version, see http://www.cfour.de.
  37. M. Piancastelli, T. Lischke, G. Prümper, X. Liu, H. Fukuzawa, M. Hoshino, T. Tanaka, H. Tanaka, J. Harries, Y. Tamenori, Z. Bao, O. Travnikova, D. Céolin and K. Ueda, J. Electron Spectrosc. Relat. Phenom., 2007, 156–158, 259–264 CAS.
  38. S. Turchini, N. Zema, S. Zennaro, L. Alagna, B. Stewart, R. D. Peacock and T. Prosperi, J. Am. Chem. Soc., 2004, 126, 4532–4533 CrossRef CAS PubMed.
  39. K. C. Prince, R. Richter, M. de Simone, M. Alagia and M. Coreno, J. Phys. Chem. A, 2003, 107, 1955–1963 CrossRef CAS.
  40. L. Alagna, S. D. Fonzo, T. Prosperi, S. Turchini, P. Lazzeretti, M. Malagoli, R. Zanasi, C. Natoli and P. Stephens, Chem. Phys. Lett., 1994, 223, 402–410 CrossRef.
  41. V. Carravetta, O. Plachkevytch, O. Vahtras and H. Ågren, Chem. Phys. Lett., 1997, 275, 70–78 CAS.
  42. K. M. Lange and E. F. Aziz, Chem. Soc. Rev., 2013, 42, 6840–6859 RSC.
  43. K. Atak, N. Engel, K. M. Lange, R. Golnak, M. Gotz, M. Soldatov, J.-E. Rubensson, N. Kosugi and E. F. Aziz, ChemPhysChem, 2012, 13, 3106–3111 CrossRef CAS PubMed.

This journal is © the Owner Societies 2019