The Location of Transition States:

A Comparison of Cartesian, Z-Matrix,

and Natural Internal Coordinates

JON BAKER* and FORA CHAN

BIOSYM Technologies, lnc., 9685 Scranton Road, San Diego, California 92121-3752

Received 21 March 1995; accepted 5 ]uly 1995

A comparison is made between geometry optimization in Cartesian coordinates,

in Z-matrix coordinates, and in natural internal coordinates for the location of
transition states. In contrast to the situation with minima, where all three
coordinate systems are of comparable efficiency if a reliable estimate of the
Hessian matrix is available at the starting geometry, results for 25 different
transition states covering a wide range of structural types demonstrate that in
practice Z-matrix coordinates are generally superior. For Cartesian coordinates,
the commonly used Hessian update schemes are unable to guarantee
preservation of the necessary transition state eigenvalue structure, while current
algorithms for generating natural internal coordinates may have difficulty
handling the distorted geometries associated with transition states. The widely
used Eigenvector Following (EF) algorithm is shown to be extremely efficient for
optimizing transition states. 0 1996 by John Wiley & Sons, Inc.

matrix formalism, with the Z-matrix serving both

Introduction as a definition of the coordinate system and a
means of inputting the molecular geometry.
The principal factor in the renewed use of
T here has been increasing interest in recent
Years in geometry Optimization
Cartesian coordinates as Opposed to the per-
in
Cartesian coordinates has been the-now wide-
spread-availability of graphical model builders,
which are being utilized to construct not just large
haps more traditional-certainly among quantum molecules, such as proteins necessary in drug de-
chemists-use of internal coordinates. The latter
are usuallY via the well-known '- sign, but also smaller systems amenable to
investigation using quantum mechanics. Because
the final geometry resulting from the model-build-
*Author to whom all correspondence should be addressed
at Department of Chemistry, University of Arkansas, Fay- ing process-Often a preliminary
etteville, Arkansas 72702. mization using a molecular mechanics force field

Journal of Computational Chemistry, Vol. 17, No. 7, 888-904 (1996)

0 1996 by John Wiley & Sons, Inc. CCC 0192-8651 196 1070888-17
 LOCATION OF TRANSITION STATES

-is a set of Cartesian coordinates, it is clearly connects the transition state to the reactants and
advantageous if the higher-level (quantum) opti- products of interest).
mization be done directly in Cartesians. It is now Location of transition states is also hindered by
well established that minimization in Cartesian and the fact that, unlike minima, there are no generally
Z-matrix coordinates is of comparable efficiency available force fields suitable for determining tran-
provided a reliable estimate of the Hessian (second sition state geometries or, indeed, model builders
derivative) matrix is available at the (reasonable) capable of building suitable starting structures with
starting geometry.’ A simple mechanics Hessian is the same ease as for normal, stable molecules. In
usually perfectly adequate for this purpose. many cases an initial geometry is constructed us-
An alternative to both Cartesian and Z-matrix ing chemical intuition, and the search proceeds
coordinates are the so-called natural internal coor- from there. This inevitably means that starting
dinates, proposed in this context by P ~ l a y .These
~,~ geometries for transition state searches are typi-
involve linear combinations of bond angles and cally much worse (i.e., farther from the final con-
torsions as deformational coordinates constructed verged structure) than are those for minimization.
(using approximate local symmetry around certain A commonly employed tactic is to select a suitable
atomic centers or in certain regions of the molecu- (internal) coordinate as the reaction coordinate and
lar system, e.g., rings) in such a way so as to do a limited potential scan along this coordinate,
minimize the coupling between the various coordi- with a full minimization in the space of the re-
nates. One clear advantage internal coordinates maining variables at each point. The highest en-
have always possessed over Cartesians is that a ergy point along this ”reaction path” is then taken
well-chosen set of internals is much more decou- as the starting guess for the transition state. No
pled than the corresponding Cartesian coordinates, matter what method is used to derive an initial
and natural internal coordinates exploit this ad- guess for the transition state geometry, a reliable
vantage to the full, providing coordinates that are Hessian must be provided, at least as far as the
essentially decoupled to the maximum extent pos- active site is concerned.
For natural internal coordinate optimizations,
sible. What this means in practice is that optimiza-
the geometrical distortions from standard struc-
tions in natural internals are much less sensitive to
tures associated with transition states are a poten-
the quality of the starting Hessian than are Carte-
tial source of problems. Natural internals were
sian optimizations and are especially insensitive to
designed to exploit local symmetry (e.g., the bond
off-diagonal (coupling) Hessian matrix elements.
angle around a formally sp3 hydridized carbon
Furthermore, natural internal coordinates can be
center is approximately tetrahedral regardless of
generated directly from Cartesian coordinates
the functional groups attached), and the fact that
themselves, and so-although they are somewhat
this no longer holds around the active site in a
complicated-no user input is required to define transition state could cause algorithms designed
them. However, it has again been demonstrated for optimizing minima to generate an inappropri-
(exactly analogous to Z-matrix coordinates) that, ate or unsuitable set of natural internal coordinates
with a reliable starting Hessian, minimization in for optimizing transition states. Additional atom
Cartesian and natural internal coordinates is also connectivity data may also be needed (e.g., for a
of comparable effi~iency.~ Additionally, there are dissociation reaction) in order to generate a full set
certain advantages in using Cartesian coordinates of coordinates in cases where distances between
when searching for local (metastable) minima: atoms are so long as to preclude the assignment of
Transition states are typically far more difficult a formal bond.
to locate than minima due almost entirely to their Given a starting geometry and a suitable Hes-
much greater sensitivity to the Hessian eigenvalue sian, this article presents a comparison between
structure; this is true regardless of the coordinate optimizations carried out in Cartesian, Z-matrix,
system used to carry out the optimization (Carte- and natural internal coordinates, contrasting their
sians or internals). At a stationary point corre- relative efficiencies in locating the final optimized
sponding to a transition state, the Hessian matrix transition state geometry for a test suite of 25
must have one and only one negative eigenvalue; reactions. In this regard, this work is a follow-up
furthermore, it must be the right negative eigen- to a previous article published in this journal4
value (i.e., the corresponding eigenmode must be which presented a similar comparison between
such that deformation of the molecular structure Cartesians and natural internals for minimization.
along this direction on the potential energy surface As in the previous study,4 all starting geometries

JOURNAL OF COMPUTATIONAL CHEMISTRY 889

BAKER AND CHAN

(as Z-matrices) and initial and final energies are rearrangements and the dissociation of formalde-
reported along with the number of cycles required hyde), and the remainder were culled from the
to reach convergence. The test suite encompasses a (mainly recent) literature.
wide range of reactions and transition structures Four reactions in particular-rearrangement of
and, it is hoped, will provide a standard by which the methoxy radical, ring opening of the cyclo-
to judge the usefulness of future developments in propyl radical, ring opening of bicycle[ 1.l.Olbutane,
algorithms for locating transition states. and 1,2-migration in the P-(formyloxy) ethyl radi-
cal-were taken from a recent publication by
Bofill." In ref. 11, Bofill uses a restricted step
Methodology algorithm to locate transition states and compares
the efficiency of this method with that of the EF
algorithm as implemented in the MOPAC'2 and
All optimizations, for all coordinate systems,
GAMESSI3 programs. He also proposes an alterna-
were carried out using the Eigenvector Following
tive Hessian update scheme-a linear combination
(EF) algorithm5 at the Hartree-Fock (HF) level (re-
of the well-known Powell'4 and M~rtagh-Sargent'~
stricted for closed-shell systems, unrestricted for
update formulas-which is claimed generally to
open-shell) with the 3-21G basis set. The EF algo-
perform better than either of the two standard
rithm, based on the work of Cerjan and Miller6
updates separately.
and Simons and co-workers? has been imple-
Two of the four reactions taken from ref. 11
mented in virtually all of the major computational
were studied by Bofill at the UHF/3-21G level,
quantum packages and is widely used to locate
both transition states and minima. and we have used Bofill's starting geometries in
Calculations were performed using the general- order to make a direct comparison with his work.
purpose optimization package OPTIMIZE' in con- We also investigate Bofill's hybrid Hessian update
junction with TURBOMOLE? Natural internal scheme, comparing its performance with that of
coordinates were generated automatically from the Powell and Murtagh-Sargent updates alone.
starting Cartesians using the algorithm built into
OPTIMIZE (see ref. 4 for more details). Although
they have been well tested for minima, natural Results and Discussion
internals are not commonly used for transition
state optimizations and-as noted in the Introduc- The reactions in the test suite cover a range of
tion-the geometrical distortions from standard different types, including transition states for dis-
bond lengths and angles encountered in transition sociation, insertion, rearrangement (e.g., hydrogen
states could cause problems, both in generating shifts), ring opening, and rotation. Most have low
and maintaining a suitable coordinate set. This or no symmetry.
will be particularly apparent for dissociation reac- For each reaction transition state optimizations
tions, in which one molecule forms two (or more) were carried out in Z-matrix, natural internal, and
fragments, with subsequent bond breaking and Cartesian coordinates. Within each coordinate set,
loss of formal bond connectivity. Consequently, for four optimizations were performed on each sys-
some of the systems studied, partial or tem, each of which varied in the method used to
forming/breaking bonds were formally taken to update the Hessian (second derivative) matrix: (1)
be full bonds and were added to the atom connec- An exact Hessian matrix was calculated at each
tivity list; this was done so that OPTIMIZE could cycle (no updates were applied); (2) the Hessian
successfully generate a full set of natural internal matrix was updated using the Powell updateI4; (3)
coordinates. (Without such additional connectivity the Hessian was updated using the Murtagh-Sar-
data, natural internals would be generated based gent (MS) updateI5; and (4) the Hessian was up-
on standard bond lengths and there would be an dated using the hybrid Powell-MS scheme pro-
insufficient number of formal bonds to generate a posed by Bofill." In every case an exact Hessian
full set of internals.) was always calculated at the starting geometry
Of the reactions in the test suite, several were (cycle 1). A total of 12 transition state optimiza-
taken from work done as part of a general valida- tions were thus undertaken for each of the 25
tion of density functional methods"; there are reactions in the test suite.
three commonly used standards (the hydrogen For each system, the Z-matrix optimizations
cyanide/isocyanide and acetylene/ethenylidine were done first. In several cases starting geome-

890 VOL. 17, NO. 7

 LOCATION OF TRANSITION STATES

known transition structures to prevent too rapid quired to reach convergence is essentially a "best-
convergence and hence little or no difference in case" scenario which can be used to judge the
optimization performance between the various co- performance of the algorithm/updating scheme in
ordinate systems. Starting Cartesian coordinates the other cases.
were generated directly from the Z-matrices (with Looking at the Z-matrix results (Table 10, we
appropriate reorientation of axes where necessary see that all optimizations converged in eight cycles
to exploit molecular symmetry), and natural inter- or less and typically around five cycles. This was
nal coordinates were in turn generated directly the case even for the five systems with an initially
from the resultant Cartesians. As mentioned in the incorrect Hessian eigenvalue structure (Table 111).
previous section, additional connectivity data was Furthermore, once the optimization algorithm had
supplied in several cases in order to generate a full moved the system into a region of the potential
set of natural internals. Starting Z-matrices, initial energy surface (PES) with the required one nega-
energy, and final energy (at convergence) for each tive eigenvalue, it never left it (i.e., once the correct
transition state are given in the Appendix. Hessian eigenstructure was achieved, it was main-
Convergence criteria were the standard defaults tained to convergence). (This was also a feature of
of 0.0003 au on the maximum gradient component exact Hessian optimizations in Cartesian and natu-
(Cartesian or internal where appropriate) and ei- ral internal coordinates as well.)
ther an energy change from the previous cycle of The natural internal coordinate results are gen-
less than l o p 6 hartree or a maximum predicted erally very similar to the Z-matrix ones; however,
displacement of less than 0.0003 au per coordinate. there were a couple of systems (13 and 15, Table
Optimizations in internal coordinates-both Z- 11) for which the natural internal optimization re-
matrix and natural internals-involve transforma- quired three more cycles to converge than the
tions of gradients and Hessian matrices from corresponding Z-matrix optimization and three
Cartesian coordinates-in which they are initially systems (6, 16, and 24) where the optimization
calculated -to the appropriate internal coordinate failed.
set. Such transformations require construction of Natural internal coordinate optimizations can
fail in essentially three ways (other than simply
the Wilson B-matrix.I6 Transformation of the Hes-
failing to locate a stationary point in the maximum
sian matrix to internal coordinates included the
allowed number of cycles), all of which have to do
gradient term derived from the derivative of the
with the coordinates themselves:
B-matrix. tnclusion of this term could change the
eigenvalue structure of the transformed Hessian;
in particular, initial starting Hessians in different 1. The assigning algorithm used to generate the
coordinate systems-although derived from the natural internals from the input Cartesians is
same original Cartesian Hessian-need not pos- unable to generate a full set of linearly inde-
sess the same number of negative eigenvalue~.'~ pendent coordinates due, for example, to a
The reactions investigated in this work together complicated or undetermined molecular
with notes and references are shown in Table I. topology.
Optimization results (i.e., number of cycles to reach 2. During the optimization a bond or dihedral
convergence) are given in Table 11. Table I11 gives angle becomes ill defined and there are prob-
the number of negative eigenvalues in the starting lems forming the B-matrix.
Hessian within each coordinate system. This shows 3. The iterative procedure for transforming back
whether or not the optimization was started in a from natural internal to Cartesian coordi-
region of the potential energy surface with the nates fails to converge and a new set of
right local curvature (transition states have one Cartesian coordinates is unavailable for the
and only one negative eigenvalue in the Hessian next cycle.
matrix).
We turn first to the results with an exact Hes- The first of these-the inability to generate a com-
sian matrix calculated at every cycle. These results plete coordinate set-can be regarded as a weak-
provide an excellent standard since-with an ex- ness of the assigning algorithm which could poten-
act Hessian-we are directly comparing the coor- tially be eliminated by adding more options to the
dinate system uninfluenced by weaknesses or id- code, but even if coordinates could be generated in
iosyncracies in the various Hessian updating every possible situation, one would still have
schemes; additionally, the number of cycles re- problems 2 and, especially, 3 to deal with. The

JOURNAL OF COMPUTATIONAL CHEMISTRY 891

BAKER AND CHAN

TABLE I.
Sources and Referencesfor the 25 Reactions Investigated.

-
Reaction Investigated Sources and References

1. HCN HNC Starting geometry from Baker5

2.
3.
4. CH,O
--
HCCH-CCH,
H,CO H, +
CH,OH
CO
Starting geometry from Baker5

Starting geometry from Bofill”

5. ring opening cyclopropyl Starting geometry from Bofill”
6. ring opening bicyclo[l . I .O]butane TS 1 Starting geometry from Bofill”
7. ring opening bicyclo[l .I.Ol butane TS 2 Starting geometry from Bofill”

- -
8. 1,2-migration -(formyloxy) ethyl Starting geometry from Bofill”
9. +
butadiene ethylene cyclohexene Parent Diels-Alderreaction; Baker et aLO
i
10. s-tetrazine 2HCN N, + Baker et al.”; Scheiner et a1.I’
11. trans-butadiene ++ cis-butadiene Baker et al.”
12.
13. CH,CH,F --
CH,CH, cf CH,CH, + H,
CH,CH, + HF
Wolfe and KimIg
Wolfe and Kim”

-
14. vinyl alcohol acetaldehyde Keto-enol tautomerism; Baker et al.”
15. HCOCI cf HCI + CO Jasien”
16. H,O + PO; H,PO; Ma et al.”
17. CH,CHCH,-0-CHCH, ~fCH,CHCH,CH,CHO Claisen rearrangement;
starting geometryfrom Peng and SchlegelZ2
18.
19.
20.
HNCCS
HCONH;
--
SiH, + CH,CH, cf SiH,CH,CH,
HNC + CS
NH,‘ + CO
Silylene insertion; Bach et al.’,
Flammang et al.24
Lin et

--
21. rotational TS in acrolein
22. HCONHOH c-* HCOHNHO Ventura et a1.26
23. +
HNC H, H,CNH Part of methylenimine PES; Andzelm et al.’7
24. H,CNH HCNH, Part of methylenimine PES; Andzelm et aLZ7
25. HCNH, cf HCN H, + Part of methylenimine PES; Andzelm et

essential problem here-as already mentioned in package is to switch from internal to Cartesian
the Introduction-is that a perfectly good set of coordinates in such cases and continue the opti-
natural internals generated at the starting geome- mization in Cartesians. Thus in real production
try can easily become unsuitable and badly be- work, OPTIMIZE will usually find a stationary
haved as the geometry distorts away from stan- point irrespective of any problems with the coordi-
dard structures. Of the transition states studied nate system.
here, system 16 failed to generate a suitable set of Looking now at the optimizations in Cartesian
internals to begin with, and both systems 6 and 24 coordinates, we see that in no case did any Carte-
failed during the iterative back-transformation to sian optimization converge in fewer cycles than
Cartesian coordinates. the corresponding Z-matrix optimization, and typ-
Note that the very definition of the Z-matrix-in ically there is a slight increase in the number of
terms of simple single internal variables- cycles required to achieve convergence. In several
eliminates 1 as a problem, and 3 is also avoided, cases (9, 10, 11, 15, 18, and 19) the Cartesian
since Cartesians can be generated directly from a optimization took three or more cycles more than
given Z-matrix without an iterative procedure; the Z-matrix optimization, with systems 10 (the
however, 2 can often occur if some thought is not dissociation of s-tetrazine) and 11(rotation in buta-
given to Z-matrix construction (e.g., a bond angle diene) taking eight and seven more cycles, respec-
going above 180” causing the optimization to tively.
abort). Why should the Cartesian optimizations take
Although situations 1-3 can occur during any more cycles to converge than the internal coordi-
optimization in natural internal coordinates and nate optimizations (more than double the number
will-for the purposes of this work-cause the of cycles for systems 10 and 11, the two worst
optimization to fail, the default in the OPTIMIZE cases)? One possible reason could be the tendency

892 VOL. 17, NO. 7

 LOCATION OF TRANSITION STATES

TABLE II.
Number of OptimizationCycles to Reach Convergence for the Systems Shown in Table I Using 2-Matrix,
Natural Internal, and Cartesian Coordinates with (1) Exact Hessian Calculated at Each Cycle; (2) Powell
Update; (3)Murtagh-SargentUpdate; and (4) Hybrid Powell-MS Update (see Text for More Details).

Exact Hessian Powell Update MS Update Powell-MS Update

System Z-Mat. Nat. Int. Cart. Z-Mat. Nat. Int. Cart. Z-Mat. Nat. Int. Cart. Z-Mat. Nat. Int. Cart.

1. c, 6 6 7 9 9 10 13 13 9 9 10
2. c, 5 5 6 8 7 9 6 6 8 7 9
3. c, 7 7 7 17 16 13 11 12 13 16 12
4. c, 5 4 5 12 7 7 7 6 11 6 7
5. c, 8 8 9 12 13 21 - 17 11 13 21
6. C, 5 - 5 8 43 8 9 - 7 44 8
7. c, 6 6 7 9 19 24 10 13 9 15 20
8. C, 5 5 6 31 - 18 12 11 26 - 17
9. c, 6 7 10 13 16 24 - -
13 16 25
10. c, 6 6 14 17 10 19 9 - 14 10 19
11. c, 7 6 14 8 8 31 8 8 8 8 29
12. c, 5 5 6 12 10 9 - - 12 11 9
13. C, 6 9 8 12 12 11 - - 11 11 11
14. C, 8 8 9 14 13 18 20 14 13 13 16
-
15. C, 7 10 10 13 28 17 ~

12 20 17
16. C, 7 - 8 24 - - - -
23 23 19
17. C, 5 5 7 9 7 25 8 14 8 7 16
18. C, 6 5 9 7 10 21 - 8 7 11 21
19. c, 6 6 10 10 - 34 12 - 10 - 33
20. c, 6 6 7 10 10 14 - 14 10 10 14
21. c, 4 4 4 4 4 5 5 5 4 4 5
22. c, 4 4 5 5 5 6 6 5 5 5 6
23. C, 4 4 5 8 8 8 8 8 8 8 8
24. C, 8 -
8 14 - 16 __ - 14 - 18
25. C, 5 5 5 9 9 10 9 11 9 9 10

The last three systems (23-25; see Table I) actually have C, symmetry although the starting geometries were in all cases C,. A
dash indicates failure to converge within 50 cycles (the maximum allowed); optimizations in natural internal coordinates usually
failed due to problems with the coordinate set (see text for more details).

for step sizes in internal coordinate space to be 0.263) in Cartesian space than for the Cartesian
much larger when converted into Cartesian space optimization (again the maximum allowed value
(this has already been commented on in ref. 4 for of 0.3 a d .
natural internals, and the same holds for Z-matrix The main reason for the differences in the con-
coordinates). Thus the internal coordinate opti- vergence rate between Cartesians and internals is
mizations could simply take larger steps than the simply-as previously discussed by Pulay and
corresponding Cartesian optimizations and hence co-workers3-the generally much reduced cou-
get to their destination in fewer cycles. While this pling between the coordinates in the latter case. In
may be a contributory factor for the rotation in the quadratic region of the PES (i.e., near the
butadiene (system 11)-for which, for example, transition state), then-since we have an exact
the first four steps in natural internal coordinates, Hessian-Cartesian coordinates are as good as any
converted into Cartesian displacements, are 0.705, other coordinate set, but farther away from the
0.858, 0.985, and 0.670, respectively, compared to transition state anharmonic terms will become
only 0.3 au (the maximum allowed default) in the more important and their effect will be greater in
Cartesian optimization-it is definitely not a factor Cartesian coordinates due to the increased cou-
for the tetrazine decomposition (system 101, for pling. It is no coincidence that the two worst cases
which-due to the composite nature of the natural we have highlighted both have starting Hessians
internal set-the first four steps in internal coordi- with all positive eigenvalues (see Table 111) (i.e.,
nates are actually less (0.248, 0.246, 0.257, and nominally closer to minima than to transition

JOURNAL OF COMPUTATIONAL CHEMISTRY 893

BAKER AND CHAN

TABLE 111. between the various internal coordinates, with

Number of Negative Eigenvalues in the Original several large off-diagonal Hessian matrix elements
(Exact) Starting Hessian for the Systems Shown in often of greater magnitude than the diagonal ele-
Table I.
ment in the same column. Additionally, the lowest
Number of Negative Eigenvalues positive Hessian eigenvalue (when it was positive)
in Starting Hessian Matrix was very small and hence did not require too
System Z-Matrix Nat. Int. Cartesians
much error in the update to change sign.
We can make a fairly general observation that
8. 2 2 2 an updated (as opposed to an exact) Hessian is
10. 0 0 0 likely to be much less efficacious when (1) the
11. 0 0 0 coordinate set is highly coupled (several large off-
14. 1 1 2 diagonal Hessian matrix elements); and (2) there
15. 1 1 0 are one or more low energy (floppy) modes (Hes-
16. 2 - 3 sian eigenvalues with small magnitude which
18. 1 1 2 can easily change sign during the update). The p-
23. 2 2 1
(formyloxy) ethyl radical is a fairly extreme exam-
24. 1 1 2
25. 1 1 2
ple (it has both of these characteristics).
Since Cartesian coordinates are usually more
If no values are given for a particular reaction, then the coupled than internals, we would expect Cartesian
starting Hessian had the correct local curvature (1 negative optimizations using updated Hessians to show an
Hessian eigenvalue).
even worse performance vis 2 vis an exact Hessian
than optimizations in internal coordinates, and this
is generally borne out by the results in Table 11.
states). Thus, for the Cartesian optimizations, we Compared to the exact Hessian results, although
are simply taking poorer quality steps on the PES there are many Cartesian optimizations using the
than in internal space and hence, not surprisingly, Powell update that show an increase in the num-
we have to take more of them to locate the desired ber of cycles to reach convergence by 50-loo%,
stationary point. there are also several that show a larger increase
Turning now to the Z-matrix optimizations with than this, and there are more of these than for
the Powell update, we see-not unexpectedly-a the Z-matrix optimizations. Of particular note are
significant increase in the number of cycles re- systems 7, 8, 11, 16, 17, and 19. All of these sys-
quired to reach convergence, reflecting the approx- tems show exactly the same symptoms (although
imate nature of the Hessian. The increase is typi- not as marked) as the Z-matrix optimization of
cally 50-100%, but in a few cases it is considerably p-( formyloxy) ethyl-namely, a change in the Hes-
more than this; the worst examples are the 1,2- sian eigenvalue structure (from one to two nega-
migration in the p-(formyloxy) ethyl radical (sys- tive eigenvalues) several times during the course
tem 8, increasing from five cycles with an exact of the optimization with a consequent reduction in
Hessian to 31 with the update) and the addition of the rate of convergence due to the poor steps taken
water to the phosphate anion (system 16; increas- using a Hessian which (often wrongly) had the
ing from seven to 24 cycles) (i.e., by factors of wrong number of negative eigenvalues.
more than 5 and 3, respectively). Both these sys- Comparing the Powell-update Z-matrix and
tems had starting Hessians with two negative Cartesian optimizations with each other, we see
eigenvalues, and examining the Hessian eigen- that although there are several systems for which
value structure at each optimization cycle shows the two coordinate sets show a similar perfor-
that this never really stabilized to having the de- mance, in general the Z-matrix optimizations are
sired one negative eigenvalue until very late in the significantly better, and in some cases dramatically
optimization. This was particularly marked for so. There are five systems for which the Cartesian
system 8, which oscillated continuously between optimization takes more than twice as many cycles
having one and two negative Hessian eigenvalues to converge as the corresponding Z-matrix opti-
until cycle 26 (on 16 cycles out of the first 25 there mization (systems 7, 11, 17, 18, and 19).
were two negative eigenvalues in the Hessian ma- Having said that, however, there are three sys-
trix). Examination of the starting Hessian for this tems for which the Z-matrix optimization takes
system revealed that, for the particular Z-matrix four or more cycles more to converge than the
used (see the Appendix), there was major coupling equivalent Cartesian optimization. One of these is

894 VOL. 17, NO. 7

 LOCATION OF TRANSITION STATES

the P-(formyloxy) ethyl radical, already discussed However, it doesn't take too much chemical intu-
(in which the Cartesian optimization performs a ition to realize that as the system dissociates and
lot better); the other two are the decomposition of the CS fragment leaves, the HNC angle-which
formaldehyde (system 3) and the rearrangement of has an initial value of 14C-i~ likely to head
the methoxy radical (system 4). For these two toward 180" since the HNC molecule is linear.
systems, optimizations in both sets of coordinates Bond angles very close to 180" are bad news for
(Z-matrix and Cartesian) have starting Hessians geometric conversion algorithms, and what actu-
with the required one negative eigenvalue. The ally happens in this case with the updated Hessian
Cartesian optimizations converge smoothly with is that a step is taken during the optimization that
no change in the Hessian eigenvalue structure. The takes the HNC angle-in the sense shown in the
Z-matrix optimizations, on the other hand, show diagram in the Appendix-above 180", causing the
smooth convergence initially, but midway through back conversion to Cartesian coordinates to fail.
the optimization there is a bad Hessian update Consequently, the natural internal coordinate opti-
which changes the eigenvalue structure (from one mization aborts at the end of cycle 3. When con-
to zero negative eigenvalues on cycle 6 for the structing the initial Z-matrix, this possibility is
methoxy radical, and from one to two negative allowed for by introducing a dummy atom to
eigenvalues on cycle 10 for formaldehyde). The define the HNC bond angle, and the Z-matrix
optimization then takes three or four cycles to optimization converges smoothly.
recover from the bad step taken with the wrong Looking at the optimizations using the
signature Hessian, and it is this one bad update Murtagh-Sargent update alone (Table 111, we can
that is responsible for the poor performance of the quickly dismiss this update as being unsuitable for
Z-matrix optimization compared to the corre- general use during transition state searches; there
sponding optimization in Cartesian coordinates. are simply too many cases which fail to converge
The natural internal coordinate optimizations (13 out of 25 Cartesian optimizations failed to
with the Powell update generally show a similar converge within the maximum allowed 50 cycles;
performance to the corresponding Z-matrix opti- more than half of the systems studied). These
mizations, although there are some systems that failures are due entirely to the poor-quality Hes-
perform better (e.g., 4 and 10) and some that are sian update. Interestingly, for some of the Z-matrix
significantly worse [ 7, 15, and especially 6-this optimizations, the MS update performs better than
latter system failed to converge with an exact the Powell update in a few cases (most notably
Hessian calculated at each cycle; it did manage to systems 3, 4, 10, and the pathological system 8,
converge with the Powell update albeit in an enor- which takes only 12 cycles, down from 31). but the
mous number of cycles (43 as compared to 8 in many failures overshadow the improved perfor-
both Z-matrix and Cartesian coordinates) showing mance in these few cases (nine of the Z-matrix
that the natural internal coordinates generated in optimizations fail with the MS update).
this example were very poor]. Four systems failed Comparing now the results using the combined
to converge: 16 and 24, which both failed even Powell-MS update proposed by Bofill" with the
with an exact Hessian, and 8 and 19. Powell update alone, there is a clear, albeit not
This last system-the dissociation of HNCCS- especially dramatic, improvement. For the Z-
is instructive as it illustrates what can go wrong matrix optimizations, none of the systems studied
with an initially perfectly reasonable set of natural took more cycles to converge with the former
internal coordinates as the molecular geometry update than with the latter, and 11 of them took
changes. A diagram of this system and the starting less. The reduction is typically only one or two
Z-matrix are given in the Appendix. The set of cycles, but for three systems (3, 8, and 10) it was
natural internals generated for HNCCS are partic- three cycles or more. With the Cartesian optimiza-
ularly simple, being the four obvious interatomic tions, although there were two systems (9 and 24)
distances appropriate to the given connectivity- that took more cycles to converge with the com-
note that the C . . . C "bond" was added to the bined update (one and two cycles more, respec-
formal connectivity list in order to generate a full tively), there were eight that took less, including
set of coordinates, as discussed in the Introduction reductions by four cycles (systems 7 and 16) and
-and the individual HNC, NCC, and CCS bond nine cycles (system 17). There were also some
angles (i.e., probably precisely the internal coordi- reductions for the natural internal coordinate opti-
nates one would use if one sat down and con- mizations, most notably for the HCOCl decompc-
structed the first Z-matrix that came to mind). sition (system 15, reduced from 28 to 20 cycles).

JOURNAL OF COMPUTATIONAL CHEMISTRY 895

BAKER AND CHAN

Overall, the hybrid Powell-MS update performs TABLE IV.

better than the Powell update alone and has conse- Comparison of the EF Algorithm in OPTIMIZE with
quently been adopted as the default for transition the Default Algorithm in GAUSSIAN 90 and the
state searches with OPTIMIZE. However, it is clear Work of Bofill."
that better updating schemes are urgently needed Number of Cycles to Reach Convergence
for transition state optimizations, especially up-
System This Worka GAUSSIAN 90b Bofill'
dates that can preserve the correct Hessian eigen-
value structure once it is found. Some authors 1. 9 11
recommend a step-reduction scheme28in conjunc- 2. 8 7
tion with standard Hessian updates (see, e.g., refs. 3. 13 10
7 and 11);we have not found this to be a particu- 4. 11 - 12
larly advantageous strategy with the EF algorithm. 5. 11 18 19
For many years Schlegel has used an updating 6. 7 11 11
scheme in his optimization algorithms which uses 7. 9 13 9
data from several previous geometries, not just the 8. 26 26 22
9. 13 14
last one, to improve the Hessian matrixz9;in com-
10. 14 21
mon with the more traditional updates, though,
11. 8 9
this is not guaranteed to preserve the Hessian 12. 12 17
eigenvalue structure. 13. 11 16
Before concluding, we present in Table IV a 14. 13 16
comparison between the performance of the EF 15. 12 13
algorithm, as implemented in the OPTIMIZE pack- 16. 23 __
age? and the default algorithm in GAUSSIAN 903' 17. 8 18
for all of the systems studied in this work. Also 18. 7 10
given are comparisons with the work of Bofill" for 19. 10 10
the four reactions common to both studies (al- 20. 10 13
21. 4 4
though only two of these are really appropriate).
22. 5 6
The GAUSSIAN optimizations used the 3-21G ba- 7
23. 8
sis and the Z-matrices given in the Appendix, and 24. 14 27
an exact Hessian was calculated on the first cycle; 25. 9 11
hence the results are directly comparable with this
work. All optimizations started from the Z-matrices given in the
Appendix; an exact Hessian matrix was calculated on the
As can be seen, the EF algorithm with the (now) first cycle only.
default Powell-MS update performs very well and aUsing the Powell-MS update; numbers as per 2-matrix
is clearly superior to the standard TS algorithm in column of Table II.
GAUSSIAN 90. The pathological P-(formyloxy) bA dash denotes failure to converge in 50 cycles.
ethyl radical (system 8) took the same number of 'Only systems 5 and 8 are directly comparable with Bofill's
work; the other systems reported used the same starting
cycles to converge with GAUSSIAN 90 as in this geometry but a different Hamiltonian.
work, and many of the other systems took signifi-
cantly more; two systems failed to converge
(surprisingly, one of them was the seemingly in- cellent performance when exact Hessian data is
nocuous methoxy radical rearrangement, which available (Table II), show that the EF algorithm is
converged in 11 cycles with the EF algorithm). extremely efficient for optimizing transition states.
Of the two systems that are directly comparable
with Bofill's work," one of them took less cycles to
converge than in his case (system 5) and the other Conclusions
more (system 8). For the former system-ring
opening of the cyclopropyl radical-Bofill claims Although previous work',4 has demonstrated
that the EF algorithm implemented in the GAMESS that, provided a reliable estimate of the Hessian
program13 took 36 cycles to converge (compared to matrix is available at the (reasonable) starting ge-
only 11 in this work) from the same starting geom- ometry, minimization in Z-matrix, natural internal,
etry and with a Hessian calculated in full on the and Cartesian coordinates is of comparable effi-
first cycle. Something seems suspicious here. ciency, for transition state searches a good-quality
The preceding comparisons, along with the ex- Z-matrix is generally superior.

896 VOL. 17, NO. 7

 LOCATION OF TRANSITION STATES

That this is the case is due in part to the fact associated with transition states make it hard to
that starting geometries for transition state opti- construct and maintain a good set of natural inter-
mizations are usually much farther from the final nals (which were designed primarily for minimiza-
converged structure than is typically the case for tion). In particular, perfectly adequate natural in-
minimization (for which molecular mechanics force ternal coordinates constructed at the starting ge-
fields can provide both reliable starting geometries ometry could become inappropriate as the geome-
and Hessians), and the increased coupling in try distorts during the optimization, hindering
Cartesian coordinates at such poor starting geome- convergence or causing the optimization to fail.
tries slows down the convergence. This is true A viable strategy for transition state searches
even with exact second derivatives available at might be to describe the atoms involved in the
each optimization cycle. active site (those where most of the geometrical
This situation is exacerbated when updated (ap- changes are taking place) in terms of individual,
proximate) Hessians are used in Cartesian opti- Z-matrix-like, internal coordinates, or even Carte-
mizations as the coupling makes it difficult to sians if this is deemed appropriate, and any side
maintain the desired Hessian eigenvalue structure chains or other structures which are not part of the
(one and only one negative eigenvalue) and, once active site in terms of natural internals. Hessian
this is lost, the optimization can take poor or information would then only be needed for the
misguided steps on the potential energy surface, active site atoms; the side chains-described using
severely inhibiting convergence. This can occur natural internals-could be successfully optimized
even with internal coordinate optimizations, but it without second derivative data. Work along these
is much less likely due to the reduced coupling lines is currently in progress.
with a good set of internals. Such effects are partic-
ularly marked in systems with one or more floppy
modes (modes associated with small Hessian
eigenvalues). Appendix
Although natural internal coordinates might be
expected to reduce the coupling between the indi- Z-matrices with starting parameters, initial en-
vidual internal coordinates even more than with a ergies, and final energies (at convergence) for the
Z-matrix, the distorted, nonstandard geometries test suite of 25 transition states given in Table I.

c1
N2 8 _____-- 0H3
c1
N2
H3
1
2
L1
L2 1 A1
L1 1.14838
L2 1.58536
A 1 90.0

1. HCN<--> HNC Einit -92.20273 Efinal -92.24604

.** 3 %

H5
c1
c2
x3
H4
H5
1
1
1
1
L1
1.0
L2
L3
2
2
3
90.0
A1
A2
3
2
180.0
180.0
L1 1.24054
L3 1.65694
L2 1.06318

A 1 60.3568
A2 60.3568
x3

2. HCCH C--> CCH2

JOURNAL OF COMPUTATIONAL CHEMISTRY 897

BAKER AND CHAN

H4
L1 1 . 2 5
c1 L2 1 . 3
02 1 L1 L3 1 . 5
H3 1 L2 2 A1 A1 100.0
H4 1 L3 2 A2 3 0.0 A2 1 5 0 . 0

c!?
’H3
02

3. H2CO <--> H 2 + CO Ejnjt -1 12.98005 Efjnal - 1 1 3 . 0 5 0 0 3

H3 L1 1 . 4 2 3
01 L2 1 . 4 8 4
Q. c2 c2
H3
1
2
L1
L2 1 A1
L3 1 . 0 8 7
A1 42.7
H4 2 L3 1 A2 3 D1 A2 1 1 7 . 5
H5 2 L3 1 A2 3 -D1 D1 1 0 6 . 4

4. CH3O <--> C H 2 0 H

H5s H4
c1
c2
C3
H4
1 L1
2 L2
1 L3
1
2
A1
A2 3 D1
L1
L2
L3
L4
1.454
1.454
1.106
1.106
A1 80.0
A2 1 1 9 . 6
A3 119.6
A4 1 1 9 . 6
H5 1 L4 2 A3 3 D2 L5 1.106 A5 1 1 9 . 6
H6 3 L5 2 A4 1 D3 L6 1.106 A6 1 4 8 . 5
L7 1.064 D 1 40.0
H7 3 L6 2 A5 1 D4
D2 - 1 6 0 . 0
H8 2 L7 3 A6 1 D5
D 3 108.0
D4 -108.0
D 5 -170.0
5. ring opening cyclopropyl

L1 1.495 A3 136.0
H5 c1 L2 1.418 A4 1 2 3 . 5
c2 1 L1 L3 1.463 A5 122.4
C3 2 L2 1 A1 L4 1.093 A6 1 2 4 . 7
H7
C4 1 L3 2 A2 3 D1 L5 1.111 A7 1 2 6 . 7
H5 1 L4 2 A3 3 D2 L6 1.098 A8 1 1 7 . 9
H6 2 L5 1 A4 4 D3 Ll 1.097 D 1 -120.4
H7 3 L6 2 A5 1 D4 L8 1.110 D2 4 . 4
H8 3 L7 2 A6 1 D5 L9 1.106 D3 108.8
H9 4 L8 1 A7 2 D6 A1 92.1 D4 - 1 0 7 . 5
H10 4 L9 1 A8 2 D7 A2 6 2 . 1 D5 8 4 . 2
D6 1 0 9 . 3
D7 - 1 0 6 . 1
6. bicyclo[l.l.OJ butane TS 1 Ejnit -153.88777 Efinal - 1 53.90494

898 VOL. 17, NO. 7

 LOCATION OF TRANSITION STATES
A3 118.7
L1 1.489 A4 109.3
c1 L2 1.374
c2 1 L1 A5 124.9
L3 1.384 A6 122.4
c3 2 L2 A1 L4 1.108
c4 1 L3 A2 A7 125.5
3 D1 L5 1.119 A8 126.8
H8 H5 1 L4 A3 3 D2 L6 1.098
H6 2 L5 A4 D1 -94.6
4 D3 L7 1.097 D2 150.5
H 10 H7 3 L6 A5 1 D4 L8 1.106
H8 3 L7 A6 1 D5 D3 107.0
L 9 1.114 D4 26.9
H9 4 L8 A7 2 D6 A1 136.3
W H10 4 L9 A8 2 D7 D5 -166.0
H9 A2 74.5 D6 82.9
D7 -115.6
7. bicyclo[l.l.O] butane TS 2 Einit -153.87578 Efinal - 1 5 3 . 8 9 7 5 4

L1 1.487 A3 94.8
c1 L2 1.458 A4 118.7
c2 1 L1 L3 1.365 A5 118.1
03 2 L2 1 A1 L4 1.812 A6 106.2
c4 3 L3 2 A2 1 D1 L5 1.095 A7 104.3
05 1 L4 2 A3 3 D2 L6 1.096 A8 116.3
H6 1 L5 2 A4 3 D3 L7 1.119 D1 8.6
H7 1 L6 2 A5 3 D4 L8 1.120 D2 -20.2
H8 H8 2 L7 3 A6 4 D5 L9 1.094 D3 -123.0
H9 2 L8 3 A7 4 D6 A1 108.2 D4 86.0
H10 4 L9 3 A8 2 D7 A2 111.4 D5 -112.9
D6 130.0
HSU D 7 -155.7
8. &
, -(formyloxy) ethyl Einit 264.63426 Efinal -2 64.6475 7

L1 1.2
x1 L2 2.0
H14 x2 1 1.0 L3 1.4
x3 1 L1 2 90.0 L4 0.77
X4 1 L2 2 A1 3 180.0 L5 0.7
c5 1 L3 3 90.0 2 90.0 L6 1.08
C6 1 L3 3 90.0 2 -90.0 L7 1.08
c7 3 L4 1 90.0 2 90.0 L8 1.081
C8 3 L4 1 90.0 2 -90.0 L9 1.079
c9 4 L5 1 90.0 2 90.0 L10 1.08
c10 4 L5 1 90.0 2 -90.0 A1 150.0
H11 7 L6 6 A2 1 D1 A2 120.0
H12 8 L6 5 A2 1 -D1 A3 118.0
H13 5 L7 8 A3 3 D2 A4 118.5
H14 6 L7 7 A3 3 -D2 A5 118.0
H15 5 L8 8 A4 3 D3 A6 118.5
H16 6 L8 7 A4 3 -D3 D1 175.0
H17 9 L9 4 A5 1 D4 D2 175.0
H18 10 L9 4 A5 1 -D4 D3 -6.0
H19 9 L10 4 A6 1 D5
HI7 H20 10 L10 D4 95.0
4 A6 1 -D5 D5 -96.0

9. parent Diels-Alder Einit -231-48964 Efinal - 2 3 1 . 6 0 3 2 1

N1 L1 1.20
N2 1 L1 L2 1.50
C3 1 L2 2 A1 L3 1.20
H8 C4 2 L2 1 A1 0.0 L4 1.08
N5 3 L3 1 A2 0.0 A1 120.0
N6 4 L3 2 A2 0.0 A2 120.0
H7 3 L4 5 A3 180.0 A3 120.0
H8 4 L4 6 A3 180.0

10. s-tetrazine <--> 2HCN + N2 Einit -292.84863 Efinal - 2 9 2 . 8 1 0 2 6

JOURNAL OF COMPUTATIONAL CHEMISTRY 899

BAKER AND CHAN
H7
L1 1.467
L2 1.320
c1 L3 1.08
c2 1 L1 L4 1.081
C3 1 L2 2 A1 L 5 1.082
H8 C4 2 L2 1 A1 3 D1 A1 124.0
H5 1 L3 2 A2 4 D2
A2 1 2 1 . 0
H6 2 L3 1 A2 3 D2 A3 119.0
H7 3 L4 1 A3 2 D3 A4 121.0
H8 3 L5 1 A4 2 D4 D1 160.0
H9 4 L4 2 A3 1 D3 D2 -20.0
H10 4 L5 2 A4 1 D4 D3 180.0
D4 0.0
H9

11. rotational TS in butadiene Einit -1 54.05460 Efinal - 1 54.05046

L1 1.43
L2 1.50
H4
0.........0..
H3 c1
c2
H3
1
1
L1
L2 2 A1
L3
L4
L5
1.50
1.09
1.09
H4 2 L3 1 A2 3 0.0 A1 120.0
H
m
5 H8
A2 60.0
H6
H5 1 L4 2 A3 3 -D1
D1 A3 115.0
HI 2 L5 1 A4 4 D2 A4 115.0
H6 H7 H8 2 L5 1 A4 4 -D2 D1 104.0
D2 104.0

L1 1.43

6-_ - - --
F4
-0
c1
c2
H3
F4
1
1
2
L1
L2
L3
2 A1
1 A2 3 0.0
L2
L3
L4
L5
1.50
1.90
1.08
1.08
H5 1 L4 2 A3 4 D1 A1 100.0
H6 1 L4 2 A3 4 -D1 A2 100.0
H7 2 L5 1 A4 4 D2 A3 115.0
H8 2 L5 1 A4 4 -D2 A4 115.0
D1 105.0
D2 105.0

L1 1.43 120.0
L2 1.30 A2 110.0
c1 L3 1.09 A3 110.0
c2 1 L1 L4 1.09 A4 120.0
03 2 L2 1 A1 L5 1.09 A5 60.0
H4 1 L3 2 A2 3 D1 L6 1.35 Dl 115.0
H5 1 L4 2 A3 3 D2 D2 -.120.0
H5 H6 2 L5 1 A4 3 D3 D3 175.0
H7 3 L6 2 A5 1 D4 D4 5.0

14. CH2CHOH <--> C H @ O Einit -151.87756 Efinal -1 51.9131 0

900 VOL. 17, NO. 7

 LOCATION OF TRANSITION STATES

01 0.
.. . -
01
c2 1 L1
L1 1.17
L2 2 . 3 3 5
L3 1.127
H4 2 L3 3 A2 1 180.0 A 1 90.0
A2 9 0 . 0

1 5. HCOCI <--> HCI + CO Einit -569.87865 Efinal -569.89752

P1 L1 1.46
A1 109.47
02 1 L1 L2 1 . 4 6 A2 109.4'7
03 1 L2 2 A1 L3 1 . 9 8 A3 125.02
04 1 L3 2 A2 3 D1 L4 1 . 5 4 A4 109.5
05 1 L4 2 A3 3 D2 L5 0.95 A5 109.5
05 D1 120.0
H6 4 L5 1 A4 2 D3 L6 1 . 2 7
H7 5 L6 1 A5 2 D4 D2 -172.35
D3 0.74
H7 H6 D4 -84.48

16. H20 + PO3' <--> H2PO4- Einit -637.78814 Efinal-637.92388

L1 1.420 AH1 115.0

c1 L2 1.405
H8 c2 1 L1 AH2 1 1 5 . 0
L3 2.008 AH3 1 2 0 . 0
H9
c3 2 L2 1 A2 L4 1.310
04 3 L3 2 A3 1 D3 AH4 1 1 5 . 0
L5 1.410 AH5 1 1 5 . 0
c5 4 L4 3 A4 2 D4 R1 1.09 AH6 1 1 5 . 0
C6 5 L5 4 A5 3 D5 R2 1.09 AH7 1 1 5 . 0
H7 1 R1 2 AH1 3 DH1
R3 1.09 AH8 1 2 0 . 0
H8 1 R2 2 AH2 3 DH2 R4 1.09 D 3 -66.8
H9 2 R3 1 AH3 3 DH3 R5 1.09
H10 3 R4 2 AH4 1 DH4 D4 56.6
R6 1.09
H11 3 R5 2 AH5 1 DH5 D5 - 6 7 . 8
H14 R7 1.09
H12 6 R6 5 AH6 4 DH 6 DH1 -30.0
R8 1.09
H13 6 R7 5 AH7 4 DH 7 A2 DH2 1 8 0 . 0
120.0
H14 5 RE 6 AH8 4 DH8 DH3 180.0
A3 101.1
A4 DH4 1 8 0 . 0
103.5
A5 DH5 3 0 . 0
120.0
DH6 1 8 0 . 0
DH7 - 3 0 . 0
1 7. Claisen rearrangement Einit - 2 6 7 . 2 1 899 Efinal -267.23859 DH8 1 8 0 . O

22-
L1 1.52 A4 108.63
c1 L2 2.30 A5 108.12

JJ?
c2 L1
1 L3 1.10 A6 108.93
S13 2
L2 1 A1 L4 1 . 0 9 A7 108.93
H4 1 L3 2 A2 3 D1 L5 1.11 A8 41.00
H5 1 L4 2 A3 3 D2 L6 1 . 0 9 A9 109.12
----_ H6 1 L5 2 A4 3 D3 L7 1 . 1 0 D1 -65.13
H6 H7 2 L6 1 A5 3 D4 L8 1 . 4 8 D2 62.27
H10 HE 2 L7 1 A6 3 D5 L9 1 . 5 0 D3 178.92
H9 H9 3 L8 2 A7 1 D6 L10 1 . 4 8 D4 121.91
H10 3 L9 2 A8 1 D7 A1 119.26 D5 -122.55
H5 H4 H 1 1 3 L10 2 A9 1 D8 A2 1 1 4 . 1 7 D6 -55.52
A3 115.06 D7 65.18
D8 -174.11
1 8. silylene insertion Einit -367.19458

JOURNAL OF COMPUTATIONAL CHEMISTRY 901

BAKER AND CHAN

H'O H1
L1 1.01
L2 1.2
N2 1 L1 L3 1.86
x3 2 1.0 1 90.0 L4 1.57
C4 2 L2 3 A1 1 180.0 A1 56.0
C5 4 L3 2 A2 3 0.0 A2 128.5
S6 5 L4 4 A3 2 180.0 A3 124.9

19 . HNCCS C-->HNC + CS Ejnit -525.42284 Efinal - 5 2 5 . 4 3 0 4 0

L1 2.0
L2 1.13
N1 L3 1.03
c2 1 L1 L4 1.08
$ 03
l # 03 2 L2 1 A1 L5 1.03
H4 1 L3 2 A2 3 0.0 A1 120.0
H7 H5 2 L4 1 A3 4 180.0 A2 109.5
H6 1 L5 2 A4 3 D1 A3 60.0
H7 1 L5 2 A4 3 -D1 A4 109.5
H4 D1 120.0

20. HCONH3' <--> NH4' + CO Ejnit -1 68.23208 Efinal - 1 6 8 . 2 4 7 5 2

L1 1.34 A3 120.0
c1 L2 1.45 A4 120.0
c2 L1 A5 120.0
L3 1.22
cJlH6 Hi3 c3 L2 1 A1 L4 1.08 A6 120.0
04 L3 2 A2 D1 L5 1.08 D1 90.0
H5 L4 2 A3 D2 L6 1.08 D2 180.0
H6 L5 2 A4 D3 L7 1.08 D3 0.0
H7 L6 1 A5 D4 A 1 120.0 D4 180.0
04 D5
H8 L7 2 A6 A2 120.0 D5 180.0
H7

21 . rotational TS in acrolein Ejnjt -189.67124 Efinal -1 8 9 . 6 7 5 7 4

H7
01 L1 1.3 A1 110.0
c2 L1 L2 1.3 A2 120.0
N3 L2 A1 L3 1.4 A3 120.0
04 L3 A2 180.0 L4 1.1 A4 109.6
H5 L4 A3 0.0 L5 1.0 A5 75.0
H6 L5 A4 180.0 L6 1.5
H7 L6 A5 0.0

22. HCONHOH <-> HCOHNHO Einit - 2 4 2 . 2 4 9 12 Efjnal - 2 4 2 . 2 5 5 2 9

902 VOL. 17, NO. 7

 LOCATION OF TRANSITION STATES

H1 L1 1.0
L2 1.2
HI. L3 1.0
N2 1 L1 L4 1.2
c3 2 L2 1 A1 A1 120.0
H4 3 L3 2 A2 1 D1 A2 150.0
H5 4 L4 3 A3 2 D2 A3 90.0
D1 170.0
D2 10.0

23. HNC + H2 <--> HzCNH Einit -93.30097 Efinal - 9 3 . 3 1 11 4

L1 1.0
L2 1.2
H4 H1 L3 1.3
N2 1 L1 L4 1.0
Q c3 2 L2 1 A1 A1 120.0
H4 3 L3 2 A2 1 D1 A2 60.0
H5 3 L4 2 A3 1 D2 A3 120.0
D1 -120.0
D2 0.0

24. H2CNH <--> HCNH2 Einit -93.29435 Efinal -93.33296

L1 1.35
L2 1.0
c1 L3 1.4
H5 N2 1 L1 L4 1.4
H3 1 L2 2 A1 A1 105.0
H4 2 L3 1 A2 3 D1 A2 105.0
H5 2 L4 1 A3 3 D2 A3 110.0
H3 D1 150.0
D2 -160.0

25. HCNH2 <--> HCN + H2 Einit -93.25640 Efjnal -93.28172

JOURNAL OF COMPUTATIONAL CHEMISTRY 903

BAKER AND CHAN

16. E. B. Wilson, J. C. Decius, and P. C. Cross, Molecular

References Vibrations, McGraw-Hill, New York, 1955.
17. P. Pulay, In Applications of Electronic Structure Theory, H. F.
1. J. Baker and W. J. Hehre, 1. Comput. Chem., 12, 606 (1991). Schaefer, Ed., Plenum, New York, 1977, p. 153.
2. P. Pulay, G. Fogarasi, F. Pang, and J. E. Boggs, 1. Am. Cheni. 18. A. C. Scheiner, G. E. Scuseria, and H. F. Schaefer 111, J. Am.
Soc., 101, 2550 (1979). Ckem. Soc., 108,8160 (1986).
3. G. Fogarasi, X. Zhou, P. W. Taylor, and P. Pulay, 1. Am. 19. S. Wolfe and C-K. Kim, Israel J . Clzem., 33, 295 (1993).
Chem. Soc., 114, 8191 (1992). 20. P. G. Jasien, 1. Phys. Clzem., 98, 2859 (1994).
4. J. Baker, 1. Comput. Chem., 14, 1085 (1993). 21. B. Ma, Y. Xie, M. Shen, P. v. R. Schleyer, and H. F. Schaefer
5. J. Baker, 1. Conzpnt. Chem., 7, 385 (1986). 111, J . A m . Chenz. Soc., 115, 11169 (1993).
6. C. J. Cerjan and W. H. Miller, 1. Chem. Phys., 75, 2800 22. C. Peng and H. B. Schlegel, Israel. 1. Chem., 33, 449 (1993).
(1981). 23. R. D. Bach, M-D. Su, E. Aldabbagh, J. L. Andres, and H. B.
7. A. Banerjee, N. Adams, 1. Simons, and R. Shepard, 1. Phys. Schlegel, 1. Am. Cheni. Soc., 115, 10237 (1993).
Chenz., 89, 52 (1985). 24. R. Flammang, D. Landu, S. Laurent, M. Barbieux-Flammang,
8. J. Baker, OPTIMIZE, version 1.0 beta, Biosym Technologies, C. 0. Kappe, M. W. Wong, and C. Wentrup, 1. Am. Chem.
San Diego, CA 1993. Soc., 116, 2005 (1994).
9. R. Ahlrichs, M. Bar, M. Ehrig, M. Haser, H. Horn, and 25. H-Y. Lin, D. P. Ridge, E. Uggerud, and T. Vulpius, 1. A m .
C. Kolmel, TURBOMOLE, version 2.2 beta, Biosym Tech- Cliem. Soc., 116, 2996 (1994).
nologies, San Diego, CA, 1993. 26. 0. N. Ventura, J. B. Rama, L. Turi, and J. J. Dannenberg,
10. J. Baker, M. Muir, and J. Andzelm, J . Chew. Phys., 102, 2063 1. Am. Cheni. Soc., 115, 5754 (1993).
(1995). 27. J. Andzelm, J. Baker, A. C. Scheiner, and M. Wrinn, Znt. 1.
11. J. M. Bofill, 1. Comput. Chem., 15, 1 (1994). Q u a ~ i t Chcni.,
. 56, 733 (1995).
12. J. J. P. Stewart, MOPAC, version 5.0, program 455, Quan- 28. R. Fletcher, Practical Methods uf Optimization: Unconstrained
tum Chemistry Program Exchange, Indiana University, Optimization, Vol. 1, John Wiley & Sons, New York, 1980.
Bloomington, IN. 29. H. B. Schlegel, 1. Coviput. Cheni., 3, 214 (1982).
13. M. Dupuis, D. Spangler, and J. J. Wendoloski, GAMES, 30. M. J. Frisch, M. Head-Gordon, G. W. Trucks, J. B. Foresman,
Nat. Resour. Comput. Chem. Software Cat., program QGO1, H. B. Schlegel, K. Raghavachari, M. A. Robb, J. S. Binkley,
1980; extended by M. W. Schmidt and S. T. Elbert, 1992. C. Gonzalez, D. J. Defrees, D. J. Fox, R. A. Whiteside,
14. M. J. D. Powell, Math. Prog., 1, 26 (1971). R. Seeger, C. F. Melius, J. Baker, R. L. Martin, L. R. Kahn,
15. B. A. Murtagh and R. W. H. Sargent, Compiit. /., 13, 185 J. J. P. Stewart, S. Topiol, and J. A. Pople, Gaussian 90,
(1970). Revision I, Gaussian Inc., Pittsburgh PA, 1990.

904 VOL. 17, NO. 7