Bond Order Mayer06

Bond Order and Valence Indices: A Personal Account
I. MAYER
Chemical Research Center, Hungarian Academy of Sciences, H-1525 Budapest,
PO Box 17, Hungary
Received 10 April 2006; Accepted 8 May 2006
DOI 10.1002/jcc.20494
Published online 26 October 2006 in Wiley InterScience (www.interscience.wiley.com).
Abstract: The paper accounts for the authors activity in developing bond order and valence indices since the early 80s.
These indices represent an important conceptual link between the physical description of molecules as systems of electrons
and nuclei and the chemical picture of molecules consisting of atoms kept together by bonds. They are also useful for a
systematization and interpretation of the results obtained in the quantum chemical calculations, by permitting to extract
from the wave function different pieces of information that may be assigned chemical signicance. In some cases they
can have some predictive power, too. Historically, the prototypes of such indices were introduced in the semiempirical
quantum chemistry; the most important developments were Coulsons chargebond order matrix in the simple Hckel
theory and the Wiberg index in the CNDOframework. (Valence indices were also introduced in the semiempirical theory.)
The denition of the ab initio bond order index emerged from the asymptotic term of the exchange energy component of
the partitioning performed in the framework of the authors so-called chemical Hamiltonian approach using a mixed
second quantization formalism for overlapping basis sets. They can also be introduced by studying the exchange part of
the two-particle density (or of the second-order density matrix). Some properties of the bond order indices are discussed
and the authors (until now unpublished) proof is also presented, showing the sufcient conditions under which the bond
order index of a homonuclear diatomics is equal to the chemists bond order, i.e., the half of the difference between
the number of electrons occupying bonding and antibonding orbitals. The ab initio valence indices are also introduced
and discussed, and it is stressed that for correlated wave function the same exchange only denition of the bond
order and valence indices should be used, which was introduced for the SCF case. The recent concept of the atomic
decomposition of identity is also discussed and it is utilized for introducing bond orders and valences in the framework of
the 3D analysis, when atoms are dened not by their basis orbitals but as regions of the three-dimensional (3D) physical
space. Two versions of the 3D analysis are consideredthe AIM (atoms in molecules)-type decomposing the space into
disjunct atomic domains and the fuzzy atoms scheme in which there are no sharp boundaries between the atoms but
they exhibit a continuous transition from one to another.
2006 Wiley Periodicals, Inc. J Comput Chem 28: 204221, 2007
Key words: bond order indices; valence indices; fuzzy atoms; chemists bond order; non-orthogonal second
quantization; chemical Hamiltonian approach
Introduction
Chemists consider molecules as consisting of atoms; physicists treat
themas systems of electrons and nuclei. Undoubtedly, the ingenious
idea of Lewis, identifying chemical bonds with electron pairs shared
between the bonded atoms, still represents the fundamental link
between these completely different descriptions. It provides one a
well-established way of describing molecules, which is relatively
easy and straightforward, as far as a qualitative picture is concerned.
However, when turning to the quantitative theory, it appeared by far
not trivial to connect the results of ab initio calculations with the
genuine chemical concepts of atoms connected by single, double,
etc., bonds, and with the electron pairs forming these bonds.
The multiplicity of a chemical bond, called also bond order,
is a quantity of fundamental importance in practical chemistry.
Obviously, if one wishes to discuss molecules on both chemical
and quantum mechanical (quantum chemical) levels, then one has
tonda quantumchemical counterpart of this fundamental chemical
concept.
When one started to use the concept of molecular orbitals (MOs),
it became obvious that there are too much valence electrons to
assume that each pair of electrons occupying a two-center MO in
Correspondence to: I. Mayer; e-mail: mayer@chemres.hu
Contract/grant sponsor: Hungarian Scientic Research Fund; contract/grant
number: OTKA T43558
Contract/grant sponsor: The SpanishHungarian intergovernmental joint
project; contract/grant number: HH2004-0010Magyar-Spanyol TT
E18/2004
2006 Wiley Periodicals, Inc.
Bond Order and Valence Indices: A Personal Account 205
a diatomics corresponds to a chemical bondalthough obviously
each such MO bears a pair of electrons delocalized between the two
atoms. It became clear that one has to distinguish between bond-
ing and antibonding orbitals, and one arrived to the denition of
chemists bond order
B =
N
bond
N
antibond
2
(1)
where N
bond
and N
antibond
are the number of electrons occupying
bonding and antibonding orbitals respectively. For diatomics one
could study the character of the orbitals obtained in a calculation and
count the electrons on bonding and antibonding ones. Nonetheless,
denition (1) cannot be considered a quantumchemical quantity in a
narrow sense, as it is not directly calculated from the wave function
as would, say, an expectation value of an operator.
The rst quantity called bond order in quantum chemistry was
the off-diagonal matrix element of Coulsons chargebond order
matrix,
1
which was identied with the -component of the bond
order between two atoms of a conjugated (i.e., -electron) system.
This was an extremely useful quantity, characterizing very well
indeed the degree of -bonding between the centers involved. It
could be considered as the -electron bond order, because it reaches
its possible maximum value equal to one for ethylene, and gives
larger values for atoms with larger delocalization of -electrons
between them. (Excellent correlations can be observed between
Coulsons bond orders calculated at the simplest Hckel level of
the theory and the experimental C C distances.) In accord with
the formula dening Coulsons bond orders in terms of the orbital
coefcients C
i
,
D
= 2
occ.
i
C
i
C
i
, (2)
it can indeed be related to the degree to which the different -
orbitals have signicant simultaneous (and in phase) contributions
from the basis orbitals of both atoms in question; thus they may
be related to Lewiss shared electron concept. This is the case
despite the fact that the number of -electrons was considered a
sum of purely atomic contributions equal to the diagonal elements
of the chargebond order matrix; thus there is no -electron
charge shared between the atoms. (The basis orbitals were consid-
ered orthogonal in most simple -electron models.) The use of the
off-diagonal matrix elements D
as bond orders is, however, lim-

ited to models with one basis orbital per atom. Note, however, that
in the early literature the name bond order had been often applied
for the off-diagonal D
matrix elements in other cases, too.

Contrary to Coulsons bond order, Mullikens overlap popula-
tion
2
assigns a part of the electronic charge directly to the pair of
atoms considered. It characterizes the accumulation of the electrons
in the region between the chemically bonded atoms, and is a very
useful quantity often characterizing well the bond strength. How-
ever, it cannot be called bond order, because it does not represent
numbers that are close to one, two, and three for systems with sin-
gle, double, and triple bonds respectively. An important property
of Mullikens overlap population is that it possesses the correct
rotational-hybridizational invariance that one should require for
any quantity assigned a physical signicance. (The same holds for
Mullikens net and gross atomic populations; for an explicit proof,
see ref. 3.)
Wiberg
4
had observed that neither Coulsons bond order nor
Mullikens overlap population could be applied for Poples CNDO-
type all-valence-electrons semiempirical theory, which was in gen-
eral use at that time. The reason was that neither the individual
elements of the density matrix (the name of matrix D usually
applied for theories other than the -electron ones) nor any of their
simple combinations have the correct invariance properties when the
molecule is rotated as a whole, while Mullikens overlap population
simply vanishes because the basis orbitals are assumed orthonor-
malized. One sometimes considers the orthogonal basis orbitals of
the semiempirical theories as Lwdin-orthogonalized counterparts
of some original ones; then one may perform a deorthogonaliza-
tion, too, and get nonzero values of Mullikens overlap population.
Fortunately enough, that was not yet used in Wibergs time, so he
had to look for a new parameter. For that reason, he introduced a
new bond index (now bearing his name) which is quadratic in the
density-matrix elements and has the proper invariance:
W
AB
=
B
|D
|
2
. (3)
Moreover, it appears that for singlet states most of rst rowhomonu-
clear diatomics (but not for C
2
) (K. Jug, personal communication,
1985) the CNDO Wiberg index is equal to the ideal integer value
one usually assigns to the bond order in the given molecule.
This point had been investigated in detail by Borisova and
Semenov.
5
They gave a strict proof, according to which one can
derive the equality (1) for the CNDO wave functions by consider-
ing the hybrid atomic orbitals (AOs) of which the individual MOare
built up. Owing to the symmetry of the homonuclear diatomics, each
MO is the (normalized) sum or difference of the hybrids of the two
atoms, which are symmetry pairs of each other. Usually there are
pairs of bonding and antibonding MOs formed of the same hybrids,
and either both of them are occupied in the SCF wave function, or
only the bonding combination is occupied and the antibonding one is
empty. Borisova and Semenov proved that in the rst case these two
MOs (and the respective hybrids) do not contribute to the resulting
Wiberg index (Borisova and Semenov called it bond multiplicity
and dened in terms of spin-orbitals), while in the second case one
gets a contribution equal to unityand there are no contributions
originating from cross terms between different such pairs of MOs.
The rst result can simply be explained by utilizing the invariance of
determinant wave functions with respect to unitary transformations
of the occupied orbitals. In every case when the bonding and the anti-
bonding combinations of some atomic hybrids are both occupied,
one has to perform the unitary transformation leading to the sum
and difference of these MOs. In this manner one obtains a pair of
localized orbitals, each of which is fully concentrated on one atom.
These strictly atomic lone pair orbitals (hybrids) do not, of course,
contribute to the resulting Wiberg index, and the later becomes equal
to the number of doubly occupied bonding orbitals the antibonding
counterparts of which are empty.
5
This result is in full accord with
the original Lewis electron pair pictureelectrons fully localized
at one of the atoms need not be counted.
Journal of Computational Chemistry DOI 10.1002/jcc
206 I. Mayer Vol. 28, No. 1 Journal of Computational Chemistry
Valence is another important classical chemical conceptit
measures the ability of the atom to form chemical bonds in its
actual state, and gives the number of bonds (counted with their
multiplicities) formed by the atom in a closed-shell system. The
CNDO denition of valence index had been introduced indepen-
dently by Borisova and Semenov
5
and Armstrong et al.
6
with a few
months priority of the former authors. Unfortunately, Borisova and
Semenov published their very important papers in a journal with
a limited circulation, and so I became aware of them only thanks
to a favor by Dr. Obis Castaowho was their former graduate
studentafter I had already published my rst papers in the eld.
Valence can probably best introduced started from the observa-
tion made by Wiberg
4
in a footnote, according to which the quantity
that we shall denote b
= 2D
D
2
(4)
measures well the covalent bonding capacity of the basis AO
for the given wave function (density matrix D), because it reaches
its maximum, equal to one, for a unit electron population on the
given orbital (D
= 1), and falls to zero for both an empty orbital

(D
= 0) and a nonbonded pair (D
= 2). At the same time, it

follows from the idempotency property D
2
= 2D of the density
matrix that for closed-shell single determinant wave functions
b
(=)
D
2
(5)
i.e., b
is equal to the sum of all partial Wiberg indices between

orbital
and all the other orbitals in the molecule. Summing up

the parameters b
for all the orbitals of the atom, but extracting the

intraatomic partial Wiberg indices lacking a chemical signicance,
6
we arrive to the valence of atom A:
V
A
= 2
A
D
,A
D
2
. (6)
Alternatively, as it was done by Borisova and Semenov,
5
one can
turn to the natural hybrids for which the intraatomic block of the
matrix Dis diagonal, sumup the quantities b
in that basis, and then

return to the original one, and obtain the same result.
Borisova and Semenov
5
stressed the importance of the fact that
the valence index V
A
is determined by matrix elements referring
to the given atom only, i.e., by the actual valence state of the atom
in the given molecule.
It follows from the idempotency of the density matrix that for
closed-shell determinant wave functions the valence of an atom
equals to the sum of its Wiberg indices:
5, 6
V
A
=
A
(A=B)
W
AB
. (7)
As the Wiberg index measures bond multiplicity, this relationship
indeed corresponds very well to the chemical notion of valence.
While Armstrong et al.
6
considered closed-shell systems only
(They only made an important remark that in the open-shell case
the equality (7) does not hold and the difference of the two sides
should be a measure of the reactivity of the atom. This was the
line along which I have later introduced the free-valence index
F
A
), Borisova and Semenov
5, 7
used spin-orbitals, permitting them
to treat open-shell (UHF) systems, too. Their denitions would give
a correct value of
1
2
for the bond order in H
+
2
. However, in the two
papers
5, 7
they gave different expressions for the valence index of
open-shell systems in terms of the spatial orbitals. It is my opinion
that the rst is the correct one, because the use of the formula given
in the second paper
7
would give a value very close to 3 for the
valence of carbon atom in the methyl radical, which I do not think
is chemical: it is more correct to consider this carbon to be four-
valent, with one of its valences being actually free. I had a similar
objection to the denition used by Gopinathan and Jug;
8
also see
ref. 9. (Chemists usually indicate this free valence by putting a dot
to the radical center on the structural formula.)
The Chemical Hamiltonian Approach and the Deni-
tion of the An Initio Bond Order Index
In the late 1970s I got interested in the conceptual relation-
ships between the strict ab initio quantum chemical theory and
the genuine chemical concepts. It appeared to me that one of the
main difculties is due to the fact that
10
the BornOppenheimer
Hamiltonian does not reect the pronounced pairwise character
of interatomic interactions in the molecular system. In order to
discuss this problem, a special treatment has been introduced
10
in
which the main elements were a mixed second quantization for-
malism for treating the overlap problem and a special projection
technique permitting to get rid (at least in some sense) of the three-
and four-center integrals. This formalism got the name Chemical
Hamiltonian Approach (CHA). It is a typical case of the Hilbert
space analysis according to Halls terminology,
11
in which the atom
is identiedwiththe nucleus andthe basis orbitals centeredonit. The
alternative is the 3D analysis in which the atom is identied with
the nucleus and a part of the three-dimensional (3D) physical space
around the nucleus; we shall consider it under the Section Atomic
Resolution of Identity and 3D Analyses. (As atoms are not true
quantum mechanical observables, one has to chose how to dene
them in the quantum mechanical framework.)
The Mixed Second Quantization Formalism
In practice we use atom-centered basis sets, and then the quantum
mechanical problem of describing the molecular electronic struc-
ture is solely dened by the one- and two-electron integrals over the
basis AOs. The overlap of the basis orbitals centered on different
atoms reects chemically very important interactions, but makes
difcult the calculation of the matrix elements of different opera-
tors. (That caused ab initio VBschemes to be not competitive to the
HFR approach.) The mixed second quantization formalism permits
to work in a relatively simple way with wave functions built up of
nonorthogonal orbitals, by using essentially the same techniques
that one applies in the case of an orthonormalized basis set, with the
expense that the annihilation operators are dened in a special man-
ner, and do not coincide with the adjoints of the respective creation
operators.
In what follows we shall use superscripts + for the creation
operators and superscripts for the annihilation ones. However, if
the overlap matrix of the spin-orbital basis {
} is not a unit matrix,

i.e.,
= S
(8)
then the operator
=(
+
is only formally an annihilation oper-

ator, because its anticommutator with the creation operator
+
is
not
as it were the case for an orthonormalized basis, but

{
+
} =
+
= S
. (9)
The true (or effective) annihilation operators
, for which
fermion anticommutation rules apply, can be constructed by using
the biorthogonal set of spin-orbitals
S
1
(10)
where S
1
is a short-hand notation for the element (S

1
)
of
the inverse overlap matrix S
1
. As orbitals and creation operators
transform analogously, one can write

= (
+
S
1
. (11)
Then, as it is easy to see,
{
+
} =
(12)
which means that the annihilation operator
dened with respect

to the biorthogonal basis of spin-orbitals is that operator which
acts in the nonorthogonal case exactly in the same manner as the
usual annihilation operators do in the orthogonal case. Note that we
invoke the biorthogonal basis {
} only to dene the true annihi-

lation operators
corresponding to the creation operators

+
in
the nonorthogonal case, and no other reference to the biorthogonal
basis is necessary.
The Operator of Atomic Population
In order to express different operators in terms of operators
+
and
, one introduces an auxiliary Lwdin-orthogonalized set of

spin-orbitals
(S
1
2
)
(S
1
2
)
(13)
and the respective creation and annihilation operators

,
which can be presented as
(S
1
2
)
(S
1
2
)
. (14)
The operator of the number of electrons

N has the usual form in
the orthonormalized spin-orbital basis {
}:
N =
. (15)
By substituting here the relationships (14) we get by a single algebra
N =
. (16)
Grouping here the terms according to the atoms on which the spin-
orbitals
are centered, we may write
N =
N
A
(17)
where
N
A
=
A

+
(18)
is the operator of atomic population for atom A. In my previous
papers
10, 12
I had given two independent proofs (one for the sin-
gle determinant wave functions, anothersee Appendix Afor the
general case) for the expectation value of the operator string
+
= (PS)
(19)
where P is the density matrix in terms of spin-orbitals. That means

N
A
=
A
(PS)
= Q
A
(20)
i.e., the expectation value of the operator of atomic population is
Mullikens gross atomic population Q
A
on the atom in question.
This result means thatirrespective of its large basis dependence
or other possible disadvantagesMullikens gross atomic popula-
tion has a privileged importance: it is that atomic population which
is consistent with the internal mathematical structure of the theory
using atom-centered basis orbitals. One could group the terms by
the atoms in the expression (15), too, and get the Lwdin popula-
tions corresponding to the individual atoms. However, the Lwdin
populations refer to the Lwdin orbitals
, which are not strictly

atomic entities and vary in dependence of the chemical composition
and geometry of the systemstudied; in addition, it has recently been
shown
13, 14
that the Lwdin populations are not necessarily invari-
ant with respect to the rotation of the molecule as a whole, and
equivalent atoms may be assigned different Lwdin populations.
Decomposition of the Hamiltonian
In terms of the auxiliary basis {
} one has the known expression

of the Hamiltonian
H =
A<B
Z
A
Z
B
R
AB
+
h|
+
1
2
,,,
[
(21)
where

h is the one-electron part of the Hamiltonian
h =
1
2

A
Z
A
r
A
, (22)
all the integrals refer to spin-orbitals (i.e., the integrations include
summations over the spins) and the convention [12|12] is used for
the two-electron integrals.
Substituting the expressions (13) and (14) into (21), we obtain
the BornOppenheimer Hamiltonian in the mixed second quantized
form containing only integrals over the original spin-orbitals and
the creation and true annihilation operators corresponding them:
H =
A<B
Z
A
Z
B
R
AB
+
,,
S
1
h|
+
1
2
,,,,,
S
1
S
1
]
+
. (23)
The elements of the inverse overlap matrix present in the Hamil-
tonian (23) reect global effects in the molecule, and one needs
a further effort to separate out the true intraatomic and diatomic
interactions. For that reason a special projection technique had
been introduced.
10
As the simplest example, let us consider the
intraatomic part of the one-electron Hamiltonian corresponding to
atom A:
h
A
=
1
2

Z
A
r
A
. (24)
For every atom A we may write
h =

h
A
B
(B=A)
Z
B
r
B
(25)
thus the one-electron part of eq. (23) can be rewritten as
H
(1)
=
,
S
1
h
A
|
B
(B=A)
S
1
|
Z
B
r
B
. (26)
As A, the function

h
A
|
entering the integral in the rst term

is of intraatomic character: the atomic Hamiltonian

h
A
acts on a
basis orbital centered on the same atom. If the basis on A were
complete (or if |
were an exact eigenvector of

h
A
as is the case
for Schrdingers orbitals for a hydrogen atom), then the function
h
A
|
could be exactly expanded in the atomic basis. This is not

the case, in general, and the function

h
A
|
has components both

in the subspace of the basis orbitals assigned to atom A and in the
orthogonal complement tothat subspace. That fact maybe expressed
by introducing a resolution of identity as
I =

P
A
+ (1

P
A
) (27)
where
P
A
=
,A
|
S
1
(A)
| (28)
is the projection operator on the atomic basis. (In eq. (27) S
1
(A)
is
a short-hand for a matrix element of the inverse intraatomic overlap
matrix.) One may write
h
A
|
P
A
h
A
|
|(1

P
A
)
h
A
|
. (29)
Obviously,

P
A
h
A
|
is that component of the function

h
A
|
which
enters the problem of the free atom A and which is transferable to
any chemical environment in which atom A occurs. Therefore, it is
of meaning to introduce the approximation
h
A
|
P
A
h
A
|
,A
S
S
1
(A)
h|
(30)
and to consider
|(1

P
A
)
h
A
|
as a nite basis correction term.

(It is quite analogous to the terms causing the so-called basis set
superposition error in the theory of intermolecular interactions.
15
)
Note that eq. (30) becomes a strict equality if both , A.
Substituting the approximation (30), the rst sum on the right-
hand side of expression (26) becomes
,,A
S
1
(A)
h
A
|
(31)
an expression that contains only one-center quantities, and rep-
resents a sum of effective atomic one-electron Hamiltonians. (They
are effective, because expression (31) is written down in the
global many-atomic basis, and contains therefore the effective
annihilation operators
.)
Similarly, we may consider the function (Z
B
/r
B
)|
in the sec-
ond term of the expression (26) as a diatomic entity ( A; A = B),
which may be approximated by projecting it on the union of atomic
subspaces AB = A
B. Furthermore, one may perform analogous

manipulations with the two-electron function (1/r
12
)
(1)
(2)
entering the two-electron integrals in the Hamiltonian (23), by
introducing a projector for each electron. (One has to introduce
the projectors

P
A
(1)
P
A
(2) or

P
AB
(1)
P
AB
(2) depending on whether
orbitals
and
are centered on the same atom A or on different

atoms A and B.)
In this manner one gets the expression of the Hamiltonian in the
following form
H =
A
+
A<B
AB
+

H
n bas
(32)
where

H
A
and

H
AB
are effective atomic and diatomic operators
(These are effective operators because they contain the effective
annihilation operators
. As a consequence, they are not Hermi-

tian) and

H
n bas
collects the nite basis corrections of either atomic
or diatomic nature, which are connected with the remainders of the
projective integrals approximations. Obviously, they are expected
to be really negligible only if very large basis sets are used.
The effective atomic and diatomic Hamiltonians contain only
terms related to the given atom and diatomic fragment respectively
(The remainders of the projective approximations leading to the
effective atomic Hamiltonians

H
A
have components that can be
expanded in the basis sets of diatomic fragments. One could regroup
these terms from

H
n bas
to the effective diatomic operators

H
AB
, and
get an expansion in which the sum of atomic and diatomic Hamilto-
nians recovers the exact one for diatomic molecules. Terms of such
type were considered in the energy decompositions,
16, 17
but are of
no relevance as far as our present subject is concerned):
A
=
,,A
S
1
(A)
h
A
|
+
1
2
,,,,,A
S
1
(A)
S
1
(A)
[
]
+
(33)
AB
=
Z
A
Z
B
R
AB
,AB
S
1
(AB)
_
|
Z
B
r
B
|
|
Z
A
r
A
|
_
+
1
2
,,,AB
S
1
(AB)
S
1
(AB)
_
_
B
[
]
+
A
[
]
+
_
_
(34)
and we refer to ref. 10 for an explicit expression of

H
n bas
.
Hierarchy of Diatomic Interactions
The rst term of the diatomic Hamiltonian (34) describes the
internuclear repulsion; it need not be discussed. In the terms describ-
ing electron-nuclear attraction and interelectron repulsion, it is of
meaning to separate out those contributions that correspond to the
electrostatic (and exchange) interactions and those that are due to
differential overlap densities
(r)
(r) of the orbitals centered

on different atoms. If there were no differential overlap, then the
sums on the right-hand side of eq. (34) would not run on the whole
diatomic basis AB but only on one of the atoms A or B and the
intraatomic blocks of the inverse overlap matrix S
1
AB
would coin-
cide with the respective atomic inverse overlap matrix S
1
A
or S
1
B
.
One may write, therefore
AB
=

H
el stat
AB
+

H
overlap
AB
(35)
where
H
el stat
AB
=
,,A
S
1
(A)
|
Z
B
r
B
|
,,B
S
1
(B)
|
Z
A
r
A
|
+
1
2
,,A
,,B
S
1
(A)
S
1
(B)
[
]
+
+
1
2
,,B
,,A
S
1
(A)
S
1
(B)
[
]
+
+
Z
A
Z
B
R
AB
(36)
and

H
overlap
AB
is the difference between

H
AB
and

H
el stat
AB
. (We do not
need its explicit expression
10
here.)
Now, we consider the asymptotic behaviour of the integrals enter-
ing the electrostatic component (36) of the Hamiltonian for large
interatomic distances; it is easy to see that they behave asymp-
totically as the interactions of point charge(s) proportional to the
respective overlap integral(s):
|
Z
B
r
B
|
Z
B
R
AB
; [
]
S
R
AB
. (37)
By substituting the asymptotic expressions (37) into (36) and
performing the summations leading to Kronecker deltas, we get, by
using the denition of the operator of atomic population (18), the
Hamiltonian of electrostatic interactions in the point-charge approx-
imation, reecting the overall electrostatic balance of the molecule
(One gets the operator string as
+
, which obviously
equals
+
because
and
are centered on different

atoms; therefore = ):
H
point
AB
=
1
R
AB
(

N
A
Z
A

N
B
Z
A
+

N
A

N
B
+ Z
A
Z
B
)
=
1
R
AB
(
N
A
Z
A
)(
N
B
Z
B
). (38)
Expectation Value of

H
point
AB
and the Bond Order Index
Now, we calculate the expectation value of operator

H
point
AB
by using
the equality (20) for the expectation value of the atomic popu-
lation operator

N
A
. For calculating the expectation value of the
product

N
A

N
B
we need the expectation value of the operator string

+
=
+
. This can be obtained as a special case

of the general relationship

+
= (PS)
(PS)
(PS)
(PS)
(39)
valid for single determinant wave functions.
10
(Here matrices P and
S refer to the basis of spin-orbitals.) The proof of this relationship
it has not been published previouslyis outlined in Appendix B.
Using this result, we get

N
A

N
B
=
B

+
= Q
A
Q
B

B
(PS)
(PS)
.
(40)
As could be expected, the right-hand side contains a direct, or
Coulombic, termand a termof exchange type; note that the matrix
element (PS)
can differ from zero only if the spin-orbitals
and
are of the same spin.

Thus the expectation value of the operator

H
point
AB
is
10

H
point
AB
=
1
R
AB
_
_
q
A
q
B

B
(PS)
(PS)
_
_
(41)
it differs from the electrostatic interaction of the resulting atomic
charges q
A
= Z
A
Q
A
by a term, proportional to the exchange
component on the right-hand side of eq. (40)
B
(PS)
(PS)
. (42)
Returning home by train from the International Congress of
Quantum Chemistry held in June 1982 in Uppsala (I will be for-
ever grateful for the invitation to this congress I got from Professor
Per-Olov Lwdin; it had a decisive impact on my scientic carrier),
suddenly I realized that this quantity, if multiplied with a constant,
will give a value 1, 2, and 3 for the simplest systems with single, dou-
ble, and triple bonds. Then I also found that the newparameter has an
obvious analogy with the Wiberg index of the CNDO-type theories:
in the orthonormalized basis it simply reduces to the Wiberg index.
(There is a full analogy also in the sense that the CNDO energy
partitioning mentioned in the work of Fischer and Kollmar
18
also
exhibits an exchange energy component proportional to the Wiberg
index.
12
) Thus I had arrived at the denition of the ab initio bond
order index in terms of spin-orbitals:
B
AB
= 2
B
(PS)
(PS)
. (43)
In terms of the spatial orbitals, it can be rewritten as
B
AB
= 2
B
_
(P
S)
(P
S)
+ (P
S)
(P
S)
_
. (44)
Here P
and P
are the density matrices for the orbitals occupied

with spins and , respectively. They have the usual expression in
terms of the orbital coefcients of the occupied orbitals:
P
=
n
i=1
C
i
C
i
; P
=
n
i=1
C
i
C
i
. (45)
In the closed-shell case P
= P
and the denition (44) reduces

simply to
12
B
AB
=
B
(DS)
(DS)
(46)
where D = P
+ P
.
After publication of my paper,
12
I got a letter from Ms. Giambi-
agi (Rio de Janeiro) with a copy of their paper
19
in which essentially
the same denition (46) was proposed for the all valence electron
semiempirical theories with overlap (practically extended Hckel)
as a formal generalization of the Wiberg index. Their paper
probably because of the use of unusual and rather cumbersome
notations and too lapidary a presentationdid not receive the proper
attention in the literature.
In my rst paper about the subject,
12
I have suggested to use
denition (46) also to the closed-shell case, and only somewhat later
(roughly when I had read the papers of Borisova and Semenov,
5, 7
and probably not without being inuenced by them) I have realized,
that the general denition (43) expressed in terms of spin orbitals in
the open-shell case should be rewritten to spatial orbitals either as
eq. (44) or, introducing the spin-density matrix P
s
= P
, as
B
AB
=
B
_
(DS)
(DS)
+ (P
s
S)
(P
s
S)
_
(47)
and not as eq. (46), which is valid in the closed-shell case only (The
corrections (44) and (47) had been published
20
as an addendum
to my paper.
12
)
Some Properties of the Ab Initio Bond Order Index
A very important property of the bond order index dened earlier
is that it is invariant with respect to the most general rotational-
hybridizational transformations mixing the basis orbitals on the
individual atoms.
3
In fact, one may consider Mullikens overlap
population and the bond order index as the simplest (if not the only)
invariant quantities, representing linear and quadratic combinations
of the interatomic density matrix elements, respectively.
It follows from the equality (40) that the emergence of the bond
order index B
AB
in the case of single determinant wave functions
may directly be connected with the fact that the expectation value of
the operator product

N
A

N
B
differs fromthe product of the respective
expectation values:

N
A

N
B
=

N
A

N
B
= Q
A
Q
B
. In fact, one has
according to the equalities (40) and (43)
B
AB
= 2
_

N
A

N
B

N
A

N
B
_
. (48)
Soon after publication of my paper,
12
de Giambiagi et al.
21
rewrote the relationship (48) to the equivalent form
B
AB
= 2
__
N
A

N
A
_ _
N
B

N
B
__
(49)
which permits to give a statistical interpretation to the bond order
index: it measures the degree inwhichthe uctuations of the electron
populations on the two atomsi.e., their deviations from the mean
(expectation) valuesare correlated with each other: if the atoms
are connected with a covalent bond, then the decrease of the electron
density on the one atom involves the increase of it on the partner
atom, and vice versa.
It is to be noted that some authors use instead of the bond order
indexas denedineqs. (44) and(47) the Wibergindexcalculatedina
Lwdin-orthogonalized counterpart of the actual basis set. This pos-
sibility was already stressed in the fundamental papers of Borisova
andSemenov,
5, 7
andlater it was independentlyproposedbyNatiello
and Medrano.
22
It has the disadvantage that the different quantities
(including atomic populations) calculated in a Lwdin basis are
not necessarily rotationally invariantin particular, they are not if
the popular 6-31G** basis set is used.
13, 14
The invariance problem
can be excluded
13, 14
if one uses Davidsons version of Lwdin-
orthogonalization,
23
in which the orbitals on the individual atoms
are preorthogonalized, and one uses the Lwdin-orthogonalization
only for treating interatomic overlap. However, this scheme results
in completely different numbers than the conventional variant of
Lwdin-orthogonalization.
Another argument is connected with the delocalized and global
character of the Lwdin-orthogonalized basis functions, which may
be a source of unphysical effects. Thus, one can articially change
the bond order values by changing the position of a basis function
that is completely unoccupied in the given wave function,
24
but
overlaps with some of basis functions having an actual importance.
This indicates that well-pronounced local effects may be distorted
by the presence of some (nearly) empty basis orbitals centered in
other parts of the system.
Nonetheless, the use of a Lwdin-orthogonalized basis (prefer-
ably in the Davidsons version) may be the only possibility of a
Hilbert-space analysis in the cases when the basis contains diffuse
functions lacking any pronounced atomic character and therefore
Mullikens atomic populations and the related bond order indices
dened earlier become ill-behaved. However, for basis sets of rea-
sonably atomic nature the use of denitions eqs. (44) or (47) is
clearly preferable.
We may note here that there are cases in which one uses a plane
wave basis set in the calculations, and then performs a Hilbert-space
analysis by projecting the MO obtained on an auxiliary AO basis
set. Conceptually one can imagine a similar approach also in the
cases when the basis contains some off-centered orbitals, e.g., bond
functions.
Exchange Density and Bond Order
In the 1980s I had got
9, 25
a possibly deeper understanding of the
bond order indices by considering the normalization integral of the
exchange part of the second-order density matrix
2
(1, 2; 1
, 2
).
Later I have understood
3
that exactly the same considerations may
be accomplished by using a mathematically much simpler entity,
the exchange density. Exchange density is the diagonal part of the
exchange component of the second-order density matrix, which one
obtains if the primed and unprimed quantities are set equal (or Fermi
hole).
The electron density (r) gives the probability density of nding
an electron around the point r; it can be calculated as the expecta-
tion value of the operator (r) =

N
i=1
(r
i
r). Analogously, the
two-particle density
2
(r
1
, r
2
) gives the probability density of nd-
ing one electron around the point r
1
and, simultaneously, another
electron around the point r
2
. It represents the expectation value of
the operator
3

2
(r
1
, r
2
) =
i,j
(i=j)
(r
i
r
1
)(r
j
r
2
) (50)
2
(r
1
, r
2
) differs fromthe product of (r
1
)(r
2
) because of the anti-
symmetry of the wave function, and of electron correlation, if the
latter is also taken into account. For single determinant wave func-
tion only the antisymmetry (exchange) plays a role and we may
dene the exchange density
x
2
(r
1
, r
2
) through the relationship
2
(r
1
, r
2
) = (r
1
)(r
2
)
x
2
(r
1
, r
2
). (51)
Assuming that we are using a single determinant wave function
built up of n
orbitals a
i
(r) lled with spin and n
orbitals b
i
(r)
lled with spin , then one can easily calculate the expectation
value of operator
2
(r
1
, r
2
) by using the general formulae for matrix
elements of two-electron operators, and obtain after some simple
manipulations
3
the expression for
x
2
(r
1
, r
2
) as
x
2
(r
1
, r
2
) =
n
i,j=1
a
i
(r
1
)a
j
(r
1
)a
j
(r
2
)a
i
(r
2
)
+
n
i,j=1
b
i
(r
1
)b
j
(r
1
)b
j
(r
2
)b
i
(r
2
). (52)
Integrating
x
2
(r
1
, r
2
) over both variables, we have
__

x
2
(r
1
, r
2
) dv
1
dv
2
=
n
i,j=1
a
i
|a
j
a
j
|a
i
+
n
i,j=1
b
i
|b
j
b
j
|b
i
=
n
i,j=1
ij
+
n
i,j=1
ij
= n
+ n
= N. (53)
By substituting here the LCAO expansion of the orbitals and
performing trivial manipulations, we get
3
m
,=1
[(P
S)
(P
S)
+ (P
S)
(P
S)
] = N. (54)
We can group the terms on the left-hand side according to the atoms
on which the basis orbitals are centered:
A,B
B
[(P
S)
(P
S)
+ (P
S)
(P
S)
] = N. (55)
Comparing this expression with the denition in eq. (44), we see that
the bond order B
AB
between atoms A and B is the diatomic contribu-
tion to the integral of the exchange density
x
2
(r
1
, r
2
). (The factor 2
present in the denition in eq. (44) is also recovered, because the sum
for Aand Bin expansion (55) runs over all the atoms independently.)
We may note that eq. (54) could formally be obtained also as
a trivial consequence of the idempotency of the density matrices
(P
S)
2
= P
S valid for the single determinant wave functions and

of the property Tr(P
S) = n
. Based on this, it was also pos-

sible to introduce the product of three (and even more) density
matrices and dene
26
some genuinely three-center (or many-center)
bond order indices that may be used to identify true three-center
(many-center) chemical bondslike the two-electron three-center
bonds in diborane molecule. (Note that the existence of a three-
center bond manifests also by the appearance of some conventional
two-center bond orderand the respective attractive exchange
interactionbetween the two external atoms, even if they are too
far apart to have any direct interactions with each other.
27
This effect
may be important for the stability of diborane molecule and similar
systems.)
Bond Orders in Homonuclear Diatomics
The derivation of the relationship (1) for homonuclear diatomics,
given by Borisova and Semenov for an orthogonal basis, could
not be generalized for the overlapping case, although the practice
indicates that this relationship does hold to a good accuracy pro-
vided that a minimal basis set is applied in the ab initio calculations
(c.f. Table 1). Therefore I have looked for another treatment.
28
Table 1. Bond Orders of Singlet Homonuclear Diatomics Calculated by
STO-6G Basis Set at the Equilibrium Bond Distances.
Molecule Bond order
H
2
1.0000
Li
2
0.9980
Be
2
1.9987
B
2
2.9994
C
2
3.3328
N
2
3.0000
O
2
2.0000
F
2
1.0000
We should utilize the invariance of the bond order indices with
respect to the rotationalhybridizational transformations of the basis
orbitals and Lwdins pairing theorem
29
originally proved by
Amos and Hall
30
(also see ref. 3). This means that we replace the
original bases of spatial orbitals {
a
(r)} and {
b
(r)} on the two
atoms by orthonormalized ones {
a
(r)} and {
b
(r)}, where sub-
scripts a and b indicate that the orbitals in question are localized
on atoms A and B respectively, and subject the latter to unitary
transformation, providing the new basis orbitals to be also paired
a
|
b
= s
. (56)
These transformations do not change the subspace of the orbitals
assigned to the individual atoms, and so should leave invariant every
physically meaningful quantity. (In what follows we shall omit the
primes for the sake of simplicity.)
Relationship (56) means that a given basis orbital has a nonzero
overlap at most with one orbital of the other atom. (In this con-
struct the pairs of orbitals that have nonzero overlap are often
called corresponding orbitals.) It follows from the symmetry of
the homonuclear diatomics that the pairs of corresponding orbitals
transform into each other under the interchange of the atoms (or
may be selected so if there are degenerate s
values). We construct
the normalized sums and differences of the pairs of corresponding
orbitals as
= [2(1 + s
)]
1
2
(
a
+
b
)
(57)
= [2(1 s
)]
1
2
(
a

b
).
Here
b
and
a
are the -th bonding and antibonding orbitals, con-

structed of the pair of corresponding orbitals
a
and
b
. (It is
supposed that the pairing of the orbitals is performed by an algo-
rithm
3
inwhichthe phases of the orbitals are selectedsoas toprovide
the overlap integrals s
to be nonnegative real numbers.)

Now we consider single determinant wave functions built up
of these orbitals lled with the respective spins = or . We
assume that the order of the basis orbitals is selected as
a1
,
b1
,
a2
,
b2
, . . . ; then both the overlap matrix S and the density matrices P
are block-diagonal with 2 by 2 nonzero blocks on the main diagonal.

As a consequence, the bond order index (44) will represent a sum
of the contributions originating from the individual pairs of basis
orbitals
a
,
b
. We may study them independently of each other,
for which it is sufcient to consider the respective 2 by 2 blocks of
the matrices S and P
. We shall denote these blocks as S
and P
,
respectively.
Now, let us rst consider the case when the bonding orbital
b
is occupied with spin in the wave function, but its antibonding

counterpart is not. In that case, as it is easy to see,
P
=
1
2(1 + s
)
_
1 1
1 1
_
(58)
and the matrix product P
is
P
=
1
2(1 + s
)
_
1 1
1 1
__
1 s
1
_
=
1
2
_
1 1
1 1
_
. (59)
It follows from this result that (P
)
12
(P
)
21
=
1
4
, and
by substituting into the denition (44) we see that the orbital
b
contributes
1
2
to the bond order B
AB
if it is occupied once. (There
is a common factor of 2 in the equation.) If this bonding orbital is
doubly occupied, its contribution to the bond order index is unity:
we get a contribution of
1
2
for both spins = and .
One gets exactly the same result if the antibonding orbital
a
is singly or doubly occupied (but

b
is empty). Although for the

ground state of a homonuclear diatomics that can hardly be the
case, this fact indicates that the bond order indices (unlike the energy
components
16, 17, 31
) not always are able to distinguish between the
bonding and antibonding situations.
If both orbitals
b
and
a
are occupied with the spin , then we

get for matrix P
=
1
1 s
2
_
1 s
1
_
(60)
and for P
=
_
1 0
0 1
_
. (61)
Therefore, if both the bonding and antibonding combinations of a
pair of corresponding orbitals are occupied with a given spin, then
there is a complete cancellation and the given pair of corresponding
orbitals does not contribute to the bond order index B
AB
.
These results indicate that in the homonuclear diatomics the
sufcient condition of getting a half-integer or integer value for
the bond order index is that the occupied orbitals coincide with the
bonding and antibonding combinations of corresponding orbitals
obtained in the pairing procedure. One may formally reduce to this
case also all wave functions in which there is only a single occupied
orbital in every symmetry species. In that case we may omit from
the basis all the hybrid orbitals forming the virtual MOs; as they are
empty, their omission changes neither the wave function nor any
parameter computed from it. Then we may perform the pairing of
the remaining basis orbitals and ensure the required block-diagonal
character of the matrices.
The conditions of the derivation were exactly fullledeven if
a minimal basis is usedonly if one could neglect completely the
interaction (overlap) of the core orbitalsactually the 1s orbitals
(K shell) for the rst row atomswith all the orbitals of the part-
ner atom. In that case the 1s AO would not be mixed with any other
orbitals in the pairing procedure, and their sumand difference would
coincide with the 1
g
and 1
u
canonic MOs. After the core orbitals
are separated out, the remaining part of the minimal basis is so small
that the sufcient conditions discussed earlier are usually satised
because of the symmetry considerations. (This is not the case, how-
ever, for the singlet C
2
molecule (K. Jug, personal communication,
1985)). As the atomic number increases in the series from Li
2
to F
2
,
the 1s orbitals become relatively more compact and more separated
from the valence shells, and thus the bond orders deviate less from
the ideal integer values.
For larger basis sets the conditions of the derivation are not ful-
lled and the bond order indices differ from the respective classical
valuesbut not too much. Usually there is a signicant interac-
tion only between strongly overlapping basis orbitals, and for the
homonuclear diatomics this will lead to the consequence that the
hybrids building up the SCF orbitals do not signicantly deviate
from those that one would get in the pairing procedure. (The most
important factor is the similarity of the hybrids making up the pairs
of bonding and antibonding orbitals corresponding to each other.
This requirement is not fullled for the C
2
molecule even in a min-
imal basis set, and its bond order is far from an integer.) These
qualitative considerations hold also for heteronuclear diatomics and
for molecules containing more than two atoms. This means that the
bond orders will not deviate too much from their classical values
although for such systems one does not get strict integers even in
minimal basis sets, as the bond orders are inuenced by different
bond polarity and delocalization effects, too.
It may be of interest to note that for the homonuclear diatomics
treated at the minimal basis levelassuming that the core orbitals
may be considered fully separatedone obtains exactly the same
bond order values also by turning to the Lwdin-orthogonalized
basis and calculating the Wiberg indices for it. (This is not strictly
true in any other case.) The explanation is as follows. We start from
an orthonormalized and paired basis on the two atoms; then Lwdin
orthogonalization will mix only pairs of the corresponding orbitals,
owing to the block-diagonal character of the overlap matrix. Then
it is enough to observe that due to symmetry reasons the bonding
and antibonding orbitals coincide with the bonding and antibonding
combinations of the Lwdin-orthogonalized AOs, and the Wiberg
indices are also equal to the ideal bond multiplicity values, as it
was proved in the work of Borisova and Semenov
5
discussed in
the introduction. It may be of interest to observe the following.
If the given occupied orbital and its antibonding counterpart are
both occupied, then one may replace the orbitals
b
and
a
by their
normalized sumand difference (1/
2)(
b
), without changing
the determinant wave function. Now, owing to the difference of the
normalization factors in eq. (57), this sum and difference do not
recover the individual AO-s
a
and
b
, butas it is discussed in
ref. 3, give their Lwdin-orthogonalized counterparts. Thus, the
wave function in which both
b
and
b
are occupied is equivalent

to a wave function in which there are two strictly local orbitals in
terms of the Lwdin basis, which do not contribute to the Wiberg
index, of course.
The Ab Initio Valence Indices
We have discussed in the introduction the quantity b
dened in
the spirit of Wibergs footnote.
4
In the ab initio case we have to
take into account overlap and should dene it through Mullikens
gross orbital populations (DS)
instead of simply D
used in
eq. (4). A disadvantage of Mullikens gross orbital populations is
that they are not strictly limited to the interval between 0 and 2. As
a consequence, it is possible, at least in principle, that one gets a
negative b
value for some basis orbitals. Thus, we have

b
= 2(DS)
(DS)
2
. (62)
In order to get valence, we shall sum the quantities b
for the given

atom and extract from the sum the partial bond orders between the
orbitals within the given atom. In the closed-shell case this means
that we have to subtract
,A
(=)
(DS)
(DS)
(63)
and arrive to the denition
V
A
= 2
A
(DS)
,A
(DS)
DS)
. (64)
Alternatively, one may turn to the (generalized) natural hybrids
32
in
which the intraatomic block of the matrix DS is diagonal, sum up
the quantities b
in that basis and then return to the original one,

and obtain the same result.
It follows fromthe idempotency properties of the matrix DS that
in the closed-shell SCF case we have the relationship
V
A
=
B
(B=A)
B
AB
(65)
quite similar to that which was valid in the semiempirical case.
Following the scheme of Armstrong et al. used in the CNDO
case,
6
we accept the denition(64) alsofor the open-shell case. Then
equality (65) does not hold and we may dene the free-valence index
F
A
as the difference (The denition (66) gives the free valence in
terms of the actual wave function and does not refer to any external
quantity. Therefore, it conceptually differs from the denitions of
the quantity also called free valence either in the old Hckel theory
or by Gopinathan and Jug
8
in a semiempirical all-valence electrons
theory. In those cases free valence measures the deviation of the
actual sum of bond orders from some theoretical (maximal or ideal)
value)
F
A
= V
A

B
(B=A)
B
AB
. (66)
In the RHF case the free-valence index of all atoms vanishes, F
A
=
0, while in the UHF one it can be expressed via the spin-density
matrix P
s
as
F
A
=
,A
(P
s
S)
P
s
S)
(67)
(For an explicit derivation, see ref. 3.).
In light of this expansion, the square root of the free-valence
index may be in some sense considered as the number of spins
on the given atom. (Note that the sum of the free valences for all
the atoms is equal to 1 only in the simplest casese.g., one electron
delocalized along a regular polygonbut usually exceeds 1 because
the spin polarization phenomenon.)
Correlated Wave Functions
All the above-mentioned considerations were related to the SCF
wave functions. In the DFT case one can apply them to the single
determinant built up of the KohnSham orbitals (see e.g., ref. 33)
although, strictly speaking, the latter are not attributed any denite
physical meaning. It is an important question, how one should gen-
eralize the bond order and valence indices for the case in which
electron correlation is taken into account explicitly.
There are two conceptually different approaches to the denition
of bond order indices in the correlated case. They are based on two
different expressions, which, however, give coinciding results in the
single determinant case. The rst starts from the expression (49) of
the correlation between the uctuations of the atomic populations,
anduses it as a general denitionof the bondorder. That is equivalent
of using the whole difference
xc
2
(r
1
, r
2
) =
2
(r
1
, r
2
) (r
1
)(r
2
) (68)
between the actual pair density
2
(r
1
, r
2
) and the product (r
1
)(r
2
)
of one-electron densities, and decomposing its integral according to
the different atoms. The notation
xc
2
in eq. (68) indicates that it
reects both exchange (antisymmetry) and the correlation effects
in other words it includes both the Fermi-hole, which is due to the
antisymmetry requirement, and the Coulomb-hole due to electron
correlation. This denitionis seeminglyveryattractive, andnodoubt
the quantity calculated in this manner may be of some interest.
However, this quantity has serious drawbacks if used for dening
the bond order, and I cannot recommend its use.
In my opinion, the most important argument against using a
bond order denition based on the relationships (49) or (68) is the
fact
25
that such a denition gives the value 0.39 for Weinbaums
classical wave function for H
2
. This wave function represents the
solution of the full CI problem for the minimal basis of Slater-
type orbitals with optimized exponents. It accounts for 84% of
the whole binding energy of the H
2
molecule, which is pretty fair,
especially if one takes into account that the free hydrogen atoms are
described exactly in this basis. One simply must not call bond order
a quantity that is only 0.39 for the prototype single chemical bond in
the H
2
molecule, described with Weinbaums prototype correlated
wave function. That is simply not chemical. I think that this is the
case, even if other correlated calculations of H
2
give results
34
in
which the deviation from unity is much less dramatic. The same
conclusion can be drawn from the results presented in ref. 35 in the
3D AIM (atoms in molecules) framework: bond indices exceeding
5 have been obtained for N
2
and F
2
by using the denition including
the Coulomb hole, while the exchange only denition discussed
below produced chemically reasonable numbers. It appears that the
authors of ref. 35 misunderstood the message of my paper,
25
and
assumed that I am proposing the use of the relationship (49) and
have overlooked that I had proved there its inadequacy. Essentially
the exchange only scheme has been rediscovered also in ref. 36.
In one of my papers
25
I had proposed to connect bond order
index with only the exchange part of the second-order density matrix
which is formally constructed from the rst-order density matrix
exactly in the same manner as in the single determinant case. The
same can be done for the simpler entity, the exchange density, used in
Figure 1. C C and C H bond orders, and total and free valences of carbon for the dissociation of the
ethylene molecule into two triplet methylenes, treated at the (4,6) CAS level of theory by using 6-31G**
basis set.
the present paper. One has simply to observe that the expansion (52)
is nothing else than a particular case of a more general expression
x
2
(r
1
, r
2
) =
i,j=1
n
i
n
j
a
i
(r
1
)a
j
(r
1
)a
j
(r
2
)a
i
(r
2
)
+
i,j=1
n
i
n
j
b
i
(r
1
)b
j
(r
1
)b
j
(r
2
)b
i
(r
2
) (69)
where a
i
and b
i
are the natural spin-orbitals of spins and respec-
tively. (One obtains eq. (52) from eq. (69) by observing that in the
single determinant case the occupation numbers n
i
( = or )
are equal to either 1 or 0.)
The use of the exchange density eq. (69) means that one has to
use exactly the same eqs. (44) or (47) and (64), (66) for dening
bond orders, and total and free valences in the correlated case, too.
Thus there is no need to work with the second-order density matrix,
as were the case if we used the uctuation type denition, and so
the calculation of bond orders and valences may be accomplished
in the correlated case exactly in the same manner in terms of the
rst-order matrix alone, as one does for single determinant wave
functions. The only difference is that the equality (65) does not hold
any more even for closed-shell molecules, and thus F
A
= 0 in the
correlated case.
It is my opinion that this way of dening the bond order and
valence indices for the correlated case is very chemical, and is
applicable not only near the equilibrium distances, but is able to
describe the whole process of bond formation/dissociation. This is
illustrated well in Figure 1, which displays some results for the ethy-
lene molecule dissociating into two triplet methylenes, as calculated
with a (4,6) CAS wave function by using 6-31G** basis set. One
may see that the C C bond order that is nearly 2 at the equilibrium
distance gradually decreases and tends to 0 at the large distances
as it should. Simultaneously with this, there appears a free valence
on the carbon, tending to a limit close to 2 at the large distances, in
agreement with the fact that there are two unpaired electrons in the
triplet methylene. The sum of the C C bond order and of the car-
bon free valence is almost constant, thus the carbon atom remains
practically four-valent during the whole dissociation. (Note that the
ground state of methylene is the triplet; contrary to the CAS scheme,
the RHF method is only able to describe dissociation of ethylene
into two singlet methylenes and results in divalent carbons with no
free valences.) (The C Hbond order stays nearly constant at a value
close to 1.)
Atomic Resolution of Identity and 3D Analyses
Atomic Resolution of Identity
As already noted, the concept of atoms within a molecule is not a
well cut one. Strictly speaking, the atoms do not directly appear in
the quantum mechanical description of molecules: the Schrdinger
equation is written down for the individual particles (electrons
and nuclei). Thus one has to introduce some denition of the
atom from outside, which necessarily means some arbitrariness.
The Hilbert space analysis considered earlier is based on the fact
that practical quantum chemistry mostly uses atom-centered basis
sets. However, neither the use of basis atom-centered basis sets nor
the Hilbert-space approach to the analysis is obligatory. A widely
used alternative is the 3D analysis in which the physical space is
decomposed into regions attributed to the individual atoms.
Most recently
36
we have proposed the atomic decomposition
of identity, permitting to treat different denitions of the AIM on
equal footing, i.e., in a framework of a common formalism. It is
applicable for analyzing different physical quantities in terms of
contributions coming from the individual atoms or pairs of atoms,
including population analysis, bond orders, and energy partitioning
schemes. In this approach one presents the identity operator

I as
a sum of some operators
A
corresponding to the individual atoms
according to the scheme of analysis selected:
I =
A

A
. (70)
Note that the individual operators
A
are not necessarily Hermitian.
The use of this approach is very simple. For instance, the total
number of electrons may be expressed in terms of the natural spin-
orbitals a
i
and b
i
having occupation numbers n
i
and n
i
introduced
earlier as
N =
_
(r) dv =
i
n
i
a
i
|a
i
+
i
n
i
b
i
|b
i
. (71)
By inserting an atomic resolution of identity to each ket, one can
present the number of electrons as a sum over the atomic electron
populations
N =
A
Q
A
(72)
where
Q
A
=
i
n
i
a
i
|
A
|a
i
+
i
n
i
b
i
|
A
|b
i
. (73)
Alternatively, by using the LCAO expansion of the rst-order
density matrix, one can write
N =
(DS)
,
D
(74)
where, as in Section Expectation Value of

H
point
AB
and the Bond Order
Matrix, D = P
+ P
. By substituting the atomic resolution of

identity (70) to the ket in each S
, and introducing the

atomic overlap matrices S
A
with the elements
S
A
|
A
|
(75)
we get the expansion (72) with the denition of the gross atomic
populations
Q
A
=
A
(DS
A
)
S
A
. (76)
It is trivial to check that the denitions (73) and (76) lead to identical
results.
By introducing two atomic resolutions of identity into each inte-
gral in the expansion (71), one to the bra and one to the ket, we
can get the decomposition of the total electron charge into the sum
of net atomic populations q
AA
and overlap populations q
AB
N =
A
q
AA
+
A,B
(A=B)
q
AB
(77)
with
q
AA
=
i
n
i
a
i
|
A

A
|a
i
+
i
n
i
b
i
|
A

A
|b
i
(78)
and
q
AB
=
i
n
i
a
i
|
A

B
|a
i
+
i
n
i
b
i
|
A

B
|b
i
(79)
respectively. It is easy to see that the equality
Q
A
= q
AA
+
B
(A=B)
q
AB
(80)
holds.
By integrating the two sides of the expression (69) inserting one
atomic decomposition of identity to each ket we may write
__

x
2
(r
1
, r
2
) dv
1
dv
2
=
n
i,j=1
n
i
n
j
a
i
|a
j
a
j
|a
i
+
n
i,j=1
n
i
n
j
b
i
|b
j
b
j
|b
i
A,B
_
_
n
i,j=1
n
i
n
j
a
i
|
A
|a
j
a
j
|
B
|a
i
+
n
i,j=1
n
i
n
j
b
i
|
A
|b
j
b
j
|
B
|b
i
_
_
. (81)
Thus, by picking up the contribution to this sum, corresponding
to a given pair of atoms Aand B, we get the most general expression
of the bond order as
B
AB
= 2
_
_
n
i,j=1
n
i
n
j
a
i
|
A
|a
j
a
j
|
B
|a
i
+
n
i,j=1
n
i
n
j
b
i
|
A
|b
j
b
j
|
B
|b
i
_
_
(82)
where the factor 2 is introduced because in eq. (81) the sum over
atoms Aand Bruns independently and all terms are symmetric with
respect to the interchange of A and B. Substituting here the LCAO
expansion of the natural orbitals, this can simply be rewritten as
B
AB
= 2
,
_
P
S
A
S
B
)
+ (P
S
A
)
(P
S
B
)
_
. (83)
Analogously, the general denition of the total valence can be
given as
V
A
= 2
DS
A
)
,
(DS
A
)
(DS
A
)
(84)
and eq. (66) applies for the free valence.
Atomic Operators: Hilbert-Space Analysis
The Hilbert-space analysis discussed in the previous sections may
be recovered if one denes the atomic operators as

A
=
A
|
| (85)
where |
is the biorthogonal counterpart of the basis orbital |
,
introduced in eq. (10). Then, according to the denition (75) the
elements of the atomic overlap matrix become
S
A
|
A
|
=
_
S
if A
0 otherwise
. (86)
This result means that the atomic overlap matrix consist of the
intraatomic columns of the original overlap matrix, while all the
other columns are zeroed. (Thus the atomic overlapmatrices trivially
sumto the original one, as they should.) It is easy to see that by using
the atomic overlap matrix given by the equality (86) we recover
immediately Mullikens gross, net, and overlap populations from
the general expressions (76), (78), and (79), respectively, and the
expressions (83) and (84) of the bond order and valence also become
identical with the denitions in eqs. (44) and (64) respectively.
Atomic Operators: 3D Analysis
In the 3D analysis one introduces a weight function w
A
(r) for each
atom and every point r of the 3D space, which should satisfy the
conditions
w
A
(r) 0;
A
w
A
(r) = 1 (87)
everywhere. Then the atomic operator
A
may be dened as

A
= w
A
(r
)|
r
=r
. (88)
Here the notation r

= r indicates that one should replace r

by
r after the action of all the operators on the wave functions (as
functions of r ) had been evaluated, but before the integration over r
is carried out. Thus, quantum mechanical operators act only on the
electronic wave functions but not on the weight function w
A
(r).
The AIM Case
Two conceptually different types of the weight functions are possi-
ble. In the rst case the 3Dspace is decomposed into disjunct atomic
domains
A
, which means that in each domain one weight function
is equal to 1 and all the others vanish:
w
A
(r) =
_
1 if r
A
0 otherwise
. (89)
The most widely used scheme of this type is Baders Atoms in
Molecules (AIM) Theory,
37
in which the dividing surfaces between
atomic domains are determined based on the topological proper-
ties of the electron density. Owing to the disjunct character of the
domains, the atomic overlap integrals are given simply as
S
A
=
_
w
A
(r)
(r)
(r) dv =
_
(r)
(r) dv (90)
i.e., the integration is restricted to the given atomic domain
A
.
Bond orders, and total and free valences are then given by eqs. (83),
(84), and (66) mentioned earlier.
Besides bond order and valence calculations, the AIM scheme
has a great number of other useful applications, which are out of our
present scope. The fact that the atomic domains are dened on the
basis of the electron density calculated from actual wave function
represents a very important advantage of the AIM method, espe-
cially because it reduces the arbitrariness in dening the individual
atoms. It is also the source of some minor drawbacks. Thus, the
atomic domains have complex (and physically not always appeal-
ing) forms, which make costly their determination and the numerical
integrations according to eq. (90). (But that is a more serious prob-
lem if one wishes to do energy partitioning
17, 38
than for bond order
calculations.) Also, in some cases there appear so-called nonnuclear
attractors i.e., domains containing no nucleie.g., in the middle of
the acetylene triple bond. They reect some real physical peculiar-
ities of the electron density and are not artifact in that sense, but
it is rather difcult to nd any chemical interpretation for them. For
some systems, e.g., for boron compounds, the total atomic charges
are also difcult to interpret.
In the AIM scheme all the overlap populations vanish (q
AB
= 0
if A = B) because of the disjunct character of the domains, and
therefore the net and gross atomic populations coincide:
Q
a
= q
AA
=
_
w(r)(r) dv =
_
A
(r) dv (91)
where (r) is the electron density
(r) =
,
D
(r)
(r). (92)
Of course, one can also apply eq. (76).
The bond order index has been introduced in the AIM theory by
ngyn et al.
39
Somewhat later, Fradera et al.
40
introduced a delo-
calization index which in the single determinant case is identical
with the index of ngyn et al.,
39
and most authors who are using
it for the the Hartree-Fock or DFT cases are not aware of the orig-
inal paper.
39
The delocalization index of Fradera et al.
40
is dened
in such a way that for the explicitly correlated case it involves the
use of the total
xc
(r
1
, r
2
) in eq. (68) mentioned earlier, which,
as already noted, I cannot recommend. Bader and Stephens
41
have
introduced a parameter similar to bond order as early as in 1975, but
not for atoms but for the so-called two-electron loges, i.e., spatial
domains housing an electron paira concept, which was introduced
byDaudel inthe 60s70s andhas became completelyobsolete nowa-
days. As already mentioned, the use of the denition in ref. 40 leads
to chemically meaningless bond indices exceeding 5 for the N
2
and
F
2
molecules,
35
while no such problem occurs if one uses the de-
nition advocated here (and also by ngyn et al.
39
). Thus the use of
denition (83) based on the exchange only part (69) of the second-
order density matrix is more chemical. (As already noted, it is much
more economical, as well, as it requires the use of the rst-order
density matrix only, and not that of the second-order one.)
The Fuzzy Atoms Case
An alternative to using disjunct atomic domains is to use so-called
fuzzy atoms, i.e., such divisions of the 3Dspace into atomic regions
in which the regions assigned to the individual atoms have no sharp
boundaries but exhibit a continuous transition from one to another.
Fuzzy atoms were rst introduced by Hirshfeld,
42
who dened the
hypothetical promolecule consisting of unperturbed and nonin-
teracting free atoms placed at the atoms actual positions in the
molecule, and then the weight function w(r) is calculated by using
the atomic electron densities of these free atoms as
w
A
(r ) =

0
A
(r )
0
K
(r )
. (93)
This is often called the stockholders principle. Obviously def-
inition (93) satises conditions (87). This is a transparent and
easy-to-apply denition; it has, however, serious drawbacks, too.
17
The hydrogen atomhas no core electrons and, as a consequence, the
w
A
(r) function of the heavy atom to which the hydrogen is attached
usually has a signicant value at the protons position. That means
the electron density near the proton is partly assigned to the adjacent
heavy atom, and thus the individual atoms are not well cut.
Recently we have proposed to use Beckes weight function
43
for calculating bond order and valence indices,
44
as well as to use
it in energy partitioning.
45
Becke originally proposed that function
for performing effectively the numerical integrations necessary in
the DFT framework. This is a relatively simple algebraic function,
which is calculated in an iterative manner. We shall not repeat here
the algorithm, which is almost easier to program
46
than to describe
(see the original paper
43
and the appendix in our paper
44
), but only
mention that it satises requirements (87) and has the value strictly
equal to 1 at the position of own nucleus (therefore the weight
functions corresponding to all the other atoms vanish there). The
nature of the atoms is introduced by the use of empirical atomic
radii; actually only the ratio of the radii of atoms that are close
to each other is of importance. There is also a stiffness parame-
ter determining the speed of transition from one atomic region to
another. It may be also mentioned that one may locate the station-
ary point of charge density along the interatomic axis connecting
bonded atoms and use its position for determining the ratio of
the atomic radii. In this manner one gets a scheme that combines
fuzzy atoms with some advantages (and disadvantages) of the AIM
method.
In the fuzzy atoms scheme one has nonzero values of the over-
lap densities, which may be calculated either by using eq. (79), or
simply as
q
AB
=
_
w
A
(r)w
B
(r)(r) dv. (94)
This fuzzy atoms overlap population is a 3D analogue of
Mullikens overlap populations used in the Hilbert-space analysis,
and similarly, it reects the degree to which the electronic charge
is shared between the atoms A and Bbut it cannot be called bond
order either. The fuzzy atoms bond orders and valences may be
calculated by using eqs. (83), (84), and (66) with the values of the
atomic integrals calculated as
S
A
=
_
w
A
(r)
(r)
(r) dv. (95)

The fuzzy atoms scheme permits the numerical integrations to
be performed very easily, and has also the important advantage that
the results exhibit quite moderate basis set dependence and, unlike
the results of Hilbert-space analysis, do have a well-dened basis
set limit, and thus may be used also for large basis sets, or those
including diffuse orbitals. The AIM results do have a basis set limit,
too.
Our program FUZZY
46
it may be downloaded freely and is
applicable after a Gaussian runperforms calculations of both
Hilbert space and fuzzy atoms bond orders and valences and fuzzy
atoms overlap populations, for the case of both SCF and correlated
wave functions. Programs BORDER and BO-VIR, applicable also
for large systems, are presently designed for the SCF (RHF. UHF,
may be DFT) case only, but would require only minor changes
input of the density matrices and not orbital coefcientsfor the
correlated calculations.
Discussion
Bond order and valence indices may have two different roles. First,
they help to link the quantum mechanical description of molecules
as systems of particles with their classical chemical picture of atoms
connected by bonds, and to nd quantum mechanical counterparts
of such genuine chemical parameters as the multiplicity of a bond
and the actual valence of an atom in a molecule. It is my opinion
that this possibility has a great conceptual importance.
The bond order and valence indices have also different practi-
cal applications. First of all, they may be used for interpretation
and systematization of the results, obtained in quantum chemical
calculations. In this respect, bond orders and valences represent the-
oretical tools permitting to extract from the wave function different
pieces of information that may be assigned chemical signicance.
(As the wave functions usually represent very large sets of num-
bers for a human, this is also useful to get a comprehensible picture
of the molecule, and may be considered a sort of data compres-
sion.) Probably this is the manner in which the indices have been
applied in most cases (for e.g., in ref. 47). One may identify chem-
ical bonds, intramolecular hydrogen bonds, or hydrogen-bond like
interaction, get an idea about the intensity of nonbonded interac-
tions, etc. (As noted earlier, bond orders cannot always distinguish
between bonding and antibonding character of the secondary inter-
actions; energy components are more sensitive (and expensive)
parameters.
16, 17, 31, 45
), decide whether the atoms in question may be
considered bonded at all (I myself am not a chemist and probably
could not survey properly the applications of this type.)
As there are strong correlations between the strengths of the dif-
ferent bonds and their bond orders calculated by a given basis set of
sufciently atomic character (e.g., the classical 6-31G** basis), in
some cases one may use bond orders not only for interpretations but
also for predictions: one could, for example, predict which bonds
will undergo rupture in a pyrolysis experiment. (That was possible
even by using Wiberg indices at the CNDO/2 semiempirical level
of theory.
48
) Of course, weaker bonds are usually longer and vice
versa, and so in some cases it may be difcult to decide whether the
bond is weak because some external (e.g., steric) effects force it to
be longer or it is long because it is weak due to some features of
the electronic structure. Our method for predicting the primary bond
cleavages in the electron impact mass spectroscopic experiment
49, 50
is truly predictive because it is based on effects of this latter type.
In that method we consider a Koopmans-type vertical ionization
of the molecule, deleting one electron from a canonic HF orbital,
and comparing the bond orders (and often also energy components)
calculated for the ion and the parent neutral molecule. If there is a
bond the bond order index of which is dropping dramatically as a
consequence of ionization, then one may expect that it will be rup-
tured in the MS experiment. There are systems for which ionization
from the HOMO does not lead to any bond cleavagethey usually
exhibit a big peak corresponding to the original molecular ionand
fragmentation occurs only if an electron is removed from a lower
lying orbital. This is the case e.g., for molecules in which the HOMO
is a -orbital of an aromatic ring, ionization from which does not
lead to fragmentation. In such cases characteristic fragments may
be due to ionization from, e.g., a lone pair of a heteroatom in a side
chain. However, it if often difcultor even impossibleto get
converged SCF solutions for such hole states; our experience indi-
cates that it is not necessary either: in order to see what will happen
in the MS apparatus, it is sufcient to consider the Koopmans-type
vertical ionization of the neutral molecule without considering any
orbital relaxation. (Moreover, that also usually works at the semiem-
pirical MNDO level, tooin particular, if combined with energy
decomposition.
51, 52
)
Although the predictions discussed earlier are related to the pos-
sible outcome of some dynamic events, they are based on static
parameters, and must be considered with due care, as they are giv-
ing some information only about the starting point of the process
in question. It is, however, also possible to follow some dynamic
processes in some detail, by investigating different points of the
reaction path on the potential surface. Thus, for instance, in one
of our works
53
we could detect (by computing fuzzy atoms bond
orders) how the presence of a Ni atom assists in a process of form-
ing a new C C bond by rst developing a transient Ni C bond
along the reaction path, which is then replaced by the newly formed
C C one. Very instructive are the studies of Lendvay,
54, 55
who
studied the changes of bond orders along the reaction paths of dif-
ferent reactions. He demonstrated that by calculating bond orders
one may follow in detail how one bond is breaking and another is
forming along the reaction path, check the bond order conserva-
tion rule, study whether the electronic structure of the transition
state is more similar to the reactants or to the products, reveal
whether the reaction is synchronous or asynchronous, and so on,
i.e., the use of bond order and valence indices permits to answer the
different qualitative questions a chemist may put forward about a
reaction.
Acknowledgments
I amdeeply indebted to my coworkerscoauthors of several papers
on the subjectDr. Andrea Hamza (Budapest) and Dr. Pedro
Salvador (Girona), who in the recent years participated in this
research. I am very grateful to Dr. G. Lendvay for many useful
discussions.
Appendix A: Proof of the Relationship Eq. (19)
For an arbitrary wave function (it need not be a single determinant)
which can be expanded by using the AO basis {
} of spin-orbitals,
the spatial distribution of the electron charge and spin density is
given by the expansion
(r, s) =
,
P
(r, s)
(r, s). (A1)

As (r, s) is real, the density matrix P is Hermitian.
(r, s) is a quantum mechanical observable, it is equal to the
expectation value of the respective operator
(r, s) =
N
i=1
(r r
i
)
ss
i
. (A2)
As any symmetric sum of the one-electron operators, (r, s) has
also a second quantized representation in terms of the auxiliary
Lwdin-orthonormalized basis of spin-orbitals {
}:
(r, s) =
(1)|(r r
1
)
ss
1
|
(1)

. (A3)
Substituting here the expansion (13) of the functions and express-
ing operators

and

through the operators

+
and
by using
the relationships (14), we get
(r, s) =
,,
S
1
(1)|(r r
1
)
ss
1
|
1
+
. (A4)
Owing to the presence of the delta-function and of the Kro-
necker symbol, one can perform explicitly the integration and the
summation over the spins:
(r, s) =
,,
S
1
(r, s)
(r, s)
+
. (A5)
By taking the expectation vale we have
(r, s) =
,,
S
1
(r, s)
(r, s). (A6)

Now, by comparing with the general formula (A1) we have
P
S
1
(A7)
or, after trivial manipulations:

+
= (PS)
. (A8)
Appendix B: Proof of the Relationship Eq. (39)
Let
| =

+
1

+
2

+
N
|vac (B1)
be an N-electron single determinant wave function built up of the
orthonormalized spin-orbitals
i
=
C
i
. (B2)
According to the anticommutation relationship (12) we have for the
anticommutator
{
+
i
,
} =
C
i
{
+
} =
C
i
= C
i
. (B3)
As a consequence,

+
| =
i,j
C
i
C
j
|(
i

;
j

) (B4)
where |(
i

;
j

) is the determinant wave function,

which can be obtained from| by replacing spin-orbitals
i
and
j
with the basis spin-orbitals
and
, respectively. It follows from

the general expression
4
for the overlap of two determinant wave
functions built up of nonorthogonal orbitals, that
|(
i

;
j

) =
i
|

i
|
j
|

j
|
. (B5)
Expanding the 2 by 2 determinant in (B5) by substituting the
integrals as
k
|
k
S
(B6)
and using the expansion (B4) we get
|
+
|
=
i,j
C
i
C
j
_
i
S
j
S
i
S
j
S
_
= (PS)
(PS)
(PS)
(PS)
. (B7)
References
1. Coulson, C. A. Proc R Soc Lond A 1939, 169, 413.
2. Mulliken, R. S. J Chem Phys 1955, 23, 1833, 1841, 2338, 2343.
3. Mayer, I. Simple Theorems, Proofs, and Derivations in QuantumChem-
istry; Kluwer/Plenum: New York, 2003.
4. Wiberg, K. A. Tetrahedron 1968, 24, 1083.
5. Borisova, N. P.; Semenov, S. G. Vestn Leningrad Univ 1973, 16, 119.
6. Armstrong, D. R.; Perkins, P. G.; Stewart, J. J. P. J Chem Soc Dalton
Trans 1973, 838, 2273.
7. Borisova, N. P.; Semenov, S. G. Vestn Leningrad Univ 1976, 16, 98.
8. Gopinathan, M. S.; Jug, K. Theor Chim Acta 1983, 63, 497, 511.
9. Mayer, I. Theor Chim Acta 1985, 67, 315.
10. Mayer, I. Int J Quant Chem 1983, 23, 341.
11. Hall, G. G. Chairmans remarks, Fifth International Congress on
Quantum Chemistry, Montreal, 1985.
12. Mayer, I. Chem Phys Lett 1983, 97, 270.
14. Bruhn, G.; Davidson, E. R.; Mayer, I.; Clark, A. E. Int J Quant Chem
2006, 106, 2065.
15. Mayer, I. Int J Quant Chem 1998, 70, 41.
17. Mayer, I. Phys Chem Chem Phys, in press.
18. Fischer, H.; Kollmar, H. Theor Chim Acta 1970, 16, 163.
19. Giambiagi, M.; de Giambiagi, M. S.; Grempel, D. R.; Heynmann, C. D.
J Chim Phys 1975, 72, 15.
21. de Giambiagi, M. S.; Giambiagi, M.; Jorge, F. E. Theor ChimActa 1985,
68, 337.
22. Natiello, M. A.; Medrano, J. A. Chem Phys Lett 1984, 105, 180.
23. Clark, A. E.; Davidson, E. R. J Chem Phys 2001, 115, 7382; Int J Quant
Chem 2003, 93, 384.
25. Mayer, I. Int J Quant Chem 1986, 29, 73, 477.
26. Ponec, R.; Mayer, I. J Phys Chem A 1997, 101, 1738.
27. Mayer, I. J Mol Struct THEOCHEM 1989, 186, 43.
28. Mayer, I. D.Sc. Thesis; Hungarian Academy of Sciences, Budapest,
1986 (unpublished).
29. Lwdin, P.-O. J Appl Phys Suppl 1962, 33, 251.
30. Amos, A. T.; Hall, G. G. Proc R Soc Lond A 1961, 263, 483.
32. Mayer, I. J Phys Chem 1996, 100, 6249.
33. Poater, J.; Sol M.; Duran, M.; Fradera, X. Theor Chem Acc 2002, 107,
362.
34. ngyn, J. G.; Rosta, E.; Surjn, P. R. Chem Phys Lett 1999, 299, 1.
35. Bochicchio, R. C.; Lain, L.; Tore, A. Chem Phys Lett 2003, 374,
567.
36. Mayer, I.; Hamza, A. Int J Quant Chem 2005, 103, 798.
37. Bader, R. F. W. Atoms in Molecules: A Quantum Theory; Oxford
University Press: Oxford, UK, 1990.
38. Salvador, P.; Duran, M.; Mayer, I. J Chem Phys 2001, 115, 1153.
39. ngyn, J. G.; Loos, M.; Mayer, I. J Phys Chem 1994, 98, 5244.
40. Fradera, X.; Austen, M. A.; Bader, R. F. W. J Phys Chem 1999, 103,
304.
41. Bader, R. F. W.; Stephens, M. E. J Am Chem Soc 1975, 97, 7391.
42. Hirshfeld, F. L. Theor Chim Acta 1977, 44, 129.
43. Becke, A. D. J Chem Phys 1988, 88, 2547.
44. Mayer, I.; Salvador, P. Chem Phys Lett 2004, 383, 368.
45. Salvador, P.; Mayer, I. J Chem Phys 2004, 120, 5046.
46. Mayer, I.; Salvador, P. Program FUZZY, Girona, 2003. Available at
http://occam.chemres.hu
47. Bridgeman, A. J.; Cavigliasso, G.; Ireland, L. R.; Rothery, J. J Chem
Soc Dalton Trans 2001, 2095.
48. Rvsz, M.; Blazs, M. Acta Chim Hung 1987, 124, 851.
49. Mayer, I.; Gmry, . Chem Phys Lett 2001, 344, 553.
50. Mayer, I.; Hamza, A. ProgramAPOST-MS, Budapest, 2001. Available
at http://occam.chemres.hu
51. Mayer, I.; Gmry, . J Mol Struct Theochem 1994, 311, 331.
52. Mayer, I.; Gmry, . Program MNDO-MS, Budapest, 1993/2001.
Available at http://occam.chemres.hu
53. Ppai, I.; Schubert, G.; Mayer, I.; Bessenyei, G.; Aresta, M. Organo-
metallics 2004, 23, 5252.
54. Lendvay, G. J Phys Chem 1989, 93, 4422.
55. Lendvay, G. J Phys Chem 1994, 98, 6098.

Bond Order Mayer06

Uploaded by

Document Information

Original Description:

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Bond Order Mayer06

Uploaded by

Copyright:

Available Formats

Bond Order and Valence Indices: A Personal Account

as bond orders is, however, lim-

matrix elements in other cases, too.

= 1), and falls to zero for both an empty orbital

= 0) and a nonbonded pair (D

= 2). At the same time, it

is equal to the sum of all partial Wiberg indices between

and all the other orbitals in the molecule. Summing up

for all the orbitals of the atom, but extracting the

in that basis, and then

} is not a unit matrix,

is only formally an annihilation oper-

as it were the case for an orthonormalized basis, but

is a short-hand notation for the element (S

dened with respect

} only to dene the true annihi-

corresponding to the creation operators

, one introduces an auxiliary Lwdin-orthogonalized set of

are centered, we may write

, which are not strictly

} one has the known expression

entering the integral in the rst term

were an exact eigenvector of

could be exactly expanded in the atomic basis. This is not

has components both

is that component of the function

as a nite basis correction term.

B. Furthermore, one may perform analogous

are centered on the same atom A or on different

. As a consequence, they are not Hermi-

(r) of the orbitals centered

are centered on different

. This can be obtained as a special case

can differ from zero only if the spin-orbitals

are of the same spin.

are the density matrices for the orbitals occupied

and the denition (44) reduces

S valid for the single determinant wave functions and

. Based on this, it was also pos-

are the -th bonding and antibonding orbitals, con-

to be nonnegative real numbers.)

are block-diagonal with 2 by 2 nonzero blocks on the main diagonal.

. We shall denote these blocks as S

is occupied with spin in the wave function, but its antibonding

is singly or doubly occupied (but

is empty). Although for the

are occupied with the spin , then we

are occupied is equivalent

value for some basis orbitals. Thus, we have

for the given

in that basis and then return to the original one,

. By substituting the atomic resolution of

, and introducing the

is the biorthogonal counterpart of the basis orbital |

(r) dv. (95)

(r, s). (A1)

through the operators

(r, s). (A6)

) is the determinant wave function,

, respectively. It follows from

You might also like