Upload
mercedes-alonso
View
212
Download
0
Embed Size (px)
Citation preview
A Universal Scale of Aromaticity for
p-Organic Compounds
MERCEDES ALONSO, BERNARDO HERRADON
Instituto de Quımica Organica General, CSIC, Juan de la Cierva 3, 28006 Madrid, Spain
Received 8 May 2009; Revised 10 June 2009; Accepted 10 June 2009DOI 10.1002/jcc.21377
Published online 27 July 2009 in Wiley InterScience (www.interscience.wiley.com).
Abstract: Aromaticity is an essential concept in chemistry, invented to account for the stability, reactivity, molec-
ular structure, and properties of many organic and inorganic compounds. In recent years, numerous methods to quan-
tify aromaticity based on the energetic, magnetic, structural, and electronic properties of molecules have been pro-
posed but none of them is universal. The inability of establishing a universal scale of aromaticity based on a single
parameter is due to the multidimensional character of this phenomenon. Consequently, aromaticity analyses should
be carried out by employing a set of aromaticity descriptors on the basis of different physical manifestations of aro-
maticity. Here, we report a universal scale of aromaticity for p-organic compounds based on the Euclidean distance
between neurons in a self-organizing map. The most widely used aromaticity indicators have been used as molecular
descriptors, and so our approach provides the first scale of aromaticity which contains the energetic, magnetic, and
structural aspects of this property. The method is applicable to a wide variety of unsaturated organic compounds and
allows quantification of both aromaticity and antiaromaticity. Additionally, the position of a compound on the bi-
dimensional map determinates immediately the following: (a) the group (aromatic, nonaromatic, or antiaromatic) to
which the system belongs, (b) their degree of p-electronic delocalization, and (c) the similarity in aromaticity/anti-
aromaticity between different compounds. This new scale of aromaticity is able to indicate the expected order of
aromaticity of analogues of fulvene and heptafulvene, heteroaromatic species, substituted benzenes, and functional-
ized cyclopentadienyl compounds.
q 2009 Wiley Periodicals, Inc. J Comput Chem 31: 917–928, 2010
Key words: aromaticity; neural network; aromaticity indices; antiaromaticity; NICS scan
Introduction
Aromaticity is one of the most important concepts in chemis-
try.1–5 Although it was first studied with organic molecules, it
has been recently applied to inorganic compounds.6,7 Similar to
other fundamental chemical properties such as charge, electrone-
gativity, or bond order, aromaticity is not an observable one and
it has to be defined by convention and measured by a relative
scale.8,9
Although aromaticity is a unique physical phenomenon, it is
manifested in multiple ways. Some consequences of aromaticity
are energetic stabilization as well as particular chemical reactiv-
ity, geometry, magnetic and electronic properties.10,11 Its differ-
ent manifestations have lead to multiple ways for assessing the
aromatic or antiaromatic character of molecules.12 Thus, several
aromaticity indices have been proposed in the literature, which
give distinct orders of aromaticity.13,14 Accordingly, most
authors advise to use a set of aromaticity descriptors based on
different physical properties to characterize aromatic com-
pounds.1,14,15 This fact has prompted a lively discussion on the
monodimensional or multidimensional character of aromatic-
ity.16–19 As pointed out earlier, we consider that aromaticity is a
single physical phenomenon with multiple manifestations, that
is, a multifaceted property, which might be measured by a single
scale that takes into account the different dimensions.
Why aromaticity has to be measured? First, by a pure scien-
tific reason because each property has to be quantified. From a
more practical point of view, arenes are key components in bio-
logically active compounds and technological useful materials
which have deeply influenced on human welfare.20 The most
fundamental property of arenes is aromaticity which is widely
used for understanding the molecular structure, chemical bond-
Additional Supporting Information may be found in the online version of
this article.
Correspondence to: M. Alonso or B. Herradon; e-mail: mercuea@
iqog.csic.es or [email protected]
Contract/grant sponsor: The Spanish Ministry of Education and Science;
contract/grant number: CTQ2007-64891/BQU
Contract/grant sponsor: Spanish Ministry of Education and Science
(MEC) (FPU fellowship)
q 2009 Wiley Periodicals, Inc.
ing, and properties as well as to predict stability and reactivity
of numerous organic and inorganic compounds. Therefore, a
quantitative scale of aromaticity is highly valuable for the design
of new molecules and materials.1–3,8 This is the goal of this
research.
Aromaticity has been measured using a variety of indices
based on geometrical,11,21 energetic,22,23 electronic,24–26 and
magnetic27,28 criteria. However, no fully satisfactory result has
yet been achieved because of its multiple manifestations, con-
cluding that there is not yet a single indicator of aromaticity that
works properly for all cases.9 Although, in principle, we can
consider that aromaticity might be expressed by a lineal combi-
nation of an arbitrary number of indices, this approach has met
with limited success13,14,29 which likely indicates that the rela-
tionship between aromaticity and the different criteria (or mani-
festations of the phenomenon) is nonlinear (see Supporting In-
formation).
Neural networks are useful mathematical tools for modeling
nonlinear functions in many real-world applications, without
knowing the analytic forms in advance.30 These artificial sys-
tems emulate the function of the brain, where a very high num-
ber of information-processing neurons are interconnected and
are known for their efficiency in self-organization, pattern recog-
nition, and dimensionality reduction. They are capable of learn-
ing complex interactions among the input variables even when
they are difficult to find and describe. Consequently, neural net-
works can provide solutions for problems that do not have an
algorithmic solution. They have been intensively used in differ-
ent fields of knowledge such as biology, engineering, and econ-
omy. In chemistry, the use of neural networks has further
expanded into the analysis of spectral data, drug design, predic-
tion of chemical reactivity and physical properties as well as the
development of quantitative structure–activity relationship.31–36
Some advantages of neural networks, useful for the purpose
of this work, are as follows: (i) they are adaptative, i.e., they
can take data and learn from it; (ii) they are essentially nonlin-
ear; (iii) they are capable of generalization, i.e., they can cor-
rectly process information that only broadly resembles the origi-
nal data training; (iv) they are fault-tolerant being capable of
properly handling noisy or incomplete data.
Recently, we reported the successful applications of Koho-
nen’s self-organizing maps (SOMs), an unsupervised neural net-
work, to classify five-membered heterocycles according to their
aromatic/antiaromatic character.37 Some useful characteristics of
the proposed method for measuring aromaticity are as follows:
(1) it is very fast being practically instantaneous to place a new
compound in the map; (2) the placement of the different com-
pounds is conveniently visualized; (3) it has predictive power.
In this article, we expand this methodology to a full set of
organic compounds ranging from highly aromatic to highly anti-
aromatic (Chart 1). The method is applicable to molecules of
different sizes (five-, six-, seven-, and eight-membered rings),
both carbocyclic and heterocyclic, including neutral, cationic, or
anionic species, and it allows us to quantify both aromaticity
and antiaromaticity. Since the input data in the neural network
are diverse energetic, magnetic, and structural descriptors of aro-
maticity, the SOM represents the multidimensional character of
aromaticity and, more importantly, the Euclidean distance
between neurons is a universal scale of aromaticity for p-organiccompounds which takes into account the different physical mani-
festations of this phenomenon.
Computational Details
All calculations have been performed with Gaussian03 package
of program at the MP2 level of theory with the 6-3111G(d,p)
basis set.38 The geometries of five- and six-membered rings,
penta- and heptafulvene and analogues, substituted benzenes,
and substituted cyclopentadienyl compounds were fully opti-
mized and characterized by harmonic vibrational frequency com-
putation, which showed that all structures were minima on the
potential energy surface. Four well strain-balanced homodes-
motic reactions based on cyclic reference compounds were used
to estimate the aromatic stabilization energy (ASE) of five- and
six-membered rings (Scheme 1).39,40 The energies were cor-
rected by MP2/6-3111G(d,p) zero-point energies. Systems with
largely enhanced aromatic stabilization energies (positive values
of ASE) are aromatic, whereas those with strongly negative
ASE values are considered to be antiaromatic.22
For the evaluation of aromatic stabilization energies of hepta-
fulvene and analogues, cycloheptatrienyl cation, and cycloocta-
tetraenyl dication, we have used the isomerization method (ISE),
which is based on the energy differences between a methyl de-
rivative of the conjugated system and its nonconjugated exo-
cyclic methylene isomer (Scheme 2).41 As one closely related
reference compound is involved, ISE provides excellent esti-
mates of aromatic stabilization energies that minimize perturbing
influences such as strain.
The magnetic susceptibility exaltation (L) is defined as the
difference between the magnetic susceptibility of a compound
(vM) and a reference one without cyclic electron delocalization
(vM0) [eq. (1)].42,43 The exaltations were obtained from the reac-
tions indicated in Schemes 1 and 2. The magnetic suscepti-
bilities were computed using the CSGT method44 at the HF/
6-3111G** level of theory. The exaltations are negative (dia-
magnetic) for aromatic compounds and positive (paramagnetic)
for antiaromatic compounds.
K ¼ vM � vM0 (1)
As additional magnetic indices of aromaticity, we have used
three derivations of the nucleus-independent chemical shift
(NICS).45 This index is defined as the negative value of the
absolute magnetic shielding computed at ring centers or another
interesting point of the system.46 NICS values have been com-
puted at the ring centers determined by the nonweighted mean
of the heavy atoms coordinates. The NICS(1) values calculated
1 A above the molecular plane are considered to better reflect
the p-effects. Another descriptor is the out-of-plane component
of the tensor NICS computed at 1 A above the ring center,
denoted as NICSzz(1), which was recently found to be a good
measure for the characterization of the p system of the ring. The
GIAO/HF/6-3111G(d,p) method were used for the NICS calcu-
lations.47 Rings with highly negative values of NICS are quanti-
918 Alonso and Herradon • Vol. 31, No. 5 • Journal of Computational Chemistry
Journal of Computational Chemistry DOI 10.1002/jcc
fied as aromatic, whereas those with positive values are antiaro-
matic. For 25 compounds of the training and validation dataset,
the NICS values have been also evaluated in the ring critical
point (rcp),48 as suggested by Morao et al.49 The results at the
rcp are practically identical to those calculated at the geometri-
cal ring center and they are reported in the supporting informa-
tion (Supporting Information Table S1 and Figure S1).
Isotropic NICS values, used as descriptors in the former
classification,37 do not describe correctly the aromatic/anti-
aromatic character for most of the derivatives of cyclo-
pentadienyl anion and cation.50 As previously reported, these
magnetic indices contain large influences from the r system
and from all three principal components of the NICS tensor.51
For the NICS scans, we have plotted the NICS values along
the z-axis to the ring plane beginning on the geometric center
of the molecular ring up to 5.0 A at intervals of 0.1 A. The
eigenvalues of the chemical shift tensors were used to separate
the isotropic NICS values into their in-plane and out-of-plane
components.52,53
As a structure-based measure, we have employed the har-
monic oscillator model of aromaticity (HOMA) defined by Krus-
zewski and Krygowski as follows54,55:
Chart 1. Five- and six-membered ring compounds used in the training of the neural network. Com-
pounds marked with an asterisk were used in the validation of the self-organizing map. The structures
of additional test compounds are indicated in Table 2.
919Scale of Aromaticity for p-Organic Compounds
Journal of Computational Chemistry DOI 10.1002/jcc
HOMA ¼ 1� an
Xn
i¼1
ðRopt � RiÞ2 (2)
where n is the number of bonds taken into the summation, and a is
an empirical constant fixed to give HOMA 5 0 for a model nonar-
omatic system and HOMA 5 1 for a system with all bonds equal
to an optimal value Ropt, assumed to be realized for fully aromatic
system. Ri is the running bond length. This index was found to be
one of the most effective structural indicators of aromaticity.11
Once the aromaticity descriptors have been obtained, we
have generated a family of input vectors which represent each
compound of the training dataset.
The Kohonen networks were obtained with the SOM_PAK
program.56 About 50 SOMs were trained varying both the map
size (number of neurons) and training parameters. Two different
topologies, namely rectangular and hexagonal, were also tested
finding that the hexagonal lattice was better for visual inspec-
tion. The quantization error was used to evaluate the perform-
ance of the network, which is the average distance between the
descriptor vectors and their respective map neurons:
Error ¼
PP
p¼1
PN
i¼1
ðxpi � wjiÞ2
P(3)
The best map is expected to yield the smallest average quan-
tization error with respect to the training and validation set. In
this case, a hexagonal lattice with 26 3 16 neurons using Bub-
ble function as neighborhood kernel was selected. Training was
done in two phases: an ordering phase with 2000 steps and a
self-organizing phase with 20,000 steps. During the first cycle, it
carried out 2000 training steps and the learning rate and initial
neighborhood were set to 1 and 10, respectively. The parameter
values for the second cycle were 0.05 and 1, respectively.
Results and Discussion
In this pattern, each compound is represented by four widely
used aromaticity descriptors based on energetic, magnetic, and
structural manifestations of aromaticity: the ASE, the magnetic
susceptibility exaltation (L), the out-of-plane component of the
NICS tensor computed at 1 A above the ring center [NICSzz(1)],
Scheme 2. Isodesmic reaction used to evaluate the isomerization
energies (ISE) and magnetic susceptibility exaltation (L) of hepta-
fulvene and analogues, C7H71 and C8H8
21.
Scheme 1. Homodesmotic reactions used to evaluate the aromatic stabilization energies (ASE) and
magnetic susceptibility exaltation (L) of five-membered heterocyclic compounds (a), substituted cyclo-
pentadienyl compounds (b), six-membered heterocycles (c), and substituted benzenes (d).
920 Alonso and Herradon • Vol. 31, No. 5 • Journal of Computational Chemistry
Journal of Computational Chemistry DOI 10.1002/jcc
and the HOMA. We have previously demonstrated that these
indices are the most suitable for describing the aromaticity/anti-
aromaticity of a wide variety of organic compounds50 and, more
importantly, they are readily calculated and implemented in
many quantum chemistry programs. In the former classification
we used the isotropic values of NICS, but NICSzz(1) was found
to be the best NICS-based indicator of aromaticity for substi-
tuted cyclopentadienyl compounds, because the local r-bondingcontributions are diminished and the effects of the p-electronring current are dominant.
Chart 1 shows the molecular structures of the 150 five- and
six-membered rings used in the training and validation of the
network. This comprehensive dataset contains a wide range of
aromatic, nonaromatic, and antiaromatic compounds. The values
of the ASE, L, NICSzz(1), and HOMA for each system are
given in Supporting Information Table S2. These descriptors are
used as a four-dimensional input vector for the SOM, which
creates a nonlinear projection onto a two-dimensional space pre-
serving topological relations as faithfully as possible. SOM
automatically clusters similar compounds based on these
descriptors. A 21 3 16 output grid with hexagonal boundaries
and the lowest quantization error (0.92) was chosen. The ratio of
the size of the network when compared with the size of the
dataset is bigger than the previously reported one37 in order to
increase the interpolation capability of SOM for new com-
pounds. A visually effective way for showing the results is a
Sammon map (Fig. 1), a nonlinear projection generated through
an iterative procedure.57,58 In this map, the four-dimensional
input vectors are reduced to points with coordinates (x, y), sothe distances among points corresponds to Euclidean distances
between neurons in the SOM. Accordingly, Sammon map allows
the visualization of the shape of the clusters and the relative
distance between them.
The map shows fairly well the quantitative character of the
classification obtained. The neuron activated by benzene (144)
represents the highest degree of aromaticity, whereas the neuron
activated by singlet cyclopentadienyl cation (35) represents the
highest degree of antiaromaticity. Aromaticity decreases gradu-
ally from benzene to cyclopentadienyl cation. Accordingly, the
Euclidean distance to the neuron activated by benzene (dj), thereference aromatic compound, can be used as a quantitative
measurement of aromaticity/antiaromaticity in five- and six-
membered rings. It is important to note that many aromaticity
descriptors have a series of disadvantages when comparing rings
of different size. For example, magnetic susceptibility exaltation,
and to a lesser extent NICS, depends heavily on the ring area59
and ASE is strongly dependent on the reaction scheme which is
necessarily different for five- and six-membered rings.22,40
The U-matrix, the matrix of distances between adjacent units,
was used to analyze the distribution of classes of the trained
SOM (Fig. 2). The distances between the neighboring neurons
are visualized by gray levels, so the color difference indicates
dissimilarities or similarities between weight vectors of neigh-
borhood neurons in the SOM. If color changes between adjacent
neurons are smooth the neurons are more similar than when
color changes are sharp. Therefore, clusters are white zones sur-
rounded by black boundaries. Figure 2 shows three large clusters
on the left-hand side, middle region, and right-hand side.
Compounds placed on the left cluster are stabilized energetically
and exhibit diamagnetic susceptibility exaltation and highly neg-
Figure 1. Sammon map obtained for the training dataset, showing the relative distances between the
input variables [ASE, L, NICSzz(1), and HOMA] in the original space. The color scale indicates the
Euclidean distances between the weight vector of each neuron and the neuron activated by benzene.
921Scale of Aromaticity for p-Organic Compounds
Journal of Computational Chemistry DOI 10.1002/jcc
ative values of NICSzz(1) as well as practically equalized bond
lengths (HOMA values ranges from 0.6 to 1), that is, aromatic
compounds. Benzene (144), pyridine (145), pyrrole (3), thio-
phene (2), furan (1), and substituted cyclopentadienyl anions
(96–119) are some compounds located in this region. On the
contrary, right cluster includes antiaromatic compounds, which
are destabilized and exhibit paramagnetic susceptibility exalta-
tion and the single and double bond lengths are localized, such
as borole (31), 1H-silolylium (43), and all the substituted deriva-
tives of cyclopentadienyl cation (120–143). Rings located on the
middle side can be classified as nonaromatic and show interme-
diate values for ASE, L, NICSzz(1), and HOMA. Some repre-
sentative compounds belonging to this class are phosphole (4),
arsole (53), cyclohexadiene, fulvene, and all the substituted
cyclopentadienes (96–119).
Consequently, from energetic, magnetic, and structural
descriptors, SOM automatically classifies the extensive dataset
of 150 five- and six-membered rings into three classes: aromatic,
nonaromatic, and antiaromatic. The border between these three
classes is well defined, so the challenging problematic classifica-
tion of borderline compounds is overcome by using the Kohonen
neural network. According to the neural network, the Euclidean
distance ranges from 0 to 33.2 for aromatic systems, from 35.3
to 68.3 for nonaromatic compounds, and from 72.6 to 170.6 for
antiaromatic molecules.
Additionally, SOM places compounds with similar but non-
identical aromatic character in neighboring neurons, creating a
smooth transition of different aromaticity degrees over the whole
map. Figure 3a shows the trained SOM with the neurons color
coded. The color scale indicates the Euclidean distances between
the weight vectors of each neuron and the neuron activated by
benzene (144). White neurons represent the cluster boundaries
determined by the U-matrix map as discussed earlier.
Table 1 collects the Euclidean distance values for compounds
1–150 used in the training and validation of the network. In con-
trast to our former classification (used mostly for five-membered
rings and not trained for highly aromatic and antiaromatic sys-
tems),37 the new SOM is able to quantify the aromaticity for six-
membered ring as well as antiaromaticity for derivatives of cyclo-
pentadienyl cation. According to the new scale of aromaticity,
benzene and pyridine are the most aromatic six-membered rings
(dj 5 0.0), followed by pyrazine (dj 5 10.2), pyridazine (dj 511.4), and finally, pyrimidine (dj 5 16.8). To discern among ben-
zene and pyridine, a larger map is necessary which assigns a dj 54.1 to the latter (see below). In relation to derivatives of cyclopen-
tadienyl cation, the dj values indicate that the substituents signifi-
cantly reduce the antiaromaticity of 35, especially the electron-
donating hydroxyl group, decreasing dj from 170.6 to 78.0. In
fact, SOM places the 2,5- (129) and 2,3,5- (137) hydroxylated
derivatives in the border with nonaromatic compounds.
Figure 2. U-matrix representation of the Kohonen network: the distances between neighboring neurons
are visualized by gray levels. Darker hexagons indicate a large distance.
Journal of Computational Chemistry DOI 10.1002/jcc
922 Alonso and Herradon • Vol. 31, No. 5 • Journal of Computational Chemistry
Figure 3. (a) Kohonen network obtained for the classification of 150 five- and six-membered rings
(1–150) according to their degree of aromaticity/antiaromaticity. The numbers in brackets represent the
neuron coordinates. Neurons are colored based on the Euclidean distance between the weight vectors
of each neuron and the neuron activated by benzene (144) and pyridine (145). White numbers repre-
sent the compounds used in the validation of the network and white neurons represent the cluster
boundaries. (b) Extension of the region of highly aromatic compounds with dj between 0 and 10,
obtained from a 14 3 12 map training only with aromatic compounds.
Figure 4. NICS values as a function of distance of benzene (144), 2,3,4,5-tetrahydroxycyclopenta-
dienyl anion (117), 3-phospaphosphole (12), gallole (47), 2,5-dihydroxycyclopentadienyl cation (129),
and cyclopentadienyl cation (35): (black circles) Isotropic chemical shift; (green circles) Out-of-plane
component; (red circles) In-plane component.
We have carried out additional tests to check the universality
of our method for quantifying aromaticity. Sola and coworkers
have recently established several tests to assess the performance
of the existing indices of aromaticity (Table 2).60 Only the tests
related to the aromaticity of monocycles have been proved.
According to the nature of the aromaticity indices used as
descriptors in the input vectors of the SOM, the Euclidean dis-
tance is a measure of the global aromaticity of the molecule.
Interestingly, the possibility of introducing compounds lacking
some of the descriptors allows to apply the neural network to
analyze also local aromaticity based on the HOMA and
NICSzz(1) values. It would be advisable to generate another
Kohonen network for measuring local aromaticity which takes
into account, apart from the structural and magnetic indicators,
several indices of electronic such as the aromatic fluctuation
index or the multicenter indices. However, the high computa-
tional cost of electronic indices involves a drawback for the
method functionality.61
The first test involves the study of aromaticity in a set of
substituted analogues of fulvene (Table 2A). It is well known
that the p-electron delocalization in analogues of fulvene
depends strongly on the electronegativity of the X substituents.
Electronic effects of these systems are usually rationalized in
terms of their differing weights of the zwitterionic resonance
structures. Electron-donating substituents enhance the aromatic-
ity of pentafulvenes by increasing the weight of the 6p-electronstructure A, whereas electron withdrawing substituents increase
the antiaromaticity of the ring favoring 4p-electron structure B
(Scheme 3).62 Therefore, a good aromaticity measure should
give the following order of aromaticity for the analogues of
fulvene: BH22 [ CH2 [ NH [ O [ NH2
1. This is exactly the
order provided by the Euclidean distance values, ranging from
18.5 to 82.0 (Table 2A). Additionally, SOM places compound
151 (X 5 BH22) in the region of the highly aromatic com-
pounds while 153 (X 5 NH21) is mapped into an antiaromatic
neuron.
Table 1. Euclidean Distance Values (dj) and Coordinates of the Winner Neuron for Compounds 1–150.a,b,c
Aromatic Nonaromatic Antiaromatic
dj Compound Neuron dj Compound Neuron dj Compound Neuron dj Compound Neuron dj Compound Neuron
0.0 104 (0, 0) 18.8 26 (4, 2) 35.9 12 (6, 9) 46.0 70* (11, 12) 72.6 129 (13, 0)
0.0 145 (0, 0) 18.8 27 (4, 2) 36.2 49 (8, 5) 46.3 56 (12, 11) 78.0 137 (14, 2)
9.9 107 (0, 2) 18.9 112 (5, 3) 36.6 4 (8, 6) 46.3 69 (12, 11) 87.6 50 (20, 11)
10.2 146 (0, 3) 19.0 21 (4, 4) 37.4 8 (7, 7) 46.3 73 (12, 11) 89.0 31 (17, 6)
10.3 99 (1, 2) 19.3 15* (3, 6) 38.6 16 (8, 8) 46.4 80 (11, 7) 90.8 43 (20, 9)
11.4 45 (0, 4) 19.3 110 (3, 6) 41.0 53 (10, 4) 46.4 88 (11, 7) 99.6 121 (15, 4)
11.4 52 (0, 4) 19.4 106 (1, 8) 41.5 33 (6, 11) 46.4 92* (11, 7) 101.4 130 (15, 0)
11.4 148* (0, 4) 20.1 101 (2, 7) 41.6 63 (8, 9) 47.9 87 (11, 15) 101.7 141 (20, 7)
12.3 18 (2, 2) 20.3 17 (6, 3) 42.0 64 (8, 10) 49.3 149 (11, 2) 103.3 138 (15, 3)
12.4 24 (2, 1) 20.3 11 (9, 0) 42.1 71 (7, 12) 49.3 150* (11, 2) 113.2 128 (16, 3)
12.7 28 (2, 0) 20.5 105 (0, 9) 42.2 67 (5, 13) 49.5 86 (13, 14) 116.1 125 (18, 7)
13.0 22 (3, 0) 20.5 14 (2, 8) 42.3 76 (9, 10) 49.7 95 (13, 15) 116.1 142* (18, 7)
13.1 34 (0, 5) 20.7 114 (1, 9) 42.7 65 (8, 12) 50.0 82 (13, 13) 122.7 122 (17, 3)
14.2 20 (2, 3) 21.0 23 (4, 5) 42.7 66 (8, 12) 50.5 93 (14, 13) 122.7 136* (17, 3)
14.3 3 (4, 0) 21.0 108* (4, 5) 43.2 38 (6, 15) 50.7 90 (15, 15) 131.1 120 (18, 3)
14.3 96 (4, 0) 21.3 1 (7, 4) 43.2 32 (10, 6) 50.9 48 (11, 5) 132.2 140 (19, 4)
15.1 37 (1, 5) 22.7 19 (6, 5) 43.2 83 (9, 12) 51.2 81 (14, 12) 132.3 132 (18, 5)
15.5 98 (0, 6) 22.9 116 (5, 5) 43.4 60 (9, 9) 52.2 41 (13, 4) 134.4 134 (19, 5)
15.7 103 (5, 0) 23.5 9 (4, 7) 43.5 77 (8, 14) 52.2 51* (13, 4) 137.5 133 (20, 5)
16.3 2 (6, 0) 23.6 5 (5, 8) 44.1 68 (10, 8) 52.4 89 (17, 14) 146.5 124 (20, 3)
16.3 115 (6, 0) 24.2 109 (2, 11) 44.1 72 (10, 8) 53.6 44 (13, 9) 146.5 126 (20, 3)
16.4 6 (5, 1) 24.2 118 (2, 11) 44.5 84 (10, 9) 53.7 94 (18, 15) 149.7 139 (17, 0)
16.4 104 (5, 1) 24.8 111 (0, 11) 45.0 75 (9, 13) 54.0 54 (14, 10) 153.8 131 (18, 0)
16.6 7* (4, 1) 25.3 119 (0, 13) 45.2 61 (11, 10) 54.5 46 (15, 10) 156.1 135 (18, 1)
16.8 147* (7, 0) 25.6 113 (4, 12) 45.2 62 (11, 10) 59.7 39 (13, 7) 160.6 143 (19, 0)
17.2 10 (0, 7) 25.8 42* (1, 14) 45.7 85 (10, 11) 62.9 36 (16, 11) 167.8 123 (20, 0)
17.2 102 (0, 7) 27.5 25 (3, 9) 45.7 78 (9, 15) 65.4 29* (14, 6) 167.8 127 (20, 0)
17.8 97 (2, 6) 29.0 13 (5, 10) 45.7 79 (9, 15) 66.7 58 (16, 9) 170.6 35 (20, 1)
18.1 100 (5, 2) 30.4 30* (0, 15) 45.9 91 (10, 14) 67.6 57 (18, 10)
18.4 55 (7, 1) 33.1 117 (3, 14) 46.0 74 (11, 12) 67.7 59 (15, 8)
68.2 47 (15, 6)
aThe structures of compounds 1–150 are indicated in Chart 1.bThe compounds marked with an asterisk were used in the validation of the neural network. Six-membered rings are
highlighted in gray.c(x, y) represent the neuron coordinates, where x is the column number and y is the row number (Figure 2).
924 Alonso and Herradon • Vol. 31, No. 5 • Journal of Computational Chemistry
Journal of Computational Chemistry DOI 10.1002/jcc
Opposite variation of aromaticity with the electronegativity of
the X substituent is expected for analogues of heptafulvene
because the dipolar structures with 6p electrons are oppositely
polarised for the five- and seven-membered rings.63 Accordingly,
dj values indicate a gradual decrease of aromaticity from the
aromatic 6p tropylium cation C of 158 (X 5 NH21) to the antiaro-
matic 8p anionic ring D of 156 (X 5 BH22), as shown in Table 2B.
A third test analyzes a series of six five-membered hetero-
cycles (Table 2C). Electron counting indicates that systems with
X 5 CH2, NH, O, and S are aromatic and it is expected that
the inclusion of heteroatoms decreases the aromaticity with
regard to the cyclopentadienyl anion in the order NH \ S \
Table 2. Calculated ASE (kcal mol21), L (ppm cgs), NICSzz(1) (ppm), HOMA, and Euclidean Distance (dj)
for Test Compounds.
Compound ASE L NICSzz(1) HOMA dj Compound ASE L NICSzz(1) HOMA dj dja
(A) Analogues of fulvene (D) Six-membered heterocyclesb
X
BH22 14.04 211.10 229.59 0.483 18.5 Pyridine 31.63 212.53 230.86 0.978 0.0 4.1
CH2 23.06 1.01 26.97 20.142 46.3 Pyridazine 22.17 29.91 230.98 0.962 11.4 13.7
NH 29.29 4.97 2.55 20.592 58.8 Pyrimidine 19.39 212.82 228.64 0.991 16.8 15.9
O 214.65 8.00 10.96 21.255 67.6 Pyrazine 23.29 210.19 230.95 1.000 10.2 12.5
NH21 227.27 11.75 25.71 20.662 82.0 s-Triazine 9.63 27.90 225.63 0.995 25.8 25.8
(B) Analogues of heptafulvene (E) Substituted benzenes
X
X
BH22 0.27 34.10 74.07 0.175 123.7 N2
1 31.50 213.39 228.84 0.955 0.0 4.7
CH2 6.78 2.27 14.97 0.335 55.9 CN 33.72 213.89 230.70 0.951 0.0 1.9
NH 9.52 2.30 2.23 0.390 48.2 F 34.16 213.66 231.07 0.974 0.0 1.9
O 12.71 25.65 27.93 0.421 30.3 H 34.83 214.95 231.84 0.962 0.0 0.0
NH21 21.32 210.09 219.12 0.862 20.3 CH3 33.70 213.36 230.67 0.955 0.0 2.3
NH2 34.06 211.20 228.54 0.957 0.0 5.1
OH 33.31 212.88 229.85 0.966 0.0 2.8
(C) Five-membered heterocycles (F) 6p-electron systems
X
X
CH2 22.05 210.13 235.98 0.740 13.1 benceno 36.85 215.35 231.84 0.962 0.0 0.0
NH 20.57 26.48 232.41 0.876 14.3 C7H72 34.17 212.46 229.00 0.954 0.0 3.5
S 18.57 27.00 229.41 0.891 16.3 C8H822 5.85 26.27 227.33 0.808 25.8 23.2
O 14.77 22.90 226.79 0.298 21.3
CH2 0.00 0.00 211.83 20.778 41.5
BH 222.49 16.09 31.30 20.595 89.0
CH1 257.88 35.54 112.21 21.152 170.6
aEuclidean distance values obtained from the larger map, trained only with aromatic compounds.bThe structures of the rest of six-membered heterocycles are shown in Chart 1.
Scheme 3. Resonance structures for analogues of fulvene and hepta-
fulvene.
925Scale of Aromaticity for p-Organic Compounds
Journal of Computational Chemistry DOI 10.1002/jcc
O.3,64 Cyclopentadiene (X 5 CH2) is anticipated to be a nonaro-
matic compound, whereas compounds with X 5 BH and CH1
should be antiaromatic. The Kohonen network classifies properly
all the five-membered heterocycles and the Euclidean distance
values give the expected order of aromaticity for this set of mol-
ecules: 34 (X 5 CH2) [ 3 (X 5 NH) [ 2 (X 5 S) [ 1 (X 5O) [ 33 (X 5 CH2) [ 31 (X 5 BH) [ 35 (X 5 CH1). By
contrast, the inclusion of nitrogen atoms at position 2 in the pyr-
role ring increases the aromaticity, being 1H-1,2-diazole (18, dj5 12.3), 1H-1,2,3-triazole (24, dj 5 12.4), and 2H-1,2,3,4-tetra-zole (28, dj 5 12.7) more aromatic than the cyclopentadienyl
anion (34, dj 5 13.1).
For six-membered heterocycles, it is expected that the
replacement of a CH fragment by nitrogen decreases aromatic-
ity. According to the classification pattern of the SOM, benzene
and pyridine are the most aromatic in this series, followed by
diazines and, finally, triazine. It has been stated that, for rings
with the same number of nitrogen atoms, the most aromatic are
those having the largest number of N��N bonds.65 However, we
observe pyrazine (146) � pyridazine (148) [ pyrimidine (147)
failing only in the ordering of pyrazine (Table 2D). It is impor-
tant to note that the expected order is not followed by any of
the 10 indices analyzed by Sola and coworkers60 and, therefore,
the initial presumption should be revisited. Interestingly, the ring
size dependence of the NICSzz(1) indices is avoided using our
aromaticity scale dj, which shows that benzene and pyridine are
more aromatic than any of the five-membered rings considered.
The following set analyzes the aromaticity of several mono-
substituted benzenes (Table 2E). Krygowski et al. have showed
that the substituents influence only very weakly the p-electrondelocalization in the ring.66 The major changes in the geometry
of 74 monosubstituted benzenes concern the ring angles and
they are related to the electronegativity of substituents.67 Only
substituents CH21 and CH2
2 cause substantial bond length
changes and HOMA decreases to � 0.7.68 Moreover, all sub-
stituents induce a loss of aromaticity independently of their elec-
tron donating or accepting character. SOM places all the ben-
zene derivatives in the neuron which represent the highest aro-
maticity indicating high resistance of the p-electron to the
substituents effect. A larger map is necessary to distinguish com-
pounds with very similar degrees of aromaticity. An extension
of the region of the highly aromatic compounds with dj between0 and 10, obtained from a 14 3 12 map training only with aro-
matic compounds, is shown in the Figure 3b. This larger map is
able to discriminate benzene, pyridine, and the substituted ben-
zenes, indicating correctly that benzene is the most aromatic
(144, dj 5 0.0). All the benzene derivatives possess dj valuesranging from 1.9 to 5.1, being the N2
1 and NH2 substituents that
lead to the largest decrease in aromaticity in agreement with
other indicators of aromaticity such as para-delocalization index,
aromatic fluctuation index, and multicenter indices.60
Finally, we have analyzed the dependence of the Euclidean
distance on the size of the ring (Table 2F). For this purpose, we
evaluated the aromaticity of three 6p-electron systems with dif-
ferent ring sizes: benzene, cycloheptatrienyl cation (C7H71), and
cyclooctetraenyl dication (C8H821). As the ring size increases it
is expected that the aromaticity decreases, giving the following
order of aromaticity: benzene [ C7H71 [ C8H8
21. This is
indeed the order provided by the Euclidean distance. To distin-
guish between benzene and C7H71, we employed the larger map
shown in Figure 3b to estimate dj. It is remarkable that the posi-
tion of benzene in the bidimensional map does not change using
different values for the aromatic stabilization energy and the ex-
altation calculated from homodesmotic/isodesmic reactions (Sup-
porting Information Table S3). Therefore, the neural network is
hardly sensitive to the reactions employed for evaluating the
energetic and magnetic descriptors of aromaticity. It is noticed
that two methods have been recently reported in the literature
which evaluates stabilization energies without using homodes-
motic/isodesmic equations. We are regarding the possibility of
including these descriptors in future developments of the
SOMs.23,69,70
Additionally, NICS scans have been calculated for borderline
compounds of the three classes (aromatic, nonaromatic, and anti-
aromatic) in order to study in depth the presence of diamagnetic
and paramagnetic ring currents (Fig. 4). The NICS scan proce-
dure is based on scanning NICS values over the distance and
separating them into in-plane and out-of-plane contributions.52
The shape of the NICS scans for benzene (144) and 2,3,4,5-tet-
rahydroxycyclopentadienyl anion (117) is typical for aromatic
systems, showing a minima for the out-of-plane component and
isotropic values which points out the presence of diamagnetic
ring current. The minimum value for the out-of-plane compo-
nent is observed at r 5 1.0 A, although the magnitude of
NICSzz decreases around 9 ppm from benzene to 117, indicat-
ing a smaller diamagnetic ring current. It is noteworthy that the
diamagnetic contribution is larger than the paramagnetic
contribution to the out-of-plane component in both compounds.
However, in 3-phospaphosphole (12) and gallole (47), classi-
fied as nonaromatic compounds by the neural network, the dia-
magnetic contribution is larger than the paramagnetic contribu-
tion to the out-of-plane component at short distances, becoming
equal at 1.3 and 2.7 A, respectively. As Euclidean distance
increases, the minima of the out-of-plane curve become less
negative and appear at larger distances from the molecular
plane, indicating an enhancement of the paramagnetic contribu-
tion and a smaller diatropic ring current. This minimum disap-
pears in the NICS scan curves of the antiaromatic compounds
129 and 35 because the paramagnetic contribution dominates
over the diamagnetic counterpart in the out-of-plane component.
Additionally, the NICS scans shows that the isotropic curves are
governed by the out-of-plane component in the aromatic and
antiaromatic compounds and about equally by the in-plane and
out-of-plane components in the nonaromatic systems. Therefore,
on the basis of the NICS scan results for borderline compounds,
we conclude that the neural network classifies correctly an
extensive dataset of organic compounds into aromatic, nonaro-
matic, and antiaromatic.
Conclusions
The results presented in this article demonstrate that the neural
network fulfils the expected aromaticity trends based on the
chemical knowledge. In addition, the Euclidean distance
between neurons (dj) indicates the expected order of aromaticity
of penta- and heptafulvenes, heteroaromatic species, substituted
926 Alonso and Herradon • Vol. 31, No. 5 • Journal of Computational Chemistry
Journal of Computational Chemistry DOI 10.1002/jcc
benzenes, and substituted cyclopentadienyl compounds. More
important is that the Euclidean distance values (dj), as a measure
of aromaticity, overcomes the limitations of the structural, mag-
netic, and energetic indicators of aromaticity such as the ring
size dependence of L and NICS or the dependence of the ASE
values on the formulation of the reaction scheme.
As a conclusion, we can state that the Euclidean distance (dj)between neurons in the SOM is the first aromaticity scale reported
in the literature which contains the energetic, structural, and mag-
netic aspects of this phenomenon. The results reported for a wide
dataset of p-organic compounds clearly show that the multifac-
eted nature of aromaticity can be only apprehended by a multidi-
mensional method. Additionally, the quantification of aromaticity/
antiaromaticity through neural networks has several advantages.
In contrast to linear regression models, the trained SOM can be
used to classify new compounds and predict its degree of p-elec-tronic delocalization with a short computational time. Further-
more, the position of a new compound on the bidimensional map
is conveniently visualized, indicating successfully the following:
a. the group (aromatic, nonaromatic, or antiaromatic) to which
the new system belongs;
b. their degree of aromaticity/antiaromaticity based on the Eu-
clidean distance;
c. the similarity in aromaticity between different compounds.
On the other hand, a two-dimensional map can much better
reflect the results of the various influences on the aromaticity/antiaro-
maticity: different directions in the map represent different kinds of
similarity and different distances indicate distinct levels of similarity.
Work is underway to expand the present methodology to
other kinds of chemicals as well as to unveil structure–property
relationships.
Acknowledgments
This work was taken in part from the Ph.D. thesis of M.A. The
authors thank Dr. A. Chana for helpful discussions on neural
networks, and Dr. J.D. Guillen for helpful assistance on statisti-
cal analyses. The authors also thank the Centro de Supercompu-
tacion de Galicia (CESGA) for computational time at the SVGD
supercomputer.
References
1. Krygowski, T. M.; Cyranski, M. K.; Czarnocki, Z.; Hafelinger, G.;
Katritzky, A. R. Tetrahedron 2000, 56, 1783.
2. Randic, M. Chem Rev 2003, 103, 3449.
3. Balaban, A. T.; Oniciu, D. C.; Katritzky, A. R. Chem Rev 2004,
104, 2777.
4. Balaban, A. T.; Schleyer, P. v. R.; Rzepa, H. S. Chem Rev 2005,
105, 3436.
5. Minkin, V. I.; Glukhovtsev; M. N.; Simkin, B. Y Aromaticity and
Antiaromaticity: Electronic and Structural Aspects; Wiley: New
York, 1994.
6. Chen, Z.; King, R. B. Chem Rev 2005, 105, 3613.
7. Boldyrev, A. I.; Wang, L.-S. Chem Rev 2005, 105, 3716.
8. Frenking, G.; Krapp, A. K. J Comput Chem 2007, 28, 15.
9. Stanger, A. Chem Commun 2009, 1939.
10. Schleyer, P. v. R.; Jiao, H. J. Pure Appl Chem 1996, 68, 209.
11. Krygowski, T. M.; Cyranski, M. K. Chem Rev 2001, 101, 1385.
12. Lazzeretti, P. Phys Chem Chem Phys 2004, 6, 217.
13. Katritzky, A. R.; Jug, K.; Oniciu, D. C. Chem Rev 2001, 101, 1424.
14. Cyranski, M. K.; Krygowski, T. M.; Katritzky, A. R.; Schleyer, P. v.
R. J Org Chem 2002, 67, 1333.
15. Feixas, F.; Matito, E.; Poater, J.; Sola, M. J Phys Chem A 2007,
111, 4513.
16. Aihara, J. Bull Chem Soc Jpn 2008, 81, 241.
17. Aihara, J. J Am Chem Soc 2006, 128, 2873.
18. Bultinck, P. Faraday Discuss 2007, 135, 347.
19. Fias, S.; Fowler, P. W.; Delgado, J. L.; Hahn, U.; Bultinck, P. Chem
Eur J 2008, 14, 3093.
20. Ball, P. Designing the Molecular World: Chemistry at the Frontier;
Princeton University Press: Princeton, NJ, 1994.
21. Krygowski, T. M.; Cyranski, M. K. Phys Chem Chem Phys 2004, 6,
249.
22. Cyranski, M. K. Chem Rev 2005, 105, 3773.
23. Mo, Y. R.; Schleyer, P. v. R. Chem Eur J 2006, 12, 2009.
24. Poater, J.; Duran, M.; Sola, M.; Silvi, B. Chem Rev 2005, 105,
3911.
25. Merino, G.; Vela, A.; Heine, T. Chem Rev 2005, 105, 3812.
26. Bultinck, P.; Mandado, M.; Mosquera, R. A. J Math Chem 2008, 43,
111.
27. Gomes, J. A. N. F.; Mallion, R. B. Chem Rev 2001, 101, 1349.
28. Heine, T.; Corminboeuf, C.; Seifert, G. Chem Rev 2005, 105, 3889.
29. Katritzky, A. R.; Karelson, M.; Sild, S.; Krygowski, T. M.; Jug, K. J
Org Chem 1998, 63, 5228.
30. Kohonen, T. Self-Organizing Maps, 3rd ed.; Springer: Berlin, 2001.
31. Zupan, J.; Gasteiger, J. Neural Networks for Chemists: An Introduc-
tion; Wiley-VCH: Weinheim, 1993.
32. Gasteiger, J.; Teckentrup, A.; Terfloth, L.; Spycher, S. J Phys Org
Chem 2003, 16, 232.
33. Guha, R.; Serra, J. R.; Jurs, P. C. J Mol Graphics Model 2004, 23,
1.
34. Panek, J. J.; Jezierska, A.; Vracko, M. J Chem Inf Mod 2005, 45,
264.
35. Darley, M. G.; Handley, C. M.; Popelier, P. L. A. J Chem Theory
Comput 2008, 4, 1435.
36. de Molfetta, F. B.; Angelotti, W.; Romero, R.; Montanari, C.; da
Silva, A. R. J Mol Model 2008, 14, 975.
37. Alonso, M.; Herradon, B. Chem Eur J 2007, 13, 3913.
38. Frisch, M. J.; Trucks, G. W.; Schlegel, H. B.; Scuseria, G. E.; Robb,
M. A.; Cheeseman, J. R.; Montgomery, J. A., Jr.; Vreven, T.; Kudin,
K. N.; Burant, J. C.; Millam, J. M.; Iyengar, S. S.; Tomasi, J.;
Barone, V.; Mennucci, B.; Cossi, M.; Scalmani, G.; Rega, N.;
Petersson, G. A.; Nakatsuji, H.; Hada, M.; Ehara, M.; Toyota, K.;
Fukuda, R.; Hasegawa, J.; Ishida, M.; Nakajima, T.; Honda, Y.;
Kitao, O.; Nakai, H.; Klene, M.; Li, X.; Knox, J. E.; Hratchian, H.
P.; Cross, J. B.; Bakken, V.; Adamo, C.; Jaramillo, J.; Gomperts, R.;
Stratmann, R. E.; Yazyev, O.; Austin, A. J.; Cammi, R.; Pomelli, C.;
Ochterski, J. W.; Ayala, P. Y.; Morokuma, K.; Voth, G. A.;
Salvador, P.; Dannenberg, J. J.; Zakrzewski, G.; Dapprich, S.;
Daniels, A. D.; Strain, M. C.; Farkas, O.; Malick, D. K.; Rabuck, A.
D.; Raghavachari, K.; Foresman, J. B.; Ortiz, J. V.; Cui, Q.; Baboul,
A. G.; Clifford, S.; Cioslowski, J.; Stefanov, B. B.; Liu, G.;
Liashenko, A.; Piskorz, P.; Komaromi, I.; Martin, R. L.; Fox, D. J.;
Keith, T.; Al-Laham, M. A.; Peng, C. Y.; Nanayakkara, A.;
Challacombe, M.; Gill, P. M. W.; Johnson, B.; Chen, W.; Wong, M.
W.; Gonzalez, C.; Pople, J. A. Gaussian 03, Revision C.02;
Gaussian, Inc.: Pittsburgh, PA, 2004.
927Scale of Aromaticity for p-Organic Compounds
Journal of Computational Chemistry DOI 10.1002/jcc
39. Cyranski, M. K.; Schleyer, P. v. R.; Krygowski, T. M.; Jiao, H. J.;
Hohlneicher, G. Tetrahedron 2003, 59, 1657.
40. Wheeler, S. E.; Houk, K. N.; Schleyer, P. v. R.; Allen, W. D. J Am
Chem Soc 2009, 131, 2547.
41. Schleyer, P. v. R.; Pulhofer, F. Org Lett 2002, 4, 2873.
42. Dauben, H. J.; Wilson, J. D.; Layti, J. L. J Am Chem Soc 1968, 90, 811.
43. Dauben, H. J.; Wilson, J. D.; Layti, J. L. J Am Chem Soc 1969, 91,
1991.
44. Keith, T. A.; Bader, R. F. W. Chem Phys Lett 1993, 210, 223.
45. Chen, Z.; Wannere, C. S.; Corminboeuf, C.; Puchta, R.; Schleyer, P.
v. R. Chem Rev 2005, 105, 3842.
46. Schleyer, P. v. R.; Maerker, C.; Dransfeld, A.; Jiao, H. J.; Hommes,
N. J Am Chem Soc 1996, 118, 6317.
47. Wolinski, K.; Hinton, J. F.; Pulay, P. J Am Chem Soc 1990, 112, 8251.
48. Bader, R. F. W. Atoms in Molecules: A Quantum Theory; Claren-
don: Oxford, 1990.
49. Morao, I.; Lecea, B.; Cossio, F. P. J Org Chem 1997, 62, 7033.
50. Alonso, M. Ph.D. Thesis; Universidad Complutense: Madrid, 2008.
51. Corminboeuf, C.; Heine, T.; Seifert, G.; Schleyer, P. v. R.; Weber,
J. Phys Chem Chem Phys 2004, 6, 273.
52. Stanger, A. J Org Chem 2006, 71, 883.
53. Stanger, A. Chem Eur J 2006, 12, 2745.
54. Kruszewski, J.; Krygowski, T. M. Tetrahedron Lett 1972, 13, 3839.
55. Krygowski, T. M. J Chem Inf Com Sci 1993, 33, 70.
56. Kohonen, T.; Hynninen, J.; Kangas, J.; Laaksonen, J. SOM Pak: The
Self-Organizing Map Program Package, Rev. 3.1; Laboratory of
Computer and Information Science, Helsinki University of
Technology: Finland, 1995. Available at http://citeseer.ist.psu.edu/
kohonen96som.html.
57. Sammon, J. S. IEEE Trans Comput 1969, C-18, 401.
58. Lazzaroni, S.; Dondi, D.; Fagnoni, M.; Albini, A. J Org Chem 2008,
73, 206.
59. Mills, N. S.; Llagostera, K. B. J Org Chem 2007, 72, 9163.
60. Feixas, F.; Matito, E.; Poater, J.; Sola, M. J Comput Chem 2008, 29,
1543.
61. Mandado, M.; Gonzalez-Moa, M. J.; Mosquera, R. A. J Comput
Chem 2007, 28, 127.
62. Havenith, R. W. A.; Fowler, P. W.; Steiner, E. J Chem Soc Perkin
Trans 2002, 2, 502.
63. Najafian, K.; Schleyer, P. v. R.; Tidwell, T. T. Org Biomol Chem
2003, 1, 3410.
64. Schleyer, P. v. R.; Freeman, P. K.; Jiao, H.; Goldfuss, B. Angew
Chem Int Ed Engl 1995, 34, 337.
65. Mandado, M.; Otero, N.; Mosquera, R. A. Tetrahedron 2006, 62,
12204.
66. Krygowski, T. M.; Ejsmont, K.; Stepien, B. T.; Cyranski, M. K.;
Poater, J.; Sola, M. J Org Chem 2004, 69, 6634.
67. Campanelli, A. R.; Domenicano, A.; Ramondo, F. J Phys Chem A
2003, 107, 6429.
68. Krygowski, T. M.; Stepien, B. T. Pol J Chem 2004, 78, 2213.
69. Fernandez, I.; Frenking, G. Faraday Discuss 2007, 135, 403.
70. Fernandez, I.; Frenking, G. Chem Eur J 2007, 13, 5873.
928 Alonso and Herradon • Vol. 31, No. 5 • Journal of Computational Chemistry
Journal of Computational Chemistry DOI 10.1002/jcc