A Dimensionless Reaction Coordinate for Quantifying the

Lateness of Transition States

THOMAS A. MANZ, DAVID S. SHOLL

School of Chemical and Biomolecular Engineering, Georgia Institute of Technology, 311 Ferst
Drive N.W., Atlanta, Georgia 30332-0100
Received 8 April 2009; Revised 4 September 2009; Accepted 13 September 2009
DOI 10.1002/jcc.21440
Published online 11 November 2009 in Wiley InterScience (www.interscience.wiley.com).

Abstract: The Hammond-Lefer postulate asserts that transition states of exothermic reactions are reactant-like
(early), whereas transition states of endothermic reactions are product-like (late). Related postulates have been proposed to describe the sensitivity of activation barriers for reactions occurring on catalytic surfaces to the catalyst
structure. To evaluate the validity of these postulates for different chemical reactions, a general method for classifying transition states as either early or late is needed. One can envision a dimensionless reaction coordinate that
changes continuously and monotonically from 0 to 1 along a minimum energy reaction pathway. The value of the
dimensionless reaction coordinate for the transition state (WTS) classies transition states as (a) early when WTS
\ 0.5, (b) late when WTS [ 0.5, and (c) equidistant between reactants and products when WTS 5 0.5. In this
article, we derive such a dimensionless reaction coordinate and illustrate its usefulness for several different chemical reactions.
q 2009 Wiley Periodicals, Inc.

J Comput Chem 31: 15281541, 2010

Key words: chemical reactions; reaction coordinate; transition state lateness; Hammond postulate; dimensional
analysis

Introduction
A transition state that resembles the reactant state more than the
product state is said to be early, whereas a transition state that
resembles the product state more than the reactant state is said
to be late. Whether a transition state is early or late can sometimes be inferred from a plot of energy versus reaction coordinate along the minimum energy reaction pathway. This direct
approach may not work, however, if atoms in the reactants or
products are separated by an arbitrarily large (innite) distance
or if changes in reaction coordinate are complicated. The development of a general method for classifying transition states as
either or late is challenging because of the different reaction
coordinate types that exist for different types of chemical reactions. For example, the reaction coordinate for NH3 inversion
is the dihedral angle, but the reaction coordinate for H2
dissociation is the interatomic distance. Here, we use the abbreviation MERP to denote the minimum energy reaction pathway
for an elementary reaction within the Born-Oppenheimer
approximation.1
Several heuristic principles relating properties of classes of
reactions to the lateness of transition states have been advanced.
The best known is the Hammond-Lefer postulate (HLP), which
asserts that for an elementary reaction step between two ground

states the transition state will geometrically resemble whichever

ground state is less stable. That is, endothermic reactions are
expected to have late transition states, whereas exothermic reactions are expected to have early transition states.2,3 Sola and
Toro-Labbe found that the relative hardness of reactants compared with products is an important factor in determining
whether or not a reaction follows the HLP.4,5 Gellman et al.
have suggested related ideas for classifying the sensitivity of
barriers for catalytic reactions occurring on solid surfaces to the
structure of these surfaces.68 This proposal involves two separate hypotheses. The structure sensitivity postulate (SSP) asserts
that the activation barriers for catalytic reactions occurring on
solid surfaces are more sensitive to the structure of the surface
if the transition state is late rather than early. The reactant sensitivity postulate (RSP) asserts that the activation barriers for catalytic reactions are more sensitive to the structure of nonreactive

Correspondence to: T. A. Manz and D. S. Sholl; e-mail: thomasamanz@

gmail.com, david.sholl@chbe.gatech.edu
Contract/grant sponsor: National Science Foundation; contract/grant
numbers: CHE 0651182
Contract/grant sponsor: National Center for Supercomputing Applications; contract/grant numbers: TG-CHE090019

q 2009 Wiley Periodicals, Inc.

Quantifying the Lateness of Transition States

groups on the reactant if the transition state is late compared to

early. According to the RSP, if one or more F atoms are substituted for H atoms on nonreactive parts of the reactant, for example, the activation barrier for the forward reaction will change
more than the activation barrier for the reverse reaction if the
transition state is late.
Several reports in the literature tested the validity of the HLP
by constructing theoretical measures of whether a transition state
is early or late. Cioslowski, Zou et al., Amat et al., and Sola
et al. introduced measures of transition state lateness that apply
only to unimolecular reactions and used these to study the HLP
for selected unimolecular systems.912 Haddon and Chow proposed a descriptor of transition state lateness based on orbital
hybridization changes and used this descriptor to study the HLP
for reverse Diels-Alder cycloaddition reactions.13 Although orbital hybridization changes are important for many types of reactions, their descriptor cannot be applied to reactions in which
orbital hybridization changes do not occur. Nalewajski and Broniatowska introduced a measure of information distance from
a transition state to reactants or products by analyzing electron
density distributions in these states.14 Using this information distance to study the collinear reaction H2 1 X ? [H---H---X]{ ?
H 1 HX, they were able to quantify the change from early to
late transition state predicted by the HLP as X varies across the
series F, Cl, Br, and I. In these calculations, they set the center
of mass distance between H2 (HX) and X (H) in the reactants
(products) equal to the corresponding distance in the transition
state. This adjustment of reactant and product positions would
not be possible if the reactants or products were adsorbed on different kinds of surface sites; therefore, the method of Nalewajski
and Broniatowska is not applicable to surface reactions. Another
method used to quantify transition state lateness and test the
HLP is to compare computed bond orders for transition states
with those for reactants and products.1517 In yet another
approach, Arteca and Mezey used a topological analysis of the
van der Waals surface to characterize the relative lateness of
transition states.18 This method is not applicable for condensed
phases that lack a van der Waals surface. In summary, there is
not yet a theoretical measure of transition state lateness that
applies to all types of chemical reactions.
Computational studies of the SSP and RSP have compared
bond distances and angles in the transition state to those in the
reactant and product states, to decide whether the transition state
is early or late.1922 This process was partially ambiguous
because some of the angles and bonds in the transition state
resembled the reactant state more, whereas other angles and
bonds in the transition state resembled the product state more.
In this article, we propose a dimensionless reaction coordinate W for elementary reactions that has a value of 0 for the
reactant state R, has monotonically increasing value along the
MERP, and has a value of 1 for the product state P. Letting WTS
denote the dimensionless reaction coordinate of the transition
state, then if WTS \ 0.5 the transition state is early, if WTS [
0.5 the transition state is late, and if WTS 5 0.5 the transition
state is equidistant between reactants and products. The difference (WTS[A] 2 WTS[B]) would provide a quantitative measure of
how much later the transition state for reaction A lies along
MERP A compared to where the transition state for reaction B

1529

lies along MERP B. We show that this approach can overcome

some of the difculties associated with the methods described
earlier.
It may seem natural to approach this problem by considering
a bond index array. This idea is problematic because the elements of this array must in some cases change discontinuously
along the MERP. This occurs for reactions in which the spin
multiplicity of the product state is different from that of the
reactant state. Where the change in spin multiplicity occurs on
the MERP, the wavefunction and electron density can also
change discontinuously; therefore, bond orders based on the
wavefunction or electron density can have a discontinuity along
the MERP. Because molecular geometry always changes continuously along the MERP, it is preferable to base a dimensionless
reaction coordinate on molecular geometry changes rather than
the wavefunction, electron density, or bond orders, which may
change discontinuously.

Theory
Basic Form of the Dimensionless Reaction Coordinate

Consider a dimensionless reaction coordinate that can be computed using only molecular geometries along a MERP. Let W(Y)
denote the value of this dimensionless reaction coordinate for
geometry Y on the MERP, where Y is the set of Cartesian coordinates specifying the atomic positions. W(Y) should be invariant
to rotations and translations of the whole system; therefore, the
dimensionless reaction coordinate should be a function of the set
of relative distances between atoms in the system. The relative
distance Dij between atom i and atom j is dened as
Dij

q
xi
xj 2 yi
yj 2 zi
zj 2 ;

(1)

where (xi, yi, zi) is the Cartesian coordinate of atom i and (xj, yj,
zj) is the Cartesian coordinate of atom j. A reaction coordinate
measure is said to be dimensionless when it does not depend on
the scale used to measure distances. A homogeneous function is
a function f(x) for which f(gx) 5 gkf(x) for any constant g, and
the exponent k is called the homogenous degree.23 A function
that does not depend on the scale used to measure the independent variable x has homogenous degree zero. Because a dimensionless reaction coordinate is unitless, W(Y) cannot depend on
the distance measurement scale and thus must be a homogenous
function of degree zero in the distance measurement scale.
The eld of differential geometry provides some insights on
how to construct distance functions. Typically, differential distance ds on a differential manifold is described by an equation
of the form
ds

rX
X
gab dXa dXb ;
a

(2)

where gab is a metric tensor and {Xa} are the independent coordinates. A distance function of form (2) is not normalized. To
achieve normalization, the scaled distance from R to state Y on

Journal of Computational Chemistry

DOI 10.1002/jcc

1530

Manz and Sholl Vol. 31, No. 7 Journal of Computational Chemistry

the MERP should be expressed as a fraction of the total distance

along the MERP from R to P. To insure the dimensionless reaction coordinate depends only on changes in relative atomic positions, the independent variables involved in dening the dimensionless distance along the reaction path must be some set {vij}
such that dvij 5 0 if and only if (iff.) dDij 5 0. The constraint
that dvij 5 0 iff. dDij 5 0 implies that Dij is the only independent variable of the dimensionless reaction coordinate vij. Taking
the derivative of vij 5 f(Dij) gives dvij 5 f 0 (Dij ) dDij, which satises this constraint as long as f 0 (Dij) = 0, which in turn
requires f(Dij) to be monotonically increasing or decreasing as
Dij increases. The constraint that f(Dij) is homogeneous in the
distance measurement scale requires ln(f) to be proportional to
ln(distance). To the independent distance variable Dij, we can
add some (possibly zero) constant offset distance Gij to give ln
(f) ! ln (Dij 1 Gij). Exponentiating both sides gives

p

vij Dij Cij ;

(3)

where Gij is an offset distance that is constant for each pair of

atoms and the constant 2p = 0 is the homogeneous degree of
the independent coordinates. (Gij can be a different constant for
different atom pairs.). These considerations suggest the construction of a dimensionless reaction coordinate of the form
R Yq
P P P P
i>j
j
k>h
h gij;kh dvij dvkh
R
WY R q
:
P
P
P
P
P
g
dv
dv
ij;kh
ij
kh
i>j
j
k>h
h
R

(4)

The conditions i [ j and k [ h insure summation over the

N(N 2 1)/2 distinct atom pairs (i,j) and (k,h), respectively,
where N is the number of atoms in the system. All terms in the
numerator and denominator of eq. (4) have the same homogeneous degree with respect to distance measurement scale; this fullls the requirement that the dimensionless reaction coordinate
must have homogeneous degree zero. The integrals in eq. (4)
refer to integration along a MERP.
For convenience, each distinct atom pair can also be denoted
by a single index:
i; j ! a and k; h ! b ;

(5)

where the indices a and b take on one of the values in {1,2,3

. . . N(N 2 1)/2}. This allows us to rewrite (4) as

R YpP
P
a
b gab dva dvb
R
WY R PpP

:
P
a
b gab dva dvb
R

XX
a

(7)

This condition will be satised iff. gab is a positive denite

matrix.24 As only the symmetric part of gab contributes to the
sums in eq. (6), without any loss of generality, we can choose
gab to be a symmetric matrix. Moreover, we seek a solution for
which p and gab are real valued. A real, symmetric matrix is
positive denite iff. all of its eigenvalues are positive.24 We
seek a solution in which all reactions and all atom pairs are
treated on an equal basis; this gives rise to only two constants
gaa 5 c and gab 5 b (a = b) for the diagonal and off-diagonal
components, respectively. As the dimensionless reaction coordinate is normalized, without any loss of generality, we can take
c 5 1. The set of eigenvalues of the matrix g is equal to the set
of scalar values k that satisfy the characteristic equation
detg
kI 0 ;

(8)

where I is the identity matrix. Evaluating this determinant by

cofactor expansion along any row or column gives the following
perturbation series in b:


h
NN
1
NN
1
NN
1
detg
kI 1
k 2


1 b2 1
k 2
2
2
i
NN
1
order b3 1
k 2
3    : 9

When |b|  1 the rst two terms in the series dominate and
inserting (9) into (8) yields
s


NN
1

1 :
(10)
k1b
2

Equation 10 has one negative solution (i.e., k \ 0) for

1
N>
2

s
2 9
:
b2 4

(11)

If b2 [ 0, then there exists some chemical reaction with a

sufciently large number N of atoms to satisfy eq. (11); in this
case, one of the eigenvalues of g is negative (i.e., k \ 0), so g
is not positive denite. On the other hand, if b2 5 0 the term on
the right side of (11) is innite, so that no solution k \ 0 exists
for a nite number of atoms and in such case g is always positive denite. As g is required to be positive denite for a system
containing any nite number of atoms, the value of b must
clearly be set to zero. With these considerations, the dimensionless reaction coordinate takes the form

(6)

To fulll the requirement that W(Y) varies monotonically

from 0 to 1 over the MERP, we require that the differential
distance be positive for any change in independent coordinates:

gab dva dvb > 0 if fdva g 6 f0g:

R Yq
P P
R Yq
P
2
2
i>j
j dvij
R
a dva
R
WY R q
R q
:
P
P P
2
P P
2
i>j
j dvij
R
a dva
R

(12)

The numerator and denominator of eq. (12) have the form of

the most commonly used distance metric, which means that this

Journal of Computational Chemistry

DOI 10.1002/jcc

Quantifying the Lateness of Transition States

common distance metric is the correct one to use in conjunction

with the independent coordinates {vij}.

1531

should roughly approximate the bond length associated with a

state halfway between a covalent single bond and nonbonding
contact. In terms of the variable vAB,

Integral Approximation

Most computational chemistry studies locate a nite number of

geometries along the MERP, and we will refer to known geometries on the MERP as images. Approximating changes in geometry between successive images on the MERP as linear
changes in the set of independent coordinates {vij} allows the
integrals in eq. (12) to be approximated as sums:

PnY q
P P
2
j2
i>j
j vij j
1
vij j

;
WY P qP
P
np
2
j2
i>j
j vij j
1
vij j

bond
bond
p
cov
p
vsingle
CAB Dsingle
CAB Rcov
AB
AB
A RB
(18)

and
half
p
vhalf

AB CAB DAB

p
vdw
cov
cov
CAB Rvdw
A RB RA RB =2 : 19

(13)

where 1, nY, and nP are the image numbers of geometries R, Y,

and P, respectively. The numerator involves the sum of scaled
distances between successive images from R to Y, whereas the
denominator involves the sum of scaled distances along the full
MERP. The dimensionless reaction coordinate of the transition
state, WTS 5 W(TS), is estimated by summing over a series of
images along the MERP from the reactant state to the transition
state.
Offset Distances

How can the constant offset distances {Gij} in eq. (3) be determined? We rst consider the case of diatomic molecules and
then show that the offset distances for these examples extend to
all chemical systems. Consider a diatomic molecule that contains
a covalent bond between atoms A and B. Let f be dened as the
ratio of the equilibrium bond length Dbond
AB and the sum of covalent radii:

For dissociation of a diatomic molecule with a covalent

single bond
single bond: (a) the reactant state has vAB % vAB
, (b) vAB %
half
vAB represents the normalized independent coordinate associated
with a state in which half of the single bond has been dissociated,
and (c) vAB 5 0 represents the limit DAB ? 1. For this type of
reaction, the dimensionless reaction coordinate of eq. (12) has the
simple form
WvAB

vAB R
vAB
vsingle bond
v
ABsingle bond AB :
vAB R
vAB P
vAB

0

(20)

The dimensionless reaction coordinate at which approximately half of the single bond is broken divides the early and
late reaction phases:
bond


vsingle

vhalf
AB
AB
W vhalf
:

0:5

AB
single bond
vAB

0

(21)

Inserting (18) and (19) into (21) and solving gives

Dbond
f cov AB cov :
RA RB

(14)
CAB

p
cov
p
vdw
cov
Rvdw
A RB 1
2 2RA RB
p

p
2 2
1

(22)

The covalent radii are dened such that

f 1

(15)

for a single bond. For higher order bonds, f is experimentally

found to be between about 0.7 and 1.25 The characteristic minimum distance for molecular systems in which atoms A and B
are in contact without forming any bond equals the sum of the
atomic van der Waals radii:
vdw
vdw
Dvdw
AB RA RB :

(16)

We propose that

Dhalf
AB

cov
vdw
vdw
Rcov
A RB RA RB
2

(17)

As the chemical state (charge, oxidation, spin, etc.) of an element can vary during the course of a chemical reaction, while
the {Gij} are constant by denition, a covalent radius independent of the chemical state of the element must be used for the
construction of a dimensionless reaction coordinate. For the rst
96 elements, the recommended covalent radii are those reported
by Cordero et al. based on statistical analysis of more than
228,000 experimental bond lengths from the Cambridge Structural Database.25 For four elements (C, Mn, Fe, and Co), they
reported different covalent radii for separate hybridization or
spin states; for these elements, we recommend averaging their
reported covalent radii over hybridization or spin states to give
.
Rcov equal to 0.73 (C), 1.50 (Mn), 1.42 (Fe), and 1.38 (Co) A
Unfortunately, a set of van der Waals radii has not been
reported that includes most of the elements in the periodic table.
To provide a comprehensive set of van der Waals radii, we

Journal of Computational Chemistry

DOI 10.1002/jcc

1532

Manz and Sholl Vol. 31, No. 7 Journal of Computational Chemistry

recommend Bondis equation estimating the van der Waals

radius from the covalent radius:26
cov

Rvdw
A RA 0:76 A:

(23)

To provide a completely self-consistent set of van der Waals

radii, we recommend using eq. (23) even for elements with tabulated estimates of the van der Waals radii. With this simplication, eq. (22) becomes

0:76 A

Cij p

Rcov

Rcov
i
j :
p
2
1

(24)

Because the product of one reaction can be the reactant for

another reaction, the offset distances for the product of one reaction and the reactant for another reaction can be equal. From
this, we infer that the offset distances for a given value of
p should be the same for both reactions. This means that a pair
of elements must have the same offset distance irrespective of
the chemical reaction. Because for some reactions the chemical
states of i and j may change, while {Gij} must remain constant
during the course of reaction, we infer that {Gij} should be set
equal to a universal constant independent of the chemical states
of elements i and j. Consequently, the offset distances {Gij} can
be estimated via (24) regardless of the molecule or solid under
investigation.
The Value of p

Specifying a value for p is all that remains to x the precise

form of the dimensionless reaction coordinate in eq. (12). Figure
1 shows the independent coordinate value vij as a function of
the distance Dij between atoms i and j for different values of p,
where the coordinate values and distances are referenced to the
covalent radii Ri and Rj of atoms i and j. For small Dij, changes
in p only weakly affect the values of vij and log(vij). For large
Dij, changes in p only weakly affect vij but strongly affect
log(vij) as vij  0. That is, p affects only the long-range scaling
of the independent coordinates {vij}. As the independent coordinate vij decays more slowly with increasing Dij for lower p values, lower p values give comparatively higher weight to large
interatomic distances than higher p values. As shown in the following example, the main implication of this is that for systems
containing an innite number of atoms, a sufciently high value
of p (i.e., p [ 1.5) must be chosen to keep the long-range contributions nite.
A methodology must be devised to set a more specic value
for p. One possibility is to choose a reference reaction in which
the energy changes monotonically along the reaction coordinate
from R to P and dene W(Y) for this reaction such that the slope
of E(Y) versus W(Y) is constant, where E(Y) is the energy along
the MERP. (The chosen reference reaction must have self-similar scaling behavior to enable the construction of a constant
E(Y) versus W(Y) slope.). According to this denition, the
dimensionless reaction coordinate for the reference reaction is
just the fractional change in energy between the reactant and
product states. The only problem with this approach is that two

Figure 1. Effect of p on the value of the independent coordinate vij.

Linear and log axes are used for vij to highlight the behavior of this
coordinate for different atom separations.

different reference reactions exhibiting two different self-similar

scaling behaviors would yield two different values of p.
An approximate upper bound on the value of p can be
derived in the following manner. For two molecules or atoms
separated by a sufciently large distance r and approaching each
other, the van der Waals interaction energy is approximately
proportional to 1/r6. In this case, the dimensionless reaction
coordinate is proportional to 1/rp. For sufciently large r, the
van der Waals interaction energy becomes proportional to the
dimensionless reaction coordinate iff. p 5 6. As among all longrange molecular interactions that scale proportional to 1/rm, the
van der Waals dispersion interaction exhibits one of the highest
values of m,
p6

(25)

is an approximate upper bound on the value of p.

As a reference reaction, we consider the approach of an isolated atom of charge q to a grounded, ideal conductor that occupies the x \ 0 portion of an x,y,z-Cartesian coordinate system.
The MERP corresponds to the ion starting at x 5 1 and traveling parallel to the x-axis until it becomes adsorbed at some distance x 5 min on the conductors surface, where min corresponds to some effective hard-sphere ionic radius. The ion is
attracted to the surface by the formation of an image charge in
the conductor. To a rst approximation, the equilibrium positions of the conductors atoms are not a function of location of
the ion, and this approximation holds whether the conductor is
in a crystalline, amorphous, or liquid state. Furthermore, we consider the case for which the offset distance associated with the
ion and any atom in the conductor is some constant G. Approximating the conductor as a continuous jellium with q atoms per
unit volume, the dimensionless reaction coordinate when the ion
is a distance x 5 from the surface is given by

Journal of Computational Chemistry

DOI 10.1002/jcc

Quantifying the Lateness of Transition States

1533

0s1

p 2
 q
R1 R1 R0
@
k
x2 y2 z2 C
q d
dxdydzA
dk

1
1
1

dk

0s1
:
W

p 2
 q
R min
R1 R1 R0
2
d
2
2
@
A
k
x y z C
dxdydz
dk

1
1
1 q dk
1

(26)

The atomic density q in the numerator and denominator

cancels. Expanding the derivative:

1
0v
!
u

1
uR 1 R 1 R 0
p2 k
x2 k
x2 y2 z2
t
@
p
2p2 dxdydzA
dk
1

1
1
1
2

R
W

k
x y2 z2 C

1
:
0v
!
u

1
2 k
x2 k
x2 y2 z2
R min u
R
R
R
p

@t 1 1 0
p
2p2 dxdydzA
dk

1
1
1
1
2

(27)

k
x y2 z2 C

The following variable substitution

x; y; z; k; s  X; Y; Z; K; L

(28)

is used to rewrite the integral:

0v
1
0
1
u


1
u
2
2
C
Z Z Z
Z Bu
2
2
2
C
Bu 1 1 0 B
C
Bp k
x k
x y z
C
Bu
C
dk
dxdydz
B
C


q
Bu
C
2p2 A
@
1 @t
1
1
1
A
2
k
x y2 z2 C
:

1
0v
0
1
u


1
u
2
2
C
Z Z Z
Z =s Bu
2
2
2
C
C
Bu 1 1 0 B
3
Bp K
X K
X Y Z
C
C
dK:
Bu
s2
p
dXdYdZ
B
C
q
2p2
C
Bu
A
1 @t
1
1
1 @
A
2
C
2
2
K
X Y Z s

As (X,Y,Z,K) are integration (dummy) variables, for the special

case G 5 0 eq. (29) reduces to
r


R
R1 R1 R0 
p2 k
x2
dxdydz

dk
2
p2
1

1
1
1 k
x y2 z2
3
s2
p :
r


2
R1 R1 R0 
R =s
p2 k
x
1

1
1
1 k
x2 y2 z2 p2 dxdydz
dk

Thus, in the case G 5 0 this reaction and its dimensionless reaction coordinate exhibit self-similar scaling; that
is, the approach of an ion to a distance L from the surface

(29)

(30)

is a magnied copy of the approach of an ion to a distance 5 sL from the surface. Comparing eqs. (27) and
(30) gives

Journal of Computational Chemistry

DOI 10.1002/jcc

Manz and Sholl Vol. 31, No. 7 Journal of Computational Chemistry

1534


W

32
p

min

(31)

could be as low as 2. Insomuch as the value of the dimensionless reaction coordinate is not unduly sensitive to the value of p,
an intermediate value of p 5 4 should give good results in
nearly all cases.

As the reactant state corresponds to ? 1 and W(R) 5 0, we

have

lim W lim

!1

!1

min

32
p

0;

(32)

which gives the absolute lower bound

3
p> :
2

(33)

Equation (33) is the constraint required to keep the integrals in

eq. (30) nite. Using the method of image charges, the theoretical energy along the reaction coordinate is given by
E

q2
for > min :
4pe0

(34)

Comparing eqs. (31) and (34), the slope of E(Y) versus W(Y) is
constant iff. p 5 5/2.
The lower bound p [ 1.5 also applies to other systems containing an innite number of atoms. Consider a system with q
atoms per unit volume. Let r be an independent variable equal
to the distance from some reactive atom A in the system. For
image Y on the MERP, we have
Z

Y
R

s
X X 2 Z
dvij
i>j

s
Z Z Z
 
2
q d r Cij
p dxdydz:

Summary

In summary, for an elementary reaction step one can dene a

dimensionless reaction coordinate W(Y) that varies from 0 to 1
along the MERP. This is done by computing a series of geometries
(images) along the MERP. For each pair of atoms, an offset distance is computed via eq. (24) using the covalent radii. The purpose
of the offset distance is to account for the different sizes of different elements. The offset distances, an exponent p, and the Cartesian
coordinates (X,Y,Z) for each atom in each image are then used to
compute the scaled interatomic distances via eq. (3). The value of
p is chosen based on the scaling of long-range interactions, or a
standard value p 5 4 is used. For systems containing an innite
number of atoms, p [ 1.5 is necessary to keep the sums nite.
Among long-range interactions, the dispersion interaction, which
scales as 1/r6, exhibits one of the highest scaling exponents; this
places a practical upper bound of p  6. If the reaction is dominated by short-range interactions, such as covalent bond rearrangement, then the value of p has little effect on the computed W(Y)
values. Once the scaled interatomic distances are computed, W(Y)
is computed by eq. (13). The early reaction phase is characterized
by W(Y) \ 0.5, whereas the late reaction phase is characterized by
W(Y) [ 0.5. For the reactant state W(Y) 5 0, and for the product
state W(Y) 5 1. It follows directly from eq. (12) that Wreverse(Y) 5
1 2 W(Y), where W(Y) and Wreverse(Y) are the dimensionless reaction coordinates of Y for the forward and reverse reactions, respectively. Consequently, the early phase of the forward reaction
becomes the late phase of the reverse reaction, and vice versa. The
dimensionless reaction coordinate of the transition state is called
WTS. A program for calculating this dimensionless reaction coordinate is freely available from http://sourceforge. net/projects/drcs/.

(35)

Results
Dissociation of Covalent Diatomic Molecules

As the volume enclosed within radius r is proportional to r , the

integral under the square root sign scales proportional to r322p
for r Gij. The requirement that this integral remains nite as
r approaches innity gives the general lower bound p [ 1.5 for
systems containing an innite number of atoms.
The Ewald summation method used for calculating the interactions in periodic systems divides energies into short-range and
long-range interactions.27 Terms contributing to the long-range
interaction scale as 1/rm for m
2, because the m 5 1 term
diverges. For long-range dispersion interactions, the scaling goes
as 1/r6. As described earlier, the p value only affects the longrange component of the dimensionless reaction coordinate. The
range of p values described earlier is consistent with the longrange part of the Ewald sum, and the optimal value of p should
be about the same as the exponent in the leading term of the
long-range part of the Ewald sum. For dispersive interactions,
this exponent would be 6, whereas for dipole interactions it

The dissociation of covalent diatomic molecules provides a

straightforward test of the theory. Although the dissociation of
some diatomic molecules may have an intermediate transition
state, more commonly the transition state for this process is the
product state. According to the theory described earlier, dissociation of a covalent diatomic molecule AB should have the following general characteristics:
1. If A and B are joined by a covalent single bond (e.g., H2),
the bond length should be approximately equal to the sum of
the covalent radii of the two atoms [see eqs. (14) and (15)].
If A and B are joined by a covalent multiple bond (e.g., CO),
the bond length should be slightly shorter than the sum of
covalent radii. If A and B are weakly bound (e.g., He2), the
bond length should be greater than the sum of covalent radii.
2. If A and B are joined by a covalent single bond, the dimensionless reaction coordinate should reach 0.5 irrespective of the p

Journal of Computational Chemistry

DOI 10.1002/jcc

Quantifying the Lateness of Transition States

1
1
1
1
3
3
3
5
1
3
3
3

DOI 10.1002/jcc

Spin multiplicity of the low energy state. The isolated atoms had the following spin multiplicities: H (2), He(1), Li (2), Na (2), Cu (2), Ag (2), Au (2),
C (3), and O (3).

99
94
84
83
99
98
99
100
83
99
99
99
0.84
0.75
0.84
0.85
0.86
0.86
0.86
0.86
0.85
0.75
0.75
0.75
0.80
0.62
0.80
0.83
0.86
0.85
0.84
0.84
0.81
0.76
0.75
0.75
0.76
0.51
0.76
0.81
0.86
0.85
0.82
0.82
0.78
0.78
0.76
0.76
1
1
1
1
1
1
1
5
1
1
1
1
94
84
61
60
86
89
96
99
61
88
90
90
0.73
0.62
0.72
0.74
0.75
0.75
0.74
0.74
0.73
0.75
0.75
0.75
0.69
0.50
0.69
0.73
0.76
0.76
0.74
0.74
0.71
0.76
0.75
0.75
0.66
0.40
0.66
0.72
0.78
0.77
0.74
0.74
0.69
0.78
0.76
0.76
1
1
1
1
1
1
1
1
1
1
1
1
61
52
25
24
46
49
57
64
25
47
49
48
0.49
0.40
0.49
0.51
0.52
0.52
0.51
0.51
0.50
0.52
0.52
0.52
0.48
0.31
0.48
0.52
0.56
0.55
0.53
0.53
0.50
0.56
0.55
0.55
0.47
0.24
0.47
0.53
0.61
0.59
0.56
0.56
0.50
0.61
0.59
0.58
1
1
1
1
1
1
1
1
1
1
1
1
4.71
9e-4
1.03
0.72
2.03
1.72
2.29
10.92
0.86
1.87
2.47
2.20
1.20
5.38
1.05
0.96
0.84
0.87
0.91
0.82
1.00
0.85
0.87
0.88
H2
He2
Li2
Na2
Cu2
Ag2
Au2
CO
LiNa
CuAg
CuAu
AgAu

0.74
3.01
2.70
3.18
2.21
2.53
2.48
1.14
2.95
2.36
2.33
2.48

DE/Ed
(%)
W
p56
W
p 5 2.5
DE/Ed
(%)
W
p56
W
p 5 2.5
D0
)
(A

Ed
(eV)

(2s11)a

W
p 5 1.5

D0 1 0.76 A

(2s11)a

W
p 5 1.5

W
p 5 2.5

W
p56

DE/Ed
(%)

(2s11)a

W
p 5 1.5

D0 1 2.28 A

D0 1 1.52 A

Journal of Computational Chemistry

D0

Table 1. Computed Parameters for Dissociation of Covalent Diatomic Molecules.

Computations were performed to test the validity of these

approximations for a series of covalent diatomic molecules.
Energies and geometries were computed using the CCSD(T)
method in Gaussian 03.28 All-electron aug-cc-pVTZ basis
sets2932 were used for H, He, Li, C, O, and Na, and aug-ccpVTZ-PP basis sets33,34 including relativistic small core pseudopotentials were used for Cu, Ag, and Au. Coefcients for
these basis sets and their diffuse functions were obtained from
the EMSL basis set exchange.35,36 As described by Feller and
Sordo, the accuracy of the CCSD(T) method for modeling dissociation of covalent diatomic molecules depends upon the
extent of the multicongurational nature of the molecules.37 If
the perturbative triples contribution to the correlation energy is
large, the CCSD(T) method may not adequately represent the
multicongurational nature of the wavefunction; therefore, we
restricted the set of diatomic molecules in Table 1 to those for
which the perturbative triples contribution to the correlation
energy was 2 eV.
Table 1 shows the computed equilibrium bond length (D0),
bond dissociation energy (Ed), and ratio (f) of D0 to the sum of
covalent radii for 12 covalent diatomic molecules. For molecules
with a single bond, the value of f was between 0.84 and 1.20.
For CO, which has a triple bond, the value of f was slightly
smaller at 0.82. For He2, which is weakly bound, the value of f
was much larger at 5.4. These results conrm the validity of
characteristic 1 above.
Table 1 also gives values of the dimensionless reaction coordinate W when the bond length has been increased by 0.76,
. Note that W 5 0 for D0 (the reactant) and
1.52, and 2.28 A
W 5 1 for innite separation (the product). Table 1 shows that
for these reactions W did not depend signicantly on the specic
value of p in the range 1.5 \ p  6. When the bond length was
, the value of W was 0.5 for each of the
increased by 0.76 A
diatomic molecules with a covalent single or multiple bond.
These results conrm the validity of characteristic 2 above.
For each bond elongation distance, the percentage of bond
, the
energy lost is listed under DE/Ed. For elongation by 0.76 A
bond energy loss was 46 (avg.) 6 14 (st. dev.)%. Thus, for dis
sociation of covalent diatomic molecules, elongation by 0.76 A
results in loss of approximately half the bond energy on average
and separates the early and late reaction phases. This result conrms characteristic 3 above. Self-consistency is obtained because
both the average percentage of bond energy lost and the com sepaputed W values indicate that bond elongation by 0.76 A
rates early from late reaction phases for covalent diatomic disso-

2s11a

beyond
value when the bond length has been extended 0.76 A
its equilibrium value [see eqs. (19), (21), and (23)].
3. A signicant fraction of the bond energy should be lost and a
signicant fraction of the bond energy should be retained
beyond its equilibwhen the bond length is extended 0.76 A
rium value. (Both the early and late reaction phases should
be accompanied by substantial bond energy changes.).
4. The bond energy is mostly lost when the bond length has
beyond its equilibrium value.
been extended 1.52 A
(According to #2 above this amount of bond extension should
occur late in the reaction. Thus, most of the bond energy
should be lost at this point in the reaction.).

1535

Manz and Sholl Vol. 31, No. 7 Journal of Computational Chemistry

1536

Table 2. WTS Values for the Reaction H2 1 X ? H 1 HX.

)
[H1


H2


X]{ distances (A
X5
H
F
Cl
Br
I

)
H


X D0 (A
0.74
0.92
1.28
1.41
1.61

[0.77]b
[0.94]
[1.30]
[1.45]
[1.65]

Erxn (ev)

Ea (eV)

H1


H2

H2


X

H1


X

WTSa

0
21.33 [21.04]
0.16 [0.28]
0.62 [0.82]
1.22 [1.33]

0.42
0.08 [0.32]
0.37 [0.59]
0.65 [0.99]
1.21 [1.43]

0.93
0.77 [0.79]
0.99 [1.01]
1.27 [1.26]
1.53 [1.46]

0.93
1.52 [1.41]
1.43 [1.43]
1.46 [1.49]
1.63 [1.67]

1.86
2.06 [2.20]
2.42 [2.44]
2.73 [2.75]
3.16 [3.13]

1
0.05 [0.04]
1.7 [1.8]
11 [12]
40 [35]

0.50 (exact by symmetry)

0.380.40 [0.410.43]
0.540.56 [0.550.57]
0.630.65 [0.620.64]
0.680.70 [0.660.68]

WTS values over range 1.5 \ p  6.

Values enclosed in [ ] are based on the GVB1112 energies and geometries reported by Dunning.38

ciation; moreover, this is the amount of bond elongation

predicted by the theory for division between early and late reaction phases of covalent diatomic dissociation.
, the weaker sinTable 1 shows that when elongated by 0.76 A
gle bonds tended to lose a smaller percentage of the bond energy
than the stronger single and multiple bonds. For example, Na2
lost only 24% of its bond energy, whereas CO lost 64% of its
bond energy. The stronger bonds tend to lose a higher percentage
of energy earlier (as measured by the W values) in the dissociation
process, whereas the weaker bonds tended to lose their energy
more evenly over a wide range of W values. When elongated by
, all 12 molecules had lost the majority of their bond
1.52 A
energy. For this elongation, the strongest bond (e.g., CO) lost
99% of its energy, the weakest single bond (e.g., Na2) lost 60% of
its bond energy, and the very weakly bound He2 lost 84% of its
bond energy. These results conrm characteristic 4 above and
is late in the dissociation process.
show that elongation by 1.52 A
Self-consistency is achieved because the W values also indicate
is late in the reaction process. Only He2 for
elongation by 1.52 A
, but we know that
p 5 1.5 had W \ 0.5 for elongation by 1.52 A
this is not the correct value of p for the van der Waals interaction
(which dominates the He2 bond); with p 5 6 [see eq. (25)] the
correct behavior W [ 0.5 is recovered. For diatomics with single
or multiple covalent bonds, the value of p had very little effect on
the computed W values, which is consistent with the fact that
these bonds are dominated by short-range interactions. Elongation
led to even higher bond energy losses, states even more
by 2.28 A
product-like, and higher W values.
In this section, we have demonstrated that for a single pair of
atoms the designation of reactant-like (W \ 0.5) or product-like
(W [ 0.5) is reasonable. For reactions involving many different
atoms, each pair of atoms inuences whether the transition state
is more reactant-like or more product-like. The relative inuence
of each atom pair depends on how much the scaled distance
between two atoms changes along the MERP. As shown in eq.
(13), the dimensionless reaction coordinate is just the sum of all
these scaled distance changes along the MERP, normalized to
the total scaled distance change along the MERP.
Test of HLP for Atom Exchange Reactions

Nalewajski and Broniatowska developed information distance

descriptors based on electron densities in the reactant, product,

and transition state to determine the relative lateness of transition states for the reactions H2 1 X ? H 1 HX, where X 5 F,
Cl, Br, and I.14 In agreement with HLP, they found that the transition state for X 5 F (exothermic) was early, whereas the transition states for X 5 Cl, Br, and I (endothermic) were late. To
perform their analysis, Nalewajski and Broniatowska used the
GVB 11 1 2 geometries reported by Dunning,38 which are
displayed in [] in Table 2. Also displayed in Table 2 are the
energies and geometries we computed at the CCSD(T) level of
theory using all-electron aug-cc-pVTZ basis sets2932 on H, F,
and Cl and aug-cc-pVTZ-PP basis sets33,34 including relativistic
small core pseudopotentials on Br and I. Coefcients for these
basis sets and their diffuse functions were obtained from the
EMSL basis set exchange.35,36 The CCSD(T) geometries were
similar to those previously reported by Dunning using the generalized valence bond method. We computed [H


H


F]{ to be
slightly nonlinear, whereas all of the remaining [H


H


X]{
{
were linear. The bond distances in [H


H


X] allow one to
unambiguously classify the transition states as early or late for
these four reactions. Specically, we can dene a parameter
c

DHH TS
DHH R
DH1 X TS
DHX P

(36)

such that c \ 1 indicates an early transition state, whereas c [

1 indicates a late transition state. As expected, the values of c
displayed in Table 2 show that the transition state is early for
X 5 F, equidistant for X 5 H, and late for X 5 Cl, Br, or I.
These ve reactions provide a falsiable test of WTS. If WTS
is a valid descriptor of transition state lateness, then WTS should
be less than 0.5 for X 5 F, equal to 0.5 for X 5 H, and more
than 0.5 for X 5 Cl, Br, and I. As shown in Table 2, the computed WTS values correctly capture this trend.
Table 3 shows the sensitivity of WTS to changes in the covalent and van der Waals radii using the standard value p 5 4.
The second column shows the WTS values when the covalent
radii are decreased by 10% and the difference between the cova . The third collent and van der Waals radii are held at 0.76 A
umn is similar except the covalent radii are increased by 10%.
In the fourth column, the Cordero et al. covalent radii values are
larger than the
used with van der Waals radii that are 0.86 A
covalent radii. The fth column is similar except the van der
larger than the covalent radii. In all
Waals radii are only 0.66 A

Journal of Computational Chemistry

DOI 10.1002/jcc

Quantifying the Lateness of Transition States

Table 3. Sensitivity of WTS to Changes in Atomic Radii.

X5

10% smaller
covalent
radii

10% larger
covalent
radii

Rvdw 2 Rcov

5 0.86 A

Rvdw 2 Rcov

5 0.66 A

H
F
Cl
Br
I

0.50
0.39
0.54
0.62
0.68

0.50
0.39
0.57
0.66
0.71

0.50
0.41
0.55
0.62
0.67

0.50
0.37
0.56
0.66
0.72

of these cases, the changes in atomic radii had little effect on

the computed values of WTS. Most importantly, transition states
that were early remained early, transition states that were equidistant remained equidistant, and transition states that were late
remained late. From this, we conclude that WTS is not unduly
sensitive to the values of the atomic radii.

Test of HLP for Rearrangement of Vinylidenes to

Acetylenes

Isomerization of vinylidene (H2C

C) to acetylene (HC B CH)
is reported to violate HLP, because the reaction is highly exothermic but the transition state is approximately halfway
between reactant and product.39,40 This qualitative assessment of
a transition state approximately halfway between reactant and
product was based on computed bond distances and angles.
Loh and Field computed the transition state energies and geometries for the isomerization of H2C
C, F2C
C, and
HFC
C.39 Using various levels of theory, the computed barriers
for F migration were 3040 kcal/mol, whereas those for H
migration were 13 kcal/mol. The much larger barriers for F
migration were attributed to the migrating F being unbound,
whereas the migrating H remained covalently bound during
migration. Using various levels of theory, all of these reactions
were exothermic with computed reaction enthalpies of 3045
kcal/mol.
WTS is now applied to quantitatively determine whether these
reactions have early or late transition states. The B3LYP-optimized geometries of Loh and Field were used. These are close
to geometries computed from other levels of theory and the
experimental geometries where available.39 The WTS values for
p 5 2.5 are shown in Table 4. This analysis shows that the

1537

transition states for H2C

C and F2C
C isomerization are
slightly late. For HFC
C, the transition state is slightly late
when F migrates and very early when H migrates.
These results can be understood by comparing the X


C


C
angles in the reactant, transition state, and product, where X is
the migrating atom. The reactant has sp2 hybridization and an
X


C


C angle circa 1208. The product has sp hybridization
and an X


C


C angle of 08. In the three late transition states,
the X


C


C angle was less than half that of the reactant, indicating closer resemblance to the product. In the one early transition state, the X


C


C angle was greater than half that of the
reactant, indicating closer resemblance to the reactant. Because
each of these reactions is exothermic, the three reactions with
late transition states violate the HLP, whereas the one reaction
with an early transition state follows the HLP. These ndings
conrm earlier reports of HLP violation for vinylidene isomerization.39,40
The isomerization of HFC
C to form HC B CF highlights a
key limitation of HLP. Namely, HLP predicts whether a transition state is early or late based solely on the reactant and product energies with no regard for the reaction pathway. In this
case, there are two elementary reaction pathways connecting the
same reactant and product states; specically, either the H or F
atom may migrate across the double bond to form the product.
HLP predicts that the transition state for this reaction should be
early irrespective of which pathway is chosen, but the WTS values show that whether the transition state is early or late
depends upon the reaction mechanism. Because different reaction pathways have different transition states, one of the transition states can be early, whereas another can be late even if the
different transition states connect the same reactant and product
states. For isomerization of HFC
C to form HC B CF, the transition state for H migration is early, whereas that for F migration
is late.
b-Hydride Elimination of Alkyl and Alkoxy Species on Cu
and Pt Surfaces

Li et al. used plane wave DFT calculations to study (i) b-H

elimination of ethyl on Cu(100), Cu(110), Cu(111), and Cu(221)
and (ii) b-H elimination of alkoxy species on Cu(111) and
Pt(111)21,22,41 (Fig. 2). Gellman et al. performed experimental
studies of b-H elimination from alkyl and alkoxy species on
Cu(111) and Pt(111).68,4245 By extracting structure-reactivity
coefcients from the experimentally measured activation

Table 4. Vinylidene Isomerization and the HLP.

X


C


C (8)39
Reactant

Product

C
H2C
HFC
C

HC B CH
HC B CF

F2C
C

FC B CF

Migrating
atom (X)

WTSa

rxn type

Agrees
HLP?

TS

react.

H
H
F
F

0.530.55
0.190.21
0.570.59
0.560.57

Exothermic
Exothermic
Exothermic
Exothermic

No
Yes
No
No

54.9
81.5
57.5
58.5

120.4
109.8
133.0
123.0

WTS values over range 1.5 \ p  6.

Journal of Computational Chemistry

DOI 10.1002/jcc

Manz and Sholl Vol. 31, No. 7 Journal of Computational Chemistry

1538

Figure 2. b-H elimination reactions studied by Li et al.21,22 (The

lled rectangle represents the metal surface). (a) b-H elimination
from ethyl and (b) b-H elimination from uorinated ethoxy species.

barriers, Gellman et al. concluded that b-H elimination transition

states on Cu(111) are late and contain a partial positive charge
on the b-carbon. The transition state for b-H elimination of
methoxy was reported to be late on both Cu(111) (endothermic
reaction) and Pt(111) (exothermic reaction), but the transition
state on Cu(111) was found to be later than on Pt(111).46,47 Li
et al. reported that the transition states for b-H elimination of
uorinated ethoxides are late on Cu(111) but could not be
clearly classied as early or late on Pt(111). They also concluded that the transition state was late for b-H elimination of
ethyl on different Cu surfaces (endothermic reaction), but the
structure sensitivity of the forward and reverse reactions was
found to be similar. Specically, for Cu(100), Cu(110), Cu(111),
and Cu(221), the computed activation barriers for the forward
reaction had an average of 0.69 and a standard deviation of 0.05
eV, whereas the activation barriers for the reverse reaction had
an average of 0.56 and a standard deviation of 0.07 eV.
To more explicitly quantify the relative lateness of these
transition states, WTS was applied to the DFT-optimized

geometries given by Li et al.21,22,41 In computing WTS for surface reactions, periodic copies of the surface atoms are generated along the surfaces two lattice directions, whereas periodic
copies of the adsorbate atoms are not used. This corresponds to
the actual periodicity in the real chemical system, but differs
from that used in the DFT slab model where all atoms are periodically replicated along three lattice directions. In general, WTS
should be computed using the periodicity of the real system.
The results are shown in Table 5. The WTS values indicate that
b-H elimination of ethyl over Cu(100), Cu(110), Cu(111), and
Cu(221) has an early transition state. Because the reaction is
endothermic, this disagrees with HLP. Because the span in Ea is
slightly greater for the reverse reaction (0.15 eV) than the forward reaction (0.11 eV), an early transition state is in agreement
with Gellmans structure sensitivity postulate (SSP). As the
structure of the reactant was not varied in this example, Gellmans RSP does not apply.
The structure of the reactant was varied for b-H elimination
of OCH2CH3-xFx over Cu(111), thus providing a test of the RSP.
Because these reactions were endothermic, a late transition state
is predicted by HLP. The WTS values indicate an equidistant
transition state for x 5 0 and a late transition state for x 5 2 or
3. (The geometry for x 5 1 was not available for analysis.)
Because the transition state was not early, the RSP predicts that
Ea for the forward reaction will vary more than Ea for the
reverse reaction as x is varied. This was indeed the case as the
span in computed Ea for the forward reactions was 0.92 eV,
whereas that for the reverse reaction was only 0.07 eV.
A similar set of reactions was studied over Pt(111). In this
case, the reaction was exothermic so HLP predicted an early
transition state. The WTS values also indicated an early transition
state. According to the RSP, the reverse reaction should have a
larger variation in Ea than the forward reaction as x is varied.
The span in computed Ea values for the forward reaction was
0.32 eV, whereas that for the reverse reaction was 0.14 eV.
However, the data point x 5 3 has substantially higher Ea,f and
substantially lower Erxn than x 5 0, 1, and 2. If x 5 3 is

Table 5. Computed WTS Values for b-H Elimination of Ethyl and Fluorinated Ethoxy Species.

Surface

Reactant

Ea,fa (eV)

Erxna (ev)

WTSb

TS
type

Agrees
SSP?

Agrees
HLP?

Agrees
RSP?

Cu(100)
Cu(110)
Cu(111)
Cu(221)
Cu(111)
Cu(111)
Cu(111)
Pt(111)
Pt(111)
Pt(111)
Pt(111)

CH2CH3
CH2CH3
CH2CH3
CH2CH3
OCH2CH3
OCH2CF2H
OCH2CF3
OCH2CH3
OCH2CFH2
OCH2CF2H
OCH2CF3

0.76
0.65
0.69
0.66
1.29 [1.26]c
1.58 [1.52]c
1.75 [1.83]c
0.84
0.80
0.83
1.12

0.12
0.16
0.10
0.13
0.69
1.01
1.10
20.33
20.27
20.20
20.04

0.410.43
0.440.46
0.420.44
0.410.42
0.490.51
0.550.57
0.540.56
0.320.36
0.340.38
0.310.35
0.360.42

Early
Early
Early
Early
Equidistant
Late
Late
Early
Early
Early
Early

Yes
Yes
Yes
Yes
Yes
n.a.
n.a.
Yes
n.a.
n.a.
n.a.

No
No
No
No
No
Yes
Yes
Yes
Yes
Yes
Yes

n.a.
n.a.
n.a.
n.a.
Yes
Yes
Yes
n.a.
n.a.
n.a.
n.a.

DFT computed energies from Li et al.21,22,41

WTS values over the range 2.5  p  6.
c
Experimental data in [ ] from Gellman et al.8
b

Journal of Computational Chemistry

DOI 10.1002/jcc

Quantifying the Lateness of Transition States

1539

Figure 3. The b-boryl elimination process.48

omitted, the span in computed Ea values for the forward reaction

(0.04 eV) is less than that for the reverse reaction (0.14 eV).
Therefore, we conclude the data set was not substantial enough
to conclusively determine whether the forward or reverse reaction is actually more sensitive to the reactant substituents.
A comparison of the Ea values for b-H elimination of
OCH2CH3 on Cu(111) and Pt(111) provided a test of SSP.
Because the transition state was not late on either catalyst, the
SSP predicts a larger difference in Ea for the reverse reaction
compared with the forward reaction. In agreement with the SSP,
the differences in computed Ea were 0.45 and 0.57 eV for the
forward and reverse reactions, respectively. The SSP did not
apply for OCH2CF2H and OCH2CF3 because the transition state
switched between early and late on Pt(111) and Cu(111).
There are reasons to believe that SSP and RSP are more fundamental principles than HLP. As discussed earlier, HLP does
not take into consideration the reaction mechanism. On the other
hand, the reaction mechanism is taken into consideration in the
SSP and RSP. If two elementary reaction pathways connect the
same reactant and product states, it is conceivable that one of
these reaction pathways may have an early transition state,
whereas the other may have a late transition state. This possibility is allowed by SSP and RSP but not HLP. SSP and RSP predict more sensitivity in forward Ea compared with reverse Ea for

the pathway with late transition state as the structures of the catalyst and nonreactive substituents, respectively, are varied.

b-Boryl Elimination Catalyzed by Ru and Os Complexes

Although the SSP has been developed for and applied to catalytic reactions occurring on solid metal surfaces, there does not
appear to be any reason why it could not also be applied to reactions occurring on other kinds of catalysts. Here, we consider
whether SSP correctly describes structure sensitivity for homogeneous catalysis for b-boryl elimination catalyzed by Ru and
Os complexes. DFT-optimized geometries for the reactions
shown in Figure 3 were reported by Lam et al.48 Table 6 shows
the computed WTS values for reactions (a) ? (c) and (d) ? (f).
For both Ru and Os catalysts, the transition state from (a) ? (c)
was early. This reaction violates HLP because it is endothermic
with an early transition state. Because the transition state for this
reaction is early, SSP predicts a larger structure sensitivity for
the reverse reaction. For this reaction, Ea values for the forward
reaction differed by 3.4 (3.4) kcal/mol for the two metals and
differed by 4.7 (4.2) kcal/mol for the reverse reaction. (Free
energies are in parentheses, whereas the electronic energies are
not.) Thus, the computed Ea values agree with the SSP.

Journal of Computational Chemistry

DOI 10.1002/jcc

Manz and Sholl Vol. 31, No. 7 Journal of Computational Chemistry

1540

Table 6. Computed WTS Values for b-Boryl Elimination by Ru and Os Complexes.

Relative energies (kcal/mol)a

Metal
Ru
Os

(a)

(b)

(c)

(d)

(e)

(f)

(a) ? (c) WTSb

(d) ? (f) WTSb

0.0 (0.0)
0.0 (0.0)

15.7 (14.5)
19.1 (17.9)

3.6 (4.0)
11.7 (11.6)

27.4 (24.2)
21.8 (1.4)

12.8 (14.9)
17.8 (20.3)

21.8 (0.2)
22.6 (20.2)

0.400.42
0.440.48

0.600.62
0.440.46

Relative electronic energies and free energies (in parentheses) from Lam et al.48
WTS values over the range 1.5  p  6.

For the reaction of (d) ? (f), the WTS values indicate a late
transition state for Ru and an early transition state for Os. This
result is in agreement with the HLP because the reaction over
Ru is endothermic, whereas that over Os is exothermic. Because
a change in catalyst resulted in a change between early and late
transition state, SSP does not make any prediction in this case
as to whether the forward or reverse reaction has higher structure sensitivity.

Conclusions
The main conclusion of this work is that it is possible to develop
a generally applicable measure of transition state lateness based
on geometry changes along the MERP. Descriptors of transition
state lateness based on the wavefunction, electron density,
atomic charges, or bond orders may not change continuously
along the reaction coordinate, especially if the MERP involves a
change in spin multiplicity. Because molecular geometry
changes continuously along the reaction coordinate for all chemical reactions, a descriptor of transition state lateness that can be
computed directly from molecular geometry changes is desirable. We developed such a descriptor, WTS, that can be computed
directly from a series of images along the MERP. A minimum
of three images is required: the reactant, transition state, and
product geometries. This quantitative measure of transition state
lateness is based on a dimensionless reaction coordinate that
changes continuously and monotonically from 0 for the reactant
to 1 for the product. For early transition states WTS \ 0.5, for
late transition states WTS [ 0.5, and for WTS 5 0.5 the transition state is equidistant between reactants and products. The usefulness of WTS was demonstrated for a series of reactions.
For dissociation of covalent diatomic molecules, the dimensionless reaction coordinate was found to reasonably describe
the transition between early (reactant-like) and late (productlike) geometry. Bond elongation by signicantly less than 0.76
resulted in more than 50% retention of the bond energy on
A
average, and for this displacement range the dimensionless reaction coordinate was less than 0.50 indicating the early reaction
phase. On the other hand, bond elongation by signicantly more
resulted in less than 50% retention of the bond
than 0.76 A
energy on average, and for this displacement range the dimensionless reaction coordinate was more than 0.50 indicating the
resulted in
late reaction phase. Bond elongation by 0.76 A
50% retention of the bond energy on average, and for this dis-

placement the dimensionless reaction coordinate was 0.50 indicating the transition between early and late reaction phases. The
dimensionless reaction coordinate values were found to be relatively insensitive to the value of the exponent p in eq. (3) over
the range 1.5 \ p  6.
Some postulates in chemistry and catalysis are based on the
idea that reactions with early transition states behave differently
than reactions with late transition states. The HLP asserts that
endothermic reactions have a late transition state and exothermic
reactions have an early transition state. The SSP asserts that
activation barriers for surface-catalyzed reactions with late transition states are more sensitive to changes in the surface structure than reactions with early transition states. The RSP asserts
that activation barriers for catalytic reactions are more sensitive
to the structure of nonreactive groups on the reactant if the transition state is late compared to early. A reliable method for classifying transition states as either early or late is needed to use or
evaluate the validity of these postulates.
The reaction H2 1 X ? H 1 HX provided an interesting
test case because with X 5 H the reaction is thermoneutral,
with X 5 F the reaction is exothermic, and with X 5 Cl, Br, or
I the reaction is endothermic. As this reaction is known to follow the HLP,14 a switch from early to equidistant to late transition state occurs when X is varied, and WTS captured this trend.
The rearrangement of vinylidene to acetylene has been
reported to violate HLP,39,40 and WTS conrmed this. We also
showed that F migration in HFC
C and F2C
C violates HLP,
whereas H migration in HFC
C follows HLP. Our ndings are
conrmed by a comparison of angles in the reactant, transition
state, and product geometries.
In some cases, SSP and HLP are in conict. Specically, b-H
elimination of ethyl on different Cu surfaces has been reported as
slightly endothermic and with slightly less structure sensitivity for
the forward reaction than for the reverse reaction.22 Accordingly,
HLP predicts a late transition state, but SSP is valid if the transition state is early but not late. The computed WTS values for this
reaction indicated an early TS, which supports the SSP. A similar
situation was encountered with b-boryl elimination catalyzed by
Ru and Os organometallic complexes. The reaction of (a) ? (c)
in Figure 3 was slightly endothermic and had slightly less structure sensitivity for the forward reaction than for the reverse reaction. The computed WTS values for this reaction indicated an early
TS, which supports the SSP but not the HLP. This suggests that
the SSP may be applied to other types of catalytic reactions
besides those occurring on solid metal surfaces.

Journal of Computational Chemistry

DOI 10.1002/jcc

Quantifying the Lateness of Transition States

Finally, we considered whether reactions with early transition

states are less sensitive to changes in the structure of nonreactive
groups on the reactant than reactions with late transition states.
This was done by examining previously published activation
barriers for b-H elimination of a series of uorinated ethoxy
compounds on Cu(111) and Pt(111). On Cu(111), the computed
WTS values indicated an equidistant TS for ethoxy reactant and
a late transition state for uorinated ethoxy reactants. In agreement with the RSP, activation barriers for the forward reaction
were more sensitive to the nature of the nonreactive substituents
than the reverse reaction. On Pt(111), the computed WTS values
indicated an early transition state, which agrees with HLP.
Because the transition state energies for OCH2CF3 were signicantly different than for the partially uorinated analogs, it could
not be determined whether the RSP is valid for b-H elimination
of uorinated ethoxy reactants on Pt(111).

References
1.
2.
3.
4.
5.
6.
7.
8.
9.
10.
11.
12.
13.
14.
15.
16.
17.
18.
19.
20.
21.
22.
23.
24.

Fukui, K. J Phys Chem 1970, 74, 4161.

Hammond, G. S. J Am Chem Soc 1955, 77, 334.
Lefer, J. E. Science 1953, 117, 340.
Sola, M.; Toro-Labbe, A. J Phys Chem A 1999, 103, 8847.
Toro-Labbe, A. J Phys Chem A 1999, 103, 4398.
Gellman, A. J. J Phys Chem B 2002, 106, 10509.
Gellman, A. J. Acc Chem Res 2000, 33, 19.
Gellman, A. J.; Buelow, M. T.; Street, S. C.; Morton, T. H. J Phys
Chem A 2000, 104, 2476.
Cioslowski, J. J Am Chem Soc 1991, 113, 6756.
Zou, J. W.; Chen, W. C.; Kao, C. L.; Yu, C. H. Chem Phys Lett
2004, 383, 40.
Amat, L.; Carbo-Dorca, R.; Cooper, D. L.; Allan, N. L. Chem Phys
Lett 2003, 367, 207.
Sola, M.; Mestres, J.; Carbo, R.; Duran, M. J Am Chem Soc 1994,
116, 5909.
Haddon, R. C.; Chow, S. Y. J Am Chem Soc 1998, 120, 10494.
Nalewajski, R. F.; Broniatowska, E. Chem Phys Lett 2003, 376, 33.
Lendvay, G. J Phys Chem 1989, 93, 4422.
Maity, D. K.; Majumdar, D. K.; Bhattacharyya, S. P. Theochem
J Mol Struct 1995, 332, 1.
Ponec, R.; Yuzhakov, G.; Pecka, J. J Math Chem 1996, 19, 265.
Arteca, G. A.; Mezey, P. G. J Phys Chem 1989, 93, 4746.
Crawford, P.; Hu, P. J Chem Phys 2007, 126, 194706.
Crawford, P.; Hu, P. Surf Sci 2007, 601, 341.
Li, X.; Gellman, A. J.; Sholl, D. S. Surf Sci 2006, 600, L25.
Li, X.; Gellman, A. J.; Sholl, D. S. J Chem Phys 2007, 127, 144710.
Wesstien, E. W. CRC Concise Encyclopedia of Mathematics;
Chapman & Hall/CRC: Baco Raton, FL, 2003.
Strang, G. Linear Algebra and Its Applications; Thomson Brooks/
Cole: Belmont, CA, 2006.

1541

25. Cordero, B.; Gomez, V.; Platero-Prats, A. E.; Reves, M.; Echeverria,
J.; Cremades, E.; Barragan, F.; Alvarez, S. Dalton Trans 2008, 2832.
26. Bondi, A. J Phys Chem 1964, 68, 441.
27. Smit, B.; Frenkel, D. Understanding Molecular Simulation; Academic Press: London, 2001.
28. Gaussian 03, Frisch, M. J.; Trucks, G. W.; Schlegel, H. B.; Scuseria,
G. E.; Robb, M. A.; Cheeseman, J. R.; Montgomery, Jr., J. A.;
Vreven, T.; Kudin, K. N.; Burant, J. C.; Millam, J. M.; Iyengar, S.
S.; Tomasi, J.; Barone, V.; Mennucci, B.; Cossi, M.; Scalmani, G.;
Rega, N.; Petersson, G. A.; Nakatsuji, H.; Hada, M.; Ehara, M.;
Toyota, K.; Fukuda, R.; Hasegawa, J.; Ishida, M.; Nakajima, T.;
Honda, Y.; Kitao, O.; Nakai, H.; Klene, M.; Li, X.; Knox, J. E.;
Hratchian, H. P.; Cross, J. B.; Bakken, V.; Adamo, C.; Jaramillo, J.;
Gomperts, R.; Stratmann, R. E.; Yazyev, O.; Austin, A. J.; Cammi,
R.; Pomelli, C.; Ochterski, J. W.; Ayala, P. Y.; Morokuma, K.;
Voth, G. A.; Salvador, P.; Dannenberg, J. J.; Zakrzewski, V. G.;
Dapprich, S.; Daniels, A. D.; Strain, M. C.; Farkas, O.; Malick, D.
K.; Rabuck, A. D.; Raghavachari, K.; Foresman, J. B.; Ortiz, J. V.;
Cui, Q.; Baboul, A. G.; Clifford, S.; Cioslowski, J.; Stefanov, B. B.;
Liu, G.; Liashenko, A.; Piskorz, P.; Komaromi, I.; Martin, R. L.;
Fox, D. J.; Keith, T.; Al-Laham, M. A.; Peng, C. Y.; Nanayakkara,
A.; Challacombe, M.; Gill, P. M. W.; Johnson, B.; Chen, W.; Wong,
M. W.; Gonzalez, C.; Pople, J. A.; Gaussian, Inc., Wallingford CT,
2004.
29. Dunning, T. H. J Chem Phys 1989, 90, 1007.
30. Kendall, R. A.; Dunning, T. H.; Harrison, R. J. J Chem Phys 1992,
96, 6796.
31. Woon, D. E.; Dunning, T. H. J Chem Phys 1993, 98, 1358.
32. Woon, D. E.; Dunning, T. H. J Chem Phys 1994, 100, 2975.
33. Peterson, K. A.; Figgen, D.; Goll, E.; Stoll, H.; Dolg, M. J Chem
Phys 2003, 119, 11113.
34. Peterson, K. A.; Puzzarini, C. Theor Chem Acc 2005, 114, 283.
35. Feller, D. J Comput Chem 1996, 17, 1571.
36. Schuchardt, K. L.; Didier, B. T.; Elsethagen, T.; Sun, L. S.; Gurumoorthi, V.; Chase, J.; Li, J.; Windus, T. L. J Chem Inf Model
2007, 47, 1045.
37. Feller, D.; Sordo, J. A. J Chem Phys 2000, 113, 485.
38. Dunning, T. H. J Phys Chem 1984, 88, 2469.
39. Loh, Z. H.; Field, R. W. J Chem Phys 2003, 118, 4037.
40. Petersson, G. A.; Tensfeldt, T. G.; Montgomery, J. A. J Am Chem
Soc 1992, 114, 6133.
41. Li, X. PhD Dissertation, Chemistry Department, Carnegie Mellon
University: Pittsburgh, PA, 2007; p. 1.
42. Dai, Q.; Gellman, A. J. J Phys Chem 1993, 97, 10783.
43. Forbes, J. G.; Gellman, A. J. J Am Chem Soc 1993, 115, 6277.
44. Ye, P. P.; Gellman, A. J. J Phys Chem B 2006, 110, 9660.
45. Ye, P. P.; Gellman, A. J. J Am Chem Soc 2008, 130, 8518.
46. Greeley, J.; Mavrikakis, M. J Am Chem Soc 2002, 124, 7193.
47. Greeley, J.; Mavrikakis, M. J Catal 2002, 208, 291.
48. Lam, K. C.; Lin, Z. Y.; Marder, T. B. Organometallics 2007, 26,
3149.

Journal of Computational Chemistry

DOI 10.1002/jcc