New Basis Sets for Lanthanide and Actinide Energy-consistent Small-core Pseudopotentials Xiaoyan Caoa b* (曹晓燕) a Institut für Theoretische Chemie, Universität zu Köln, D-50939,Germany b Biochemistry Department, Zhongshan University, Guangzhou, 510275, P.R. China * cao@thch.uni-bonn.de phone: (00)49-228-732919 fax: (00)49-228-739066 March 3, 2016 Abstract. The optimization of generalized contracted Gaussian (14s13p10d8f6g)/[6s6p5d4f3g] atomic natural orbital valence basis sets for relativistic energy-consistent small-core lanthanide and actinide pseudopotentials of the Stuttgart-Bonn variety is described. Corresponding Gaussian valence basis sets using a segmented contraction scheme (14s13p10d8f6g)/ [10s8p5d4f3g] for lanthanides and (14s13p10d8f6g)/ [10s9p5d4f3g] for actinides are also discussed. The basis sets have been proven to be stable and reliable by atomic and molecular calibration studies. Application to the electronic ground states of some selected homonuclear lanthanide dimers shows that the derived molecular constants are the best theoretical estimates available at present and have a satisfactory agreement with experimental data. Remaining discrepancies in the vibrational constants of La2, Ce2 and Pr2 are shown to be probably related to quite large Ar-matrix shifts. 1 Introduction Theoretical chemistry investigations on systems containing f elements are still a considerable challenge 1, 2, 3 . The presence of several open shells with different main quantum numbers poses difficulties to theoretical work, i.e. in neutral atoms and their ions often nf (n=4, 5 for lanthanides and actinides, respectively) as well as (n+1)d, (n+2)s and possibly (n+2)p orbitals may be partially occupied in the ground states or rather low-lying excited states. A 2S+1LJ term arising from a fm subshell may lead to a spin S as large as 7/2 and an angular momentum L as large as 12. Even more extreme values of S and L may result from the coupling of the fm subshell to other partially occupied shells of s, p, or d symmetry. In addition spin-orbit coupling usually splits each LS term into a large number of energetically adjacent electronic J states. From the experimental point of view the limited availability of most actinide elements as well as their radioactivity are additional obstacles. Therefore, up to now the experimental knowledge of the energy levels of free actinide atoms and ions is very limited and most of the higher ionization potentials are unknown for essentially all actinide elements4. For free lanthanide atoms and ions, the experimental knowledge is much wider, but still far from being complete 5. Despite the problems outlined above, first-principles methods, i.e., wavefunction-based ab initio schemes as well as density-functional approaches, are presently at the edge of successfully dealing with systems containing one or two f atoms. For rigorous wavefunction-based ab initio methods relativity has been taken into account at the all-electron Dirac-Coulomb-Breit level and electron correlation effects are included by means of coupled-cluster (CC) or configuration interaction (CI) methods. H- or even i-functions in the one-particle basis sets are usually necessary to yield accurate results 6 , 7 . Such highly-correlated state-of-the-art all-electron studies are presently only feasible for atoms by exploiting their spherical symmetry, and to our knowledge, the method is only applicable for some elements, i.e., only calculations for closed shell systems, one or two electrons outside a closed shell or one or two holes inside a closed shell are feasible. In order to be able to treat all f atoms and also to deal with molecular 2 systems, compromises have to be made with respect to the treatment of relativity and electron correlation. Among the most successful approaches in relativistic quantum chemistry is the ab initio pseudopotential (PP) method8, where the explicit quantum chemical treatment is restricted to the valence electrons and relativistic effects are only implicitly accounted for by a proper adjustment of free parameters in the valence model Hamiltonian. Several sets of such effective core potentials (ECPs), both of the energy-consistent9, 10, 11 and the shape-consistent PP variety 12, 13, 14, 15, 16, 17 , have been published for the f elements. Recently also model potentials (MPs) 18, 19 became available . About a decade ago the energy-consistent small-core PPs for all f elements except for lanthanum and lutetium have been published 11, 9, which proved to yield quite accurate results in numerous first-principles molecular applications. Corresponding optimized valence basis sets have not been derived initially. Later, preliminary (12s10p8d8f6g)/[5s5p4d3f3g] for lanthanide and (12s11p10d8f)/[8s7p6d4f] for actinide atomic natural orbital (ANO) Gaussian valence basis sets were generated and distributed through the MOLPRO PP and basis set library20, 21, 22, 23, 24 . These basis sets were not completely satisfactory in several respects, e.g., they were not flexible enough to describe accurately states with different f occupations since they have been optimized only for a single electronic configuration. In addition the variation of the exponents with the nuclear charge was partly irregular leading to difficulties in studies of trends along the series. Recently, we published the energy-optimized (14s13p10d8f6g)/[6s6p5d4f3g] ANO Gaussian valence basis sets for the whole lanthanide and actinide series 26 25, . Slightly larger basis sets (14s13p10d8f6g)/[10s8p5d4f3g] for lanthanides and (14s13p10d8f6g)/[10s9p5d4f3g] for actinides using a segmented contraction were also presented studies 27, 28 . This article summarizes the atomic and molecular calibration 25, 26, 27, 28, 29, 30, 31 and discusses the reliability and stability for the new basis sets. An application to the studies on the electronic structure of some selected lanthanide dimers (Ln2, Ln=La, Ce, Pr, Eu, Gd, Yb, Lu) reviewed. 3 32, 33 are also Methods The method of relativistic energy-consistent ab initio pseudopotentials (PPs) is described in detail elsewhere 10, 11 and will be outlined here only briefly. The valence-only model Hamiltonian for an atom or ion with n valence electrons is given as H 1 2 n i i n 1 ij ij r Va Vso Here i and j are electron indices. Va denotes a spin-orbit averaged relativistic PP in a semilocal form for a core with charge Q n Va i Q ri n A i lk exp(alk ri2)Pl l,k Pl is the projection operator onto the Hilbert subspace of angular momentum l. The spin-orbit term Vso may be written as Vso n 2l i l 0,k 2 Blk exp(blk ri 2)PllisiPl 1 The free parameters Alk, alk, Blk, and blk are adjusted to reproduce the valence total energies of a multitude of low-lying electronic states of the neutral atom and its ions34. Small-core PPs for lanthanides (actinides) have been used, i.e., the 1s-3d (1s-4d) shells were included in the PP core, while the 4f (5f) shell and all the others with main quantum number larger than 4 (5) were treated explicitly. The choice of a small core allows an accurate description of electronic states arising from all important electronic configurations, e.g., the frozen-core errors resulting mainly upon changes in the 4f (5f) occupation number are restricted to a tolerable minimum. The reference data used to determine Va have been taken from relativistic all-electron calculations using the so-called Wood-Boring (WB) scalar-relativistic Hartree-Fock (HF) approach. Spin-orbit operators Vso have been derived for the p, d, f symmetries by adjusting fine-structure splittings obtained with the valence model spin-orbit 4 Hamitonian to all-electron Dirac-Hartree-Fock (AE) (MCDHF) state-averaged data multi-configuration obtained with the Dirac-Coulomb-Hamiltonian for the one-valence electron ions35. The routine to generate the Gaussian (14s13p10d8f6g)/[6s6p5d4f3g] ANO valence basis sets using a generalized contraction scheme, as well as using a segmented contraction pattern (14s13p10d8f6g)/[10s8p5d4f3g] for lanthanides and (14s13p10d8f6g)/[10s9p5d4f3g] for actinides has been described in detail elsewhere25, 26 and will be summarized here only briefly. The primitive (14s11p8d8f) sets of exponents were energy-optimized in HF calculations for the energetically lowest LS states of the neutral atoms. Two diffuse d and p functions were then added to describe the (n+1)d and (n+2)p orbitals, yielding the final (14s13p10d8f) uncontracted basis sets. The contraction coefficients were obtained from atomic natural orbitals of state-averaged complete active space multi-configuration self-consistent field calculations (CASSCF) with subsequent 20 multi-reference configuration interaction calculations (MRCI) . The density matrices for both low-lying configurations fn+1s2 and fnd1s2 were generated and averaged, yielding the contraction coefficients describing these two states on equal footing. The generalized contraction scheme yields the (14s13p10d8f)/[6s6p5d4f] ANO valence basis sets. Sets of 6g functions with exponents identical to the 6 largest (and most important) f exponents were chosen. A generalized ANO contraction, as described above, was performed to obtain the final (14s13p10d8f6g)/[6s6p5d4f3g] Gaussian ANO valence basis sets. The segmented contraction pattern was chosen in such a way that without too much loss in accuracy relatively compact basis sets result, yielding slightly larger (14s13p10d8f6g)/[10s8p5d4f3g] lanthanide and (14s13p10d8f6g)/[10s9p5d4f3g] actinide basis sets. The segmented contraction coefficients were taken from the ANO contractions of the lowest energy pseudo-valence orbital in each symmetry in the generalized contracted basis sets. The described pseudopotentials and basis sets are available from Electronic Physics Auxiliary Publication Service (EPAPS) as well as by writing to the authors. The data will also be available in the upcoming revision of the MOLPRO 20 PP and basis set library 36 . Using both the generalized and segmented contraction for the valence basis sets and the small-core pseudopotential, calibration studies have been carried out for the first to fourth ionization potentials of all lanthanide and actinide elements 5 25, 26, 27, 28 . State-averaged complete active space self-consistent field (CASSCF) calculations with subsequent multi-reference averaged coupled-pair functional (ACPF)37 calculations were performed. The active space in the CASSCF always comprised all open-shell orbitals (4f, 5d, 6s for lanthanides and 5f, 6d, 7s, 7p for actinides) as well as 6s for lanthanides and 7s for actinides if doubly occupied. State averaging always included all components of the energetically lowest LS-term in order to avoid symmetry breaking. In the ACPF calculations excitations were allowed from all occupied valence orbitals. The MOLPRO20 program system was applied for all scalar-relativistic calculations. The large uncontracted valence basis sets have been applied to extrapolate to the basis set limit29, 31 . For the lanthanides (14s13p10d8f6g) basis sets were augmented by a (2s2p2d2f2g) diffuse set as well as (8h) and (8h8i) sets. The g, h and i exponents were chosen to be identical to the f exponents with the largest expansion coefficient in the 4f orbital. For actinides slightly smaller basis sets were applied, i.e., (14s13p10d8f6g) augmented by (6h) and (6h6i) sets, since the addition of diffuse functions was found to lead to negligible changes in the results for Ac. Again, the g, h and i exponents were chosen to be identical to the f exponents having the largest expansion coefficients in the 5f orbital. The dynamical correlation energies E(l) obtained from calculations with basis sets up to g (l=4), h(l=5) and i (l=6) functions were found to be almost linear in 1/l3, when l denotes the maximum angular quantum number in the basis set 25, 29. E(l) al3 E() The correlation coefficient was found to deviate by less than 10-4 from the ideal value of one for all cases. The extrapolations 1/l3 0 yield the basis set limit for the dynamical correlation energies E(). Spin-orbit corrections have been derived for lanthanides by a full CI within the space of fractionally occupied orbitals (COSCI, complete open shell CI) 31 25, 26, . In case of the actinides the radial shape of the two-component spinors deviates much more from the scalar-relativistic orbitals than for the lanthanides. Therefore a CASSCF within the space of fractionally occupied orbitals was performed, keeping the 5s, 5p, 5d, 6s, 6p shell frozen in their scalar-relativistic form. 6 State-averaging over all possible J-levels for a given nonrelativistic configuration was performed. The spin-orbit calculations have been carried out with a modified version of the GRASP relativistic atomic electronic structure code 35. The molecular calculations were carried out with the MOLPRO program package, using basis sets and methods as indicated in the text25, 26, 30 . The spectroscopic constants were derived by fitting a fifth-degree polynomial in the interatomic distance R times a factor 1/R for six points on the potential curve near the equilibrium distance. The stability of the derived spectroscopic constants is better 0.001 Å (Re), 1 cm-1 (e) and 0.01 eV (De). The basis set superposition error (BSSE) was accounted for by the counterpoise method38. Results and discussion Before we discuss the results for atomic and molecular calibration calculations in detail we want to emphasize that the newly developed ANO valence basis sets exhibit errors in the total valence energies which are up to a factor of 10 smaller than those for the previous basis sets. The errors, which were evaluated by comparing algebraic atomic HF calculations to corresponding finite difference results, stay below 8 milli-Hartree (0.2 eV). In addition, the variation of the exponents along the lanthanide/actinide series is much more regular than that for the previously used basis sets and therefore more reliable comparisons between systems containing different lanthanide or actinide elements are possible. Atoms The ground state configurations and LSJ labels of all atoms and ions considered here are listed in Tables 1 and 2 for lanthanides and actinides, respectively. The LSJ labels assigned in the experimental compilations of atomic energy levels 4, 5 were found to be consistent with our present AE MCDHF as well as PP CASSCF/MRACPF results, with the exception of Pa2+, where we found a 2 H state 0.16 (0.26) eV below the 4I state in scalar-relativistic CASSCF (MRCI) calculations. Both states yield J=11/2 as the lowest level in our spin-orbit corrected calculations, which also is experimentally found to be the ground state. Since unfortunately ACPF calculations did not converge for the 2H state, we used the 4I state of Pa2+ as the lowest scalar-relativistic level. In addition, ACPF 7 convergence problems were encountered for Pa+, and limiting values of IP1 and IP2 could not be derived. However, since the energy difference between Pa and Pa2+, i.e., IP1+IP2, is 17.74 eV for a (14s13p10d8f6g)/[6s6p5d4f3g] ANO standard basis set and 17.78 eV for our present basis set extrapolation, we assume that also the individual results for IP1 and IP2 are quite similar. Therefore we use the values obtained with the standard basis set in the following. The results for IP1 to IP4 of all lanthanides and actinides are compiled in Tables 3 and 4, respectively. Let us first check the applied valence basis sets. At the HF level the m.a.e. for the lanthanides are <0.01 eV for IP1, IP2 and < 0.02 eV for IP3, IP4, respectively. The corresponding mean relative errors (m.r.e.) are <0.1% in all cases. The largest relative error (0.6%) occurs for IP1 of Gd; all other relative errors are less than 0.4%. For actinides the mean absolute errors (m.a.e.) are 0.02, 0.01, 0.03 and 0.03 eV, and the corresponding m.r.e. are 0.4, 0.1, 0.2 and 0.1%. The largest error of 2.1% occurs for IP1 of Lr, all other errors are less than 0.9%. The errors were evaluated by comparing CASSCF calculations to corresponding finite difference results. Since the basis set errors are very small at the HF level, an extrapolation to the basis set limit is performed at the correlated level. Differential electron correlation effects (SCF vs. ACPF) turn out to be very important. For actinides they are at the most 1.5 eV for IP1, IP2 and 3 eV for IP3, IP4. These effects are of the same magnitude as found for lanthanides for IP1 and IP2, but by more than 1 eV smaller for IP3 and IP4. We attribute this to the more diffuse character of the actinide 5f shell compared to the lanthanide 4f shell and to the corresponding smaller electron-electron interaction in the shell. The large correlation contributions indicate that the quality of the results mainly depends on the treatment of electron correlation. Quite accurate experimental data is available for IP1 of lanthanides (maximum error bar 0.02 eV 5) and actinides (maximum error bar 0.12 eV; no value available for Lr 4, 39, 40 ). In case of the lanthanides IP2 (maximum error bar 0.08 eV for La to Yb, 0.4 eV for Lu) is also sufficiently accurate, whereas for the actinides only values for Ac and U are at hand (maximum error bar 0.19 eV). The higher ionization potentials of the lanthanides exhibit experimental uncertainties of up to a few tenths of an electron volt (maximum error bars 0.4 eV for IP3, 0.7 eV for IP4). For actinides IP3 and IP4 are available only for Th and U (maximum error bars 0.3 eV for IP3, 1.0 eV for IP4). 8 We now turn to the discussion of the results obtained for lanthanides, which are listed in Table 3. The m.a.e. of the theoretical value for IP1 (0.15 eV) and IP2 (0.13 eV) obtained with the generalized contracted standard basis sets augmented by diffuse functions (std.) are sufficiently small, so that a basis set extrapolation does not really appear to be worthwhile. For selected elements (La, Yb, Lu), the CCSD(T) leads to an improvement of the m.a.e. for IP1 (0.06 eV) and IP2 (0.13 eV) with respect to the ACPF values (0.16 eV; 0.15 eV). The results for IP3 and IP4 at the basis set limit without (ext; 0.30 and 0.37 eV) and with (est; 0.35 and 0.38 eV) correction of PP errors exhibit m.a.e. of similar magnitude. Also here for selected cases (IP3 of La, Eu, Yb, Lu; IP4 of La, Ce, Lu) the m.a.e. is considerably reduced when changing from ACPF (0.37 eV) to CCSD(T) (0.06 eV). The series of estimated values including a correction of PP errors show slightly larger m.a.e. for IP3 and IP4 than the underlying series without correction. This is most likely a result of an error compensation, i.e., the relativistic AE Wood-Boring method used to generate the reference data for the PP generation tends to overestimate IP 3 and IP4 by a few tenths of an electron volt, whereas the corresponding correlation contributions are underestimated. For actinides the theoretical values for IP1 frequently underestimate the experimental values. For a 5fn 7s2 5fn 7s1 ionization process the errors are almost constant at 0.3 eV. This behavior is similar to the one observed for the lanthanides, although the errors are slightly larger for actinides, and mainly related to the incomplete correlation treatment. Our theoretical values for IP2 of the actinides are most likely by 0.1 to 0.2 eV below the unknown experimental values, similar to the corresponding situation for the lanthanides. Unfortunately, only two experimental values are available for calibration, i.e., 11.780.19 eV for Ac and 11.590.37 eV for U. The theoretical values (11.60 eV for Ac, 11.92 eV for U) agree well with the experimental data. For IP3 and IP4 only two experimental values are at hand, i.e., IP3, 18.330.05 eV; IP4, 28.650.02 for Th and IP3, 19.800.31 eV; IP4, 36.701.00 eV for U. For Th the theoretical values (IP3, 18.21 eV; IP4, 28.21 eV) are in reasonable agreement with the experimental data. The ionization of Th3+ 5f1 was found to be a difficult case. Similar with the ionization of Ce3+ 4f1, both require an especially extensive correlation expansion at the CCSD(T) level. A graphical comparison of the quality of our results for Ce and Th is given in figure 1. The very similar trends observed for the correlation 9 contributions in Ce and Th also support that errors in our predicted values for actinides are of similar magnitude to those found for lanthanides, provided the same electronic configurations are involved. For U a strong disagreement between our values of 19.07(18.77) eV for IP3, 33.17(32.73) eV for IP4 without (with) correction of PP errors and the experimental result (IP3, 19.800.31 eV; IP4, 36.701.00 eV) exists. Previous fully-relativistic DKS DFT calculations using three different functionals with self-interaction and gradient corrections yielded values between 18.77 and 18.94 eV for IP3, 32.68 and 32.87 eV for IP441 which agree well with our present values. Refined experiments are recommended. Besides for the calibration studies using basis sets obtained with generalized contraction scheme, the (14s13p10d8f6g)/[10s8p5d4f3g] valence basis sets obtained with a segmented contraction scheme have also been used to calculate the first to fourth ionization potentials of all lanthanide and actinide elements for calibration purposes27, 28 . The values agree with those of the generalized contracted (14s13p10d8f6g)/[6s6p5d4f3g] basis sets within 0.06 eV at the CASSCF and 0.20 eV at the ACPF level. The comparison shows that segmented and generalized contracted basis sets have essentially the same quality. Diatomic molecules Molecular calibration calculations for some selected monohydrides, monoxides, monofluorides and monosulfides of lanthanides and actinides were performed using standard valence basis sets with a generalized and a segmented contraction scheme, respectively 25, 26, 27, 28, 30 . Standard augmented correlation-consistent valence quadruple-zeta basis sets (AVQZ in MOLPRO were used for H, F, O, and S. The results are summarized in Table 5. 20 ) It is also worth reviewing the applications on the study of electronic structure for Ln2 (Ln=La, Ce, Pr, Eu, Gd, Yb, Lu) 32, 33 , not only because good agreement with experimental results was obtained but also since they are typical representatives of the rather complex homonuclear lanthanide dimers containing partially occupied 4f shells on both atoms, i.e., 4f is empty (La2), partially occupied (Ce2, Pr2), half filled (Eu2, Gd2) and completely filled (Yb2, Lu2). The results are summarized in Table 6. 10 Hydrides, Oxides, Fluorides and Sulfides At the Hartree-Fock level the BSSE is negligibly small, whereas at the CI and CC level it slightly increases the bond lengths and depresses the binding energies. The BSSE for the heavier lanthanide elements is usually larger than for the lighter ones. For example, the BSSE correction to bond distances, binding energies and vibrational frequencies of LuH and LaH is respectively: 0.032 Å, 0.011 Å; 0.29 eV, 0.09 eV; 71 cm-1, 10 cm-1. The large contributions in the heavier lanthanide elements arise mainly from the more compact 4d and 4f subshells for heavier elements, which have to be correlated in accurate work. Test calculations for LuH yield counterpoise corrections of 0.009 Å, -0.12 eV, -23 cm-1 when Lu 4s-4f is inactive, 0.016 Å, -0.18 eV, -37 cm-1 when Lu 4s-4d is inactive, and 0.032 Å, -0.29 eV, -71 cm-1 when Lu 4s-4p is inactive. Note that, e.g., the counterpoise-corrected CCSD(T) bond length for inactive Lu 4s-4f is 1.930 Å, whereas the corresponding value for inactive Lu 4s-4p of 1.914 Å agrees very well with the experimental result of 1.912 Å42. The monoxides and monofluorides are the experimentally most extensively investigated diatomics of the lanthanides. Our CCSD(T) results are in good agreement with experimental data for LaO, LuO and LaF (errors in bond distances, binding energies and vibrational frequencies are at most 0.015 Å, 0.16 eV, 6 cm-1). Unfortunately a CCSD(T) treatment was not possible for LnX (Ln=Eu, Gd, Yb; X=O, F). We attribute the larger errors for these molecules, at least partly, to the deficiencies of the CI (SD)+Q correlation treatment. The electronic structure of YbO is still an open problem and a challenge for future more accurate investigations. The calculated bond distance (1.871 Å, PP) for the 4f14σ2σ2π4 1 Σ+ state is close to the DFT result (1.865 Å43), but deviates by about 0.06 Å from the experimental value (1.807 Å44). The vibrational frequency (725 cm-1, PP) agrees well with both DFT (725 cm-1 43 ) and experimental data (699 cm-1 44 ). However the PP binding energy (2.93 eV) underestimates the experimental value (4.29 eV 44 ), whereas DFT (4.65 eV 43 ) slightly overestimates it. In the case of LuF the theoretical results of 7.50 eV (PP) and 7.55 eV (DKH-DFT) indicate that the estimated experimental value of 5.93 eV 42 is substantially too low. Very little experimental information is available for the monohydrides of the lanthanide elements (see Table 5). The PP results are in good agreement with available 11 experimental data (the largest errors are: bond distances 0.019 Å, binding energies 0.06 eV, and vibrational frequencies 7 cm-1), as well as DFT results, except for the binding energies that tend to be larger by up to 1 eV in DFT. The PP results of EuS, YbS are in quite good agreement with experiments; the errors of equilibrium distances, binding energies and vibrational frequencies are 0.1 Å, 0.4 eV, 2 cm-1 for EuS and 0.01 Å, 0.18 eV, 26 cm-1 for YbS. We now turn to the discussion of the results obtained for actinides. At the SCF level, the counterpoise corrections (CPC) of the BSSE are almost negligible (Re0.001 Å, -e1 cm-1, -De0.02 eV). At the CCSD(T) level if excitations from Ac 5d are allowed the significant CPCs are obtained for AcH, AcO, and AcF, especially for De (0.015 Å, 24 cm-1, 0.53 eV) . Relatively small CPCs are obtained if the Ac 5d shell is kept frozen (0.011 Å, 14 cm-1, 0.08 eV) as it has been done in the basis set derivation. However, the corrected results show good agreement with those obtained from a relaxed Ac 5d shell. The uncontracted basis set leads to a substantially smaller CPC for De (0.016 Å, 17 cm-1, 0.16 eV). The basis sets generated including Ac 5d correlation, yields corrections of similar magnitude (0.017 Å, 14 cm-1, 0.16 eV); however, it tends to yield by 0.05 eV too low De after CPC both at the SCF and CCSD(T) level. We attribute this to a loss of flexibility of the Ac basis set in the valence region, since some ANOs tend to be more compact to account for the correlation of the Ac 5d shell. In contrast to Ac more consistent CPCs are obtained for ThO, i.e., the CCSD(T) results obtained with a Th basis set optimized without correlation of 5d (1.839/1.845 Å, 898/891 cm-1, 9.58/9.38 eV without/with CPC) are almost identical to those using a Th basis set optimized with correlation of 5d (1.840/1.845 Å, 897/890 cm-1, 9.56/9.37 eV). We attribute the higher sensitivity of Ac with respect to the optimization of the basis set to the very diffuse character of the Ac 5f and 6d shells and that the corresponding ANO contractions derived from Ac 5f1 7s2 and 6d1 7s2 less efficiently correlate the quite compact 5d shell. The experimental information for diatomics of the actinides is very limited. For ThO counter-poise corrected CCSD(T) values , including also a correction for the fine-structure splitting of the Th 6d2 7s2 3F ground state (Re=1.845 Å, =891 cm-1, D0=8.96 eV), are in excellent agreement with the experimental data (Re=1.840 Å, =896 cm-1, D0=9.00 0.09 eV Laerdahl et al. 47 42, 45, 46 ). For AcH and AcF performed fully relativistic DHF and MP2 calculations using 12 large uncontracted basis sets for AcH and AcF. Their results compare quite favorably to our data. Hong et al.48 presented DFT results for AcH, AcO and AcF using the ZORA as well as the DKH Hamiltonians and the gradient-corrected Becke exchange and Perdew correlation functionals. The agreement with the present data is satisfactory for AcH and AcF, whereas slightly larger deviations are observed for AcO (Table 5). Molecular calibration studies were also done for basis sets using a segmented contraction scheme 27, 28. At the HF and CCSD(T) level the results agree well with those of the generalized contracted basis sets, i.e., the differences amount up to at most 0.002 Å, 0.05 eV, 13 cm-1 for Re, De and e, respectively. Therefore, we conclude that the basis sets obtained from both contraction patterns are reliable tools for electronic structure calculations on lanthanide/actinide systems. Dimers The PP calculations show that the ground states of homonuclear dimers composed of lanthanide elements with ground state valence sub-configuration 5d1 6s2 (6s2 for Pr, Eu, Yb), are most likely La2 1g+ g2 u4, Ce2 1g+, 1u-, 3g-, 3u+, 1 6g, 36u 4f14f1g2u4 , Pr2 5g+ , 5u-, 510g 4f24f2 g2 u4, Eu2 u+ 4f74f7g2u2, 15 Gd2 19g- 4f74f7g2u1g1u2, Yb2 1g+ 4f144f14g2u2 and Lu2 3g- 4f144f14g2u2u2 32, 33 . At present it is difficult to clarify the ground state for Ce2 and Pr2, since all candidates listed above are virtually degenerate (below 20 cm-1 at the MRCI+Q level) and the same spectroscopic constants were obtained (within 210-4 Å and 0.04 cm-1 for Re and e). For Ce2 the practical degeneracy of singlet and triplet states is attributed to the core-like properties of the 4f orbitals (<r>4f 0.55 Å), which lead due to the long bond distance (Re 2.6 Å) to a negligible spin correlation for the two 4f electrons localized on the two Ce atoms. In contrast to Ce2 the state average CASSCF calculations for Pr2 yield the lowest triplet and singlet states at term energies of 300 cm-1 and 400 cm-1, respectively. This 13 different behavior for Pr2 may be attributed to the increasing number of 4f electrons (four 4f electrons for Pr2 and two 4f electrons for Ce2), and the concomitant increased multi-reference character of the wavefunction. Whereas the two unpaired electrons in Ce2 are essentially localized in 4f orbitals and direct coupling is not mediated by other occupied shells of symmetry, in Pr2 the four unpaired electrons are in 4f , 4f orbitals and the direct coupling is possible via the occupied valence shells having partly the same symmetry. For La2, Ce2 and Pr2 the analysis of the valence orbitals for the ground state exhibits a weak contribution (3%) of the 4f shell to chemical bonding. Therefore, the 4f electrons should be treated as valence electrons in highly accurate calculations. Unfortunately there are not many experimental results to be compared. The theoretical results are in reasonable overall agreement with the available experimental data. A notable exception is the e values of La2, Ce2 and Pr2. In order to get some idea of possible matrix effects in the experimental work (Raman spectroscopy in Ar-matrix49, 50 ), a simple linear complex between a single Ar atom and the lanthanide dimers Ln2 was geometry-optimized at the CCSD(T) level 32, 33 . Compared to free La2 (206 cm-1), Ce2 (205 cm-1), Pr2 (198 cm-1), Gd2 (136 cm-1), and Lu2 (120 cm-1) the substantially higher frequencies (228 cm-1 for La2, 223 cm-1 for Ce2 cm-1, and 216 cm-1 for Pr2) were found in the Ar-Ln2 complexes (Ln=La, Ce, Pr). In contrast to this, the Ar-Gd2 and Ar-Lu2 complexes exhibit vibrational frequencies (136 cm-1 for Gd2 and 123 cm-1 for Lu2) which are almost the same as for the free Gd2 and Lu2 systems. Conclusion Optimized atomic natural orbital valence basis sets of polarized valence quadruple-zeta quality describing both low-lying configurations fn+1 s2 and fn d1 s2 14 on equal footing were presented and tested for lanthanide/actinide small core energy-consistent ab initio pseudopotentials. The new basis sets yield significantly lower total valence energies than the ones previously in use. In atomic test calculations, mean absolute errors of less than 0.30 eV are obtained for IP1 and IP2 of lanthanides, as well as for IP1 of actinides. For IP3 and IP4 of lanthanides the mean absolute errors amount up to 0.60 eV. Calculations with the extended basis sets and extrapolation techniques reproduce IP3 and IP4 with an accuracy of 0.34 eV. For actinides the accuracy of IP2-IP4 is estimated to be better than 3%. In molecular test calculations it was found that the basis set superposition errors are considerably smaller, proving the reliability of the new basis sets. An application to the studies of electronic structure for some selected lanthanide dimers (La2, Ce2, Pr2, Eu2, Gd2, Yb2, Lu2) shows the derived spectroscopic constants are in reasonable agreement with experimental data, except for the vibrational frequencies of La2, Ce2, and Pr2. Model calculations point to possible large positive matrix shifts of e for La2, Ce2, and Pr2, but normal behavior for Gd2 and Lu2. It is concluded that La2 and Ce2 exhibit stronger bonding than Pr2, Gd2 and Lu2. Weak contributions to chemical bonding of 4f orbitals (ca. 3%) are found for La2, Ce2, and Pr2, whereas no contributions are observed for Gd2 and Lu2. For the latter systems a relatively large contribution from the 6p orbital (23% for Gd2, 32% for Lu2) is detected. Acknowledgements. The author is grateful to Prof. Michael Dolg for reading the manuscript. The financial support of Fonds der Chemischen Industrie is acknowledged. Key words: Lanthanides, Actinides, Basis sets, Pseudopotentials, Ionization potentials, Spectroscopic constants, Lanthanide dimer 15 Figure captions: Fig. 1 Error with respect to experimental values of IP4 of Ce (36.76 0.01 eV) and Th (28.650.02 eV) for PP CCSD(T) results including spin-orbit corrections. The Th results include a correction for PP errors. The highest angular momentum l included in the uncontracted basis set is indicated on the abscissa; their location is defined by 1/l3. The error of a linear extrapolation to the CCSD(T) limit based on the basis sets including g, h, and i functions are given in parentheses. 16 Table 1: Electronic ground states and configurations for the lanthanides Lnn+ (n=0-4). M1+ M La Ce Pr Nd Pm Sm Eu Gd d1s2 f1d1s2 f3s2 f4s2 f5s2 f6s2 f7s2 f7d1s2 2 D3/2 G4 4 I9/2 5 I4 6H 5/2 7F 0 8 S7/2 9 D2 1 d2 f1d2 f3s1 f4s1 f5s1 f6s1 f7s1 f7d1s M2+ 3 d1 F2 H7/2 5 I4 6 I7/2 7H 2 8F 1/2 9 S4 10 D5/2 f3 f4 f5 f6 f7 f7d1 7 H8 6I 17/2 5I 8 4H 13/2 3F 4 2S 1/2 1S 0 f9 f10 f11 f12 f13 f14 f14s1 4 f2 M3+ 2 D3/2 H4 4 I9/2 5 I4 6H 5/2 7F 0 8 S7/2 9 D2 3 M4+ 1 f1 f2 f3 f4 f5 f6 f7 S0 F5/2 3 H4 4 I9/2 5I 4 6H 5/2 7 F0 8 S7/2 2 p5 p6 f1 f2 f3 f4 f5 f6 2 P3/2 S0 2 F5/2 3 H4 4I 9/2 5I 4 6 H5/2 7 F0 1 1 Tb Dy Ho Er Tm Yb Lu f9s2 f10s2 f11s2 f12s2 f13s2 f14s2 f14d1s2 6 H15/2 5I 8 4I 15/2 3H 6 2F 7/2 1S 0 2D 3/2 f9s1 f10s1 f11s1 f12s1 f13s1 f14s1 f14s2 17 6 H15/2 5I 8 4I 15/2 3H 6 2F 7/2 1S 0 2S 1/2 f8 f9 f10 f11 f12 f13 f14 7 F6 6H 15/2 5I 8 4I 15/2 3H 6 2F 7/2 1S 0 f7 f8 f9 f10 f11 f12 f13 8 S7/2 7F 6 6H 15/2 5I 8 4I 15/2 3H 6 2F 7/2 Table 2: Electronic ground states and configurations for the actinides Ann+ (n=0-4). M1+ M Ac Th Pa U Np d1s2 d2s2 f3d1s2 f3d1s2 f4d1s2 2 D3/2 3 F2 4 K11/2 5 K6 6L 11/2 s2 d2s1 f2s2 f3s2 f4d1s M2+ 1 S0 F3/2 3 H4 4 I9/2 7L 5 4 s1 f1d1 f2d1 f4 f5 M3+ 2 S1/2 G4 4 I11/2 5 I4 6H 5/2 1 1 f1 f2 f3 f4 S0 F5/2 3 H4 4 I9/2 5I 4 2 M4+ p5 2 f1 f2 f3 P3/2 S0 2 F5/2 3 H4 4I 9/2 f4 f5 f6 f7 f8 f9 f10 f11 f12 f13 I4 H5/2 7 F0 8 S7/2 7F 6 6H 15/2 5I 8 4I 15/2 3H 6 2F 7/2 p6 1 1 Pu Am Cm Bk Cf Es Fm Md No Lr f6s2 8 f6 F0 F1/2 7 1 9 fs f7 S7/2 S4 7 1 2 7 2 9 8 fds fs f8 D2 S7/2 9 2 9 1 6 7 fs fs f9 H15/2 H8 10 2 10 1 5 6 f s f s f10 I8 I17/2 11 2 11 1 4 5 f s f s f11 I15/2 I8 12 2 12 1 3 4 f s f s f12 H6 H13/2 13 2 13 1 2 3 f s f s f13 F7/2 F4 14 2 14 1 1 2 f s f s f14 S0 S1/2 14 1 2 14 2 2 1 f ps f s f14s1 P1/2 S0 2+ 2 Our calculations found a Pa H11/2 ground state 26. f7s2 7 f6s1 8 18 7 F0 S7/2 7 F6 6 H15/2 5I 8 4I 15/2 3H 6 2F 7/2 1S 0 2S 1/2 8 f5 f6 f7 f8 f9 f10 f11 f12 f13 f14 6 H5/2 7 F0 8 S7/2 7 F6 6H 15/2 5I 8 4I 15/2 3H 6 2F 7/2 1S 0 5 6 Table 3: First to fourth ionization potentials of the lanthanides (in eV). PP CASSCF/ACPF results, corrected for spin-orbit interaction, obtained with generalized contracted standard basis sets augmented by diffuse functions (std.) and by extrapolation to the basis set limit (ext.) are compared to experimental data (Exp. [5]). Additional theoretical bases set limit estimates were corrected for PP errors (est.). The mean absolute errors (m.a.e.) are listed in the last line. La Ce Pr Nd Pm Sm Eu Gd Tb Dy Ho Er Tm Yb Lu m.a.e . std. 5.58 5.52 5.35 5.40 5.45 5.49 5.51 6.00 5.76 5.80 5.83 5.89 5.95 5.97 5.23 0.15 IP1 Exp. 5.58 5.54 5.46 5.53 5.55 5.64 5.67 6.15 5.86 5.94 6.02 6.11 6.18 6.25 5.43 std. 11.06 11.13 10.62 10.78 10.92 11.02 11.15 12.15 11.75 11.60 11.62 11.72 11.80 11.81 13.81 0.13 IP2 Exp. 11.06 10.85 10.55 10.73 10.90 11.07 11.24 12.09 11.52 11.67 11.80 11.93 12.05 12.18 13.90 ext. 18.82 20.05 21.54 21.98 22.30 23.44 24.58 20.65 21.42 22.53 22.41 22.28 22.73 24.37 20.82 0.30 19 IP3 est. 18.79 20.05 21.47 21.86 22.22 23.33 24.49 20.35 21.30 22.40 22.35 22.23 22.71 24.44 20.77 0.37 Exp. 19.18 20.20 21.62 22.10 22.30 23.40 24.92 20.63 21.91 22.80 22.84 22.74 23.68 25.05 20.96 ext. 50.01 36.15 39.04 40.70 41.23 41.64 43.12 44.83 39.15 41.08 42.46 42.42 41.80 43.33 44.87 0.34 IP4 est. 49.96 36.03 38.84 40.46 40.91 41.37 42.85 44.24 38.82 40.74 42.13 42.06 41.52 43.18 44.86 0.38 Exp. 49.95 36.76 38.98 40.40 41.10 41.40 42.70 44.00 39.37 41.40 42.50 42.70 42.70 43.56 45.25 Table 4: First to fourth ionization potentials of the actinides (in eV). PP CASSCF/ACPF results at the basis set limit corrected for spin-orbit interaction (ext.) and also for PP errors (est.) compared to experimental data (Exp. [4, 39, 40]). IP1 IP2 ext. est. Exp. ext. est. Ac 5.17 5.13 5.17a 11.60 11.60 Th 6.25 6.23 6.31 12.11 12.15 Pa 5.81 5.76 5.90a 11.96 11.96 U 6.06 6.01 6.19 11.63 11.92 Np 5.98 5.98 6.27 11.35 11.56 Pu 5.71 5.73 6.03 11.50 11.48 Am 5.71 5.69 5.97 11.71 11.68 Cm 5.68 5.59 5.99 12.17 12.42 Bk 5.90 5.90 6.20 11.96 11.95 Cf 5.96 5.96 6.28 12.03 12.06 Es 6.07 6.04 6.37 12.20 12.19 Fm 6.18 6.13 6.50a 12.38 12.35 Md 6.25 6.23 6.58 12.47 12.46 No 6.33 6.31 6.65 12.58 12.58 Lr 4.78 4.71 14.25 14.23 m.a.e. 0.24 0.27 Additional Exp. values (in eV) IP2 Ac 11.780.19a, U 19.800.31; IP4 Th 28.650.02, U 36.701.00. Due to convergence problems for Pa+ the results for IP1 (14s13p10d8f6g)/[6s6p5d4f3g] ANO basis set 26. a Semiempirical estimates. 20 IP3 ext. 17.39 18.30 17.73 19.07 19.92 21.37 22.34 20.36 21.93 22.84 23.06 23.66 24.69 26.05 21.52 est. 17.37 18.21 18.65 18.80 19.66 21.04 21.75 20.10 21.55 22.50 22.68 23.13 24.30 25.76 21.50 IP4 ext. 44.99 28.45 31.24 33.17 34.27 35.43 37.26 39.06 36.52 38.12 39.52 40.16 40.60 41.96 44.12 est. 44.78 28.21 30.91 32.76 33.84 34.95 36.77 37.66 35.99 37.64 38.85 39.30 40.03 41.54 43.64 11.590.37; IP3 Th 18.330.05, U and IP2 of Pa were obtained with a Table 5 Bond lengths Re(Å), binding energy De(eV), vibrational frequencies e (cm-1) for XH (X=La, Eu, Yb, Lu, Ac), XO (X=La, Eu, Gd, Yb, Lu, Ac, Th), XF (X=La, Eu, Gd, Yb, Lu, Ac) and XS (X=Eu, Yb) from pseudopotential (PP) calculations using basis sets (14s13p10d8f6g)/[6s6p5d4f3g] in comparison to experimental data Re (Å) a De(eV) b a LaH AcH EuH GdH YbH LuH PP 2.016/2.027 2.135/2.150 2.104/2.116 1.911/1.924 2.041/2.072 1.882/1.914 Exp. 2.032 LaO AcO EuO GdO YbO LuO ThO 1.836/1.841 1.938/1.945 1.874/1.879 1.785/1.791 1.862/1.871 1.784/1.794 1.839/1.845 1.826 LaF AcF EuF GdF YbF 2.027/2.034 2.113/2.122 2.081/2.088 1.956/1.963 2.022/2.034 2.027 1.962 2.016 6.92/6.83 7.57/7.16 5.70/5.61 6.38/6.24 5.36/5.22 LuF 1.908/1.923 1.917 7.81/7.50 2.053 1.912 1.89 1.812 1.807 1.790 1.840 PP 2.97/2.88 3.34/3.02 1.99/1.92 2.47/2.34 1.65/1.49 3.64/3.35 8.30/8.13 7.81/7.28 4.28/4.16 6.82/6.62 3.14/2.93 7.28/6.90 9.16/8.96d b Exp . 1.55, 1.93 3.47 8.29 4.88 7.39 4.29 7.04 9.000.09 8.870.15 8.790.13 814/807 773/762 734/729 884/877 736/725 857/840 898/891 813 688, 672 824 699 842 896 578/574 538/530 496/490 613/606 514/502 570 620/603 612 EuS 2.410/2.417 2.39(2.51)c 3.41/3.31 3.71 405/402 YbS 2.352/2.373 2.359 2.85/2.55 2.73 405/393 The notation …/… refers to without/with counterpoise correction of the basis set superposition error. a Results for LaH, AcH, YbH, LuH, LaO, LuO, ThO, LaF, AcF, and LuF are from CCSD(T) calculations. The results for others are from CISD+Q calculations; f elements 4s(for Ac, Th 5s), 4p(for Ac, Th, 5p) and O, F, S 1s frozen in CCSD(T) and CI calculations. Ref [30, 26, 25]. b references are: LaH [51]; YbH, LuH, LaO, LuO, LuF [42]; EuO[44, 52, 53]; GdO[52, 54, 55]; YbO [44]; LaF[42, 56]; GdF[42, 57]; YbF[42, 58, 59]; EuS[42, 60]; YbS[60, 61]; ThO[45, 46, 42]. c 2.39 and 2.51 eV for EuS were derived from different interpolations. d D , The values have been corrected for molecular (0.03 eV) and atomic (Th 0.38 eV, O 0.01 0 eV) spin-orbit energy lowerings. e empirically interpolated results. 400e 367 21 6.90 e (cm-1) PP Exp.b 1456/1446 1386/1362 1294/1272 1581/1559 1312/1256 1249 1577/1506 1500 a 5.42 6.95 4.90, >5.36, 4.83-4.89 5.93 607 502 Table 6 Bond lengths Re (Å), vibrational constants e (cm-1), and binding energies De (eV) for selected lanthanide dimers. Metal La State g2u4, 1g+ Re(Å)a 2.70 0.03 2.80 De(eV)b 2.31 0.13 2.52 0.22 e (cm-1)c 186 13 236 0.8 Ref. PPs Exp. Ce 4f1 4f1g2u4, 3u+ 2.62 0.02 1.73 0.41 2.47 0.22 201 13 245.4 4.2 PPs Exp. Pr 4f2 4f2g2u4, 510g 2.555 1.19 0.16 1.31 0.30 213 244.9 1.2 PPs Exp. Eu 4f74f7g2u2, 15u+ 4.878 0.080 0.300.17 27 35 PPs Exp. Gd 4f74f7g2u1g1u2, 19g- 2.88 0.02 1.38 0.18 1.784 0.35 149 2 138.7 0.4 PPs Exp. Yb 4f144f14g2u2, 1g+ 4.549 4.19 0.092 0.170.17 25 22 PPs Exp. 4f144f14g2u2u2, 3g- 3.07 0.03 123 1 121.6 0.8 PPs Exp. 1.40 0.12 1.43 0.34 a The estimated experimental bond lengths for La are from Ref. [62] 2 Lu b Experimental values are from Ref. [63]. Spin-orbit corrections of -0.16 eV (La2), +0.13 eV (Ce2), -0.10 eV (Pr2), -0.22 eV (Gd2) and -0.30 eV (Lu2) where added to the scalar-relativistic results 33. c The experimental values for La2, Ce2, Pr2, Eu2, Gd2, Yb2, and Lu2 are from Ref. [50], [64], [64], [65], [65], [66], and [67], respectively. 22 Figure 1 23 [1] Pepper, M.; Bursten, B. Chem. Rev. 1991, 91, 8234. [2] Dolg, M. In Encyclopedia of Computational Chemistry; Schleyer, P. v. R.; Allinger, N. L.; Clark, T.; Gasteiger, J.; Kollman, P. A.; Schaefer III H. F.; Schreiner P. R., Ed.; Wiley: Chichester , 1998; pp 1478. [3] Schreckenbach, G.; Hay, P. J.; Martin, R. L. J. Comput. Chem. 1999, 20, 70. [4] Blaise, J.; Wyart, J. -F. In International Tables of Selected Constants, Vol. 20; CNRS: Paris, 1992. [5] Martin, W. C.; Zalubas, R.; Hagen, L. In Atomic Energy Levels- The Rare Earth Elements; NSRDS-NBS 60: Washington, DC, 1978. [6] Eliav, E.; Kaldor, U.; Ishikawa, Y. Phys. Rev. 1995, A51, 291. [7] Eliav, E.; Shmulyian, S.; Kaldor, U.; Ishikawa, Y. J. Chem. Phys 1998, 109, 3954. [8] Kutzelnigg, W. Phys. Scr. 1987, 36, 416. [9] Küchle, W.; Dolg, M.; Stoll, H.; Preuss. H. J. Chem. Phys. 1994, 100, 7535. [10] Dolg, M.; Stoll, H.; Savin, A.; Preuss, H. Theor. Chim. Acta 1989, 75, 173. [11] Dolg, M.; Stoll, H.; Preuss, H. J. Chem. Phys. 1989, 90, 1730. [12] Hay, P. J. J. Chem. Phys. 1983, 79, 5469. [13] Ermler, W. C.; Ross, R. B.; Christiansen, P. A. Int. J. Quant. Chem. 1991, 40, 829. [14] Nash, C. S.; Bursten, B. E.; Ermler, W. C. J. Chem. Phys. 1997, 106, 5133. [15] Hay, P. J.; Martin, R. L. J. Chem. Phys. 1998, 109, 3875. [16] Cundari, T. R.; Stevens, W. J. J. Chem. Phys. 1993, 98, 5555. [17] Ross, R. B.; Gayen, S.; Ermler, W. C. J. Chem. Phys. 1994, 100, 8145. [18] Sakai, Y.; Miyoshi, E.; Tatewaki, H J. Mol. Struct. (Theochem) 1998, 451, 143. [19] Seijo, L.; Barandiaran, Z.; Harguindey, E. J. Chem. Phys. 2001, 114, 118. [20] Werner, H. -J.; Knowles, P. J.; Amos, R. D.; Berning, D. L.; Cooper, D.L.; Deegan, M.J.O.; Dobbyn, A.J.; Eckert, F.; Hampel, C.; Hetzer, G.; Leininger, T.; Lindh, R.; Lioyd, A.W.; Meyer, W.; Mura, M.E.; Nicklass, A.; Palmieri, P.; Peterson, K.; Pitzer, R.M.; Pulay, P.; Rauhut, G.; Schütz, M.; Stoll, H., Stone, A.J., Thorsteinsson, T., 2000; MOLPRO 2000 is a package of ab initio electronic structure programs [21, 22, 23, 24]. 24 [21] Knowles, P. J.; Werner, H.-J., Chem. Phys. Lett. 1985, 115, 5053. [22] Werner, H.-J.; Knowles, P. J. J. Chem. Phys. 1988, 89, 5803. [23] Knowles, P. J.; Werner, H.-J. Chem. Phys. Lett. 1988, 145, 514. [24] Knowles, P. J., Werner, H. J. Theor. Chim. Acta 1992, 84, 95. [25] Cao, X.; Dolg, M. J. Chem. Phys. 2001, 115, 7348. [26] Cao, X.; Dolg, M.; Stoll, H. J. Chem. Phys. 2002, 117, in press. [27] Cao, X.; Dolg, M. J. Mol. Struct.: THEOCHEM 2002, 581, 139. [28] Cao, X.; Dolg, M. to be published. [29] Cao, X.; Dolg, M. Chem. Phys. Lett. 2001, 349, 489. [30] Cao, X.; Liu, W; Dolg, M. Science in China(B) 2002, 45, 91. [31] Cao, X.; Dolg, M. Mol. Phys. in press. [32] Cao, X.; Dolg, M. Theor. Chem. Acc. 2002, 108, 150. [33] Cao, X.; Dolg, M. Mol. Phys. accepted. [34] Dolg, M.; Wedig, U.; Stoll, H.; Preuss, H. J. Chem. Phys. 1987, 86, 866. [35]Dyall, K. G.; Grant, I. P.; Johnson, C. T.; Parpia, F. A.; Plummer, E. P. Comput. Phys. Commun. 1989, 55, 425, GRASP is a relativistic atomic structure code. [36] http://www.theochem.uni-stuttgart.de/pseudopotentials [37] Gdanitz, R. J.; Ahlrichs, R. Chem. Phys. Lett. 1988, 143, 413. [38] Boys, S. F.; Bernardi, F. Mol. Phys. 1970, 19, 203. [39] Köhler, S.; Deissenberger, R.; Eberhardt, K.; Erdmann, N.; Herrmann, G.; Huber, G.; Kartz, J. V.; Nunnemann, M.; Passler, G.; Rao, G.; Riegel, P. M.; Riegel, J.; Trautmann, N.; Wendt, K. Spectrochim. Acta B 1997, 52,717. [40] Peterson, J. R.; Erdman, N.; Nunnemann, M.; Eberhardt, K.; Huber, G.; Kartz, J.V.; Passler, G.; Stetzer, O.; Thorle, P.; Trautmann, N.; Waldeck, A. J. Alloys and Compounds 1988, 271, 876. [41] Liu, W; Küchle, W; Dolg, M Phys. Rev. A 1998, 58, 1103. [42] Huber, K. P.; Herzberg, G. Molecular Spectra and Molecular Structure, vol. 4, Constants of Diatomic Molecules; Van Nostrand: New York, 1979. [43] Liu, W.; Dolg, M.; Li, L. J. Chem. Phys. 1998, 108, 2886. [44] McDonald, S. A.; Rice, R. F.; Field, R. W.; Linton, C. J. Chem. Phys. 1990, 93, 7676. [45] Ackermann, R. J.; Rauh, E. G. High Temp. Sci. 1973, 5, 463. [46] Hildenbrand, D. L.; Murad, E. J. Chem. Phys. 1974, 61, 1232. 25 [47] Laerdahl, J. K.; Faegri, K.; Visscher, L.; Saue, T. J. Chem. Phys. 1998, 109, 10806. [48] Hong, G.; Dolg, M.; Li, L. Chem. Phys. Lett. 2001, 334, 396. [49] Liu, Y.; Fang L.; Shen, X.; Lombardi, J.R. Chem. Phys. 2000, 262, 25. [50] Fang, L.; Chen, X.; Shen, X.; Lombardi, J.R. J. Chem. Phys. 2000, 113, 10202. [51] Gabelnick, S. D.; Reedy, G. T.; Chasanov, M. G. J. Chem. Phys. 1974, 60, 1167. [52] Dulick, M.; Murad, E.; Barrow, R. F. J. Chem. Phys. 1986, 85, 385. [53] Gabelnick, S. D.; Reedy, G. T.; Chasanov, M. G. J. Chem. Phys. 1974, 60, 1167. [54] Yadav, B. R.; Rai, S. B.; Rai, D. K. J. Mol. Spectrosc. 1981, 89, 1. [55] DeKock, L.; Weltner, W. J. Phys. Chem. 1971, 75, 514. [56] Hildenbrand, D. L.; Lau, K. H. J. Chem. Phys. 1995, 102, 3769. [57] Kaledin, L. A.; Bloch, J. C.; McCarthy, M. C. J. Mol. Spectrosc., 1996, 176, 148. [58] Barrow, R. F.; Chojinicki, A. H. J.Chem. Soc. Faraday Trans. II 1975, 71, 728. [59] Yokozeki, A.; Menzinger, M. Chem. Phys. 1976, 14, 427. [60] Czack, G.; Hein, H.; Kirschstein, G.; Merlert, P; Vetter, U In Gmelin Handbook of Inorganic Chemistry: Sc Y, La-Lu Rare Earth Elements; Springer: Berlin, 1983; Chapter 7. [61] Melville, T. C.; Coxon, J. A. J. Chem. Phys. 2000, 113, 1771. [62] Verhaegen, G.; Smoes, S.; Drowart, J. J. Chem. Phys., 1964, 40, 239. [63] Connor, J.A. In Metal Clusters in Catalysis, Studies in Surface Science and Catalysis, 29; Elsevier Science: Amsterdam, 1986. [64] Shen, X.; Fang, L.; Chen, X.; Lombardi, J.R. J. Chem. Phys. 2000, 113, 2233. [65] Goodfriend, P.L. Spectrochim Acta 1984, 40, 283 [66] Chen, X.; Fang, L.; Shen, X.; Lombardi, J.R. J. Chem. Phys. 2000, 112, 9780. [67] Fang, L.; Chen, X.; Shen, X.; Lombardi, J.R. J. Chem. Phys. 2000, 113, 10202. 26