3414 J. Am. Chem. Soc.

1999, 121, 3414-3420

What Do the Kohn-Sham Orbitals and Eigenvalues Mean?

Ralf Stowasser and Roald Hoffmann*
Contribution from the Department of Chemistry, Baker Laboratory, Cornell UniVersity,
Ithaca, New York 14853-1301
ReceiVed July 29, 1998. ReVised Manuscript ReceiVed February 16, 1999

Abstract: Kohn-Sham orbitals and eigenvalues are calculated with gradient-corrected functionals for a set of
small molecules (H2O, N2, CrH66-, and PdCl42-), varying basis sets and functionals. The calculated Kohn-
Sham (KS) orbital shapes, symmetries, and the order and absolute energy of the associated eigenvalues are
investigated and compared with those of Hartree-Fock (HF) and one-electron extended Hückel (eH) calculations,
as well as experimental ionization potentials. The shape and symmetry properties of the KS orbitals are very
similar to those calculated by HF and eH methods. The energy order of the occupied orbitals is in most cases
in agreement among the various methods. The order of empty orbitals of a minimal basis set is sometimes
interchanged, within that group or with some orbitals resulting from a larger basis calculation. Overall the KS
orbitals are a good basissas Baerends suggestedsfor qualitative interpretation of molecular orbitals. For the
Kohn-Sham eigenvalues we find an approximately linear dependency of |iKS - iHF| vs iHF (≈-IP) for the
occupied as well as for the unoccupied orbital eigenvalues. We suggest an ax + b scaling for quantitative
interpretation of KS eigenvalues, at least if these are calculated utilizing commonly used functionals.

1. Introduction shied away from attributing to Kohn-Sham orbitals the reality

that (we think) they deserve.
Chemists have found orbitals usefulsatomic orbitals, mo-
As with any general statement, this is an exaggeration. There
lecular orbitals, the orbitals of molecular fragments. Orbitals
are theoreticians5-14 who haVe provided a welcome place for
provide a natural language for an aufbau of the complex reality
orbitals in the density functional schemesBaerends13,14 and
of the molecules of the inorganic and organic world.
Parr10-12 are prominent examples. Baerends’ impressive work
As Hartree-Fock self-consistent-field theory and its improve- pinpoints the physical significance of KS orbitals by “splitting
ment through configuration interaction (HF-SCF and CI) the exchange-correlation part of the KS potential into a part
evolved, the “reality” of orbitals was both strengthened and that is directly related to the total energy and a so-called response
weakened. The strengthening was derived from Koopman’s part that is related to response of the exchange-correlation hole
theorem, which provided a simple and natural connection to density change”.13,14 Baerends and co-workers in fact argue
between orbital energy and ionization potential (enhanced by that the KS orbitals are Very suitable for qualitative, chemical
the development of photoelectron spectroscopy, which ef- applications. Meanwhile, without waiting for the justification
fectively allowed us to measure easily ionization potentials (IP’s) that Baerends’ arguments provide, the chemical community has
other than the first). However, the limitation of the single- begunsmainly in the past few yearssto apply KS orbitals in
configuration viewpoint weakened, so to speak, the reality of rationalizing chemical phenomena,15-22 as we are used to doing
orbitals. If many states of a molecule are not well described by with extended Hückel (eH) orbitals.
one (or two) configurations, the orbital idea loses its utility.
(5) Perdew, J. P.; Norman, M. R. Phys. ReV. 1982, 26B, 5445.
We are in a time of ascendency of density functional theory1-3 (6) Perdew, J. P.; Levy, M. Phys. ReV. Lett. 1983, 51, 1884.
for the computation of the electronic structure of molecules. (7) Kohn, W.; Vashishta, M. R. In Theory of the Inhomogeneous Electron
The methodology contains orbitals to be sure, the Kohn-Sham Gas; Lundquist, S., March, N. H., Eds.; Plenum: New York; p 79.
(8) Levy, M.; Perdew, J. P.; Sahni, V. Phys. ReV. 1984, 30A, 2745.
(KS) orbitals4 which we will denote as φi, with associated (9) Tozer, D. J.; Handy, N. C.; Green, W. H. Chem. Phys. Lett. 1997,
eigenvalues i. But the fundamental variable, which determines 273, 183.
all observables, is the total electron density ∑iφi2 ) F. (10) Kohn, W.; Becke, A. D.; Parr, R. G. J. Phys. Chem. 1996, 100,
12974.
From the beginning of the utilization of the density functional (11) Zhao, Q.; Parr, R. G. Phys. ReV. 1992, 46A, 2337.
method, the significance of the Kohn-Sham orbitals has been (12) Zhao, Q.; Parr, R. G. J. Chem. Phys. 1993, 98, 543.
deemphasized, perhaps because it its very difficult to extract (13) Baerends, E. J.; Gritsenko, O. V.; van Leeuwen, R. In Chemical
quantitative information from these orbitals. They have been Application of Density-Functional Theory; Laird, B. B., R. B., Eds.;
(14) Baerends, E. J.; Gritsenko, O. V. J. Phys. Chem. 1997, 101, 5383.
often viewed as just an auxiliary construct, a necessary but not (15) Bickelhaupt, F. M.; Baerends, E. J.; Ravenek, W. Inorg. Chem. 1990,
necessarily meaningful way to build up the all-important total 29, 350.
density. Here then is the crux of the problemschemists know (16) Rosa, A.; Baerends, E. J. New J. Chem. 1991, 15, 815.
(17) DeKoch, R. L.; Baerends, E. J.; Hengelmolen, R. Organometallics
that orbitals are useful, but the physicists and chemists who 1984, 3, 289.
use density functional theory so fruitfully have by and large (18) Sargent, A. L.; Titus, E. P. Organometallics 1998, 17, 65.
(19) Garland, M. T.; Saillard, J.-Y.; Ogliaro, F.; Otero, M.; Roman, E.
(1) Parr, R. G.; Yang, W. Density-Functional Theory of Atoms and Inorg. Chim. Acta 1997, 257, 253.
Molecules; Oxford University Press: New York, 1989. (20) Ehlers, A. W.; Baerends, E. J.; Bickelhaupt, F. M.; Radius, U. Chem.
(2) Dreizler, R. M.; Gross, E. K. U. Density Functional Theory; Springer- Eur. J., in press.
Verlag: New York, 1990. (21) Bickelhaupt, F. M.; Radius, U.; Ehlers, A. W.; Hoffmann, R.;
(3) Ziegler, T. Chem. ReV. 1991, 91, 651. Baerends, E. J. New J. Chem. 1998, 1.
(4) Kohn, W.; Sham, L. J. Phys. ReV. 1965, 140A, 1133. (22) Hoffmann, R. J. Mol. Struct. (THEOCHEM) 1998, 424, 1.

10.1021/ja9826892 CCC: $18.00 © 1999 American Chemical Society

Published on Web 03/25/1999
Kohn-Sham Orbitals and EigenValues J. Am. Chem. Soc., Vol. 121, No. 14, 1999 3415

Table 1. Parameters Used in the EH Calculations

atom orbital Hii (eV) ζ1 ζ2 C1 C2
H 1s -13.6 1.3
N 2s -26.0 1.950
2p -13.4 1.950
O 2s -32.3 2.275
2p -14.8 2.275
Cl 3s -26.3 2.183
3p -14.2 1.733
Cr 4s -8.66 1.7
4p -5.24 1.7
3d -11.22 4.95 0.5060 1.80 0.6750
Pd 5s -7.32 2.19
5p -3.75 2.152
4d -12.02 5.983 0.5535 2.613 0.6701

In this paper we confront the problem discussed above

empirically and directly, we hope. We ask two questions: (1)
Are the KS orbitals different in number, symmetry properties,
and shape from the orbitals of a HF-SCF calculation or from
those of a one-electron scheme? (2) What shall we make of the
Kohn-Sham eigenvalues given by currently popular potentials?
How are they related to ionization potentials? Do they provide
an energy ordering of the orbitals that resembles or differs from
that given by HF-SCF, one-electron schemes, or experimental
IP’s?23

2. Methodology
KS and HF-SCF calculations were performed by means of Figure 1. Calculated occupied valence and virtual orbitals for water
the Gaussian 94 program package.24 For all HF and DF by BP86/6-31G*, RHF/6-31G*, and eH methods, as well as the
experimental values.39 The Fermi level is indicated by a dotted line;
(functional BP86) calculations on first- and second-row mol-
the compression of the KS levels relative to HF levels is highlighted
ecules the 6-31G* basis set was used. For transition-metal by dashed lines.
complexes we worked with the comparable size LANL1DZ
basis set with a Los Alamos ECP. For the variation of basis
occupied leVels (the eH orbitals do not contain the core 1s
sets and functionals the STO-3G, 3-21G, 6-31G, 6-31G*,
levels). Could it be different? Could you imagine that the KS
6-311G, 6-311G*, 6-311G** basis sets and BHandH, X-R,
orbitals of, for example, a Ne atom are not 1s, 2s, 2p-like? One
BP86, BLYP, PW91, HFS, and SVWN (LDA) exchange
would justifiably think the eH orbitals are a poor approximation
correlation functionals were investigated. eH orbital energy
(to either the HF-SCF or DFT orbitals), but even as poor as
values were calculated with the program YAeHMOP25 and the
that approximation might be, the qualitative description of the
parameter set given in Table 1.
orbitals, really determined as it is by their nodal structure, should
We will study the following molecules in detail: H2O, N2,
be the same.
CrH66-, and PdCl42-. In case of the DF and HF calculations
We need to be a little more precise, however, in defining
we perform a full geometry optimization; for the hypothetical
several types of orbitals: group I, core levels (e.g. 1s on
CrH66- we looked at single-point calculation on a MP2/double-ζ
oxygen); group II, occupied valence orbitals; group III, unoc-
optimized Oh CrH6 geometry.26,27 The eH results are for the
cupied valence orbitals; group IV, higher unoccupied orbitals,
experimental geometries (H2O,28 N2,29 PdCl42-30) and for the
mainly a result of the larger basis. Figure 1 shows the one-
MP2/double-ζ optimized geometry of CrH6.
electron energies of these orbitals for eH, DFT, and HF-SCF
2.1. Orbital Shape: The Case of Water. Of course, the
calculations for a water molecule.
number and symmetry properties of the KS orbitals, the
The eH calculations by definition use a valence orbital basis
canonical HF orbitals, and eH orbitals are the same for the
set; therefore, they lack group I + IV orbitals. The HF-SCF
(23) Many of these questions have been addressed by R. G. Parr and and DFT calculations have orbitals of all kinds, the number of
co-workers,1,10-12 whose excellent work provides a quantitative account of group IV orbitals depending, of course, on this basis set size.
the relationship between KS and HF orbitals and suggests that the EH
method be viewed as a particular empiricization of the KS method. The border between group III and IV levels cannot be sharp.
(24) Trucks, G. W.; Head-Gordon, M.; Gill, P. M. W.; Wong, M. W.; In particular, there are excited states which are reasonably well-
Foresman, J. B.; Johnson, B. G.; Schlegel, H. B.; Robb, M. A.; Replogle, decribed by valence-state MO’s and others (e.g. Rydberg states)
E. S.; Gomperts, R.; Andres, J. L.; Raghavachari, K.; Binkley, J. S.;
Gonzalez, C.; Martin, R. L.; Fox, D. J.; Defrees, D. J.; Baker, J.; Stewart,
which need a large basis for an accurate description.
J. J. P.; Pople, J. A. Gaussian 94; Gaussian, Inc., Pittsburgh, PA, 1994. There is no doubt that each method gives (aside from the
(25) Landrum, G. A.: Yet Another extended Hückel Molecular Orbital core states) for water four occupied valence energy levels of
Package. YAeHMOP is freely available on the WWW at the URL http://
overlap.chem.cornell.edu:8080/yaehmop.html. (28) Herzberg, G. L. Molecular Spectra and Molecular Structure. III.
(26) The simplest conceivable 18-electron transition-metal complex Electronic Spectra and Electronic Structure of Polyatomic Molecules; Van
(CrH66-) is not stable in DF and HF geometry optimizations, presumably Nostrand Reinhold: New York, 1966.
due to the high negative charge. To calculate this prototype, we used (29) Huber, K. P.; Herzberg, G. L. Molecular Spectra and Molecular
therefore the optimized CrH6 geometry and did a single-point SCF Structure. IV. Constants of Diatomic Molecules; Van Nostrand Reinhold:
calculation for the 6- species. New York, 1979.
(27) Kang, S. K.; Tang, H.; Albright, T. A. J. Am. Chem. Soc. 1993, (30) Tables of Interatomic Distances and Configurations in Molecules
115, 1971. and Ions; Special Publication No. 11; The Chemical Society: London, 1958.
3416 J. Am. Chem. Soc., Vol. 121, No. 14, 1999 Stowasser and Hoffmann

Figure 2. Calculated contour plots (xy plane) of the a1, b2, a1, and b1 orbitals of water (for b1, Ψ ) 0 in the xy plane; this orbital is hence plotted
with an offset of 0.5 Å) with BP86/3-21G, BP86/6-31G*, RHF/3-21G, RHF/6-31G*, and eH methods.

2a1 + b2 + b1 symmetry. We will discuss the energies of these

levels and the significance of the eigenvalues in the next section;
here we want to focus on their shape.
Figure 2 shows a contour diagram of the four occupied
valence MO’s of water computed by the three methods (two
basis sets for HF-SCF and DFT).
Are these orbitals similar or different? The eH orbitals
arbitrarily exclude the core and use nodeless Slater functions;
thus, the eH orbitals are recognizably different, especially near
the nuclei. The other four sets of orbitals are, on the scale of
the figure, nearly indistinguishable (we could exaggerate the
difference with a density difference map), and away from the
nuclei they are not that different from the eH orbitals.
2.2. Orbital Energies and Their Ordering, and the
Relationship to Ionization Potentials: Water. The general
literature of density functional theory states that only the energy
i of the highest occupied KS orbital (HOMO) has physical
significance, in the sense that the i value of the HOMO is in
theory equal to the first ionization potential1-4,31,32 In practice,
with the commonly used functionals implemented in standard
quantum chemistry program packages, even HOMO energies
differ significantly from experimental data (e.g. IP ) 14.52 eV35
and HOMOLDA ) - 6.50 eV for the nitrogen atom36). This
deviation arises from the insufficient cancellation of the self-
interaction error in the Hartree term ∫∫ d3rid3rj[F(ri)][W(ri,rj)]-
[F(rj)] by terms of the opposite sign in the approximated
exchange-correlation functionals.2 New functionals and ap-
proximations for the KS potential37,38 have been developed (e.g. Figure 3. Calculated occupied valence and virtual KS orbitals for water
LSDSIC, OEP, KLI) which correct the self-interaction error and using a BP86 functional but varying the basis set. The occupied core
lead to precise agreement of the HOMO energy with the first orbitals are denoted as type I, the occupied valence orbitals as type II,
the virtual orbitals arising from a minimal basis as type III, and all
(31) The proof33 given in 1982 that HOMO ) -IP has been recently other virtual orbitals as type IV; the Fermi level is indicated by a dotted
questioned by Kleinman.34 line.
(32) Levy, M. Phys. ReV. 1997, 56B, 1133, 12044.
(33) Perdew, J. P.; Parr, R. G.; Levy, M.; Balduz, J. L., Jr. Phys. ReV. IP; hopefully these approximations will be implemented in
Lett. 1982, 49, 1691.
(34) Kleinman, L. Phys. ReV. B 1997, 56, 12042. quantum chemistry packages soon.
(35) Moore, C. E. Atomic Energy LeVels; NBS Circular; National Institute We begin our study with the influence of the basis set chosen
of Standards and Technology: Washington, DC, 1998; p 467. on the KS energy levels. In Figure 3 the KS orbital energies of
(36) Trickey, S. B. Phys. ReV. Lett. 1986, 56, 881. the water molecule, calculated with the BP86 functional and
(37) Grabo, T.; Gross, E. K. U. Chem. Phys. Lett. 1995, 240, 141-150.
(38) Chen, J.; Krieger, J. B.; Li, Y.; Lafrate, G. J. Phys. ReV. A 1996, different basis sets, are shown. The absolute KS orbital energy
54, 3939. values are (with the exception of the inadequate STO-3G
Kohn-Sham Orbitals and EigenValues J. Am. Chem. Soc., Vol. 121, No. 14, 1999 3417

Figure 5. Differences of RHF/6-31G* and DF(BP86, B3LYP)/6-31G*

calculated orbital energies plotted vs RHF/6-31G* orbital energies for
Figure 4. Calculated occupied valence and virtual orbitals for water, water.
assuming a 6-31G* basis and varying the functional. The Fermi level
is indicated by a dotted line. vertical IP’s.39 As far as the occupied orbital energies go (cf.
Figure 1; the gap between filled and unfilled levels is indicated
minimal basis set) roughly independent of the basis set for type by a dotted line), the HF results match well the experimental
II and III orbitals. The symmetries and order of all KS orbitals IP’s. The eH results are in acceptable agreement with experi-
are not influenced by expanding the basis set. mental data (cf. Figure 1) as well. However, the KS energy
Next (cf. Figure 4) we study the influence of the exchange- levels are shifted by a constant (e.g. for the HOMO ∆ ≈ 7
correlation functional used on the KS orbital energies of the eV) to higher energy, relative to the HF results.
water molecule. We included the BHandH, Xalpha, BP86, During preparation of this paper we became aware of work
BLYP, PW91, HFS, and SVWN (LDA) exchange correlation similar in motivation to ours, by Politzer and Abu-Awwad.40,41
functionals. The basis is fixed as a common 6-31G* set of These authors examine the relationship between various IP’s
double-ζ quality. and the KS and HF i’s. Their results for the H2O molecule are
The KS energy levels drawn in Figure 4 show substantial similar to ours, and in general they find that the calculated KS
absolute energy shifts, depending on the applied functional. i’s differ significantly from the experimental IP’s, as we do.
However, the relative spacing of the levels (e.g. the HOMO- It must be mentioned right away here that people who use
LUMO gap) is approximately constant within the entire set DFT calculations and are interested in IP’s or ultraviolet
(except for the hybrid functionals B3LYP and BHandH). The photoelectron spectroscopy (UPS) do not generally focus on
variation in the orbital energies may be traced to different self- the i’s but calculate IP’s as E(molecule) - E(cation radical),
interaction errors of the applied functional. If the absolute shifts choosing the appropriate cation radical state. The results are in
are corrected by a suitable scaling, e.g. by a constant energy general quite satisfactory. Still, one would like to see if a
relative to the experimentally accessible HOMO energy (as “Koopman’s-theorem like” association of IP’s with KS i’s
measured by the ionization potential), then the orbital energies calculated with common functionals might work.
of all examined functionals (except the hybrid functional B3LYP 2.3. A Scaling Relationship. We find a systematic (if as yet
and the BHandH functional) match very well. However, if no mysterious) relation between KS energies, calculated with
scaling is applied, one has to be aware that different contem- gradient corrected functionals, and IP’s (which are close to the
porary functionals may lead to orbital energies different by up HF values). In Figure 5 we plot the energy difference between
to 10 eV from each other for a given valence orbital. A reviewer the HF and KS orbital energies vs the HF orbital energies (which
(whom we thank) aptly remarks: “The finite basis set can be are close to the IP’s for the occupied levels). One can see a
arbitrarily large until its finiteness causes ‘negligible’ errorss linear energy shift42 for the occupied orbital energies (left part
given enough computing power. Improving approximate usable of the graph; rhomboid symbols).43 In calculations on extended
XC-potentials is an ongoing major theoretical challenge”.
(39) Siegbahn, K.; Nordling, C.; Johansson, G.; Hedman, J.; Heden, P.
There is more to be learned from the orbital energies F.; Hamrin, K.; Gelius, U.; Bergmark, T.; Werme, L. O.; Manne, R.; Baer,
computed by the various methods. In Figure 1 the KS orbital Y. ESCA Applied to Free Molecules; North-Holland: Amsterdam, 1969.
energies of water (calculated with the BP86/6-31G* functional/ (40) Politzer, P.; Abu-Awwad, F. Theor. Chem. Acc. 1998, 99, 83.
basis set combination) were shown, along with the orbital energy (41) Politzer, P.; Abu-Awwad, F.; Murray, J. S. Int. J. Quantum Chem.,
in press.
levels of Hartree-Fock (RHF/6-31G*) and extended Hückel (42) This linear dependency is not necessarily valid for the (yet unknown)
(YAeHMOP25) calculations and the experimentally determined true KS eigenvalues.
3418 J. Am. Chem. Soc., Vol. 121, No. 14, 1999 Stowasser and Hoffmann

Figure 6. Calculated occupied valence and virtual orbitals for the Figure 7. Calculated occupied valence and virtual orbitals for CrH66-
nitrogen molecule by BP86/6-31G*, RHF/6-31G*, and eH methods by BP86/LANDZL1, RHF/LANDZL1, and eH methods. The Fermi
and experimental values.39 The compression of the KS levels relative level is indicated by a dotted line.
to HF levels is highlighted by dashed lines.
and HF band gap energies48 and with calculations in the
systems similar linear dependencies between KS and so-called literature40 using hybrid functionals and the 6-31+G** basis
quasi-particle band energies appear.44 Most intriguing in Figure set.
5 (right part of the graph; [ symbols) is that this linear energy The calculated core orbital energy BP86core ) -510.87 eV
shift apparently also applies to the virtual orbitals! This is of the water molecule is very different from the HF result HFcore
consistent with the fact that calculated virtual KS energy levels ) 559.17 eV. Again the HF energy represents a reasonably good
are usually found to be of lower energy than those of HF reference, because the sum of the experimental determined core
calculations.45-47 energy exptlcore ) 539.89 eV49 and an estimation of the relaxation
To obtain a correspondence between KS (calculated with effect relaxcore ≈ 20 eV50 gives an energy value of ∼560 eV,
gradient-corrected functionals, e.g. BP86) and HF orbital which is close to the HF result. The calculated difference BP86core
energies, an empirical scaling of the form ax + b appears to be - HFcore lies roughly on the extrapolated line in Figure 5
necessary. b adjusts for the constant shift (self-interaction error), (aH2O,coreocc ≈ 0.1).
and a accounts phenomenologically for the linear scaling which Qualitatively, the order in energy of the calculated occupied
we find. We defined a and b in that way that aocc and avirt are and virtual H2O orbitals is consistent for all methods applied
the slopes of the interpolation lines (cf. Figure 5) for the (cf. Figure 5). Orbitals of type IV appear above the orbitals
occupied and virtual orbital energy differences, respectively, type III.
and b is the y intercept of the crossing of the two interpolation 2.4. The Case of N2. The next molecule we want to focus
on is N2, because two of the valence orbitals of N2, σg+(2pz)
lines. Specific calculation of a and b gives the following for
and π1u(2px,y), are close together in energy. Figure 6 shows
H2O: aH2Oocc ) -0.24, bH2O ) 4.68 eV, and aH2Ovirt ) 0.15.
trends similar to those found for the water molecule. The
The other data set in Figure 5 (symbolized by triangles) shows calculated HF orbital energies match best with the experimental
what transpires if ones uses the B3LYP functional. The deviation values; eH one-electron energies are acceptably close, and the
issas one expects from the hybrid character of the functional KS orbitals show again the ax + b energy shift. Clearly one
(intermediate between HF and DFT)ssmaller than that of the can also see this shift in the compression (indicated by dashed
pure density functionals. This is in agreement with the fact that lines, cf. Figure 6) of the KS orbital energy spacing relative to
band gaps of hybrid functionals lie between the calculated DF the HF energies plotted. The calculated scaling parameters aN2occ
) 0.27, bN2 ) 5.4 eV, and aN2virt are relatively close to the values
(43) An upward shift of KS eigenvalues, calculated with LDA and GGA,
is noted in: van Leuween, R.; Baerends, E. J. Phys. ReV. A 1994, 49, 41, for the H2O molecule.
2421. Interestingly the experimentally determined (and very sensi-
(44) Hybertson, M. S.; Louie, S. G. Phys. ReV. 1986, 34, 5390. tive) order of highest occupied levels “πu under σg+”39 is well
(45) Baerends, E. J.; Ellis, D. E.; Ros, P. Chem. Phys. 1973, 2, 41.
(46) Baerends, E. J.; Ros, P. Chem. Phys. 1975, 8, 412. (49) Thomas, T. D.; Shaw, R. W., Jr. J. Electron Spectrosc. Relat.
(47) Baerends, E. J.; Ros, P. Int. J. Quantum Chem. Phenom. 1974, 5, 1081.
(48) Salzner, U.; Lagowski, J. B.; Pickup, P. G.; Poirier, R. A. J. Comput. (50) Clark, D. T.; Cromarty, B. J.; Sgamellotti, A. J. Electron Spectrosc.
Chem. 1997, 18, 1943. Relat. Phenom. 1998, 424, 1.
Kohn-Sham Orbitals and EigenValues J. Am. Chem. Soc., Vol. 121, No. 14, 1999 3419

Figure 8. Calculated occupied valence and virtual orbitals for PdCl42- Figure 9. Differences of RHF/6-31G* and DF(BP86)/6-31G* calcu-
by BP86/LANDZL1, RHF/LANDZL1, and eH methods. The Fermi lated orbital energies plotted vs RHF/6-31G* orbital energies for
level is indicated by a dotted line. The compression of the KS levels PdCl42-.
relative to HF levels is emphasized by dashed lines.
LUMOsin the DF/HF calculations this orbital is not the
reproduced by KS as well as eH calculations. In the HF results51 expected eg symmetry, but a t1u orbital instead. We find the
the πu - σg+ ordering (cf. Figure 6) is interchanged. The expected52 “LUMO” (mainly dx2-y2 and dz2, eg) at much higher
symmetries and shapes of the orbitals calculated with DF, HF, energy, above the t1u, a1g orbitals of type III, and another t1u, a
and eH methods are as they should be and are consistent with higher order type IV orbital (cf. Figure 6). This is a nice
each other. examplesas mentioned at the beginning of the papersof the
The same trends are observed in calculations (not reported fact that the borderline between type III and type IV orbitals is
here) on molecules such as O2, F2, ethane, ethylene, singlet CH2, not sharp.
and triplet CH2. There is no doubt from these calculations that all chemically
2.5. A Transition-Metal Case, CrH66-. To study a bit more expected orbitals can be found in the various computational
complicated orbital ordering, where energy levels of different schemes (eH, KS, and HF). We can map all eH-like orbitals
types are still closer together, we investigated the orbital energies (which we want to assume as chemically meaningful; however,
of the simplest conceivable 18-electron ML6 transition-metal one can look at them also as an arbitrarily chosen reference
complex, the hypothetical d6 CrH66- (cf. Figure 7). For such a system) to orbitals in the KS (and HF) picture. The KS and HF
prototype octahedral organometallic complex, one would expect orbital energy positions relative to eH orbitals can be inter-
filled a1g, t1u, eg levels, localized mainly on the ligands, and a changed among themselves (e.g. the empty t1u and a1g orbitals
t2g below eg splitting of the metal 3d orbitals. The crystal field (cf. Figure 7); or by virtual type IV orbitals).
splitting should be substantial, with t2g (expected HOMO) below 2.6. A Second-Row Transition-Metal Complex, PdCl42-.
the unfilled eg LUMO. So far we have shown that KS orbitals are able to describe HF
All the methods give this general orbital pattern. A striking orbitals qualitatively and, after suitable scaling, also quantita-
difference here is that all the KS and HF levels are up (indeed tively (the HF and KS terms can be exchanged in this sentence).
at positive energy), while the EH levels are down in energy. We finish this paper by discussing the calculated energy levels
This is a consequence of the large negative charge on the of a representative second-row transition-metal complex, the
molecule. The extended Hückel method ignores electron repul- square-planar PdCl42-. The KS eigenvalues of PdCl42- (plotted
sion; therefore, its energy levels are low in energy. An isolated in Figure 8) show again, impressively, the compression and the
CrH66- is unrealistic; once the ion (even if it did exist) is ax + b shift of the KS orbitals relative to HF orbitals. This
surrounded by countercations, its levels will move down in systematic linear dependency can be nicely seen in Figure 9.
energy. We calculated the scaling parameters as aPdCl42-occ ) 0.27,
If we compare the virtual orbital levels (cf. Figure 7), we bPdCl42- ) 1.54 eV, and aPdCl42-virt ) 0.26. The value of
see that the ordering of the calculated virtual KS energy levels aPdCl42-virt differs significantly from that found for aH2Ovirt and
is identical with that of HF orbitals but is significantly different aN2virt, but the scaling parameter for the occupied orbitals,
from that of eH energies. This is illustrated dramatically by the
(52) According to the simple crystal field considerations and eH
(51) Cade, P. E.; Sales, K. D.; Wahl, A. C. J. Chem. Phys. 1966, 44, calculations. We want to be a little careful here and refrain from saying
1973. which order is “correct”.53
3420 J. Am. Chem. Soc., Vol. 121, No. 14, 1999 Stowasser and Hoffmann

aPdCl42-occ ) -0.27, is nicely close to -0.24 (H2O) and -0.28 to rationalize chemical phenomena, we have to identify the
(N2). It remains to be seen if there is meaning in this. What order, symmetry, and shape of KS orbitals. When we are dealing
orbitals would one expect for PdCl42-? Aside from core orbitals with unoccupied orbitals, we have to distinguish in some cases
of Pd and Cl (including the a1g + b1g + eu Cl 2s set) we would between interchanged virtual orbitals of type III and type IV.
expect lower lying Pd-Cl bonding orbitals of a1 + b1g + eu Once this is done, then we think one may apply KS orbitals in
symmetry, and Cl lone pairs spanning a1g + b2g + eu + a2u + a qualitative manner in MO arguments, in the way we are
b1g + eg + b2u + eu + a2g representations. The Pd levels should comfortable to doing with eH orbitals. Their number, symmetry
followsfor these we expect a characteristic 4 below 1 splitting
properties, and shape are just like those of the expected one-
of a square-planar complex. The LUMO should be the dx2-y2
electron orbitals. The situation is much like Baerends describess
orbital of b1g symmetry.
For the palladium complex, the energy ordering of the Pd- these seem to be the orbitals a qualitative, chemical analysis
Cl bonding orbitals a1, eu, and b1g (the lowest occupied orbital needs.13,14 Also, as a reviewer remarks, “since F(r) ) ∑i|φi,KS|2,
block in Figure 8) is in excellent agreement for all the methods. the shape and size of the KS orbitals is in case of a ‘good’ VKS
Some occupied orbitals of higher energy (Cl lone pairs and Pd (KS potential) more ‘physical’ than those of other single-particle
d orbitals) are interchanged in order (cf. Figure 8; for clarity, approximations”.
the symmetry of only the highest seven occupied orbitals is If we want to go a step beyond a qualitative interpretation
given). For instance the eH calculations lead to a HOMO of and look at orbital energies as rough ionization potentials, and
b2g symmetry, while the HF and DF calculations have an a2g if the DFT calculations are done with commonly used potentials,
HOMO. In contrast to the previously discussed CrH66- complex, then it appears we must take the absolute constant and linear
the virtual orbitals of type IV for the PdCl42- complex are nicely
orbital energy shift into account by applying a suitable ax + b
separated energetically from the virtual type III orbitals. With
scaling. Perhaps the situation will change with new functionals
all method we find consistently a LUMO of b2g symmetry. The
trends previously obtained for the main-group-element mol- and new methods.55
ecules thus hold up for PdCl42-.
Acknowledgment. We gratefully acknowledge financial
3. Summary support from the Deutsche Forschungsgemeinschaft for a
Let us return to the title question: What meaning is there to generous fellowship to R.S. and the Cornell University Theory
the KS orbitals and eigenvalues? If we wish to use KS orbitals Center for providing computational time. We also wish to thank
E.J. Baerends, W. Kohn, W. Glassey, E. Merschrod, and R.
(53) Cf. the case of the lowest excitation (t2g to eg calculated with eH)
in Cr(CO)6 which was recently reassigned in: Pollak, C.; Rosa, A.; Rytz for helpful discussions and comments. Also we want to
Baerends, E. J. J. Am. Chem. Soc. 1997, 119, 7324. thank Prof. Politzer for sending us preprints of his work.40,41
(54) Johnson, J. B.; Klemperer, W. G. J. Am. Chem. Soc. 1977, 99, 7132.
(55) Zhao, Q.; Morrison, R. C.; Parr, R. G. Phys. ReV. A 1994, 50, 2138. JA9826892