Ab initio and diatomics in molecule potentials for and

Jiri Vala, Ronnie Kosloff, and Jeremy N. Harvey

Citation: The Journal of Chemical Physics 114, 7413 (2001); doi: 10.1063/1.1361248
View online: https://doi.org/10.1063/1.1361248
View Table of Contents: http://aip.scitation.org/toc/jcp/114/17
Published by the American Institute of Physics

Articles you may be interested in

The electronic structure of the triiodide ion from relativistic correlated calculations: A comparison of different
methodologies
The Journal of Chemical Physics 133, 064305 (2010); 10.1063/1.3474571

Photodissociation dynamics of the triiodide anion

The Journal of Chemical Physics 113, 2255 (2000); 10.1063/1.482040

Photodissociation of gas-phase : Comprehensive understanding of nonadiabatic dissociation dynamics

The Journal of Chemical Physics 126, 204311 (2007); 10.1063/1.2736691

Intermolecular interactions in the condensed phase: Evaluation of semi-empirical quantum mechanical methods
The Journal of Chemical Physics 147, 161704 (2017); 10.1063/1.4985605

Improving the accuracy of Møller-Plesset perturbation theory with neural networks

The Journal of Chemical Physics 147, 161725 (2017); 10.1063/1.4986081

Product energy deposition of CN + alkane H abstraction reactions in gas and solution phases
The Journal of Chemical Physics 134, 214508 (2011); 10.1063/1.3595259
JOURNAL OF CHEMICAL PHYSICS VOLUME 114, NUMBER 17 1 MAY 2001

Ab initio and diatomics in molecule potentials for I2À , I2 , I3À , and I3

Jiri Vala and Ronnie Kosloff
Department of Physical Chemistry and The Fritz Haber Research Center, The Hebrew University,
Jerusalem, 91904, Israel
Jeremy N. Harveya)
School of Chemistry, University of Bristol, Cantock’s Close, Bristol BS8 1TS, United Kingdom
共Received 8 December 2000; accepted 15 February 2001兲
The electronic structure of the I⫺ ⫺
3 molecular anion and its photoproducts I2 , I2 , and I3 were studied.
Ab initio calculations were carried out using the multireference configuration interaction 共MRCI兲
method for the valence electrons together with a relativistic effective core potential. The ab initio
wave functions were also used to compute some spin–orbit coupling matrix elements, as well as
approximate valence bond wave functions, used as guidelines in the construction of a 108-state
diatomics in molecule 共DIM兲 description of the electronic structure of I⫺ 3 . In the DIM model,
spin–orbit coupling was introduced as a sum of atomic operators. For I⫺ 2 the ab initio and the DIM
ground-state potentials show excellent agreement with the experimental results. The results for I2
are also in very good agreement with experimental data. For I⫺ 3 , the MRCI calculations give a very
good description of the spectroscopic constants and agree with the vertical excitation energies,
provided spin–orbit coupling is included. The DIM description fails both quantitively by leading to
erroneous spectroscopic constants, and qualitatively by not even reproducing the MRCI ordering of
the excited-states. The failure of the DIM is attributed to the omission of ionic states. The overall
qualitative picture of the excited-state potentials shows a maze of dense avoided crossings which
means that all energetically allowed photoproducts will be present in the experiment. The ground
electronic state of I3 was calculated to be a collinear and centrosymmetric 2 ⌸ u,3/2 . The collinear
state is stabilized by spin–orbit coupling relative to a bent configuration. Calculated vertical
transition energies from the ground to low-lying excited states of the radical are in excellent
agreement with the experimental data. The spin–orbit assignment of these states is provided.
© 2001 American Institute of Physics. 关DOI: 10.1063/1.1361248兴

I. INTRODUCTION
Insight in physical chemistry is typically obtained by
comprehensive studies of a particular molecular system
where a close feedback is established between experiment
and theory. The triodide molecular anion I⫺ 3 is emerging as
one of these systems. Its strong optical absorption in the near
ultraviolet 共UV兲 makes it readily accessible to photochemical
studies. The photochemical products I⫺
2 absorb in a different
spectral region. This factor and the slow dynamics induced
by the heavy masses have enabled the photoreaction to be
followed in real time from reactants to products.1–3 The ex-
perimental observations have stimulated the development of
novel theoretical tools4 addressing the problem of dissipative
quantum dynamics in solution. The present study is part of
the large theoretical effort on the I⫺
3 system, addressing spe-
cifically the electronic structure of the parent ion and the
routes to photofragments. Obtaining the ground- and excited-
state potentials is a necessary condition for any comprehen-
sive theoretical understanding of these processes.
The first experimental studies of the I⫺
3 ion were oriented
on dissociation kinetics in solution,5–7 electronic absorption
spectroscopy in the condensed phase,8–12 and thermo-
chemistry.13 The electronic spectrum of I⫺ 3 in solution con-
a兲
Electronic mail: jeremy.harvey@bris.ac.uk
sists in two distinct bands centered at about 290 and 360 nm.
After some controversy,9,14–16 these two bands were assigned
to the spin–orbit-induced mixture of 1 ⌺ 0⫹u and 3 ⌸ 0⫹u
states,11,12,17 where the former state carries the transition di-
pole moment responsible for optical absorption.
More recently, Ruhman and co-workers have used fem-
tosecond spectroscopy to investigate the properties and dy-
namics of the I⫺ 3 anion. This research covered several
complementary viewpoints, investigating aspects such as the
light-induced vibrational dynamics on the electronic ground
state,18,19 photodissociation,1,2,4,20 and geminate
recombination.21,22 These last studies revealed several dis-
tinct recombination time scales and the possibility of mul-
tiple potential-energy surfaces being involved in solution dis-
sociation dynamics. Another aspect of this experimental
corpus was devoted to the spectroscopy and relaxation dy-
namics of the transient diiodide ion.23 A similar experimental
study of the photodissociation dynamics of I⫺ 3 was recently
carried out by Vöhringer and coworkers.24–26
The triodide anion has also been investigated using ad-
vanced frequency domain spectroscopical methods,27–29
which have brought considerable insight into the role of sol-
vent effects and the dynamical symmetry breaking of the I⫺ 3
molecule.
The gas-phase properties of the di- and tri-iodide anions

0021-9606/2001/114(17)/7413/11/$18.00 7413 © 2001 American Institute of Physics

7414 J. Chem. Phys., Vol. 114, No. 17, 1 May 2001
have been extensively investigated by Neumark and co-
workers using ultrafast photoelectron spectroscopy.3 The
photodissociation of I⫺ 2 was studied
30–34
with an emphasis on
the electronic structure, providing experimental potentials for
the X̃ 2 ⌺ ⫹
u ground state
34
and the à ⬘ 2 ⌸ g,1/2 excited state.33
Photofragmentation experiments35–37 on I⫺ 3 indicate that a
complicated set of electronic states with many curve cross-
ings is involved in the dissociation process, in agreement
with the results of solution studies.21,22 Also, the neutral I3
radical has been recently characterized for the first time.38,39
This species, which is important in the iodine atom recom-
bination reaction, has a ground state which is bound with
respect to I⫹I2 , and has a linear, essentially centrosymmet-
ric configuration.
The dynamics of triatomic species is sufficiently simple
to be described with very high accuracy by fully quantum-
mechanical methods. It is, therefore, no surprise that the ex-
perimental work on I⫺ 3 has been accompanied by extensive
theoretical work. However, dynamical computations are
critically dependent on obtaining accurate descriptions of the
potential-energy surfaces involved, and of the coupling be-
tween them where relevant. All of the dynamical studies so
far have used an empirical description of the excited state
corresponding to the blue band of the I⫺ 3 electronic spectra,
constructed in terms of the London–Eyring–Polanyi–Sato
共LEPS兲 potential.1,20 Because the surfaces have been roughly
fitted to experimental properties, reasonable agreement with
experiment has been obtained with respect to, e.g., the elec-
tronic spectra characteristics of I⫺ 3 in solution,
1,20
the time-
resolved photoelectron spectra, or the vibrational energy dis-
tribution of the I⫺ photofragment in the gas-phase
2
experiment.35–37 However, the experimental and theoretical
work is full of indications that the surfaces used are very
approximate at best. For instance, the first LEPS surface
used1,20 reproduces the upper energy band of the solution
electronic spectrum rather well, which is expected since it
was constructed to do so. Less expectedly, it also reproduces
the lower-energy band, providing the necessary shift in total
energy is introduced. However, it is not compatible with the
gas-phase experimental observations, so that a substantially
modified LEPS surface36 has been used to account for the
results of photofragmentation. A more serious problem is
that both the solution21,22 and the gas-phase experiments36,37
agree in predicting that the excited-state dynamics occurs on
multiple potential-energy surfaces.
In the case of triatomic systems, it is possible to perform
enough accurate ab initio computations 共⬃103n⫺6 ⫽1000
points兲 to fully characterize the potential-energy surface.
Given the enormous experimental interest in the I⫺ 3 system,
the attraction of determining its potential-energy surface us-
ing ab initio computations should be obvious. However,
there are also many problems for such an approach. First of
all, the dynamics of I⫺ 3 involves bond-breaking, so that non-
dynamical correlation will be important in many areas of the
potential-energy surface which need to be described. To-
gether with the expected role of dynamical correlation in
obtaining an accurate description of the potential, this means
that the ab initio method used to describe the system will
need to be an expensive multireference method with some
Vala, Kosloff, and Harvey
form of configuration interaction 共CI兲 and large one-particle
basis sets. However, this problem is general when accurate
global potential-energy surfaces are needed. The specific
problem with the present iodine-containing system is due to
the very high atomic weight of iodine, which leads to very
important relativistic effects on the wave functions, which
cannot as yet be treated in a rigorous way, even for such a
‘‘small’’ system. Whilst scalar relativistic effects can be ad-
equately reproduced using effective core potentials to de-
scribe the innermost, ‘‘core’’ electrons, the treatment of the
nonscalar effects, which are mostly due to spin–orbit cou-
pling, is much less satisfactory. Although several studies
have used methods based on the Dirac equation to study
diatomic iodine species,40,41 application to I⫺ 3 , especially at
the level of detail and accuracy needed for dynamical stud-
ies, is still quite distant.
In the present study, we present a first attempt to use ab
initio computations to characterize the excited-state surfaces
and dynamics of I⫺ 3 , including the photoionization to neutral
I3 . Because the chemistry of these species is so intimately
linked to that of diatomic I2 and I⫺ 2 , we have also performed
computations on the latter. The level of theory used, multi-
reference configuration interaction, with a large one-particle
basis set, should treat the correlation problem adequately if
not perfectly. The use of a relativistic effective core potential
to treat the innermost electrons has the dual advantage of
decreasing the computational expense and accounting for
most of the scalar relativistic effects. To treat the spin–orbit
coupling, we have used two approaches both relying on a
simplified one-electron operator. For certain important points
on the potential-energy surfaces, we are able to compute
spin–orbit coupling matrix elements between a limited num-
ber of spin–orbit-free ab initio wave functions. To get a
more global picture of the effect of spin–orbit coupling, we
have constructed a diatomics in molecules 共DIM兲 model of
the valence electronic states of I⫺ 3 , which includes spin–
orbit coupling within an atoms in molecules ansatz. Al-
though the results obtained are only of semiquantitative ac-
curacy, they should already represent an improvement on
previous semiempirical surfaces. Also, by giving for the first
time a global, nonempirical description of the surfaces and
their interaction, the study should be a valuable guideline for
future research in this area.
After describing the details of the ab initio computations
and the Diatomics in Molecules model, we present the re-
sults below for each of the species separately, namely I⫺ 2 , I2 ,
I⫺
3 , and I3 .
II. AB INITIO COMPUTATIONS
We have performed new ab initio computations on I2 ,
I⫺
2 , I⫺
3 , and I3 . All calculations used the MOLPRO 98 and
2000 program packages.42 Three types of calculations were
used to 共a兲 reproduce the energy curves in the absence of
spin–orbit coupling; 共b兲 compute spin–orbit coupling matrix
elements between certain specified electronic states; and 共c兲
obtain approximate valence-bond wave functions for aid in
the construction of and comparison with the DIM model.
The energy calculations were performed at the multiref-
J. Chem. Phys., Vol. 114, No. 17, 1 May 2001
erence singles and doubles configuration interaction level
共MRCISD, referred to as MRCI below兲, with the added
Davidson correction to provide approximately size-
consistent results. The CI wave function was built on a com-
plete active-space self-consistent field 共SCF兲 reference, with
the active space comprising all the explicitly described elec-
trons, which for the triodide anion, for example, means 22
electrons in 12 orbitals. The spin–orbit coupling matrix ele-
ments were computed between CASSCF wave functions
computed with the same active space. The approximate VB
wave functions were constructed by first performing a
CASSCF computation with a slightly smaller active space in
which the 5s orbitals were constrained to be doubly occupied
in all configurations. The resulting orbitals were then local-
ized within the reduced-symmetry C 2 v point group, and the
orbital and CI coefficients of the wave function within this
orbital set examined. Localization was found to lead to al-
most entirely pure 5p-like orbitals, localized on the indi-
vidual iodine atoms.
In all these calculations, the same 46-electron relativistic
ECP43 was used, in the form implemented in MOLPRO.42 The
valence 5s and 5p electrons were, however, described by
different basis sets. In the energy calculations, the
(7s7 p2d1 f )/ 关 3s3 p2d1 f 兴 basis associated with the ECP43
was used after decontracting the two most diffuse contracted
s and p primitives, and replacing the standard polarization
functions with a 共loosely energy-optimized for neutral and
anionic iodine atom兲 set of 3 d ( ␣ ⫽1, 0.4, and 0.16兲, 2 f
( ␣ ⫽0.632, 0.252兲, and 1 g ( ␣ ⫽0.45) uncontracted primi-
tives, to yield a (7s7 p3d2 f 1g)/ 关 5s5 p3d2 f 1g 兴 basis. In
the spin–orbit coupling calculations, supplementary spin–
orbit coupling parameters for the ECP were used, as docu-
mented in the MOLPRO manual,42,43 together with the stan-
dard (7s7p2d1 f )/ 关 3s3 p2d1 f 兴 basis set. Calculations with
this basis yield a 2 P 3/2 to 2 P 1/2 splitting of 7601 cm⫺1 , in
excellent agreement with the experimental value of 7603
cm⫺1 . The approximate valence bond 共VB兲 computations
also used the standard basis set, but with the f functions
removed.
Gradients for geometry optimization and second deriva-
tives for frequencies were obtained by numerical differentia-
tion of the Davidson-corrected MRCI energies; note that
only the stretching frequencies were considered. Fully state-
optimized orbitals were used throughout, except for compu-
tation of spin–orbit coupling matrix elements, excitation
transition dipole moments, and second or higher roots within
a given symmetry species, where state-averaged orbitals
were used. Bond dissociation energies were calculated with
respect to the appropriate supermolecular asymptote.
III. DIM
The DIM model used is very similar to the one used by
Naumkin44,45 to describe the Ar.I2 and Ar.I⫺ 2 species. For a
general description of the DIM method and the common ap-
proximations used in its construction, the reader is referred
to the literature.46–48 It should be pointed out that the DIM
model used here is considerably different from other DIM
models which have been used to describe Xen .I2 , 49
Potentials for I2⫺ , I2 , I3⫺ , and I3 7415
Arn .I⫺
2 ,
50
and the I3 radical.51,52 To be specific, these last
models rely on diatomic data for I2 and I⫺ 2 in the ⍀, ␻ rep-
resentation, that is, Hund’s case 共c兲 adiabatic potential-
energy curves including the effect of spin–orbit coupling.
The advantage is that some of these curves, the lower-lying
ones in particular, are very well known from experiment, and
of course, therefore, incorporate exactly all the effects such
as correlation, relativity, spin–orbit 共SO兲 coupling, etc.,
which are hard to obtain from computation.
However, evaluating a DIM model for I⫺ 3 constructed in
this way by comparing it to spin–orbit-free ab initio compu-
tations would be difficult. For this reason, we use instead a
DIM model based on ⌳,⌺-eigenstates of the diatomic spe-
cies. Because these are not very realistic models of the
potential-energy curves, we add a simple spin–orbit coupling
operator to the DIM Hamiltonian 共this is referred to below as
the DIM⫹SO model兲. For comparison of the DIM model
with computations, the simple spin–orbit free results can be
used 共referred to simply as DIM below兲. Naumkin44 has ar-
gued that the ⌳,⌺ approach to DIM models for iodine-
containing species is actually better than the other approach
共provided atomic spin–orbit coupling is included兲, because
the construction of the diatomic Hamiltonian requires fewer
arbitrary assumptions.
The basis set for the DIM Hamiltonian is constructed
from all the spin-singlet and 共triply degenerate兲 spin-triplet
states which can be constructed from two iodine atoms and
one iodide anion. This leads to a total of 27⫹3⫻27⫽108
primitive states to be included in the calculation. The fitting
of the ab initio data to the DIM Hamiltonian is for the most
part straightforward and follows an identical approach to that
of Naumkin.44 In particular, the correspondence Table be-
tween molecular states and atomic configurations for I2 is
mostly taken from his work. We note, however, that our
assignment of the ⌸ g and ⌸ u states is taken instead from
Ref. 53, and is thus reversed with respect to that of Ref. 44,
which appears to be incorrect.
As in both these references, we assume that there is no
mixing between the two 1 ⌺ ⫹ g configurations or between the
two 3 ⌺ ⫺ configurations. This is based on the large energy
u
gap between the two configurations.44 To test the validity of
this assumption, we computed approximate ab initio valence
bond wave functions at the I2 ground-state minimum. The
weight of atomic configurations such as Px Px in the X 1 ⌺ ⫹ g
wave function, and of the Pz Pz configuration in the II 1 ⌺ ⫹ g
wave function, are very small. These small, if nonzero,
weights justify the approximation used here.
Analytical functions fitted to the ab initio data are used
for the diatomic curves. Curves with a significant minimum
共deeper than 0.001 Hartree兲 are fitted to an extended Morse
curve of the form D 0 兵 1⫺exp(⫺␤1(r⫺r0)⫺␤2(r⫺r0)2⫺␤3(r
⫺r0)3)其2. The purely repulsive curves are fitted to extended
exponential curves of the form A exp(⫺␣r⫺␣⬘r2).54
The spin–orbit coupling is treated approximately as a
sum of atomic operators, independent of molecular geom-
etry, which is added to the normal DIM Hamiltonian. The
form of the operator for one iodine atom in terms of the P x ,
P y , P z basis set can be found in the literature.44 The ap-
proximation we use is valid when the atomic orbitals making
7416 J. Chem. Phys., Vol. 114, No. 17, 1 May 2001 Vala, Kosloff, and Harvey

It is to be noted that within the simplified treatment of spin–

orbit coupling included here, the 2 ⌸ g,3/2 state is not affected,
so that the ab initio and DIM⫹SO constants are identical.
The bond length and energy are in good agreement with
experimental values. The main effect of the spin–orbit cou-
pling on the ground state is to stabilize the atomic asymptote
and thus decrease the bond energy. At the minimum, al-
though spin–orbit coupling of the 2 ⌺ ⫹ u state with the ⌸ u
2

states is still strong, the latter are so much higher in energy

that the effect on the ground-state energy is minimal, that is,
it is lowered in energy with respect to the spin–orbit free
species by only 0.074 eV. The bond length is slightly in-
creased, by 0.024 Å, and the harmonic frequency slightly
decreased. Overall, the agreement with experiment is rather
good, and this extends to the whole potential-energy curve.
Fitting to the extended Morse curve D e 关 1⫺exp兵兺i⫽1 5
⫺i
⫺␤i⌬r 其兴 yields ␤ parameters 共in Å ) of 1.175 84,
i 2
FIG. 1. DIM⫹SO potential-energy curves for I⫺
2 .
⫺0.012 335 8, ⫺0.004 880 43, ⫺0.000 263 595, and
0.000 192 631. Apart from the slight difference in bond
up the molecular orbitals are not substantially distorted with length and energy, this curve is almost superimposable on
respect to the isolated atom orbitals, and comparison with a the experimental one.34
few ab initio computed spin–orbit coupling matrix elements As a final note on the I⫺ 2 species, we note that whilst the
suggests that the approximation performs very well. In fact, present bond length and energy are in good agreement with
the ab initio values computed with MOLPRO are also only experiment, this has not been the case in all ab initio studies.
approximate, because they are not derived from the full This is to be expected for the bond energy, which is strongly
Breit–Pauli spin–orbit coupling Hamiltonian 共which in- subject to correlation effects, but somewhat less so for the
cludes one-electron and two-electron terms兲, but from a sim- bond length, which can usually be well reproduced even with
plified one-electron operator. However, unlike the more dras- fairly low levels of theory. We found the bond length to be
tic approximation used in the DIM approach, the ab initio quite sensitive to basis set and correlation treatment, so that
computation accounts rigorously for orbital relaxation in the at the complete active space self-consistent field 共CASSCF兲
different electronic states, so the agreement between the two level with the same basis set, it is extended to 3.257 Å,
suggests that both approaches are at least qualitatively quite whereas at the MRCI level with a reduced spd basis, it is as
reliable. much as 3.291 Å. This no doubt accounts for previously
noted discrepancies. Finally, test calculations suggest even
IV. I2À better agreement can be reached—for example, MRCI calcu-
lations using an added core-polarization potential 共CPP兲40,43
The diiodide anion has been the subject of many previ- lead to a bond length of 3.164 Å and a D e of 1.259 eV.
ous computational studies,55–60 including at least two which Assuming the spin–orbit coupling correction to be the same
have calculated all the states originating from the ground- as above leads to a prediction of R e ⫽3.188 Å and D e
state atoms, with and without including spin–orbit ⫽1.017 eV. Coupled-Cluster 关CCSD共T兲兴 calculations yield
coupling.59,60 Our four MRCI curves are similar to previous almost identical results to MRCI computations with the same
results, and are not shown here. The six DIM⫹SO curves are basis.
shown in Fig. 1, and the spectroscopic constants of the two
states displaying significant minima are included in Table I.
V. I2
TABLE I. Ab initio and DIM⫹SO computed spectroscopic constants for I⫺ 2 The iodine molecule has been the object of innumerable
and I2 . DIM⫹SO results for the ⌸ states listed here are identical to the investigations, both experimentally and computationally,
MRCI results.
only a few of which can be cited here.40,41,61,62 Several stud-
Species State Method R e 共Å兲 D e 共eV兲 ␻ e (cm⫺1 ) ies have addressed all 12 spin–orbit free and 23 spin–orbit
coupled potential energy curves correlating to two ground-
2
⌺⫹ MRCI 3.192 1.226 115
I⫺ 2
⌺⫹
u
DIM⫹SO 3.216 0.984 109 state iodine atoms.63,64 Our spin–orbit free results are similar
2 u,1/2
2 ⫹
⌺ u,1/2 exp. a 3.205 1.014 110⫾2 to those found in these studies and are not shown here. The
2
⌸ g,3/2 MRCI 3.984 0.130 97 spin–orbit coupled curves are shown in Fig. 2. The com-
1 ⫹
⌺g MRCI 2.659 1.978 221 puted spectroscopic constants for the ground and excited B
I2 X̃ 0 ⫹
g
DIM⫹SO 2.672 1.444 213 states are shown in Table I for reference. The major discrep-
X̃ 0 ⫹
g
exp. b 2.666 1.55 215 ancy concerns the ground-state bond energy, which is
3
⌸u MRCI 3.060 0.414 81 slightly too low, as expected from previous results which
a
Reference 34. have shown it to converge only slowly with the size of the
b
Reference 77. one-particle basis set.40
J. Chem. Phys., Vol. 114, No. 17, 1 May 2001 Potentials for I2⫺ , I2 , I3⫺ , and I3 7417

TABLE III. Energies 共in eV兲 of selected states of I⫺

3 relative to the ground-
state minimum, at the MRCI and DIM levels.

State E rel 共MRCI兲 E rel 共DIM兲 Expt.a

1
⌺⫹g 0.000 0.000
1
⌺⫹u 4.496 2.003
1
⌸g 2.924 2.222
1
⌸u 3.620 2.455
II 1 ⌸ g 4.594 4.647
1
⌬g 5.736 4.653
3 ⫹
⌺u 2.582 1.726
3
⌸g 2.626 1.985
3
⌸u 3.517 2.277
II 3 ⌸ g 4.355 4.557
3
⌬u 5.939 4.801
I2 ⫹I⫺ 1.247 0.715 1.31⫾0.06
I⫺
2 ⫹I3/2 2.090 1.467 1.85⫾0.06
I⫹I⫹I⫺ 3.209 2.693 2.85⫾0.06
2
⌸ I3 4.414 4.226⫾0.013

Spectral bands E rel 共MRCI⫹SO兲 Expt.b

0.84 1 ⌺ ⫹
u ⫹0.16 ⌸
3
4.654 4.28
1 ⫹
0.16 ⌺ u ⫹0.84 3 ⌸ 3.661 3.44

FIG. 2. DIM⫹SO potential-energy curves for I2 . The ground 1 ⌺ ⫹

g and
a
References 32 and 37.
b
excited 3 ⌸ u ,0 states are shown in bold. Reference 9.

The spin–orbit coupling between the ground state and periment, a fact which should not be badly affected by the
the excited states is almost completely quenched at the mini- omission of spin–orbit coupling, because this is a closed-
mum, due to the large energy separation between them. This shell species which also dissociates into closed-shell frag-
leads to the residual SOC stabilization at the minimum being ments. This can be verified by comparing the values obtained
of only 0.095 eV, a value comparable to that obtained in with and without spin–orbit coupling with the DIM model.
previous studies.64 As discussed below, the absolute values at these levels are
rather incorrect, but the difference between them should pro-
vide a useful indication of the likely magnitude of spin–orbit
VI. I3À
effects in this species. As can be seen in the Table, neither
The triodide ion has been the subject of a number of the bond length, the stretching frequencies, nor the bond en-
previous computational studies. Many of these have focused ergy are much changed. For the latter quantity, this reflects
only on the electronic ground state, considering such aspects the fact that the residual stabilization by spin–orbit coupling
as its geometry, bond energy, and electronic 共hypervalent兲 at the minimum, of 0.109 eV, is almost identical to that of I2
structure.65–68 However, a number of studies have also ad- at its minimum.
dressed the vertical excitation energies11,55 and part of the Turning to the DIM results, the immediate conclusion to
dissociation pathways.16 be drawn from Table II is that whilst the bond length of the
Unlike for the diatomic species, where the DIM model ground-state I⫺ 3 is well reproduced, neither the bond energy
itself exactly reproduces the ab initio data, for I⫺
3 , the DIM nor the vibrational frequencies are correct. The suggested
results will not match the MRCI results, and the two sets of reason for this poor agreement is discussed below.
data will need to be considered separately and compared. Computed vertical excitation energies 共up to ⬃5 eV兲 of
The computed spectroscopic constants of I⫺ 3 are com- the lower-lying excited states, together with energies of the
pared in Table II to available experimental data. As can be relevant dissociation asymptotes and of the I3 neutral, are
seen, the MRCI results are in excellent agreement with ex- presented in Table III.
Starting with the ab initio results, it can first of all be
seen that I⫺3 supports a fairly large number of excited states
TABLE II. Spectroscopic constants of ground-state I⫺
3 . which are below or close to the level of the neutral. How-
Method R e 共Å兲 D e 共eV兲 ␻ e 共sym兲 ␻ e 共asym兲 ever, some of the states shown, including the important 1 ⌺ ⫹ u
one, and all of the other computed states not shown here
MRCI 2.930 1.247 114 147
because they lie at higher energy, are in formal terms reso-
DIM 2.949 0.715 100 101
DIM⫹SO 2.966 0.729 95 104 nances and not stable states.69
Expt. 2.93a 1.31⫾0.06b 112⫾1c 145d Of the states shown, the lower one with 1 ⌺ ⫹ u symmetry
a
is important because it is the only one to have an appreciable
In crystals, see Ref. 14 and reference therein.
b excitation transition dipole moment, of 11.53 Debye, with
Reference 78.
c
Reference 32. the ground state. Optical transitions to most of the other
d
In solution, Ref. 8. states are symmetry forbidden, and the transition moment for
7418 J. Chem. Phys., Vol. 114, No. 17, 1 May 2001
the allowed excitation to the 1 ⌸ u state is very weak, at 0.16
Debye.
Turning to the DIM results, it can immediately be seen
that the agreement of the spin–orbit free results with the ab
initio data is extremely poor. Part of this disagreement is due
to the poor description of the ground-state, discussed above,
which is the origin of energy in Table III. Had the dissoci-
ated atom asymptote been chosen as a zero of energy, the
overall agreement would have been slightly better. However,
this would not have altered the fact that the ordering of the
states in the DIM method is qualitatively and quantitatively
wrong. In terms of the photochemistry, it is particularly no-
ticeable that the 1 ⌺ ⫹
u state is predicted to lie close to or even
below the 3 ⌸ states, and also well below the three-atom
dissociation limit.
Analysis of the DIM and ab initio wave functions sug-
gests that the reason for the poor quality of the DIM results is
the limited size of the DIM basis set. The latter contains only
those states derived from two neutral iodine atoms and one
iodide anion. The true wave function also contains ‘‘ionic’’
states with two iodide anions and one cationic iodine, such as
I⫺ •••I⫹ •••I⫺ . If one considers the ground 1 ⌺ ⫹ g and excited
1 ⫹
⌺ u states, the extent to which this ‘‘ionic’’ configuration,
which is strongly stabilized by Coulombic interactions,
mixes into both of them can readily explain the mismatch
between the ab initio results, which include this mixing, and
the DIM results, which do not. The ground state of I⫺ 3 is best
described as an in-phase linear combination of I⫺I•••I⫺ and
I⫺ •••I⫺I, but symmetry allows other gerade configurations
such as the ionic one above, to mix in as well. The excited
state is best described as the phase-reversed combination of
these two covalent states, with the mixing-in of the ionic
configuration indicated above symmetry forbidden. It is im-
portant to realize that the DIM method implicitly attributes a
certain stabilizing effect of ionic configurations to all states
described in part by the diatomic ground-state potential-
energy curve of I2 . This is because the corresponding di-
atomic Hamiltonian matrix element is directly derived from
the MRCI potential-energy curve, which owes part of its
stability to an optimum mix of covalent and ionic states.
Therefore, DIM predicts both the ground and excited states
of triodide to be significantly stabilized by admixture of ionic
states. In fact, the ground state is more ionic than predicted
by DIM, because of the very favorable nature of the charge-
alternating structure described above. On the other hand, the
excited state is considerably less ionic than predicted, be-
cause the only symmetry allowed ionic configurations, such
as I⫺ •••I⫺ •••I⫹ , are highly destabilized. Therefore, the
ground state is more stable than predicted by DIM, and the
excited state less, as can be seen in Table III. Approximate
valence-bond wave functions generated for iodine and trio-
dide are in full agreement with this analysis. Note also that
similar arguments can be used, e.g., to understand the over-
estimated stability of the 3 ⌸ u state.
The remedy for this poor behavior of DIM is obvious:
Include ionic states in the basis.70 There is very little doubt
that, done properly, this would yield much better results,
although probably still not comparable in quality to the
MRCI data. Although the DIM Hamiltonian would be larger,
Vala, Kosloff, and Harvey
it should still be manageable computationally, and the num-
ber of extra ab initio diatomic curves required would not be
insuperable. The main problem would be the tedious and
arbitrary fitting work which would be needed to derive the
necessary Hamiltonian matrix elements for a global model of
good quality. It is essentially because of the unstraightfor-
ward nature of this task that we did not extend the model in
this way.
The overall effect of spin–orbit coupling is difficult to
discuss, because only the DIM⫹SO approach gives a global
picture, yet is unreliable due to the underlying DIM prob-
lems. However, a fairly accurate idea can be derived from
the MRCI energies, and a limited number of ab initio calcu-
lations of individual matrix elements, as to what the effect of
spin–orbit coupling is on the absorption spectrum. As could
be expected from symmetry considerations, the 1 ⌺ ⫹ u state
couples strongly only to ⌸ u states, and only one of these is
3
at low enough energy to mix with it substantially. The matrix
element with this state is computed to be of 1361 cm⫺1 . It is
to be noted that despite the poor energetics and simple spin–
orbit coupling operator, the DIM method gives a rather simi-
lar result, of 1487 cm⫺1 .
Upon diagonalizing a model Hamiltonian with the
MRCI energies as diagonal elements, and the ⌺/⌸ and ⌸/⌸
共⫽2428 cm⫺1 ) SOC matrix elements in the off-diagonal po-
sitions, one obtains one pure ⌸ state and two mixed 3 ⌸ u /
1 ⫹
⌺ u states. The lower of these lies 3.661 eV above the 共as-
sumed to be unperturbed兲 ground state, and is 0.84 by weight
3
⌸ in nature, with the 1 ⌺ contributing 0.16 to the weight.
The upper state has the reverse composition and lies 4.654
eV above the ground state. This leads to a predicted spec-
trum containing one low-energy, low-intensity band, and one
high-energy, high-intensity band. These results are in excel-
lent agreement with the experimental spectrum11,37 peaked
around 3.44 eV 共360 nm兲 and 4.28 eV 共290 nm兲. The pre-
dicted ratio of the intensities is also of the correct order of
magnitude. This suggests that the essential physics of the
absorption process is contained by considering only the mix-
ing between the two states. It is to be noted that an identical
assignment was suggested already by Okada,11 based on
Hückel calculations with added spin–orbit coupling; our cal-
culations provide a much firmer theoretical footing to this
conclusion. The same assignment was also derived in the
context of an experimental study17 of the electronic absorp-
tion spectroscopy and magnetic circular dichroism of the ion.
Turning now to the dissociation behavior of the different
states, we first show in Fig. 3 the MRCI potential along the
symmetric stretch coordinate for several of the lower-lying
states of I⫺ 1 ⫹
3 . Both the ⌺ u and ⌸ u states are quite strongly
3
repulsive at the ground-state minimum. However, the former
state barely displays a minimum along the dissociation path-
way, whereas the ⌸ state has a rather deep minimum, of
⬃0.8 eV, at a bond length close to 3.5 Å. This is rather
similar to the behavior of the LEPS excited-state potential.
During photodissociation, such a minimum might be ex-
pected to 共partly兲 ‘‘focus’’ the dissociating wave packet,
whose momentum is initially directed almost entirely to-
wards the three-body asymptote, into the two-body valleys
instead. If this is the case, photodissociation via the lower of
J. Chem. Phys., Vol. 114, No. 17, 1 May 2001
FIG. 3. MRCI computed potential energy curves along the symmetric
stretch dissociation curve, for I⫺ 1 ⫹ 3
3 . Shown in bold are the X ⌺ g , ⌸ u , and
1 ⫹
⌺ u states.
the two absorption bands would be expected to lead to some-
what less intensity in the three-body channel than that via the
upper band, which is mainly 1 ⌺ ⫹ u in nature. This rather sim-
plistic dynamical prediction ignores, however, the expected
strong effect of spin–orbit coupling.
As expected from the poor prediction of vertical ener-
gies, the corresponding diagonal curves computed with the
spin–orbit free DIM model are in medium to poor agreement
with the MRCI results. They are, therefore, not shown here.
However, the curves computed using the DIM⫹SO model
are of some interest because they should give a qualitative
idea of the ‘‘real’’ potential-energy curves. The 63 nonde-
generate curves are displayed in Fig. 4. As can be seen, the
state with maximum 1 ⌺ ⫹ u character is very different from the
MRCI-computed curve in that it has a deep minimum near
3.3 Å. This is mainly due to the incorrect underlying spin–
orbit-free DIM description, which also leads to a deep mini-
mum, so that no real conclusions can be drawn from this
observation. What is more significant is that the second 0 ⫹ u dissociation pathway leading either to I2 ⫹I⫺ or to I⫺ 2 ⫹I. The
state, which is mainly 3 ⌸ u in nature, can be seen to dissoci-
ate to the first excited asymptote (I⫺ ⫹I⫹I*兲. It thereby un-
dergoes a weakly avoided crossing with a 1 ⌺ ⫹ u state, whose
underlying electronic structure is that of a phase-reversed
combination of the diatomic II 1 ⌺ ⫹ g states of I2 . The weak
nature of the avoided crossing can, therefore, be traced back
to the weak interaction between the diatomic X 1 ⌺ ⫹ g and II
1 ⫹
⌺ g states. 共In the DIM model, these states are assumed not
to mix at all, directly, but they do mix a bit indirectly due to
the spin–orbit coupling; this should roughly match the real
coupling, which is expected to be weak, as discussed in the
DIM section.兲 In the real potentials, it is likely that either the
1 ⫹
⌺ u excited state or the 1 ⌸ u , or both, undergo weakly
avoided crossings of this type. This could explain the
Potentials for I2⫺ , I2 , I3⫺ , and I3 7419
FIG. 4. DIM potential energy curves along the symmetric stretch dissocia-
tion curve, for I⫺ 1 ⫹ ⫹
3 . Shown in bold are the ground ⌺ g state, and all 0 u states
which are of same Hund’s case 共c兲 symmetry as the 1 ⌺ ⫹ u state which has a
strong transition dipole moment with the ground state.
experimental observation37 that photodissociation via either
of the bands always leads both to ground and excited iodine
atoms.
Overall, Fig. 4 highlights just how many states are
present in the energy régime at which dissociation takes
place. This shows that nonadiabatic behavior is almost cer-
tain to occur, as is observed experimentally, so that the suc-
cess of previous modeling studies based on single surfaces is
probably due, at least in part, to a fortunate averaging out of
the multistate dynamics. It is also possible, of course, that the
DIM⫹SO surfaces are so incorrect that they also exaggerate
the density of states in the excitation and dissociation win-
dows, and thereby also exaggerate the complexity of the ex-
pected dynamics.
The final element of the triodide potential-energy sur-
faces that we shall consider is the nonsymmetrical, two body
DIM⫹SO potential-energy curves are plotted in Fig. 5 as a
function of one I–I bond length, and for two different values
of the other, fixed, bond length. In Fig. 5共a兲, one bond is
fixed at the equilibrium value in ground-state triodide 共2.95
Å兲, while in Fig. 5共b兲, the fixed bond has the extended length
of 3.3 Å. This should give a better picture of the potential
experienced during photodissociation, because the excited-
state potentials are initially very repulsive along the symmet-
ric stretch coordinate, so that a substantial extension of both
I–I bonds will occur initially whatever the excited state and
the details of the potential.
In Fig. 5共a兲, the energy ordering of the two lowest dis-
sociation asymptotes is as expected: I2 ⫹I⫺ lies below I⫺ 2 ⫹I.
These latter, excited products, are formed adiabatically from
7420 J. Chem. Phys., Vol. 114, No. 17, 1 May 2001
FIG. 5. DIM⫹SO potential-energy curves along the nonsymmetric dissociation coordinate, for I⫺ 3 . One of the bond lengths is set equal to 2.95 Å 共a兲 or 3.30
Å 共b兲. The curves of states of 0 ⫹ symmetry are shown with bold lines. The nature of the two lowest dissociation asymptotes is shown in both cases.
dissociation of the I⫺3 excited states. However, in Fig. 5共b兲,
corresponding to the case where the diatomic iodine–iodine
distance is extended, the energy ordering of the asymptotes
is reversed: The I⫺2 ⫹I products actually lie lowest in energy.
This conclusion concerning the diatomic asymptotes is actu-
ally true despite the severe shortcomings of the DIM method
in the triatomic region. This is because the potentials towards
the right-hand side of Fig. 5 are essentially diatomic. The
reversal in order of the states is due to the fact that at r
⫽3.3 Å, the neutral I2 bond is strongly extended, whereas I⫺ 2
is near its minimum, thereby compensating for the greater
intrinsic stability of the former state. From Fig. 5共b兲, it is
apparent that there must be a quite strongly avoided crossing
between the two states along the dissociation pathway, and
this is indeed clearly so from an analysis of the DIM wave
functions.
As a result of this, the observed production of I⫺ 2 ⫹I
during gas-phase and solution photodissociation experiments
becomes at once easier and more difficult to understand. On
the one hand, Fig. 5 is suggestive in explaining why the
solution experiments do not lead to the ‘‘ground-state’’ prod-
ucts, namely molecular iodine and the iodide molecule, de-
spite the expected collisions with the solvent. This appears to
be because the nature of the dynamics means that the system
only visits regions of the potential-energy surface where the
‘‘excited’’ products are actually more stable. On the other
hand, though, the gas-phase experiments must involve
surface-hopping to the lower state if ground-state I⫺ 2 and I
are formed, as is the case at least part of the time. The exact
dynamics describing this crossing are however hard to pre-
dict at present given the poor accuracy of the DIM potential
in the Franck–Condon region. However, the theoretical con-
clusions are in qualitative accord with the inferred existence
Vala, Kosloff, and Harvey
of complicated photodissociation pathways as obtained in
recent experimental observations in solution21,22 and the gas
phase.32,37
VII. I3
The triodide radical has been the object of extremely few
computational investigations, with none of the published re-
sults being at a particularly reliable level of theory. Viste71
studied the I⫹I2 system using relativistically parametrized
extended-Hückel calculations, and found a linear transition
state at about r e ⫽3.2 Å on the lowest-lying potential-energy
surface, which was assigned a 2 ⌫ 3/2 symmetry within the
C ⬁ v point group. A DIM model for the ground state of the
radical has been proposed,51,52 which has assumed that this
state has a 2 ⌺ symmetry (J⫽1/2), and some extended
Hückel computations have been reported.72 In a recent ex-
perimental study,39 unpublished higher-level computations
are cited which seem to agree with our conclusion, below,
that the triodine radical has a bent symmetric structure in the
absence of spin–orbit coupling.
We have optimized the structure of I3 within the D⬁h ,
C ⬁ v , and 共bent兲 C 2 v point groups. Starting from a linear
structure with two different bond lengths, convergence to the
symmetric geometry is observed, with r e ⫽2.836 Å and a
2
⌸ u electronic state. This state lies just 0.019 eV below the
I2 ⫹I dissociation asymptote, and has vibrational stretching
frequencies of 122 共sym兲 and 209 共asym兲 cm⫺1 . The former
value is in fair agreement with the tentatively assigned ex-
perimental value of 115⫾5 cm⫺1 . The next highest state
within the D⬁h point group is a 2 ⌺ ⫹ g state, with r e ⫽2.860 Å,
but this lies 0.150 eV above the asymptote. At the minimum
of the 2 ⌸ u state, 2 ⌸ g and 2 ⌺ ⫹
u excited states are found to lie
J. Chem. Phys., Vol. 114, No. 17, 1 May 2001 Potentials for I2⫺ , I2 , I3⫺ , and I3 7421

TABLE IV. The ground and low-lying excited states of I3 . A similar simple model for the spin–orbit coupling can
be used for the bent structures. The 2 A1 component of the
State E rel(MRCI⫹SO兲 Expt.a 2
⌸ u state lies 0.253 eV below the linear minimum at a bend-
2
⌸ u,3/2 0.000 0.000 ing angle of 145.4 degrees, whereas the 2 B1 component lies
2
⌺ g,1/2 0.276 0.279
0.470 eV higher in energy at the 2 A1 minimum, where the
2
⌸ u,1/2 0.605 0.615
2
⌸ g,3/2 0.775 0.679 two states are thus split by 0.724 eV. As well as this strong
2
⌸ g,1/2 1.624 splitting, the spin–orbit coupling between the two states de-
2
⌺ u,1/2 3.356 creases drastically, to be of only 930 cm⫺1 . This leads to a
a much smaller spin–orbit stabilization, of only 0.018 eV, at
Reference 30.
the 2 A1 minimum, compared to the 0.335 effect for the 2 ⌸ u
linear state. This leads to a reversal of the order of stability
of the linear and bent minima upon including spin–orbit cou-
at high energy, respectively, 0.731 and 2.942 eV above the pling, with the linear state lying some 0.06 eV lower. This is
dissociated asymptote. in agreement with experimental results38 which suggest that
As in the case of the Br3 radical,73 however, the lowest the I3 radical is a linear centrosymmetric species. Similar
energy is found for the bent structure, due to Renner–Teller conclusions were reached for the Cl3 species,74 which is also
coupling which splits the two components of the 2 ⌸ u state to predicted to be linear 共but not centrosymmetric兲 due to spin–
form 2 A1 and 2 B1 states. The former of these, with an r e of orbit coupling, despite the overall spin–orbit free minimum
2.808 Å, and an angle of 145.4° is found to lie 0.272 eV corresponding to a bent configuration.
below the asymptote. As mentioned before, a DIM model for I3 has been sug-
These results are all obtained without spin–orbit cou- gested, and assumes that the ground state is linear and has a
pling, which, as in the other species studied here, is expected value of 1/2 for the projection of the total angular momen-
to have major effects on the relative energies of different tum onto the molecular axis. In terms of Hund’s case 共a兲
states. To assess this effect, we have computed the spin– states, this would mean the ground state is mostly 2 ⌺ in
orbit coupling elements between the lowest 2 ⌺ g,u and 2 ⌸ g,u nature. This would be in stark contrast with the situation for
states at the linear 2 ⌸ u minimum, and diagonalized the re- both Cl3 74–76 and Br3 , 73 which both have ground states cor-
sulting matrix 共with the MRCI energies as diagonal ele- responding to 2 ⌸ states. The computations described here
ments兲. We obtain six eigenvalues, the lowest of which cor- allow an unambiguous assignment of the ground state as
responds to the 2 ⌸ u,3/2 component of the spin–orbit free being primarily 2 ⌸ u,3/2 in nature. This is because this state is
ground state, and is stabilized by the value of the coupling the lowest in energy in the spin–orbit free ab initio compu-
between the two components, i.e., by 2703 cm⫺1 or 0.335 tations, and it is the only one which is strongly stabilized by
eV. A similar limited SOC calculation for the I2 ⫹I asymp- spin–orbit coupling. The 2 ⌺ ⫹ g state does not lie much higher
tote would lower it by roughly 0.314 eV 共due to the atomic in energy, but is only coupled to very high-lying 2 ⌸ g states,
coupling, neglecting the effect on I2 ). This leaves the ⌸ state and so will not be greatly stabilized by spin–orbit coupling.
overall at 0.040 eV below the asymptote, in moderate agree- It is unclear at this stage whether the disagreement with
ment with the experimental value of 0.143 eV.38 The adia- the DIM model is due entirely to the assumption made in that
batic electron affinity of I3 can be derived from the ab initio model that the ground state is 2 ⌺ in nature, or whether there
total energies of the anion and neutral minima 共⫺34.283 42 are similar intrinsic problems with the DIM methodology
and ⫺34.123 18 a.u., respectively兲 and the approximate SOC applied to I3 to those we found to occur for I⫺ 3 . The con-
stabilization of the neutral 共0.335 eV兲. This gives a value for struction of a DIM model based on L,S atoms with added
Ea (I3 ) of 4.025 eV, in similar agreement with the experi- spin–orbit coupling, as done here for I⫺ 3 , would be relatively
mental value of 4.226⫾0.013 eV.38 The diagonalized SOC trivial 共although the basis set constructed from all states cor-
matrix also leads to predictions for the energies of the ex- relating to 3 2 P iodine atoms would be somewhat larger, with
cited states of the neutral, which are shown in Table IV. As 216 states兲, but was not attempted here.
can be seen, the relative energies of the four lowest levels
共0.00, 0.28, 0.61, and 0.77 eV兲 are in excellent agreement
with the experimental observation of states lying at 0.279, VIII. CONCLUSIONS
0.615, and 0.679 eV. These results are summarized in A theoretical understanding of the photoreactions of I⫺ 3
Table V. requires a detailed knowledge of the potential-energy sur-
faces involved. The very large nuclear charge of iodine pre-
cludes the use of all electron methods, and leads to very
TABLE V. Spectroscopic constants of ground-state I3 2 ⌸ u . important scalar relativistic and spin–orbit coupling effects,
of the same order of magnitude as the binding energy of
Method R e 共Å兲 D e 共eV兲 ␻ e (sym) ␻ e 共asym兲 many of the species studied here. Due to these difficulties
MRCI 2.836 0.019 共0.040 兲a
122 209 any electronic structure description has to be a compromise.
Expt.b 0.143 115⫾5 The main computational approach employed here is
multireference configuration interaction 共MRCI兲 using a
a
The number in brackets refers to the bond energy of the 2 ⌸ u,3/2 state when
spin–orbit effects are approximately included. large one-particle basis set and a relativistic effective core
b
Reference 38. potential 共ECP兲. Within this framework the spin–orbit cou-
7422 J. Chem. Phys., Vol. 114, No. 17, 1 May 2001
pling has been added empirically. The MRCI approach
yields extremely good results for the ground-state properties
of all species studied. The results are found to have predic-
tive value in assigning vibrational frequencies to experimen-
tal results. Also, the computed geometries agree extremely
well with the available experimental data. It is noted that a
correct prediction of the bond energy and bond length re-
quires an accurate treatment of correlation effects together
with rather large basis sets and an ECP. In the case of I⫺ 2 ,
this treatment leads to a significantly improved prediction of
both molecular descriptors compared to previous ab initio
studies 共for a summary see Ref. 36, for instance兲. In particu-
lar, our results describe the whole ground-state potential of
this ion equally well, including the parts sampled by very
highly excited vibrational states.
For the excited states, the involvement of spin–orbit
coupling is much more crucial. Due to the very large number
of states it was not realistic to treat the spin–orbit coupling
between all of them at a rigorous level. However, an empiri-
cal treatment of the spin–orbit coupling yields good agree-
ment with experiment for the potentials of the fairly well-
known diatomic species. For the triatomic species, the
identification and computation of a few key spin–orbit cou-
pling matrix elements allows us to rationalize some of the
experimental data concerning these far less well understood
species. Thus, the two bands in the absorption spectrum of
the anion are explained in terms of a higher-lying, mostly
1 ⫹
⌺ u state, to which optical transition from the ground state is
strongly allowed, and a lower-lying 3 ⌸ u state, to which tran-
sitions are formally forbidden, but which gains intensity
from its substantial spin–orbit coupling-induced mixing with
the 1 ⌺ ⫹
u state. For neutral I3 , a simple model of the SOC
leads to the ground state being predicted to be a weakly
bound, centrosymmetric and linear 2 ⌸ u,3/2 state, despite the
overall MRCI minimum being a bent structure, due to better
spin–orbit stabilization at the linear geometry. This is in
agreement with experiment.38 The energies of the lowest-
lying excited electronic states are also predicted using the
simple SOC model, and are found to be in excellent agree-
ment with experiment.39
The diatomics in molecules model for electronic struc-
ture has gained considerable popularity. Since it supplies a
wave functionlike description of both the ground and excited
potential-energy surfaces, it has been used extensively for
simulating nonadiabatic dynamics in clusters and condensed
phases. In the development of the DIM model for I⫺ 3 we had
two goals. The first was to account for the spin–orbit cou-
pling between all electronic states, thus having a model of
the excited states which would enable us to follow the maze
of potentials leading from I⫺ 3 to its photoproducts. Second,
once a DIM model was established, it could be used as a fast
interpolation scheme to obtain potential-energy surfaces on
which to run the dynamics of the photoreaction. An addi-
tional benefit of the DIM model would be the ability to in-
clude in the model solvation effects. However, comparison
with MRCI computations shows the DIM method to perform
fairly poorly in this case. This suggests that any empirical
DIM model for this system is likely to be of questionable
quality. This failure seems to be due to the inability to treat
Vala, Kosloff, and Harvey
the mixing of ionic and covalent states in a flexible way, due
to the absence of ionic basis states in the DIM model.
Nevertheless, the qualitative appearance of the
DIM⫹SO excited-state potentials for I⫺ 3 does have important
implications for the molecular dynamics. Once the system is
promoted to the excited states, the mazelike nature of the
potentials, with their numerous closely lying avoided cross-
ings, will mean that the amplitude of the excited wave func-
tion will be distributed onto many different states. This
means that the simple picture whereby the photoreaction is
modeled by following a single potential surface from reac-
tants to products is misleading. Instead, the DIM⫹SO model
predicts that all energetically allowed photofragments will be
produced, which is in accordance with existing experiments.
ACKNOWLEDGMENTS
We thank Professor Gabriel Balint-Kurti, Professor San-
ford Ruhman, and Professor Dan Neumark for many valu-
able discussions. J.V. thanks Erez Gershgoren for many in-
teresting conversations about peculiarities of experimental
work. This research was supported by the Israel Science
Foundation administered by the Israel Academy of Science.
The Fritz Haber Center is supported by the Minerva Gesell-
schaft für die Forschung, GmbH München, FRG.
1
U. Banin, R. Kosloff, and S. Ruhman, Isr. J. Chem. 33, 141 共1993兲.
2
U. Banin and S. Ruhman, J. Chem. Phys. 98, 4391 共1993兲.
3
B. J. Greenblatt, Ph.D. thesis, University of California at Berkeley, 1999.
4
G. Ashkenazi, U. Banin, A. Bartana, R. Kosloff, and S. Ruhman, Adv.
Chem. Phys. 100, 229 共1997兲.
5
J. C. Roy, W. H. Hanillm, and R. R. Williams, Jr., J. Am. Chem. Soc. 77,
2953 共1955兲.
6
L. I. Grossweiner and M. S. Matheson, J. Phys. Chem. 61, 1089 共1957兲.
7
A. Barkatt and M. Ottolenghi, Mol. Photochem. 6, 253 共1974兲.
8
W. Gabes and H. Gerding, J. Mol. Struct. 14, 267 共1972兲.
9
W. Gabes and D. J. Stufkens, Spectrochim. Acta A 30, 1835 共1974兲.
10
T. Okada, J. Phys. Soc. Jpn. 50, 582 共1981兲.
11
T. Okada and J. Hata, Mol. Phys. 43, 1151 共1981兲.
12
M. Mizuno, J. Tanaka, and I. Harada, J. Phys. Chem. 85, 1789 共1981兲.
13
L. E. Topol, Inorg. Chem. 10, 736 共1971兲.
14
W. Gabes and M. A. M. Nijman-Messter, Inorg. Chem. 12, 589 共1973兲.
15
H. L. Friedman, J. Chem. Phys. 21, 319 共1953兲.
16
P. W. Tasker, Mol. Phys. 180, 1 共1994兲.
17
H. Isci and M. R. Mason, Inorg. Chem. 24, 271 共1985兲.
18
U. Banin, A. Bartana, S. Ruhman, and R. Kosloff, J. Chem. Phys. 101,
8461 共1994兲.
19
E. Gershgoren, S. Ruhman, J. Vala, and R. Kosloff, J. Phys. Chem. A 共to
be published兲.
20
I. Benjamin, U. Banin, and S. Ruhman, J. Chem. Phys. 98, 8337 共1993兲.
21
E. Gershgoren, U. Banin, and S. Ruhman, J. Phys. Chem. A 102, 9 共1998兲.
22
Z. Wang, T. Wasserman, E. Gershgoren, J. Vala, R. Kosloff, and S.
Ruhman, Chem. Phys. Lett. 313, 155 共1999兲.
23
U. Banin, R. Kosloff, and S. Ruhman, Chem. Phys. 183, 289 共1994兲.
24
T. Kühne and Vöhringer, J. Chem. Phys. 105, 10788 共1996兲.
25
T. Kühne and Vöhringer, J. Phys. Chem. A 102, 4177 共1998兲.
26
T. Kühne, R. Küster, and Voḧringer, Chem. Phys. 233, 161 共1998兲.
27
A. E. Johnson and A. B. Myers, J. Chem. Phys. 102, 3519 共1995兲.
28
A. E. Johnson and A. B. Myers, J. Phys. Chem. 100, 7778 共1996兲.
29
A. E. Johnson and A. B. Myers, J. Chem. Phys. 104, 2497 共1996兲.
30
B. J. Greenblatt, M. T. Zanni, and D. M. Neumark, Chem. Phys. Lett. 258,
523 共1996兲.
31
M. T. Zanni, T. R. Taylor, B. J. Greenblatt, B. Soep, and D. M. Neumark,
J. Chem. Phys. 107, 7613 共1997兲.
32
M. T. Zanni, V. S. Batista, B. J. Greenblatt, W. H. Miller, and D. M.
Neumark, J. Chem. Phys. 110, 3748 共1999兲.
33
V. S. Batista, M. T. Zanni, B. J. Greenblatt, D. M. Neumark, and W. H.
Miller, J. Chem. Phys. 110, 3748 共1999兲.
J. Chem. Phys., Vol. 114, No. 17, 1 May 2001 Potentials for I2⫺ , I2 , I3⫺ , and I3 7423

34
M. T. Zanni, A. V. Davis, C. Frishkorn, M. Elhanine, and D. M. Neumark, 59
P. E. Maslen, J. Faeder, and R. Parson, Chem. Phys. Lett. 263, 63 共1996兲.
J. Chem. Phys. 112, 8847 共2000兲; 113, 8854 共2000兲. 60
S. B. Sharp and G. I. Gellene, Mol. Phys. 98, 667 共2000兲.
35
M. T. Zanni, B. J. Greenblatt, A. V. Davis, and D. M. Neumark, Proc. 61
R. S. Mulliken, J. Chem. Phys. 55, 288 共1971兲.
SPIE 3271, 196 共1998兲. 62
P. Schwerdtfeger, L. v. Szentpály, K. Vogel, H. Silberbach, H. Stoll, and
36
M. T. Zanni, B. J. Greenblatt, A. V. Davis, and D. M. Neumark, J. Chem. H. Preuss, J. Chem. Phys. 84, 1606 共1986兲.
Phys. 111, 2991 共1999兲. 63
J. Li and K. Balasubramanian, J. Mol. Spectrosc. 138, 162 共1989兲.
37
H. Choi, R. T. Bise, A. A. Hoops, and D. M. Neumark, J. Chem. Phys. 64
C. Teichteil and M. Pelissier, Chem. Phys. 180, 1 共1994兲.
113, 2255 共2000兲. 65
38
T. R. Taylor, K. R. Asmis, M. T. Zanni, and D. M. Neumark, J. Chem. G. A. Landrum, N. Goldberg, and R. Hoffmann, J. Chem. Soc., Dalton
Phys. 110, 7607 共1999兲. Trans. 1997, 3605.
39
H. Choi, T. R. Taylor, R. T. Bise, A. A. Hoops, and D. M. Neumark, J.
66
S. B. Sharp and G. I. Gellene, J. Phys. Chem. A 101, 2192 共1997兲.
67
Chem. Phys. 113, 8608 共2000兲. R. M. Lynden-Bell, R. Kosloff, S. Ruhman, D. Danovich, and J. Vala, J.
40
M. Dolg, Mol. Phys. 88, 1645 共1996兲. Chem. Phys. 109, 9928 共1998兲.
41 68
W. A. de Jong, L. Visscher, and W. C. Nieuwpoort, J. Chem. Phys. 107, A. Sanov, T. Sanford, L. J. Butler, J. Vala, R. Kosloff, and W. C.
9046 共1997兲. Lineberger, J. Phys. Chem. A 103, 10244 共1999兲.
42
MOLPRO is a package of ab initio programs written by H.-J. Werner and P. 69
When very diffuse s and p primitives ( ␣ ⫽0.015) are added to the basis
J. Knowles, with contributions from R. D. Amos, A. Berning, D. L. set, some of the higher-lying excited states are shown to be indeed reso-
Cooper et al.
43 nances by one of the occupied orbitals becoming very diffuse and con-
M. Dolg, Habilitationsschrift, Universität Stuttgart, 1997.
tinuumlike. These extra-diffuse primitives have a negligible effect on the
44
F. Y. Naumkin, Chem. Phys. 226, 319 共1998兲.
45
F. Y. Naumkin, Chem. Phys. 240, 79 共1999兲. energy of the valence electronic states studied here and were not included
46
J. C. Tully, in Potential Energy Surfaces, edited by K. P. Lawley 共Wiley, in the basis set used to produce the results described in this paper.
70
New York, 1980兲, p. 63. For an example of where ionic states improve the results from DIM, see
47
P. J. Kuntz, Ber. Bunsenges. Phys. Chem. 86, 367 共1982兲. B. L. Grigorenko, A. V. Nemukhin, and V. A. Apkarian, Chem. Phys.
48
D. Horinek and B. Dick, Phys. Chem. Chem. Phys. 1, 2667 共1999兲. 219, 161 共1997兲.
49
V. S. Batista and D. F. Coker, J. Chem. Phys. 105, 4033 共1996兲. 71
A. Viste and P. Pyykkö, Int. J. Quantum Chem. 25, 223 共1984兲.
50
V. S. Batista and D. F. Coker, J. Chem. Phys. 106, 7102 共1997兲. 72
N. N. Gorinchoi, F. Cimpoesu, and I. B. Besuker, J. Mol. Struct.:
51
C. J. Margulis, D. A. Horner, S. Bonella, and D. F. Coker, J. Phys. Chem. THEOCHEM 530, 281 共2000兲.
A 103, 9552 共1999兲. 73
P. Schuster, H. Mikosch, and G. Bauer, J. Chem. Phys. 109, 1833 共1998兲.
52
C. J. Margulis and D. F. Coker, J. Chem. Phys. 113, 6113 共2000兲. 74
A. L. Kaledin, M. C. Heaven, W. G. Lawrence, Q. Cui, J. E. Stevens, and
53
I. H. Gersonde and H. Gabriel, J. Chem. Phys. 98, 2094 共1993兲. K. Morokuma J. Chem. Phys. 108, 2771 共1998兲.
54
The parameters of these diatomic curves, as well as all the other data 75
A. L. Kaledin, M. C. Heaven, K. Morokuma, and D. M. Neumark, Chem.
described here, and the FORTRAN programs used for the DIM models, can
Phys. Lett. 306, 48 共1999兲.
be obtained upon request from JNH. 76
55
D. Danovich, J. Hrušák, and S. Shaik, Chem. Phys. Lett. 233, 249 共1995兲. J. B. Giorgi, F. Y. Naumkin, J. C. Polanyi, S. A. Raspopov, and N. S.-K.
56
M. N. Glukhovtsev, A. Pross, H. B. Schlegel, R. D. Bach, and L. Radom, Sze, J. Chem. Phys. 112, 9569 共2000兲.
J. Am. Chem. Soc. 118, 11258 共1996兲.
77
K. P. Huber and G. Herzberg, Constants of Diatomic Molecules 共Van
57
B. Braı̈da and P. C. Hiberty, J. Phys. Chem. A 104, 4618 共2000兲. Nostrand-Reinhold, New York, 1979兲.
58 78
P. W. Tasker, G. G. Balint-Kurti, and R. N. Dixon, Mol. Phys. 32, 1651 K. Do, T. P. Klein, C. A. Pommerening, and L. S. Sunderlin, J. Am. Soc.
共1976兲. Mass Spectrom. 8, 688 共1997兲.