Volume 195, number 5,6 CHEMICAL PHYSICS LETTERS 3 1 July 1992

Application of the natural population analysis

to transition-metal complexes.
Should the empty metal p orbitals be included
in the valence space?
Feliu Maseras and Keiji Morokuma
institute for Molecular Science, Myodaiji, Okazaki 444, Japan

Received 6 April 1992; in final form 13 May 1992

The performance of the natural population analysis of Weinhold and co-workers is studied for transition-metal complexes on
the model systems NiHz and [ Os( PHj).,H( Hz) ] +. The calculations demonstrate that the addition of a set of formally empty
metal p orbitals to the “valence” space of NPA has a decisive effect on the final output of the population analysis. Nonetheless,
the question of whether such orbitals should be included or not in the NPA “valence” space remains unsolved; various arguments
in favor of each of the two options are discussed.

The natural population analysis (NPA) method is also starting to be applied to inorganic systems [ 5 1,
proposed by Weinhold and co-workers in the mid- a fact which prompted us to use the NPA method to
1980s [ 1,2] #’ is currently emerging as a powerful analyze the results of our own ab initio calculations
alternative to the extensively used Mulliken popu- on transition-metal complexes. When doing this, we
lation analysis [3] for the interpretation of the re- became concerned about the appropriateness of the
sults of quantum-chemical calculations. Certainly, principle used by the NPA method to label the atomic
the NPA method seems to be excellently positioned orbitals as valence or Rydberg.
for the formidable task of substituting the Mulliken In short, the method adopts the principle of taking
method for the assignment of electronic populations as reference the atomic ground state and assigning as
to atomic orbitals. On the one hand, it shares some “valence space” all the orbitals that are in fully or
of the most appealing features of the Mulliken partly occupied shells [ 61 in this electronic state, and
method, like the low computer cost and the general as “Rydberg space” all the orbitals that are in empty
applicability to virtually any wavefunction. On the shells. With this choice, one can assign without am-
other hand, it overcomes some of its most serious biguity valence and Rydberg orbitals for all the ele-
limitations, namely the basis set dependence and the ments. This approach, which is obviously correct for
appearance of physically meaningless results when main group and organic systems, becomes arguable
non-covalent bonds are considered. when transition-metal complexes are considered.
The success of the NPA method can easily be mea- This principle applied to transition-metal atoms leads
sured by monitoring its growing number of appli- to the exclusion of the formally empty set of p or-
cations in the recent literature [4]. In particular, it bitals from the valence space of the metal atom. This
means, for instance, that in the case of a first-row
Correspondence to: K. Morokuma, Institute for Molecular Sci- transition metal like nickel, the 3d and 4s orbitals
ence, Myodaiji, Okazaki 444, Japan. are in the valence space, but the 4p are not.
*’ Ref. [ 1 ] is the fundamental reference where the natural pop-
ulation analysis is completely developed. Other papers, essen-
The exclusion of this set of p orbitals from the va-
tially focused on the closely related natural bond orbitals analy- lence space is in clear contradiction with some of the
sis, are cited in ref. [ 21. current views on transition-metal chemistry. For in-

500 0009-2614/92/$ 05.00 0 1992 Elsevier Science Publishers B.V. All rights reserved.
Volume 195, number 5,6 CHEMICAL PHYSICS LETTERS 31 July 1992

stance, from the theoretical point of view it is well swer to our doubts. Rather, we are going to present
known that the role played by p orbitals when build- more questions than answers, as will be seen in what
ing a transition-metal basis set usually exceeds by far follows.
the small effect that should be expected from mere Example 1. NiH,. The first model molecule used
diffuse or polarization basis functions #*. From the in our tests is the linear symmetric NiH2 complex in
experimental point of view, it has even been sug- the singlet electronic state. The reason for this choice
gested that the inclusion of the metal p orbitals in the is that while the system is fairly simple and has a rel-
valence space is more important than that of the d atively small number of electrons, it retains the main
orbitals, since the role of the former in the interac- feature required for our study. This is the presence
tion with the ligands may be more important [ 71. of empty orbitals in a suitable orientation to interact
Finally, the well-known 18 electron rule [ 81, which with the ligands in each of the 3d, 4s and 4p shells.
applies fairly well to most transition-metal com- If the molecule is oriented along the z axis, these are
plexes, relies on the assumption that the metal pos- 3d,2, 4s and 4p,.
sesses a valence space composed of s, p and d orbitals. The NiHz system has been studied previously in
The inclusion or not of the metal p shell in the a thorough way from a theoretical point of view [ 9 1,
“valence” space in the NPA analysis is by no means and it so happens that the electronic state considered
a mere academic discussion with no practical effects. here is not the ground state of the system, though it
On the contrary, the initial division of the atomic plays some role in its chemistry. This is not however
orbitals between “valence” and “Rydberg” spaces is a question of concern, since the calculations are not
actually one of the cornerstones of the population aimed at understanding the properties of this par-
analysis formulated by Weinhold and co-workers. In ticular complex, but using it as a probe for the meth-
their algorithm, qualitatively speaking, the electron odology. In line with this reasoning, the starting level
population is at first distributed only among the “va- of calculation used to obtain the wavefunction is
lence” orbitals, and the “Rydberg” orbitals are in- rather modest: RHF with a valence double-zeta basis
troduced later for a small correction to the initial dis- set, which will be referred to as basis set I. For the
tribution. In other words, the initial labeling of an nickel atom, an effective core potential [ lo] is used
atomic orbital as “valence” or “Rydberg” affects de- to represent the ten innermost electrons, while the
cisively the way the electron population is distributed. valence electrons are represented with the associated
The ideas discussed in the previous paragraphs valence double-zeta basis set. For the hydrogens, the
raise the question of whether the formally unoccu- well-known 3-21G basis set [ 111 is used. The Ni-H
pied p shell of the transition-metal atoms should or distance was optimized at this level of methodology,
should not be included in its NPA “valence” space. and a value of 1.6 15 8, obtained.
The rest of this Letter presents the results of some Once the wavefunction had been obtained, the
calculations we have made in order to answer this natural population analysis was performed on it with
question. In particular, two different blocks of re- different classes of valence/Rydberg partitions. For
sults are presented. The first one, devoted to the NiH2 this purpose, we made a small modification to the
system, demonstrates how the initial definition of the original code of NPA subroutines attached to the
“valence” and “Rydberg” spaces actually affects the GAUSSIAN 90 package [ 121 .
output of the NPA method. The second block, which The results obtained are presented in table 1, in
presents results on the [Os(PH,),H(H2)]+ com- which the criterion used to define different parti-
plex, shows an attempt to apply the NPA method with tions is shown simply in placing on the left of the
different valence/Rydberg partitions to the system “/” sign the valence shells, and on its right the Ryd-
with which we were really working. Perhaps it should berg shells. Several possible partitions were consid-
be advanced here that we did not find any clear an- ered, even some with little chemical sense. Even a
superficial inspection of table 1 shows the decisive
1(2Throughout this paper, the shell will be understood as a set of effect of the initial labeling of atomic orbitals on the
orbitals that share the same two principal quantum numbers. That amount of electronic charge they are awarded at the
is, the shells for a given atom are Is, 2s, 2p, 3s, 3p, 3d, ... . end of the NPA procedure.

501
Volume 195, number 5,6 CHEMICAL PHYSICS LETTERS 31 July 1992

Table 1 passing that the Mulliken population is more af-

Atomic charges and populations on the metal orbitals involved
fected by the basis set than the NPA population.
in @and-metal bonding in the NiH2 complex, obtained through
the NPA method with different valence/Rydberg partitions us- Certainly, for each one of the orbitals individually
ing the basis set I. The Mulliken population and the formal charges considered, 3dZ2, 4s and 4p,, there are roughly two
are also included for comparative purposes different possible values of NPA electronic popula-
tion, depending on their initial assignment as “va-
Partition Charge on 3d,z 4s 4Pz lence” or “Rydberg” orbitals: a larger population
Ni H when one is assigned as “valence” and a smaller pop-
ulation when one is assigned as “Rydberg”. And
3d4s4p/ +1.10 -0.55 0.24 0.45 0.20 maybe more importantly, these fluctuations in the
3d4s/4p + 1.25 -0.63 0.24 0.46 0.04
population of the metal orbitals are directly echoed
3d/4s4p +I.65 -0.82 0.22 0.10 0.03
/3d4s4p + 1.74 -0.87 0.12 0.10 0.03 on its total charge, since the electrons won by the
Mulliken +0.33 -0.16 0.32 0.68 0.67 metal orbitals when they are introduced in the va-
formal +2.00 - 1.00 0.00 0.00 0.00 lence space do not come from an intra-atomic re-
distribution, but from the hydrogens. Therefore, ta-
Table 2
bles 1 and 2 confirm that the initial assignment of
Atomic charges and populations on the metal orbitals involved orbitals as “valence” or “Rydberg” is critical in the
in ligand-metal bonding in the NiH2 complex, obtained through NPA method, and that it has to be made very care-
the NPA method with different valence/Rydberg partitions us- fully in cases where it is not obvious, as in transition-
ing the basis set II. The Mulliken population and the formal
metal complexes.
charges are also included for comparative purposes
It could be argued that the difference between the
Partition Charge on 3d,z 4s 4Pz results of the partition proposed in the original for-
mulation of the NPA method (3d4s/4p, which will
Ni H be referred to from now on as “standard” #3) and
3d4s4p/ + 1.04 -0.52 0.21 0.48 0.27 the other “chemically reasonable” possibility
3d4s/4p + 1.26 -0.63 0.21 0.48 0.05 (3d4s4p/, to be referred to as “extended valence”)
3d/4s4p +1.63 -0.82 0.19 0.12 0.05 is relatively small. Certainly, a difference of popu-
/3d4s4p + 1.72 -0.86 0.12 0.12 0.05 lation in the 4p, orbital of 0.16- 0.22 electron is not
Mulliken +0.1s -0.09 0.32 0.78 0.74
0.00
spectacular, but we think that this is not the point
formal +2.00 - 1.00 0.00 0.00
under discussion. From our point of view, the prob-
lem is rather of a qualitative nature. When the 4p,
In order to assess the generality of this result, the orbital is labeled as a “Rydberg” orbital, it is a mere
same kind of study was repeated with a better basis spectator of the repartition of the electrons between
set, which will be labeled as basis set II. In the case the other orbitals of the system, and only gets a sym-
of the metal, it is the basis set initially defined by bolic amount of 0.04-0.05 electron. On the other
Wachters [ 131, the modifications proposed by Hay hand, when it is included in the “valence” space it
[ 141 having been included. The resulting set of becomes an active participant in the process, getting
primitives for the Ni atom is thus ( 14s9p5d), con- 0.20-0.27 electron, practically as many as the 3d,2,
tracted to [ 8s6p3d]. For the hydrogen, the double- orbital, which gets 0.24-0.2 1.
zeta basis set of Dunning [ 15 ] was used, augmented According to all the arguments and evidence that
with a p shell of polarization functions (expo- have been presented so far in this Letter, it could be
nent = 1.0). The results obtained through the use of concluded that the solution to all these problems
this all-electron basis set II are presented in table 2.
The obvious similarity between tables 1 and 2 con- g3The standard partition reported in the papers by Weinhold
and co-workers and implemented in the original source code of
firms that the orbital electronic population assigned
the program is certainly 3d4s/4p. But GAUSSIAN 90 users should
by the NPA method depends strongly on the initial be warned that the version we used initially had a bug that as-
labeling of the orbitals, a behavior that is not af- signed by default the incorrect 3d/4s4p partition to the nickel
fected by the choice of the basis set. We notice on atom.

502
Volume 195, number 5,6 CHEMICAL PHYSICS LETTERS 3 1 July 1992

consists simply in changing the default selection of function with this geometry and basis set.
the valence space for transition metals in order to The charges arising from the different population
include an additional shell of p orbitals. This would analysis are collected in table 3. The first striking
break to a certain extent the uniformity and aes- feature of this table is the large difference between
thetics of the NPA standard method of labeling based the formal charge and the result of any of the cal-
in the electronic ground state of the atoms, but it culations, but this can be interpreted as a demon-
could be adopted without too many problems. Un- stration of the intrinsic inaccuracy of these formal
fortunately, things are not so simple. assignments. A result which gives much more con-
Example 2. [O.s(PH,),H(H&+. Our attempts to cern is the fact that the charges coming from the ex-
apply the natural population analysis coupled with tended valence NPA analysis are by far the ones with
the extended valence partition scheme to the chem- least resemblance to the formal charges. In fact, we
ical systems on which we were working led to rather consider that they are out of the range of what could
unpleasant results. The meaning of the previous sen- be considered a “reasonable” value. Certainly, it is
tence will be clarified with an example. In particular, possible that a metal with a formal oxidation +2
we will present the atomic charges arising from dif- bears a slightly negative charge, but we are afraid that
ferent population analyses on the molecular hydro- a charge of -2 is completely unacceptable. Apart
gen complex [Os(PH,),H(HZ) I+, the simplified from the fact that this result should look suspicious
geometry of which is presented in scheme 1. for any theoretician, no experimental chemist would
The discussion of our conclusions on the chemical accept a method with this level of performance.
properties of the system [OS(PH~)~H(H~)]+ are Still, this disappointing result of the extended va-
completely out of the aim of this paper, and will in- lence NPA analysis was somehow to be expected. It
deed be presented thoroughly elsewhere. The only is the only possible outcome of a scenario where the
details pertinent here are that its geometry was op- standard NPA assigns already a small negative charge
timized at the MP2 level with a valence double-zeta on the metal, the introduction of the p shell in the
basis set [ 11,12,16] and that the atomic charges valence space of the metal has an important effect,
being presented were obtained from the RHF wave- and the only possible net effect is an increase of the
negative charge on the metal.
Therefore, the results obtained on the [Os-
H , H (PH,),H( H,) ] + system are by no means a special
feature of this complex, but the manifestation of a
wi much more essential problem that is likely to appear
in a systematic way if the extended valence NPA was
;/
to be applied as a general method. And it constitutes
PH3
A
PH,’
PH3 at the same time the demonstration of how dramatic
the differences between the standard and extended
valence partitions can become, making therefore the
il choice of one of the two schemes especially critical.
Discussion and conclusion. As it has already been
Scheme 1. advanced before, the results and discussions pre-

Table 3
Atomic charges in the [ Os( PH,)4H (H,) ] + system according to different population analysis

OS H- HZ PHg (aver.)

NPA standard (Sd6s/6p) -0.65 -0.16 +0.14 +0.42

NPA extended valence (Sd6s6p/ ) - 1.98 +0.04 +0.37 +0.64
Mulliken - 1.03 +0.10 +0.26 +0.42
formal +2.00 -1.00 0.00 0.00

503
Volume 195, number 5,6 CHEMICAL PHYSICS LETTERS 31 July 1992

sented in this Letter cannot lead to any definite con-

clusion. On one side, both qualitative considerations
and calculations on the NiHz system demonstrate the [ 1] A.E. Reed, R.B. Weinstock and F. Weinhold, J. Chem. Phys.
justification of extending the valence shell of tran- 83 (1985) 735.
sition metals by adding an extra set of p orbitals to [ 2 ] J.P. Foster and F. Weinhold, J. Am. Chem. Sot. 102 ( 1980)
7211;
that originally defined by the natural population A.E. Reed and F. Weinhold, J. Chem. Phys. 78 ( 1983) 4066;
analysis. On the other side, the results on the 83 (1985) 1736;
[Os(PH,),H(H,) 1’ complex show that the inclu- A.E. Reed, L.A. Curtiss and F. Weinhold, Chem. Rev. 88
sion of such a set of p orbitals in the metal valence (1988) 899.
space can produce a substantial worsening of the [3] R.S. Mulliken, J. Chem. Phys. 23 (1955) 1833,1841,2338,
2343.
atomic charges, leading them eventually to nonsens-
[4] A.E. Reed and P. von R. Schleyer, J. Am. Chem. Sot. 109
ical values. And the results as a whole show that this (1987) 7362;
is likely to be a general problem intrinsically asso- A.E. Reed, P. von R. Schleyer and R. Janoschek, J. Am.
ciated to the application of the natural population Chem. Sot. 113 (1991) 1885;
analysis to transition-metal complexes. In order to M.A. Al-Laham and K. Raghavachari, Chem. Phys. Letters
187 (1991) 13;
solve this problem, we are currently considering the M. Kaupp and P. von R. Schleyer, J. Am. Chem. Sot. I 14
possibility of modifying the original method, in par- ( 1992) 492.
ticular in the symmetry averaging step, but this is an [ 5 ] H.A. Harris, D.R. Kanis and L.F. Dahl, J. Am. Chem. Sot.
idea which at its present stage requires much further 113 (1991) 8602.
[6] P.J. Hay and W.R. Wadt, J. Chem. Phys. 82 (1985) 270;
development. K.D. Dobbs and W.J. Hehre, J. Comput. Chem. 8 ( 1987
Finally, we would like to remark that the objective 880.
of the presentation of this Letter is by no means to [7] M. Gerloch, Coord. Chem. Rev. 99 (1990) 117.
discourage application of the NPA method to tran- [ 81 CA. Tolman, Chem. Sot. Rev. 1 ( 1972) 337;
R. Hoffmann, Angew. Chem. Intern. Ed. Engl. 21 ( 1982 1
sition-metal systems. On the contrary, the purpose of
711.
exposing these problems is to encourage a further [9] M.R.A. Blomberg and P.E.M. Siegbahn, J. Chem. Phys. 78
study of its methodological details, which could (1983) 5682.
eventually lead to a refinement of the natural pop- [lo] P.J. Hay and W.R. Wadt, J. Chem. Phys. 82 (1985) 299.
[ 111 J.S. Binkley, J.A. Pople and W.J. Hehre, J. Am. Chem. Sot.
ulation analysis method able to provide for inor-
102 (1980) 939.
ganic systems the same high performance the present [ 12 ] M.J. Frisch, M. Head-Gordon, G.W. Trucks, J.B. Foresman,
formulation is already giving for organic systems. H.B. Schlegel, K. Raghavachari, M. Robb, J.S. Binkley, C.
Gonzalez, D.J. DeFrees, D.J. Fox, R.A. Whiteside, R. Seeger,
We thank Professor Frank Weinhold for very stim- C.F. Melius, J. Baker, R.L. Martin, L.R. Kahn, J.J.P. Stewart,
S. Topiol and J.A. Pople, GAUSSIAN 90 Program, Revision
ulating discussions. We also acknowledge Chizuru I (Gaussian, Inc., Pittsburgh, PA, 1990).
Muguruma for discussions. FM acknowledges a fel- [ 131 A.J.H. Wachters, J. Chem. Phys. 52 (1970) 1033.
lowship from the Spanish “Ministerio de Education [ 141 P.J. Hay, J. Chem. Phys. 66 (1977) 4377.
y Ciencia” under the “Subprograma de Perfeccion- [ 151 T.H. Dunning, J. Chem. Phys. 53 (1970) 2823.
[ 16 ] M.S. Gordon, J.S. Binkley, J.A. Pople, W.R. Pietro and W.J.
amiento para Doctores y Tecnblogos”. The present Hehre, J.Am. Chem. Sot. 104 ( 1982) 2797.
research is in part supported by the Grant-in-Aid for
Scientific Research on the Priority Area of “Theory
of Chemical Reactions”.

504