Sei sulla pagina 1di 21

Journal of Molecular Structure Theochem. 461462 1999.

121

A new ONIOM implementation in Gaussian98. Part I. The


calculation of energies, gradients, vibrational frequencies
and electric field derivatives
a,2
Stefan Dappricha,1, Istvan
, K. Suzie Byuna , Keiji Morokumaa,U ,
Komaromi

Michael J. Frisch b
a

Cherry L. Emerson Center for Scientific Computation and Department of Chemistry, Emory Uni ersity, 151 Pierce Dri e,
Atlanta, GA 30322, USA
b
Lorentzian Inc., North Ha en, CT 06473, USA
Received 26 June 1998; received in revised form 2 September 1998; accepted 2 September 1998

Abstract
The IMOMM, IMOMO, and ONIOM methods have been proven to be powerful tools for the theoretical
treatment of large molecular systems where different levels of theory are applied to different parts of a molecule.
Within this framework we present a modified handling of the link atoms which are introduced to terminate the
dangling bonds of the model system. Using this new scheme the definition of the combined energy gradient, the
Hessian matrix, and the integration of higher derivatives of the energy with respect to nuclear coordinates and the
electric field vector becomes straightforward. This allows for the first time the consistent combination of vibrational
frequencies and the calculation of other molecular properties such as IR intensities, Raman intensities as well as
dipole moments, polarizabilities, and hyperpolarizabilities. Test calculations for some typical as well as unusual
examples and partitioning schemes are presented to demonstrate the power and limitations of the method and to
provide guidelines for its applicability. Users of the method are strongly advised to test, calibrate and confirm for
themselves the validity of the method combination and the model subsystem for the properties they want to
calculate. 1999 Elsevier Science B.V. All rights reserved.
Keywords: Ab initio methods; Energy gradients; Molecular properties; ONIOM

Dedicated to Professor Keiji Morokuma in celebration of his 65th birthday.


Corresponding author. Tel.: q1 404 7272180; fax: q1 404 7276586; e-mail: morokuma@emory.edu
1
Present address: Institut fur
Physikalische Chemie, Universitat
Karlsruhe TH., Engesserstrasse 7, D-76128 Karlsruhe,
Germany.
2
Present address: Department of Clinical Biochemistry and Molecular Pathology, University Medical School of Debrecen,
Nagyerdei krt. 98, H-4032 Debrecen, Hungary.
U

0166-1280r99r$ - see front matter 1999 Elsevier Science B.V. All rights reserved.
PII: S 0 1 6 6 - 1 2 8 0 9 8 . 0 0 4 7 5 - 8

S. Dapprich et al. r Journal of Molecular Structure (Theochem) 461462 (1999) 121

1. Introduction
The theoretical treatment of large molecular
systems has made tremendous progress during
the past few years. Especially the development of
reliable functionals in the framework of density
functional theory DFT. in combination with efficient schemes for the evaluation of coulomb and
exchange integrals have led to theoretical methods that scale almost linearly with the size of the
system w17x. However, the accurate ab initio
modeling of chemical systems containing a large
number of atoms is still a challenging task. Especially the theoretical description of chemical reactions, i.e. the accurate modeling of transition
states, requires methods that are usually not applicable to large molecules. There are several
remedies to circumvent this problem. Often small
model systems are used to describe the reaction
center that is usually concentrated in a particular
region of the molecule w8,9x. Another approach
uses simplified Hamiltonians like in semiempirical and molecular mechanics methods w10x. However, the validity and applicability strongly depends on the parametrization process and can not
be generalized w11,12x.
An obvious solution to this problem is the
partitioning of the system into two or more parts
or layers, where the interesting or difficult part of
the system the inner layer. is treated at a high
level of theory and the rest of the system the
outer layer. is described by a computationally less
demanding method. This idea is not new and
many different implementations can be found in
the literature w1318x. These hybrid methods differ mainly in two aspects. First, there are different ways to treat the boundary region of the
different parts of the molecule. If there is no
covalent bond between the layers, there is no
special boundary region. A typical case is a solvated system, where the solvent molecules form
the outer layer and the solute is the inner part
which is treated by a higher level method. However, if one is interested in the accurate description of a particular region of a large organic
molecule or a macromolecule, covalent bonds
have to be cut in order to generate the inner
model system. This process leaves dangling bonds

at the border of the inner layer, which have to be


saturated in order to avoid a chemically unrealistic model. Therefore, so-called link atoms usually hydrogen atoms are introduced. They are
only present in the model system and their treatment differs in the different implementations that
have been published.
The second crucial aspect in all the hybrid
schemes is the interaction between the inner and
the outer part of the system. If the total energy
E XY . of the entire system XY inner region
X, outer region Y . is defined as
E XY . s Ehigh X . q Elow Y .
q Einterlayer X ,Y .

1.

with Einterlayer X,Y . being a separate interaction


energy between the two layers w13x, this may be
referred to as a connection scheme. On the
other hand, if the total energy E XY . is calculated according to
E XY . s Elow XY . y Elow X . q Ehigh X . ,
2.
then we shall refer to it as an embedding or
extrapolation scheme w20x. In the latter case there
is no necessity for a special interaction Hamiltonian, since the interaction between the two layers
is consistently treated at the low level of theory.
Obviously, both approaches are equivalent, if
Elow Y . q Einterlayer X ,Y .
s Elow XY . y Elow X . ,

3.

i.e. if Einterlayer X,Y . corresponds to the exact


interaction energy at the respective low level.
This paper describes the unification, generalization, and extension of our recently introduced
IMOMM w19x, IMOMO w20x, and ONIOM w21x
methods that have been proven to be very valuable and promising tools for the treatment of
large molecular systems w2233x. The new implementation shows a better performance and has a
wider applicability. The ONIOM including
IMOMM and IMOMO. method is an extrapola-

S. Dapprich et al. r Journal of Molecular Structure (Theochem) 461462 (1999) 121

tion scheme. A molecular system can be divided


into up to three different layers and the three
layers do not have to be inclusive, i.e. the outermost layer can be connected directly to the innermost one. Every layer can be treated at an arbitrary level of theory. Geometry optimizations
can be performed in any system, and stationary
points can be characterized by the integrated
Hessian evaluation. The ONIOM method has
been implemented into the Gaussian98 program
system w34x. Additionally, this implementation can
also determine integrated energy derivatives with
respect to an electric field like dipole moments,
polarizabilities, hyperpolarizabilities, infrared intensities, and Raman intensities.
2. Theory
2.1. ONIOM energy definition
The basic idea behind our ONIOM approach
can be explained most easily when it is considered
as an extrapolation scheme in a two-dimensional
space, spanned by the size of the system on one
axis and the level of theory on the other axis.
Fig. 1 shows the extrapolation procedure
schematically. Our goal is to describe the real
system at the highest level of theory, i.e. the
approximation of the target E4 point 4. in a

system partitioned into the two-layer ONIOM or


E9 point 9. in a system consisting of three layers.
In the case of two layers, the extrapolated energy
EON IOM2 is then defined as:
EON IOM 2 s E3 y E1 q E2

4.

where E3 is the energy of the entire real. system


calculated at the low level method and E1 and E2
are the energies of the model system determined
at the low and high level of theory, respectively.
EON IOM2 is an approximation to the true energy
of the real system E4 :
E4 s EONIOM 2 q D.

5.

Thus, if the error D of the extrapolation procedure is constant for two different structures e.g.
between reactant and transition state., their relative energy E4 will be evaluated correctly by
using the ONIOM energy EON IOM2 .
For a system partitioned into three different
layers, the expression for the total energy EON IOM3
as an approximation for E9 reads:
EON IOM 3 s E6 y E3 q E5 y E2 q E4 .

6.

Since the evaluation of E1 the smallest model


system at the lowest level of theory. does not
require much computational effort, its value can

Fig. 1. The ONIOM extrapolation scheme for a molecular system partitioned into two left. and three right. layers. See text for
description.

S. Dapprich et al. r Journal of Molecular Structure (Theochem) 461462 (1999) 121

be used to determine the effect of the three-layer


approach as compared to a two-layer partitioning
with points 1, 4, and 6. If the energy difference
between the two- and three-layer extrapolation is
constant, a two layer partitioning with the intermediate layer omitted would give comparably accurate results.
It should be noted that the layers need not be
inclusive or contiguous. The so-called inner layer
does not have to be physically inside the outerlayer. The layers can be any part of the system.
Each layer does not have to be contiguous; it can
consist of several separate regions of the system.
2.2. Treatment of link atoms
As mentioned before, an important and critical
feature of all the combination schemes is the
treatment of the link atoms. For the following
discussion, we first introduce some useful definitions by adopting a two-layer ONIOM scheme as
an example, as illustrated in Fig. 2. The methodology in the case of a three-layer ONIOM is
exactly the same and will not be discussed explicitly.
The atoms present both in the model system
and the real system are called set 1 atoms and
their coordinates are denoted by R1. The set 2
atoms are the artificially introduced link atoms

w19x. They only occur in the model system and


their coordinates are described by R 2 . In the real
system they are replaced by the atoms described
by R 3 . Atoms that belong to the outer layer and
are not substituted by link atoms are called set 4
atoms with the coordinates R 4 . The geometry of
the real system is thus described by R1 , R 3 and
R 4 and they are the independent coordinates for
the ONIOM energy:
EON IOM s EONIOM R1 , R 3 , R 4 . .

In order to generate the model system, described by R1 and the link atoms R 2 , we define
R 2 as a function of R1 and R 3 :
R 2 s f R1 ,R 3 . .

8.

The explicit functional form of the R 2 dependency can be chosen arbitrarily. However, considering the fact that the link atoms are introduced to mimic the corresponding covalent bonds
of the real system, they should follow the movement of the atoms they replace. Therefore we
adopt the following coupling scheme. If atom A
belongs to set 1 and atom B belongs to set 3, the
set 2 link atom symbolized by H in Fig. 2. is
placed onto the bond axis AB. In terms of
internal coordinates we choose the same bond
angles and dihedral angles for set 2 atoms as for
set 3. Therefore, in the model calculations the
link atoms are always aligned along the bond
vectors of the real system. For the exact position
r 2 of a single H atom along an AB bond r 3 y
r 1 ., we introduce a fixed scale factor or distance
parameter. g. Hence,
r2 s r1 q g r3 y r1 . .

Fig. 2. Definition of different atom sets within the ONIOM


scheme.

7.

9.

If the AB bond distance < r 3 y r 1 < changes


during a geometry optimization, the AH bond
distance < r 2 y r 1 < also changes. In our previously
published implementations w1933x, we have assumed that both the AB and AH distances are
of fixed length. Although this assumption usually
does not cause many difficulties in geometry optimizations, it requires the introduction of two

S. Dapprich et al. r Journal of Molecular Structure (Theochem) 461462 (1999) 121

additional bond length parameters, while the new


scheme only needs one distance parameter. More
importantly, the older implementations lose one
degree of freedom per set 3 atom and therefore,
only 3n y 6 y n 3 n 3 being the number of set 3
atoms. normal coordinates and vibrational frequencies could be calculated, which obviously is
not satisfactory. Eq. 9. is similar to that used by
Eichler et al. in their MO q MM method w18x, but
is different in that their g is a variable whereas
our g is a constant.
How should one choose a reasonable value for
the parameter g? Of course, this crucially depends on the nature of the cut bond, the atoms A
and B, and the link atom. If we want to substitute
a CC single bond by a CH bond in the model
calculations, a reasonable value for g would be a
. divided by a
standard CH bond length 1.084 A
., and g be
standard CC bond length 1.528 A
comes 0.709. Another common example is the
. by
substitution of a PC bond typically 1.860 A

.
a PH bond 1.403 A for modeling bulky phosphine groups by using PH 3 in the model calculations, which gives 0.754 as a reasonable value for
g. It should be noted that the optimal scale factor
also depends on the levels of theory used for the
two layers which are connected by the cut bonds.
Results focusing on the choice of g for different
method combinations will be presented in a forthcoming paper. In the Gaussian98 code, if the set
3 atom is specified but the set 1 atom bonded to
this atom is not, the nearest atom in the set 1 is
automatically used as the set 1 atom bonded to
the set 3 atom. Also in the code, if no g value is
specified, the program automatically takes the
ratio of the sum of the covalent radii of the set 2
link. and the set 1 atom to the sum of the
covalent radii of the set 3 and the set 1 atom.
There are cases where one does not have to
break a covalent bond in the real system to form
a model system. For instance, the real system
contains a solute molecule and solvent molecules,
and the model system containing only the solute
molecule is formed by removing the solvent
molecules, without cutting a covalent bond. In
such cases, there is no need for introducing the
set 2 link atom.

2.3. ONIOM gradients and second deri ati es


Once we have defined an expression of the
ONIOM energy along with a certain functional
relationship between set 2 and set 1r3 atoms, the
definition of the corresponding integrated gradient expression is straightforward. For a two-layer
ONIOM system, we obtain
EON IOM 2 s E3 y E1 )J R 2 ; R1 , R 3 .
q E2 )J R 2 ; R1 , R 3 . ,

10.

where J, the Jacobian matrix, projects the forces


on all the set 2 link atoms R 2 . onto the set 1
R1 . and set 3 atoms R 3 .. When one adopts the
relationship 9. between them, differentiation of
the vector components of one set 2 atom r 2 with
respect to the components of r 1 and r 3 yields a
very simple Jacobian:

r 2,ar r 3,b s g a,b

11.

r 2,ar r 1,b s 1 y g . a,b

12.

where the indices a and b denote the cartesian


components x, y, and z, and is the Kronecker
symbol. It should be emphasized that this particular coupling scheme allows a rigorous and consistent definition of the energy derivatives without
complicated transformations. The spare Jacobian
matrix helps computer time and storage for large
systems.
The definition of the second derivatives of the
ONIOM total energy EON IOM with respect to the
nuclear coordinates, the Hessian matrix HON IOM
s 2 EON IOM can be achieved easily. The force
constant matrix of the model system at low level
H1 . and at high level H2 . have to be transformed by applying the Jacobian J and its
transposed J T. Hence we obtain for a two-layer
ONIOM system:
HON IOM 2 s H3 y J T R 2 ; R1 , R 3 .
=) H1 ) J R 2 ; R1 , R 3 .
q J T R 2 ; R1 , R 3 . ) H2 ) J R 2 ; R1 , R 3 .
13.

S. Dapprich et al. r Journal of Molecular Structure (Theochem) 461462 (1999) 121

Similar gradient and second derivative expressions can be derived easily for three- or n-layer
ONIOM partitionings. They will not be repeated
here.
2.4. Differential scaling of the Hessian matrices
Normal vibrational frequencies computed with
MO methods are well known to be overestimated
due to the incomplete consideration of electron
correlation and the harmonic approximation, in
comparison with experimental fundamental frequencies. This results in an overestimation of
about 10% compared to experimental fundamental frequencies. Therefore, it is common practice
to scale these frequencies by an empirical factor
of 0.89 leading to a very good agreement with
experiment for a wide range of systems w35x. Normal frequencies computed with correlated methods are also scaled for the anharmonicity and
incomplete inclusion of electron correlation. Of
course, the optimal scaling factors vary depending
also on the basis set w36x. In ONIOM different
methods are used for different parts of a molecule
and therefore, different scale factors should be
used for Hessians at different levels. Because
ONIOM combines the Hessian matrices at different theoretical levels, the scaling process has
to be performed for the Hessian matrix at each
level:

Thus, one-electron properties can be extrapolated in the same spirit as the energy and its
derivatives. They will be considered now.
Without additional effort the ONIOM scheme
allows for a consistent treatment of molecular
properties related to an electric field F w37x. The
ONIOM2 dipole moment is given as:

s EON IOM 2 tr F
s E3r F y E1r F q E2r F .

16.

The polarizability tensor is defined as:

s 2 EON IOM 2 tr Fa Fb .
s 2 E3r Fa Fb . y 2 E1r Fa Fb .
q 2 E2r Fa Fb .

17.

and the third order hyperpolarizability tensor


becomes:

q 3 E2r Fa Fb Fc .

R 2 ; R1 , R 3 .

=) H2 ) J R 2 ; R1 , R 3 . .

15.

s 3 E3r Fa Fb Fc . y 3 E1r Fa Fb Fc .

=) H1 ) J R 2 ; R1 , R 3 .
T

ON IOM 2 s 3 y 1 q 2 .

s 3 EON IOM 2r Fa Fb Fc .

HON IOM 2 s c32 ) H3 y c12 ) J T R 2 ; R1 , R 3 .

q c 22 )

quence, one cannot define the wave function consistent with the ONIOM energy. However, the
ONIOM density can be clearly defined as the sum
and difference of the densities, as for the energy.
For instance, for the two-layered ONIOM, the
density is given as:

14.

Since the frequency corresponds to the


square-root of the Hessian, the square of the
frequency scale factor has to be used for Hessian
scaling.
2.5. Electric field deri ati es
The ONIOM method combines molecules of
different size at the energy level. As a conse-

18.

The infrared intensities I IR of molecular vibrations are derivatives with respect to the nuclear
coordinates. Here again we have to apply the
Jacobian J:
I IR s 2 E3r R F .
y 2 E1r R F . ) J R 2 ; R1 , R 3 .
q 2 E2r R F . ) J R 2 ; R1 , R 3 . .

19.

Finally, the ONIOM Raman intensities IRaman


can be written as:

S. Dapprich et al. r Journal of Molecular Structure (Theochem) 461462 (1999) 121

IRaman s 3 E3r R Fa Fb .
y 3 E1r R Fa Fb . ) J R 2 ; R1 , R 3 .
q 3 E2r R Fa Fb . ) J R 2 ; R1 , R 3 . .
20.
The corresponding equations for three- and
n-layer ONIOM for all the derivative properties
are easily derived and implemented in the same
fashion. Even though the derivatives with respect
to magnetic field have not been tested yet, they
also should be obtained in the same way. ONIOM
applications to hyperpolarizabilities and magnetic
perturbation will be presented in a future paper.
3. Applications
There are already a number of publications
employing our integration scheme w2233x. However, in previous incarnations the handling of the
link atoms and the border region between the
layers were different. Both AH and AB distances Fig. 2. were kept frozen at reasonable
values w1921x. In the following sections, we present examples of MO q MO integrations calculated by using the new coupling scheme outlined

in Section 2. Our investigation starts with some


small and simple examples where the ONIOM
features are most visible. We shall focus on the
calculation of energies, equilibrium geometries
and vibrational frequencies in comparison with
the corresponding pure low, medium, and high
level calculations. Additionally, basic questions
related to the choice of the scale factor, the
necessity of link atoms, the empirical scaling of
vibrational frequencies, and the calculation of
electric field derivatives will be illustrated and
discussed.
3.1. Bond energies
The theoretical determination of bond energies
and barrier heights often requires methods that
include electron correlation. Since the computational effort of these methods scale as N 5 N s
size of the system. or even steeper, it is difficult
to extend these techniques to larger systems w38x.
The ONIOM method provides a convenient and
practical way to include electron correlation in a
restricted part of a molecule in which, for instance, a bond is broken. In contrast to some
localized correlation approaches w3942x, there is

Table 1
CH bond dissociation energies De in kcalrmol. for CH 3 R Rs H, CH 3 , CH 2 F, CF3 , CHO. at B3LYPr6-311q q GUU
optimized geometries a,b,c
R

HF

MP2

CCSD
T.

ONIOM
MP2:HF.

ONIOM
CC:HF.

ONIOM
CC:MP2.

H
CH3

85.1
82.7
wy2.4x

107.4
104.9
wy2.5x

108.4
105.6
wy2.8x

CH2 F

84.8
wy0.3x

106.8
wy0.6x

107.5
wy0.9x

CF3

87.4
wq2.3x

109.7
wq2.3x

110.2
wq1.8x

CHO

87.2
wq2.1x

109.9
wq2.5x

110.1
wq1.7x

105.1
wy2.3x
q0.2.
107.1
wy0.3x
q0.3.
109.7
wq2.3x
0.0.
109.6
wq2.2x
y0.3.

106.2
wy2.2x
q0.6.
108.2
wy0.2x
q0.7.
110.8
wq2.4x
q0.6.
110.6
wq2.2x
q0.5.

106.0
wy2.4x
q0.4.
107.9
wy0.5x
q0.4.
110.8
wq2.4x
q0.6.
110.9
wq2.5x
q0.8.

In the ONIOM calculations CH 4 has been used as the model system.


In all energy calculations a 6-311 q q GUU basis set has been employed.
b
Values given in brackets are substituent effects w ECH 3 R. y ECH 4 .x.
c
Values given in parentheses are errors of the ONIOM results, relative to the full high level calculations.
a

S. Dapprich et al. r Journal of Molecular Structure (Theochem) 461462 (1999) 121

no necessity of special assumptions about how to


localize the orbitals for a correlated treatment. In
the following example we calculate the influence
of substituents on CH bond energies for a set of
substituted . hydrocarbons CH 3 R Rs H, CH 3 ,
CH 2 F, CF3 , CHO. for the reaction, CH 3 R H
q CH 2 R.
First, geometry optimizations for all systems
were performed at the B3LYPr6-311q q GUU
level of theory. Then we carried out single point
energy calculations at the HF, MP2, CCSDT.,
ONIOMMP2:HF., and ONIOMCCSDT.:HF.
levels of theory employing the same basis set. For
ONIOM we divided the system into two parts
with CH 3 H being the model system the inner
part. described by the correlated methods. The
distance parameter g for the cut CC bond saturated by a hydrogen atom has been chosen to be
0.724. Table 1 shows the calculated dissociation
energies.
While the results for the MP2 and CCSDT.
methods are very close to each other 107.4 and
108.4 kcalrmol, respectively., the dissociation energy for methane calculated at the HF level 85.1
kcalrmol. is significantly underestimated. Similar
results can be found for Rs CH 3 , CH 2 F, CF3 ,
and CHO. In contrast the substituent effects show

a good agreement among all the three methods:


De is lower than in CH 4 for Rs CH 3 and CH 2 F
and higher for Rs CF3 and CHO. This observation indicates that substituent effects in this example can be appropriately treated at the HF
level, i.e. that the ONIOM extrapolation should
give reasonable results. Indeed the dissociation
energies at ONIOMMP2:HF. are in excellent
agreement with the full calculations at MP2 with
deviations ranging from y0.3 Rs CHO. to q0.3
Rs CH 2 F. kcalrmol. Similar results are obtain ed for O N IO M C C SD T . :H F . an d
ONIOMCCSDT.:MP2., although here the extrapolated energies are about 0.6 kcalrmol too
high.
How much does the quality of the low level for
the outer layer affect the results? Table 2 shows
the energies at HFr3-21G and the semiempirical
AM1 level to be compared with MP2r6-311q q
GUU . In the case of Rs CH 2 F, HFr3-21G gives
the wrong trend in the substituent effect and
predicts it to be 0.1 kcalrmol higher than for
CH 4 . Consequently the ONIOMMP2:HF. result
is also higher for fluoroethane. With AM1 the
dissociation energy in CH 3 CHO is 2.3 kcalrmol
lower than in CH 4 , in sharp contrast to the MP2
value 2.5 kcalrmol higher.. The ONIOM result

Table 2
Calculated CH bond dissociation energies De in kcalrmol. for CH 3 R Rs H, CH 3 , CH 2 F, CF3 , CHO. at B3LYPr6-311q q GUU
optimized geometries a,b
R

MP2r
6-311 q q GUU

HFr
3-21G

AM1

ONIOM
MP2:HF.

ONIOM
MP2:AM1.

H
CH3

107.4
104.9
wy2.5x

86.6
84.2
wy2.4x

90.1
84.7
wy5.4x

CH2 F

106.8
wy0.6x

86.7
wq0.1x

89.1
wy1.0x

CF3

109.7
wq2.3x

89.2
wq2.6x

90.5
wq0.4x

CHO

109.9
wq2.5x

88.6
wq2.0x

87.8
wy2.3x

105.2
wy2.2x
q0.3.
107.5
wq0.1x
q0.7.
110.1
wq2.7x
q0.4.
109.2
wq1.8x
y0.7.

101.4
wy6.0x
y3.5.
106.1
wy1.3x
y0.7.
107.8
wq0.4x
y1.9.
105.0
wy2.4x
y4.9.

In the ONIOM calculations CH 4 has been used as the model system.


Values given in brackets are substituent effects w ECH 3 R. y ECH 4 .x.
b
Values given in parentheses are errors of the ONIOM results relative to the full high level MP2 calculations.
a

S. Dapprich et al. r Journal of Molecular Structure (Theochem) 461462 (1999) 121

y2.4 kcalrmol. shows again that the quality of


the substituent effect is controlled by the lower
level for the outer layer. Of course, the best
compromise between accuracy and speed will be
obtained if the most efficient method that gives
the desired accuracy is employed for every part of
the system.
Our previous calculations w25,43,44x for accurate calculations of binding energies, ONIOM
CCSDT. or G2:MP2. is an excellent combination, where G2 is the original G2 w45,46x or
variations of the G2 method, including CBS
w4749x, G2M w50x and G2MS w25x. On the other
hand, ONIOMCCSDT. or G2:HF. has larger
errors, probably because HF cannot account for
substituent effects accurately in some cases.
3.2. Geometry optimizations and ibrational
frequencies
Now we consider some illustrative geometry
optimizations and vibrational frequency calculations using the ONIOM method. As mentioned
above, we have limited ourselves to comparably
small examples where the ONIOM performance
and the features of the border region between the
different layers are most visible. Since the outer
layer part of the system is treated only at the low
level, one expects that the ONIOM geometry of
this part would be close to that optimized at the
low level. For the inner layer part of the system,
the low level contributions of the real system
should cancel with the low level contributions of
the model system, and the ONIOM optimized
geometry of this part is expected to be similar to
that optimized at the high level. For the connecting bond AB distance between the two layers,
one expects a result between the low and high
levels, unless some serious interference takes
place between different methods.
3.2.1. Acetaldehyde
Table 3 shows the results for acetaldehyde 1.
using formaldehyde as the model system. The
definition of the geometrical parameters is given
in Fig. 3. Again, we employ a typical ONIOM
linking, where in the model the CC bond is
substituted by a CH bond, now with a distance

Table 3

angles in degrees. and


Optimized geometries distances in A,
calculated vibrational frequencies in cmy1 . of acetaldehyde
1.
HFrSTO-3G
R1
R2
R3
R4
R5
A1
A2
A3
A4
Frequencies
R1 . A
R2 . A
R3 . A
R4 . A
R5 . A
A1 . A
A3 . A

1.104 1.102.
1.217 1.217.
1.536 1.217.
1.085
1.087
121.5 122.7.
124.3
110.5
109.9

3541 3499.
2121 2100.
1063
3565
3764
1619 1767.
1704

B3LYPrD95V
1.110 1.106.
1.244 1.241.
1.512 1.241.
1.093
1.100
120.2 121.9.
124.7
111.1
109.7

2960 2966.
1707 1702.
909
3038
3179
1428 1530.
1416

ONIOM
1.111
1.242
1.535
1.085
1.087
120.4
123.8
110.3
110.0

2952
1720
989
3564
3764
1431
1691

Values for the model system, HCHO, are given in parentheses.

parameter g s 0.709. The bond lengths and bond


angles for the real system at a pure low level
HFrSTO-3G w51,52x. and high level B3LYP
w5355xrD95V w56x. are also shown in Table 3.
The CH R1 . and CO bonds R 2 . of the aldehyde group are longer at B3LYP than at HF
vs. 1.104 A
and 1.244 A
vs. 1.217 A
.. The
1.110 A

CC bond R 3 is shorter at B3LYP 1.512 A. than


.. The angle A1 is only slightly
at HF 1.536 A
different at both levels 120.2 vs. 121.5.. The
ONIOM optimization gives a geometry that is
very similar to the high level method for those
geometry parameters that belong to the model
system R1 , R 2 , A1 ., while the variables describing the outer layer geometry R 4 , R 5 , A 3 , A 4 .
show values that are very close or even identical
to the low level optimized geometry. The connecting bond length R 3 , where the borderline between
the inner and outer layer lies, is very similar to
vs. 1.536 A
.. These
the low level result 1.535 A
findings are, as mentioned above, exactly what
one would ideally expect from a combined
method.

10

S. Dapprich et al. r Journal of Molecular Structure (Theochem) 461462 (1999) 121

Fig. 3. Test systems for ONIOM geometry optimizations and frequency calculations.

What are the advantages of an ONIOM optimization compared to a high level optimization
of just the model system? Table 1 also lists the
geometries for formaldehyde in parentheses.
the ONIOM
While R1 in HCHO is 1.106 A,
. that is much
method gives a value 1.111 A
closer to the bond length of the real system 1.110

.. A similar trend can be found for the angle


A
A1. Here the ONIOM value 120.4. is closer to
the B3LYP result 120.2. than to HFrSTO-3G
121.5.. For the bond length R 2 , the ONIOM
. is very close to and in-between
value 1.242 A
. and acthe distance in formaldehyde 1.241 A
..
etaldehyde 1.244 A

S. Dapprich et al. r Journal of Molecular Structure (Theochem) 461462 (1999) 121

Now let us examine unscaled. vibrational frequencies. As in the geometry discussion, one can
expect that the frequencies of the normal modes
localized in the inner and outer layer look like
high and low level results, respectively. Those
delocalized in the two regions are expected to be
an average of the two levels. The frequencies of
acetaldehyde exhibit exactly this trend, as shown
for representative vibrational modes in Table 3.
The carbonyl stretch frequency R 2 . is calculated to be 1720 cmy1 at ONIOM, which is comparable to 1707 cmy1 , the value at B3LYPrD95V.
At low level R 2 . is significantly higher 2121
cmy1 .. The CH stretch frequency R1 . in
ONIOM 2952 cmy1 . reproduces the high level
value 2960 cmy1 . very well, while the HF value is

11

much higher 3541 cmy1 .. Even for the link atom


vibration R 3 . the ONIOM scheme improves
the result considerably: the frequency 989 cmy1 .
lies between the B3LYP 909 cmy1 . and the HF
1063 cmy1 . values.
3.2.2. Propanal
The next example is a three-layer ONIOM
optimization of propanal 2.. The smallest model
system is HCHO and is treated at the high level
B3LYPrD95V., the intermediate model system
is CH 3 CHO calculated at the intermediate level
MP2rSTO-3G., and for the real system a low
level HFrSTO-3G. is employed. Here we were
interested in the question: does it have any unexpected effects if the borderlines of different lay-

Table 4

angles in degrees. and calculated vibrational frequencies in cmy1 . of propanal 2. assuming


Optimized geometries distances in A,
Cs symmetry
HFrSTO-3G
R1
R2
R3
R4
R5
R6
R7
A1
A2
A3
A4
A5
A6
Frequencies
R1 . A
R2 . A
R3 . A
R4 . A
R5 . A
R6 . A
R7 . A
A1 . A
A2 . A
A3 . A
A4 . A
A5 . A
A6 . A

1.104 1.102.
1.217 1.217.
1.547
1.089
1.539
1.086
1.086
121.3 122.7.
123.9
106.9
110.5
110.7
112.4

3536 3499.
2111 2100.
1229
3709
1229
3754
3753
1658 1767.
580
1802
1057
1740
334

MP2rSTO-3G
1.125 1.121.
1.262 1.262.
1.572
1.107
1.557
1.103
1.103
121.7 123.0.
124.2
106.8
110.6
110.7
112.1

3283 3399.
1756 1800.
1149
3505
1149
3556
3556
1532 1608.
533
1699
982
1622
307

B3LYPrD95V
1.111 1.106.
1.244 1.241.
1.523
1.100
1.540
1.096
1.098
120.1 121.9.
124.2
105.6
110.9
111.3
112.9

2949 2966.
1696 1702.
1063
3092
1063
3141
3132
1429 1530.
527
1479
909
1448
296

ONIOM
1.112
1.240
1.559
1.107
1.553
1.086
1.086
120.5
123.3
106.7
110.2
110.6
112.8

2938
1711
1160
3510
1160
3757
3755
1416
520
1703
968
1729
305

ONIOM calculation uses HCHO and H 3 CCHO as the small and intermediate model systems, respectively. Values for HCHO are
given in parentheses.

12

S. Dapprich et al. r Journal of Molecular Structure (Theochem) 461462 (1999) 121

ers are very close i.e. no bond between them. to


each other?
The results in Table 4, as expected, show again
that the geometry parameters R1 , R 2 , and A1 for
the ONIOM calculation are close to the high
level optimization of the real system. R 4 and A 3
show values that are almost identical to the intermediate level optimization, while R 6 , R 7 , A 4 , and
A 5 resemble the results of the low level calculation. The bond lengths R 3 and R 5 and the bond
angle A 2 and A 6 are between the corresponding
pure level calculations.
The vibrational frequencies are also shown in
Table 4. In analogy to the optimized geometries
the ONIOM frequencies are in good correlation
with the results of each level of the pure level
calculations. Since the C s symmetric conformation of propanal is not a true minimum on the
potential energy surface, the frequency calculations gave one imaginary frequency. The corresponding eigenvector represents a torsional motion around the R 3 bond and leads to an eclipsed
conformation of the CO group with respect to
the C H bond. The ONIOM calculation gives
an imaginary frequency of 84 i cmy1 , which is
between 78i cmy1 B3LYPrD95V. and 94 i cmy1

MP2rSTO-3G.. Conclusively we can state that


at least for systems where CC single bonds are
substituted by CH bonds, the close neighborhood of different border regions is probably not a
problem.
3.2.3. Trifluoroacetaldehyde
In preceding examples the ONIOM method
has been applied by using hydrogen atoms to
saturate the dangling bonds of the model systems.
In principle, any atom or group that is monovalent like F, Cl, Br. may be used to saturate a
dangling single bond. Or, when cutting a CC
double bond one could possibly use a divalent
atom O, S. to catch the electrons. In the next
example, trifluoroacetaldehyde 3., we used F as a
model atom for the CF3 group; i.e. the model
system is FCHO. Our simple intention was to
mimic the electron-withdrawing influence of a
CF3 group with F as a link atom better than with
H. From a more general point of view, we cut a
polarized CC bond and introduce a polarized
link atom bond.
Table 5 shows the results of the geometry optimizations and vibrational frequency calculations. The geometry for ONIOM using H as a

Table 5

angles in degrees. and calculated vibrational frequencies in cmy1 . of trifluoracetaldehyde


Optimized geometries distances in A,
3.
HFrSTO-3G
R1
R2
R3
R4
R5
A1
A2
A3
A4
Frequencies
R1 . A
R2 . A
R3 . A
R4 . A
NIMAG

1.102 1.108.
1.215 1.210.
1.586 1.351.
1.370
1.374
123.7 125.6.
122.4
111.0
110.4

3570
2094
1365
1502
0

B3LYPrD95V
1.102 1.095.
1.229 1.209.
1.546 1.412.
1.381
1.402
124.6 129.4.
122.8
114.0
109.5

3071
1733
1225
1106
0

ONIOM with HCHO


1.107
1.239
1.597
1.370
1.373
123.2
121.5
110.7
110.5

2999
1689
1298
1535
0

ONIOM with FCHO


1.090
1.202
1.841
1.364
1.366
130.6
123.9
111.8
108.6

3195
1813
1195
1540
1 35i .

ONIOM calculations use HCHO or FCHO as model systems. Values for FCHO are given in parentheses.

S. Dapprich et al. r Journal of Molecular Structure (Theochem) 461462 (1999) 121

link atom is slightly worse than in the previous


examples. The distances R1 , R 2 , and R 3 are
somewhat longer than in the low and high level
optimizations and the angles A1 and A 2 are
smaller. For the coordinates R 4 , R 5 , A 3 , and A 4
the agreement with the low level geometry is
fairly good.
Can we get an improvement by using F as link
atom? Here, the bond lengths R1 and R 2 are
much shorter than in the HFrSTO-3G and
B3LYPrD95V optimizations, but R 3 is extremely
.. Of course, this bond lengthening
long 1.841 A
has a strong influence upon the entire CF3 group.
Therefore, R 4 , R 5 , A 3 , and A 4 also show a large
disagreement with the HF geometry. How can we
understand these findings? The fluorine atom has
a completely different electronic structure in
comparison with the carbon atom of the CF3
group. While in the real system the CC bond
R 3 . is shortened when going from HFrSTO-3G
. to B3LYPrD95V 1.546 A
., the model
1.586 A
system exhibits the reverse trend; in FCHO the
.
CF bond is longer at B3LYPrD95V 1.412 A
.. This also holds
than at HFrSTO-3G 1.351 A
true for HCHO, but in this case the difference
1.241 Ay
1.217 A,
see Table 3. is
R 3 s 0.024 A
..
less significant than in FCHO R 3 s 0.061 A
This example shows clearly that an appropriate
substitution of a cut bond requires a model bond
that shows a similar behavior with respect to both
levels of theory being employed.
The ONIOM frequencies with H as a link atom
are also shown in Table 5. As one would expect,
R1 . s 2999 cmy1 and R 2 . s 1689 cmy1 are
in good agreement with the high level calculation
3071 cmy1 and 1733 cmy1 , respectively., while
R 4 . s 1535 cmy1 is close to the low level value
1502 cmy1 .. For the CC stretching vibration
R 3 . ONIOM gives a value of 1298 cmy1 which
is between the HFrSTO-3G 1365 cmy1 . and
B3LYPrD95V 1225 cmy1 . frequencies. The result of the vibrational analysis of CF3 CHO using
FCHO as a model system is as pathologic as the
optimized geometry and even one imaginary frequency is obtained 35i cmy1 ., corresponding to a
torsional motion around the R 3 bond. Obviously,
a CF bond is not a good choice for mimicking a
CCF3 bond.

13

Table 6

angles in degrees. and


Optimized geometries distances in A,
calculated vibrational frequencies in cmy1 . of cyclobutene
4.
HFrSTO-3G
R1
R2
R3
R4
R5
A1
A2
A3
A4
Frequencies
R1 . A1
R2 . A1
R2 . B2
R3 . A1
R4 . A1
R5 . A1
A1 . A1

1.314 1.306.
1.082 1.082.
1.526
1.565
1.089
124.3 122.2.
94.7
115.7
109.1

1997
3767
3715
1336
1115
3606
1336
1188

B3LYPrD95V
1.359 1.348.
1.085 1.088.
1.538
1.590
1.097
133.6 121.7.
94.3
115.7
108.9

1616
3255
3213
1132
892
3069
1132
1009

ONIOM
1.366
1.086
1.538
1.549
1.089
133.3
93.4
115.5
108.8

1639
3262
3220
1216
1110
3604
1216
1041

Values for ethylene, the model system, are given in parentheses.

3.2.4. Cyclobutene
In the next test system, cyclobutene 4., ethylene is chosen as the model system. Therefore, we
have to cut two considerably strained CC bonds
R 3 . in order to generate it. Can this work?
The ONIOM geometry optimization Table 6.
shows again good agreement with the high and
low level structures. It is noteworthy that the
connecting bond length R 3 in ONIOM is the
same to that in the B3LYPrD95V optimization
.. It is obvious that in this particular case
1.538 A
the pure model system does not even resemble
the real system in basic geometric features. The
angle A1 is calculated by ONIOM to be 133.3
136.3 at B3LYPrD95V., while the pure model
system calculation gives a value of 121.7 at the
high level of theory. The low level method for the
real system takes care of the constrained structure.
The ONIOM frequencies show the usual trends
discussed before. The interesting vibrational mode
R 3 . s 1216 cmy1 lies between the values at

14

S. Dapprich et al. r Journal of Molecular Structure (Theochem) 461462 (1999) 121

HFrSTO-3G 1336 cmy1 . and B3LYPrD95V


1132 cmy1 ., although the bond distance is identical to the pure high level calculation.
3.2.5. Rh(CO)2 Cp
Transition metal complexes are computationally demanding systems, not only because of
their large number of electrons, but also because
of the often large ligands used in experiments
such as pentamethyl cyclopentadienyl. Thus, computational chemists dealing with this type of
molecule often have to resort to model systems,
sometimes introducing inappropriate simplifications. The ONIOM method can be an attractive
and promising alternative. However, the question
arises: what is the best way to divide the system
into inner and outer layers? Is it feasible to cut a
metalligand bond, and, if yes, what kind of link
atom should be employed? The next two examples demonstrate unusual though hopefully instructive ONIOM applications for transition metal
complexes.
Cyclopentadienyl Cp. is a ligand often needed
in organometallic chemistry. An example for such
a complex is RhCO. 2 Cp 5. and we might explore the possibility to exclude the 10 atoms of

the Cp group in the model system. By doing so we


face the problem: should the RhCp bonds
be considered as being mainly ionic, i.e.
y
RhCO.q
as the usual electron count would
2 Cp
suggest? If yes, we do not need to introduce a link
atom and the model system could be just
RhCO.q
2 . Or is the character of the RhCp bond
mainly covalent? Then, we need to introduce link
atoms or a single link atom at the center of the
Cp ring that somehow resembles the electronic
structure of the Cp ligand. Based on orbital considerations and the isolobal principle, Cl might be
a good choice for a link atom, since it has a
similar frontier orbital structure w57x. Another
heuristic approach would be the introduction of
an energy shift operator for changing the Coulomb integral of certain link atom orbitals w58x.
However, this method will not be pursued in this
paper.
Table 7 shows the results for geometry optimizations and frequency calculations of
RhCO. 2 Cp at HFrLANL2MB w59x low level.,
B3LYPrLANL2DZ w59x high level., and at different ONIOM levels. Using RhCO.q
2 as a model
system, ONIOM gives bond lengths for R1 1.842
. and R 2 1.171 A
. that are much closer to the
A

Table 7

angles in degrees. and calculated vibrational frequencies in cmy1 . of RhCO. 2 Cp 5.


Optimized geometries distances in A,
HFr
LANL2MB
R1
R2
R3
R4
R5
A1
A2
Frequencies
R1 . A1
R2 . A1
R3 . A1
R4 . A1
R5 . A1
A1 . A1
A2 . A1

2.030 2.095.
1.147 1.142.
2.088
1.209
1.080
134.7 133.6.
80.9

379 313.
2477 2536.
273
1290
3767
88 91.
922

B3LYPr
LANL2DZ
1.872 1.870.
1.180 1.160.
2.040
1.226
1.081
134.3 136.1.
88.7

493 485.
2000 2117.
277
1122
3311
96 112.
795

ONIOM
no link

ONIOM
g s 1.2

ONIOM
g s 1.3

ONIOM
g s 1.4

1.842
1.171
2.097
1.209
1.080
136.4
80.9

1.852
1.180
2.000
1.210
1.080
135.3
81.1

1.848
1.180
2.033
1.210
1.080
135.5
81.0

1.846
1.180
2.052
1.209
1.080
135.5
81.0

542
2041
286
1290
3767
106
924

.q
ONIOM calculations use RhCO.q
2 no link or Rh CO 2 Cl with different values for g as model systems. Values for Rh CO 2 are
given in parentheses.

S. Dapprich et al. r Journal of Molecular Structure (Theochem) 461462 (1999) 121

and 1.180 A
. than to
high level geometry 1.872 A

.
the low level 2.030 A and 1.147 A , while the
coordinates R 3 , R 4 , R 5 , and A 2 are expectedly
similar to the low level optimization. The angle
A1 is somewhat larger 136.4. than in both pure
MO geometries. This is due to the fact that the
B3LYP value for A1 of the model system 136.1.
is larger than HF 133.6., thereby leading to an
artificial opening of this angle. Similar trends can
be found for ONIOM vibrational frequencies.
Can one get better results by introducing a link
atom in the ONIOM calculation? Since we substitute a virtual point the midpoint of the Cp
ring by a real atom Cl., we must impose
certain constraints during the geometry optimization in order to keep the Cl atom on a position
along the RhCp bond vector. More specifically,
this means that all the geometry optimizations in
this example were carried out in Z-matrix coordinates using local C 5v symmetry for the Cp moiety.
It should also be noted that a frequency calculation is not feasible because the mass weighting
step naturally fails for the virtual atom in the real
system. We have performed ONIOM optimizations using three different distance parameters g,

15

ranging from 1.2 to 1.4. The best results are


. is identical to
obtained for g s 1.3. R 2 1.180 A
the full high level optimization, and R 3 gives a
in good agreement with 2.040 A

value of 2.033 A
at the high level. Similar to the ONIOM calculation without a link atom, A1 is slightly higher
135.5. than in both reference optimizations. For
g s 1.2 and g s 1.4 results are slightly worse, especially for the bond length R 3 . Beside the specific system the best value for g depends very much
on the method combination and it is therefore
difficult to be definite. Keeping in mind that one
should not cut the system close to the most important region of the molecule, a value around
1.3 is probably sufficient in most cases. These test
calculations suggest that in the case when high
accuracy is not required, one may perform
ONIOM calculations with a model system in
which Cpy is simply removed or replaced by Cly.
3.2.6. Ru(CO)2 (C6 H6 )
If the ONIOM replacement of Cp in a transition metal complex works reasonably well, one
might suggest a similar treatment for a benzene
ligand in transition metal complexes. Thus, the

Table 8

angles in degrees. and calculated vibrational frequencies in cmy1 . of RuCO. 2 C 6 H 6 . 6.


Optimized geometries distances in A,
HFr
LANL2MB
R1
R2
R3
R4
R5
A1
A2
Frequencies
R1 . A1
R2 . A1

R3 . A1
R4 . A1
R5 . A1
A1 . A1
A2 . A1

1.963 1.949.
1.155 1.153.
2.121
1.405
1.083
136.2 133.6.
85.3

428 423.
2377 2391.
2404 2413.
142
1127
3750
94 93.
846

B3LYPr
LANL2DZ
1.866 1.824.
1.189 1.190.
1.965
1.428
1.084
136.9 136.0.
90.2

522 552.
1886 1885.
1951 1965.
201
978
3275
102 113.
754

ONIOM
no link

ONIOM
g s 1.1

ONIOM
g s 1.2

ONIOM
g s 1.3

1.834
1.193
2.237
1.402
1.083
137.9
85.8

1.892
1.198
1.744
1.418
1.081
135.9
85.7

1.864
1.197
1.979
1.408
1.082
136.9
85.4

1.858
1.197
2.038
1.406
1.083
137.3
85.4

556
1867
1949
128
1131
3748
108
837

ONIOM calculations use RuCO. 2 no link. and RuCO. 2 Cly with different values for g . as model systems. Values for RuCO. 2
are given in parentheses.

16

S. Dapprich et al. r Journal of Molecular Structure (Theochem) 461462 (1999) 121

compound RuCO. 2 C 6 H 6 . can be modeled by


using either RuCO. 2 without link atom. or
RuCO. 2 Cly Cly as a link atom.. The results are
shown in Table 8.
In the pure model system RuCO. 2 , the bond
lengths of the carbonyl groups, R1 and R 2 , are
shifted towards those in the high level real system
in comparison with the low level. However, the
trend is slightly overestimated; R1 , in ONIOM
. is too short compared to the benchmark
1.834 A
at B3LYPrLANL2DZ. The difference
1.866 A
at
between the CO distance R 2 s 1.193 A
at the low level is exactly
ONIOM and 1.155 A
the same as the bond length shift for the model

.. ONIOM with
system RuCO. 2 1.190 A1.153
A
a chlorine anion as a link atom with a g parameter of 1.2 improves the optimized bond lengths R1
and 1.979 A,
reand R 3 considerably 1.864 A
.
spectively., but R 2 is slightly longer 1.197 A
.. The results
than at B3LYPrLANL2DZ 1.189 A
for the ONIOM frequency calculations are consistent with the corresponding optimized geometries.
3.2.7. Pt(PH3 )2 (C8 H10 )
Having demonstrated two unusual partitioning
recipes, we now move on to a transition metal
complex where the ONIOM method can be applied in the classical way. One of the most
interesting aspects of platinumolefin complexes
is the ability of PtL 2 L s PPh 3 , for example. to
stabilize strained olefins upon formation of the
complex w6062x. Our aim here is not to make
accurate predictions of the structure and stability
of the yet-to-be-synthesized complex 8, but to
demonstrate the performance of the ONIOM
method by using the ethylene complex 7 as the
model system Fig. 4..
The most important geometrical parameters
and the Ptolefin bond dissociation energies are
listed in Table 9. While in the ONIOM calcula . is
tion the length of the PtC bond 2.079 A
in-between the values for both levels, the bond
parameters, R 2 , R 3 , and A1 are similar to the
high level values. Comparison of the structures 7
and 8 reveals that the most striking differences
vs. 2.089 A
at
are the values of R1 2.136 A
at ONIOM., R 3
B3LYPrLANL2DZ and 2.079 A

Fig. 4. Two Ptolefin complexes. 7 is used as the inner part in


an ONIOM calculation of 8.

vs. 1.537 A
and 1.598 A
. and A 2 114.3
1.450 A
vs. 98.1 and 96.3.. The coordinated ethylene
ligand in 7 is much less pyramidalized than the
strained olefin C 8 H 10 in 8 and as a consequence
electron back-donation in the olefinplatinum
bond is less favored w63x. In agreement with the
geometrical trends, the bond dissociation energy
De in complex 7 is much smaller 13.9 kcalrmol
at B3LYPrLANL2DZ. than in 8 where De is
86.6 kcalrmol HF., 53.6 kcalrmol B3LYP., and
62.7 kcalrmol ONIOM.. These results demonstrate that in systems where the geometry of the
reaction center is strongly influenced by a bulky
group with steric requirements, the bulky group
Table 9

and degrees. and calculated


Optimized geometries in A
Ptolefin bond energies in kcalrmol. of the Pt complexes 7
and 8 at different levels of theory the ONIOM calculation for
8 uses 7 as the model system
HFr
LAN2MB

B3LYPr
LAN2DZ

ONIOM

R1
R2
R3
A1
A2

2.092
2.815
1.428
89.5
112.5

2.136
2.422
1.450
104.3
114.3

R1
R2
R3
A1
A2

2.069
2.757
1.500
94.5
97.7

2.089
2.425
1.537
103.5
98.1

2.079
2.420
1.598
107.7
96.3

De 7.
De 8.

17.5
86.6

13.9
53.6

62.7

S. Dapprich et al. r Journal of Molecular Structure (Theochem) 461462 (1999) 121

can be assigned to the outer layer in an ONIOM


calculation and treated by uncorrelated or even
molecular mechanics. methods.
3.3. Choice of the distance parameter g
Although we have already shown a few examples concerning the suitable choice of a distance
parameter g for a link atom bond, it is necessary
to determine the sensitivity of energies, geometries, and frequencies upon g, which in the present examples is the ratio of the model CH link
distance and the real CC distance, as defined in
Section 2. For two examples, CH 3 CH 3 and
CH 3 CF3 , we have recalculated the CH bond
dissociation energies De with values for g between
0.5 and 1.0, as shown in Table 10. Quite surprisingly, the parameter dependency is of almost no
chemical significance for both CH 3 CH 3 and
CH 3 CF3 with CH 4 being the model system. Using
the 6-311G q q GUU basis set, with the unrealistic choice of g s 1.0 the dissociation energy is

17

106.5 kcalrmol for ethane, while with all of the


values, the bond energy remains within 105.2" 0.1
kcalrmol, within an error of 0.3 kcalrmol from
the pure MP2 value of 104.9 kcalrmol Table 1..
Similar results are obtained with a smaller basis
set as well as for CH 3 CF3 .
In geometry optimizations and frequency calculations, the scale factor effect is much more
pronounced. We discuss only one example:
CH 3 CHO in Table 11. The geometry parameters
that belong to the inner system R1 , R 2 , A1 . are
not very much affected by different values for g.
While R1 monotonically decreases as g increases,
.,
R 2 exhibits a maximum value at g s 0.6 1.243 A
and A1 increases as g increases. The CC distance R 3 , which is replaced by CH in the model,
is as expected more strongly influenced by g,
g s 0.5. to 1.665 A
gs
ranging from 1.461 A
1.0.. The calculated vibrational frequencies show
a similar trend in the corresponding bond lengths
and angles. Interestingly, R 2 . has a minimum
value at g s 0.8. It should be clear from this

Table 10
Dependence of the CH bond dissociation energy in kcalrmol. of CH 3 CH 3 and CH 3 CF3 on the distance parameter g calculated
at ONIOMMP2r6-311q q GUU :HFr6-311q q GUU .

CH3 CH3
CH3 CF3

0.5

0.6

0.7

0.724

0.8

0.9

1.0

105.21
112.37.
109.87
117.52.

105.16
112.15.
109.79
117.11.

105.12
112.01.
109.73
116.84.

105.12
112.00.
109.71
116.80.

105.14
112.05.
109.71
116.79.

105.36
112.41.
109.85
117.06.

106.51
112.43.
110.79
117.93.

Geometries optimized at the B3LYPr6-311q q GUU level. Values in parentheses obtained at ONIOMB3LYPr6-31GUU :HFr321G.. The model system is CH 4 .
Table 11
Dependence of the optimized geometry and vibrational frequencies of CH 3 CHO 1. model system: HCHO. on the distance
parameter g at ONIOMB3LYPrD95V:HFrSTO-3G.
0.5
R1
R2
R3
A1

R1 . A
R2 . A
R3 . A
A1 . A

1.119
1.239
1.461
119.6
2889
1787
1098
1450

0.6
1.116
1.243
1.492
119.8
2914
1749
1054
1437

0.7
1.111
1.242
1.531
120.3
2947
1721
996
1431

0.724
1.110
1.242
1.541
120.4
2956
1717
978
1429

0.8
1.107
1.239
1.579
121.0
2990
1711
921
1423

0.9
1.103
1.235
1.628
122.0
3035
1717
862
1415

1.0
1.100
1.233
1.665
122.7
3075
1730
840
1410

18

S. Dapprich et al. r Journal of Molecular Structure (Theochem) 461462 (1999) 121

example that the distance parameter for the link


atom should be chosen carefully, if the model
system is small and the borderline bond is close
to the action center.
3.4. Empirical frequency scaling
In this section we illustrate some ONIOM calculations, where different empirical frequency
scale factors are applied, as shown in Eq. 14..
The unscaled normal frequencies of CH 3 CHO
at the ONIOMB3LYPrD95V:HFrSTO-3G.
level are listed in Table 12 together with different
scale factor combinations c H :c L .. The first example uses the scale factors 0.5:1.0., i.e. the high
level Hessian is scaled by 0.5, while the low level
Hessian remains unchanged. This chemically senseless choice is just to demonstrate how the differential scale factors work. The resulting frequencies are listed and their effective scale factor, obtained by scaledrunscaled , is given in paren-

theses. Without going into much detail, the results show the expected changes. Those normal
modes localized in the inner region reflect the
high level scaling, those localized in the outer
region reflect the low level scaling, and those
delocalized are scaled by some average between
the two. More realistic factor combinations, e.g.
0.92:0.89., are given in the remaining columns of
Table 12 and demonstrate the useful applicability
of the ONIOM differential scaling.
3.5. Electric field deri ati es
The final topic of this paper is one-electron
properties, including electric field derivatives calculated with ONIOM. As discussed in Section 2,
ONIOM can provide such properties in exactly
the same way as it gives energies and geometrical
derivatives.
3.5.1. AgCO q
The first example is one of the smallest possible ONIOM systems. It is AgCOq as shown in
Table 13, where CO is the inner layer and
Agq is the outer layer. The unscaled ONIOM
B3LYPrLANL2DZ:HFrLANL2MB. zero point
vibrational energy ZPE, 4.0 kcalrmol. is quite

Table 12
ONIOMB3LYPrD95V:HFrSTO-3G. vibrational frequencies in cmy1 . of CH 3 CHO 1. with different Hessian scale factors
employed at each layer model system: HCHO.
Symmetry

Modea

Unscaled

0.5:1.0.

0.89:1.0.

0.92:0.89.

A
A
A
A
A
A
A
A
A
A
A
A
A
A
A

2-1-4-5
2-1-4
q3
1-4
5-4-1
q1
3-1-2
5,6,7-4-1
1-2
q1,5
6-4-7
1-3
5,6,7-4
6-4; 7-4
5-4; 6,7-4

158
490
861
989
1253
1261
1430
1691
1719
1805
1816
2950
3564
3738
3764

157 0.99.
295 0.60.
477 0.55.
881 0.89.
1280 1.02.
1196 0.95.
675 0.47.
1675 0.99.
712 0.41.
1805 1.00.
1810 0.99.
1469 0.50.
3563 1.00.
3738 1.00.
3763 1.00.

159 1.00.
448 0.91.
787 0.91.
932 0.94.
1204 0.96.
1237 0.98.
1299 0.91.
1686 0.99.
1559 0.91.
1805 1.00.
1811 0.99.
2633 0.89.
3564 1.00.
3738 1.00.
3764 1.00.

141 0.89.
448 0.91.
785 0.91.
896 0.91.
1127 0.90.
1133 0.90.
1311 0.92.
1511 0.89.
1571 0.91.
1611 0.89.
1627 0.90.
2710 0.92.
3182 0.89.
3338 0.89.
3361 0.89.

Effective net factors are given in parentheses.


a
x-y: stretch; x-y-z: bend; x-y-z-w: torsion; qx: out-of-plane; x,y-z: symmetric stretch.

S. Dapprich et al. r Journal of Molecular Structure (Theochem) 461462 (1999) 121


Table 13
., vibrational frequencies in
Optimized geometries in A
y2 .,
cmy1 ., ZPEs in kcalrmol., intensities I in D 2 amuy1 A

dipole moments in D , and polarizabilities in a.u.. of


AgCOq

r CO.
r AgC.
CO.
AgC.
ZPE
I w CO.x
I w AgC.x

xx.
yy.
zz.

HFr
LAN2MB

B3LYPr
LAN2DZ

ONIOM

1.141 1.146.
2.373
2533.7 2462.6.
227.3
4.6 3.5.
3.2 40.9.
1.8
1.209 0.124.
8.4 3.5.
8.4 3.5.
15.5 6.6.

1.150 1.167.
2.197
2161.6 2028.5.
238.6
4.1 2.9.
80.2
0.0
1.306 0.033.
15.4 7.9.
15.4 7.9.
29.0 13.7.

1.160
2.374
2120.1
226.2
4.0
20.7
2.0
1.330
12.9
12.9
22.8

In the ONIOM calculation CO was treated at


B3LYPrLANL2DZ, while the outer layer Agq atom was
calculated at HFrLANL2MB. Values for isolated CO are
given in parentheses.

close to the pure high level value 4.1 kcalrmol..


The agreement between the intensities of the
CO stretch vibrations at the different levels is
much poorer: 20.7 at ONIOM, 80.2 at high level,
and 3.2 at low level, although ONIOM improves
the low level results substantially. The ONIOM
dipole moment and polarizabilities also show a
clear improvement closer to the high level values. over the low level results.
Table 14
y2 .,
ZPEs in kcalrmol., IR intensities I in D 2 amuy1 A
dipole moments in D., and polarizabilities in a.u.. of
CH3 CHO

ZPE
I w R1 .x
I w R2 .x
I w R3 .x
I w A1 .x

xx.
yy.
zz.

HFr
STO-3G

B3LYPr
D95V

ONIOM

41.3
23.7
30.0
1.5
18.6
1.78
16.6
11.3
7.7

35.0
120.0
135.5
6.5
10.2
3.28
30.1
22.1
16.8

39.3
126.6
127.8
10.7
5.7
2.99
24.4
17.1
11.6

In the ONIOM calculation HCHO is the model system treated


at the B3LYPrD95V level.

19

3.5.2. CH3 CHO


In a second example we use again acetaldehyde
as shown in Table 14. The ONIOM vibrational
intensities of R1 . and R 2 . are in reasonable
agreement with the high level results. However,
the ONIOM intensities for R 3 . 10.7. and
A1 . 5.7. seem to be overcompensated, an
effect due to the intensity differences calculated
at HF and B3LYP for HCHO. Dipole moments
and polarizabilities follow the trends already described for AgCOq.
4. Conclusions
The new ONIOM method is a generalized hybrid scheme to treat different parts of a molecular system at different levels of theory. Its implementation in the Gaussian98 program system allows the definition of up to three arbitrarily
shaped layers with the free choice of the method
for every part. All methods available in Gaussian
can be used without any modification of the
parametrization. For the situation where covalent
bonds need to be cut in order to generate a
model system, an improved link atom technique is
employed. Besides energy calculations, ONIOM
allows geometry optimizations, vibrational frequency calculations, and the evaluation of electric
field derivatives such as dipole moments, polarizabilities, hyperpolarizabilities, IR intensities, and
Raman intensities.
Although we have not described any large and
impressive examples of ONIOM calculations in
this paper, the power and general applicability of
the method should be obvious to the reader. In
contrast to other hybrid methods, ONIOM treats
the interaction between the model and the real
system consistently at a well defined level of theory. Due to the virtually free choice of method
combinations and system partitioning, ONIOM is
scalable and can be systematically improved. This
allows its successful and reliable application to a
wide variety of chemically interesting problems.
Because of the large flexibility of the ONIOM
method and the availability of the code in the
heavily used Gaussian package, a word of caution
is in order from the authorsrdeveloper of the

20

S. Dapprich et al. r Journal of Molecular Structure (Theochem) 461462 (1999) 121

method to all the users. The method may not


work, or will not produce the targeted high level
results of the real system, if the lower level methods cannot handle the difference between the
real system and the model systems, or the substituent effects properly. For example, we have
shown in single point calculations that HF may
not be enough as the low level and MP2 is needed
to obtain the reliable bond energies, in combination with the CCSDT. or G2-type methods as the
high level. If the target is a density functional
calculation, HF or semi-empirical MO method as
the low level may be quite practical. However, the
choices of the methods to be combined depends
very much on the choice of the models to describe the inner layers as well as the properties to
be calculated: optimized geometries, energies, or
other properties. Therefore, it is imperative that
each user himselfrherself tests, calibrates and
confirms the validity of hisrher choice of method
combination and model for the properties to be
obtained.
In the present paper, we discussed only
the general ONIOM scheme and the ONIOM
MO:MO. and ONIOMMO:MO:MO. combinations. Discussions on our new implementations of
the ONIOMMO:MM. and ONIOMMO:MO:
MM.will be given in the forthcoming paper of the
series w64x.
Acknowledgements
The authors are grateful to Drs. Feliu Maseras,
Stephane Humbel and Max Holtahusen for their
discussions and comments. The present work was
in part supported by grants from Gaussian, Inc.,
Pittsburgh and the National Science Foundation
Grant No. CHE-9627775..
References
w1x R.E. Stratmann, G.E. Scuseria, M.J. Frisch, Chem. Phys.
Lett. 257 1996. 213.
w2x M.C. Strain, G.E. Scuseria, M.J. Frisch, Science 271
1996. 51.
w3x J.P. Dombroski, S.W. Taylor, P.M.W. Gill, J. Phys. Chem.
100 1996. 6272.
w4x J.C. Burant, G.E. Scuseria, M.J. Frisch, J. Chem. Phys.
105 1996. 8969.

w5x C.A. White, B.G. Johnson, P.M.W. Gill, M. HeadGordon, Chem. Phys. Lett. 230 1994. 8.
w6x C.A. White, B.G. Johnson, P.M.W. Gill, M. HeadGordon, Chem. Phys. Lett. 253 1996. 268.
w7x L. Greengard, V. Rohklin, J. Comput. Phys. 60 1985.
187.
w8x N. Koga, K. Morokuma, Chem. Rev. 91 1991. 823.
w9x A. Veillard, Chem. Rev. 91 1991. 743.

w10x J. Aqvist,
A. Warshel, Chem. Rev. 93 1993. 2523.
w11x A.K. Rappe,
C.J. Casewit, K.S. Colwell, W.A. Goddard,
III, J. Am. Chem. Soc. 114 1992. 10024.
w12x J.E. Eksterowicz, K.N. Houk, Chem. Rev. 93 1993.
2439.
w13x J. Gao, in: K.B. Lipkowitz, D.B. Boyd Eds.., Reviews in
Computational Chemistry, vol. 7, VCH, New York, 1995,
p. 119.
w14x D. Bakowies, W. Thiel, J. Phys. Chem. 100 1996. 10580.
w15x D. Bakowies, W. Thiel, J. Comput. Chem. 17 1996. 87.
w16x U.C. Singh, P.A. Kollman, J. Comput. Chem. 7 1986.
718.
w17x M.J. Field, P.A. Bash, M. Karplus, J. Comput. Chem. 11
1990. 700.
w18x U. Eichler, K.M. Kolmel,
J. Sauer, J. Comput. Chem. 18

1996. 463.
w19x F. Maseras, K. Morokuma, J. Comput. Chem. 16 1995.
1170.
w20x S. Humbel, S. Sieber, K. Morokuma, J. Chem. Phys. 105
1996. 1959.
w21x M. Svensson, S. Humbel, R.D.J. Froese, T. Matsubara,
S. Sieber, K. Morokuma, J. Phys. Chem. 100 1996.
19357.
w22x T. Matsubara, F. Maseras, N. Koga, K. Morokuma, J.
Phys. Chem. 100 1996. 2573.
w23x G. Ujaque, F. Maseras, A. Lledos, Theor. Chim. Acta 94
1996. 67.
w24x G. Barea, F. Maseras, Y. Jean, A. Lledos, Inorg. Chem.
35 1996. 6401.
w25x M. Svensson, S. Humbel, K. Morokuma, J. Phys. Chem.
105 1996. 3654.
w26x T. Matsubara, S. Sieber, K. Morokuma, Int. J. Quant.
Chem. 60 1996. 1101.
w27x R.D.J. Froese, K. Morokuma, Chem. Phys. Lett. 263
1996. 393.
w28x E.L. Coitino, D.G. Truhlar, K. Morokuma, Chem. Phys.
Lett. 259 1996. 159.
w29x R.D.J. Froese, S. Humbel, M. Svensson, K. Morokuma,
J. Phys. Chem. 101 1997. 227.
w30x Y. Wakatsuki, N. Koga, H. Werner, K. Morokuma, J.
Am. Chem. Soc. 119 1997. 360.
w31x R.D.J. Froese, D.G. Musaev, K. Morokuma, J. Am.
Chem. Soc. 120 1998. 1581.
w32x R.D.J. Froese, J.M. Coxon, S.C. West, K. Morokuma, J.
Org. Chem., in press.
w33x R.D.J. Froese, K. Morokuma, Hybrid Method. The
ONIOM Method. Integration of Different Levels of
Molecular Orbital Methods andror Molecular Mechanics Methods for Large Molecular Systems and Its Appli-

S. Dapprich et al. r Journal of Molecular Structure (Theochem) 461462 (1999) 121

w34x

w35x
w36x
w37x
w38x
w39x
w40x
w41x
w42x
w43x
w44x
w45x
w46x
w47x

cations to Structures, Energies and Chemical Reactions,


in: P.v.R. Schleyer Ed.., Encyclopedia of Computational
Chemistry, Wiley, 1998, in press.
M.J. Frisch, G.W. Trucks, H.B. Schlegel, G.E. Scuseria,
M.A. Robb, J.R. Cheeseman, M.C. Strain, J.C. Burant,
R.E. Stratmann, S. Dapprich, K.N. Kudin, J.M. Millam,
A.D. Daniels, G.A. Petersson, J.A. Montgomery, V.G.
Zakrzewski, K. Raghavachari, P.Y. Ayala, Q. Cui, K.
Morokuma, J.B. Foresman, J. Cioslowski, J.V. Ortiz, V.
Barone, B.B. Stefanov, G. Liu, A. Liashenko, P. Piskorz,
W. Chen, M.W. Wong, J.L. Andres, E.S. Replogle, R.
Gomperts, R.L. Martin, D.J. Fox, T. Keith, M.A. AlLaham, A. Nanayakkara, M. Challacombe, C.Y. Peng,
J.P. Stewart, C. Gonzalez, M. Head-Gordon, P.M.W.
Gill, B.G. Johnson, J.A Pople, Gaussian98, Gaussian,
Inc., Pittsburgh, PA, 1998.
J.A. Pople, A.P. Scott, M.W. Wong, L. Radom, Israel J.
Chem. 33 1993. 345.
C.W. Bauschlicher, H. Partridge, J. Chem. Phys. 103
1995. 1788.
D.W. Davies, The Theory of the Electric and Magnetic
Properties of Molecules, Wiley, New York, 1967.
R.J. Bartlett, J. Phys. Chem. 93 1989. 1967.
P. Pulay, Chem. Phys. Lett. 100 1983. 151.
S. Saebo, P. Pulay, J. Chem. Phys. 88 1988. 1884.
C. Hampel, H.-J. Werner, J. Chem. Phys. 104 1996.
6286.
X. Assfeld, J.-L. Rivail, Chem. Phys. Lett. 263 1996.
100.
R.D.J. Froese, K. Morokuma, in preparation.
T. Vreven, K. Morokuma, in preparation.
L.A. Curtiss, K. Raghavachari, G.W. Trucks, J.A. Pople,
J. Chem. Phys. 94 1991. 7221.
L.A. Curtiss, P.C. Redfem, B.J. Smith, L. Radom, J.
Chem. Phys. 104 1996. 5148.
J.A. Montgomery, Jr., J.W. Ochterski, G.A. Petersson, J.
Chem. Phys. 101 1994. 5900.

21

w48x G.A. Petersson, M.A. Al-Laham, J. Chem. Phys. 94


1991. 6081.
w49x G.A. Petersson, T.G. Tensfeldt, J.A. Montgomery, Jr., J.
Chem. Phys. 94 1991. 6091.
w50x A. Mebel, K. Morokuma, M.C. Lin, J. Chem. Phys. 103
1995. 7414.
w51x W.J. Hehre, R.F. Stewart, J.A. Pople, J. Chem. Phys. 51
1969. 2657.
w52x J.B. Collins, P.v.R. Schleyer, J.S. Binkley, J.A. Pople, J.
Chem. Phys. 64 1976. 5142.
w53x A.D. Becke, J. Chem. Phys. 98 1993. 5648.
w54x C. Lee, W. Yang, R.G. Parr, Phys. Rev. B 37 1988. 785.
w55x P.J. Stephens, F.J. Devlin, C.F. Chabalowski, M.J. Frisch,
J. Phys. Chem. 98 1994. 11623.
w56x T.H. Dunning Jr., P.J. Hay, in: H.F. Schaefer III Ed..,
Modern Theoretical Chemistry, Plenum Press, New
York, 1976, pp. 128.
w57x T.H. Upton, A.K. Rappe,
J. Am. Chem. Soc. 112 1990.
5633.
w58x N. Koga, K. Morokuma, Chem. Phys. Lett. 172 1990.
243.
w59x P.J. Hay, W.R. Wadt, J. Chem. Phys. 82 1985. 270, 284,
299.
w60x A. Kumar, J.D. Lichtenhan, S.C. Critchlow, B.E.
Eichinger, W.T. Borden, J. Am. Chem. Soc. 112 1990.
5633.
w61x K. Morokuma, W.T. Borden, J. Am. Chem. Soc. 113
1991. 1912.
w62x A. Nicolaides, J.M. Smith, A. Kumar, D.M. Barnhart,
W.T. Borden, Organometallics 14 1995. 3475.
w63x J. Uddin, S. Dapprich, G. Frenking, B.F. Yates, J. Am.
Chem. Soc., submitted.
w64x I. Komaromi,
S. Dapprich, K.S. Byun, K. Morokuma,

M.J. Frisch, in preparation.

Potrebbero piacerti anche