Vibrational Spectroscopy 43 (2007) 26–37

www.elsevier.com/locate/vibspec

Calculating the vibrational spectra of molecules: An introduction

for experimentalists with contemporary examples
Robert J. Meier a,b
a
Institute of Chemistry and Dynamics of the Geosphere IV: Agrosphere, Forschungszentrum Jülich GmbH, D-52425 Jülich, Germany
b
DSM Research, P.O. Box 18, 6160 MD Geleen, The Netherlands
Received 3 May 2006; received in revised form 1 June 2006; accepted 7 June 2006
Available online 21 July 2006

Abstract
The theoretical evaluation of vibrational spectra has, in the past, played a crucial role regarding assignment of vibrational spectra, an aspect that
might be largely forgotten. This, however, is actually ever so true today, where despite the presence of data bases of spectra (atlas of spectra) there is
still a large number of species including, though certainly not exclusively, less stable species which have spectra lacking appropriate interpretation.
In recent years the capabilities to calculate vibrational spectra have greatly improved. In this review the basic various approaches are introduced.
For practical use, the use of comparatively simple tools including Hartree–Fock level or standard DFT type calculations often already do a very
good job at the comparative level. An important recent development includes the observation that differences between experimental and calculated
frequencies are largely due to neglect of anharmonicity rather than the theoretical method employed. Methods to include anharmonicity are
becoming available for users. The introduction of methods allowing for periodic calculations, involving full 3D molecular structures, lead to new
opportunities and final answers to long-standing issues. A number of examples are provided showing where calculated spectra have provided
unique added value to experimental work.
# 2006 Elsevier B.V. All rights reserved.

Keywords: Infrared spectroscopy; Raman spectroscopy; Vibrational spectroscopy; Computational chemistry; Molecular modelling; Vibrational intensities

1. Introduction introduction of new features, including imaging and 2D

correlation spectroscopy. Another factor is the recognition that
Following the historic context illustrating the importance of vibrational spectroscopy can play a role in new rather than only
calculating vibrational spectra, this paper is partly (in particular in the traditional fields of application, examples include new
Section 3) tutorial with respect to the calculation of vibrational applications in the life-science field (living cells, cancer
spectroscopic data and therefore intended for the practical research) and the characterisation of soil.
spectroscopist interested in methods to calculate these proper- Irrespective of the field of application, the calculation of
ties. The second part (Sections 4 and 5) provides a description vibrational spectra is an activity of sufficient relevance for
of state of the art developments and applications in the field. the practicising vibrational spectroscopist to know what
Infrared and Raman spectroscopy are among the traditional developments are taking place. Very many questions of
methods of analysis, and particularly powerful for non- interpretation remain to be addressed, where contemporary
destructive characterisation of substances including living computational chemistry can play an essential role. In
material. For reference we can definitely recommend the particular cases calculations are the only tool to obtain certain
recently issued impressive five-volume set handbook of information (see e.g., Section 5.3). However, despite sig-
vibrational spectroscopy [1] in which practically all issues nificant improvements in computational power and a large
and applications are covered to a greater or lesser extent. But number of different methods, we are still facing intriguing
compared to the past, a shift in applications has taken place problems to be dealt with before the calculation of vibrational
bringing new opportunities [2]. This is partly due to the spectra will become a standard and reliable tool for spectrum
interpretation. Therefore, at present, the proper evaluation of
vibrational frequencies and intensities by calculation is still a
E-mail address: r.meier@wxs.nl. task for the expert.
0924-2031/$ – see front matter # 2006 Elsevier B.V. All rights reserved.
doi:10.1016/j.vibspec.2006.06.003
 R.J. Meier / Vibrational Spectroscopy 43 (2007) 26–37 27

Something not always sufficiently appreciated by theore- say 1200 cm1, why is this a molecular vibration. In this
ticians is that experimentally vibrational frequencies can be context, it may be interesting to note that in the early days of
easily determined with wavenumber accuracy. For the vibrational spectroscopy applied to molecules a mechanical
fingerprint region, however, this implies an accuracy of model in which atoms were described by balls and the bonding
1:1000, which is too severe a demand for any current quantum by springs, i.e. a predecessor of the force field or molecular
mechanics method (the same, e.g., applies to the theoretical mechanics method to be outlined below, was applied in order to
evaluation of NMR shift). Therefore, one must rely on correctly understand the internal vibrations of molecules. In a 1930 paper
predicted relative frequencies. For a molecule comprising only by Kettering, Shutts and Andrews [4] these authors described
a few, say up to 10, atoms, the number of vibrational bands is the construction of such a real mechanical model with springs
limited, and therefore mutual differences in wavenumber are weighing 1.5 oz each and steel balls 1(1/8) in. diameter
sufficiently large in order to obtain correct ordering from weighing 0.2 lbs describing carbon atoms. The system was
calculations. But when a system is just a little larger, in mechanically excited using a motor and the specific displace-
particular the interesting fingerprint region will be very ments of the steel balls, mimicking the atoms, were followed by
crowded, not even to speak of, e.g., amorphous and crystalline applying a stroboscope. The authors wrote ‘‘the tuning was
components of the same molecular entity or different found to be very sharp. When the motor was running at a speed,
conformations of the same molecular species in the same which did not correspond to a characteristic frequency, the
sample. Therefore very reliable frequencies need to be model was absolutely quiet. When the motor speed came within
calculated. Moreover, because of that complication, the author about 5% of a characteristic frequency, however, the balls
himself, until more accurate methods become available, began to move over the paths characteristic of that frequency
decided to generally evaluate both vibrational frequencies and with closer tuning would generally vibrate with an
and intensities in order to arrive at a considerably more reliable amplitude of the order of 0.5 in.’’ What a beautiful example of
assignment of experimentally observed vibrational bands. In how it all started now more than half a century ago! In fact by
fact without doing so, erroneous interpretation might be the using different configurations of the springs, one compared the
result because there is no warning mechanism otherwise then experimental molecular vibrations of benzene with Kekule’s
inspection of intensities. Finally, simply transferring assign- and other suggested structures in order to find out the correct
ment from small model compounds to larger molecules description for the chemical bonding in this molecule.
comprising similar molecular entities is not without risks. Another example originates from the early days of vibrational
In this review we will first briefly discuss the relevance of the spectroscopy on polymers, when it was not uncommon that a
computation of vibrational spectra (Section 2). Then we point theoretically calculated spectrum (frequencies only) was
out the various computational methods (Section 3), whereas employed, and in fact required, to assign the vibrational bands.
Section 4 is especially devoted to anharmonic effects as they This definitely applies with respect to the vibrational spectra of
were recently shown to be the main source of ‘error’ in polyethylene and polypropylene, but also to the spectra of other
calculated frequencies. Thereafter (Section 5) we will illustrate polymers. A compilation of relevant facts has been presented by
the power of calculating vibrational spectra by showing explicit Dechant and coworkers [5]. Today it might seem surprising that
examples, whereas Section 6 provides a summary and outlook. computation of the normal modes of vibrations was necessary to
We will primarily discuss the calculation of vibrational interpret the full spectrum of a polymer (chain). Although in first
frequencies, and not of the vibrational intensities. Despite instance the interpretation of the vibrational spectrum of the
the relevance which we pointed out before, unfortunately most chemically simplest polymer, polyethylene, may be pursued by
theoretical papers on intensities focus on small molecules and comparing it to the corresponding spectra of (short) alkanes, the
high-level methods, see e.g., Galabov et al. [3], which are spectrum could not be completely assigned unambiguously. The
impractical for most real life molecules. Therefore we feel these vibrational spectra of the n-alkanes were reported as early as
are less interesting to the community of experimentalists. 1949 by Simanouti and Mizushima [6]. In the 1960’s in a series of
However, some of the examples discussed will involve papers Snyder and Schachtschneider reported a very thorough
appropriate calculation of intensities for real life systems study on the vibrational spectra of the alkanes and polyethylene
without going to high level methods of calculation. [7–10]. These studies have shown that whereas for short alkanes
often numerous vibrational bands are present within a certain
2. Why calculating vibrational spectra: the historic frequency range, there is a strong tendency for these bands to
context merge into a few when the chain length goes to infinity, i.e. for a
polymer (polyethylene). It was a landmark in the 1960’s that
Although most spectroscopists have never performed a force fields could be fitted to accurately reproduce polymer
vibrational calculation and even might never do so, it is of vibrational spectra. Regarding a correct and unambiguous
sufficient relevance to understand the basics of the underlying interpretation of the vibrational spectrum of polyethylene,
methodologies and their application. The primary reason is that Snyder [9], when discussing the infrared spectrum of poly-
such computations have formed, probably much more than ethylene, noted that ‘‘although there are abundant illustrations in
commonly presumed, the basis for our current interpretation of the literature of the dangers of ‘predicting’ or ‘confirming’
vibrational spectra of many species, in particularly also vibrational assignments using computer calculated frequencies,
polymers. Moreover, when we observe an IR absorption at the present example is singularly spectacular. What is so
28 R.J. Meier / Vibrational Spectroscopy 43 (2007) 26–37

surprising at first sight is the ease with which it is possible to fit with qi is the displacements from the equilibrium positions. The
precisely the B2g frequency to an ‘observed’ value either at zero of the vibrational energy may be conveniently chosen as
1415 cm1 or at 1381 cm1, while in either case the other 185 the state in which all vibrating atoms are at their equilibrium
calculated frequencies are maintained in equally excellent positions, and thus V0 = 0. At the minimum energy configura-
agreement with the observed values.’’ This experience should not tion the first derivatives are zero by definition, otherwise the
be interpreted as a particular weakness of the computational system would be unstable with regard to displacement of the
approach in those days: even when combining experimental data average positions of atoms. The harmonic approximation,
with computed data, an unambiguous assignment might still be which in practice turned out a very good approximation,
difficult to attain for all bands in the spectrum. implies that all terms of higher order than 2 are neglected.
The latter example well illustrates how calculated vibra- We are thus left over with the second order term in the potential
tional spectra were, historically, essential in the assignment of energy V only. In the second term coordinates of atoms are
vibrational spectra. The same applies to many other species. mixed, i.e. qi and qj, which implies that the vibrating atoms in
Although we have many books of reference available nowa- the polyatomic molecule may not be considered as indepen-
days, which can help us to assign bands on basis of similarities dently vibrating atoms. Then, from Newton’s second law, we
in molecular structure, on the other hand there are still many have:
unanswered questions. Many interpretations are incomplete or   3N 
X 
non-existing and require the help of computations [11–13], d2 qi @V @2 V
¼  ¼  qj (4)
whereas some information simply cannot be obtained directly dt2 @qi j¼1
@qi @q j
qi ¼0;q j ¼0
from experiment such as the dependence of vibrational
frequencies and intensities upon the introduction of one or The second derivatives to the potential energy are the force
more isolated or consecutive gauche bonds in an alkyl or a constants:
polymer chain [14,15]. Similar examples can be found for
 2 
simple organometallics where even up till today the inter- @ V
pretation of the spectra of only comparatively small and simple fij ¼ (5)
@qi @q j
species have been reported.
Solving the equations of motion Eq. (4) leads to a determinant
3. Calculating vibrational spectra: methods and whose eigenvalues mv2 yield the vibrational frequencies v. The
limitations eigenvectors describe the atomic displacements involved in
each of the vibrations characterized by one of the eigenvalues
3.1. General introduction v. The eigenvectors corresponding to the different eigenvalues
are mutually orthogonal, meaning that the motions are inde-
When we consider a simple homonuclear diatomic, the pendent. Vibrations with these characteristics are called normal
potential energy between the atoms reads: modes of vibrations, and the frequency is known as the normal
frequency or the fundamental frequency. The fact that solving
V harmonic vibration ¼ 12kðr  r 0 Þ2 (1) the equations of motion Eq. (4) leads to a set of independent
vibrations at frequencies vk are the cause that distinct vibra-
whereas the vibrational frequency is given by tional bands are observed in experimental vibrational spectra,
rffiffiffiffi viz. the above-described experiment of the mechanical model
v 1 k in which atoms were described by balls and the bonding by
n¼ ¼ (2)
2p 2p m springs.
So what need to be calculated to obtain the vibrational
with k is the force constant, m the mass of each atom in the frequencies are the second derivatives of the energy with
homonuclear diatomic, r the actual bond length and r0 the respect to atomic displacement. There are various ways to
equilibrium (minimum energy) bond length. For a polyatomic achieve this: the empirical force field method and quantum
system, the potential energy expression is more complex. Using mechanics based methods. Both approaches allow for the
the more convenient internal coordinates qi rather than carte- calculation of vibrational frequencies. The quantum mechan-
sian cordinates xi, as the former direcly connect to internal bond ical approach allow for the straightforward, though computa-
lengths and angles, the potential energy expression takes the tionally intensive, calculation of infrared and Raman
form: frequencies and intensities. The force field method can also
be devised for the purpose of calculating the intensities, which
X
3N6
@V

however, requires the possibility to evaluate the dipole
V ¼V 0 þ qi
i¼1
@qi moments (IR intensities) and polarizabilities (Raman inten-
qi ¼0
sities) within the context of the force field. Force fields that
X 3N6
1 3N6 X  @2 V 
þ qi q j comprise atomic charges and therefore Coulombic interactions
2 i¼1 j¼1 @qi @q j in principle allow for the evaluation of IR intensities. However,
qi ¼0;q j ¼0
force fields are generally not parametrised for that purpose, and
þ higher order terms (3) therefore intensity reporting using force field methods is rare
 R.J. Meier / Vibrational Spectroscopy 43 (2007) 26–37 29

and mostly inappropriate. Using quantum mechanics, inten- The force field approach fits the force constants by using
sities can be readily calculated, although the evaluation of experimental vibrational frequencies as input parameters.
Raman intensities requires substantial computational effort and Actually, historically, force fields were initially developed
is therefore often not reported. for the description of vibrations in molecules [17]. Rephrased,
Finally, the calculation of vibrational frequencies within molecular vibrations were the natural origin of the force field
the context of a harmonic potential only, implies a significant methodology. In those days vibrational spectra were actually
limitation. In reality molecular vibrations are not purely used in order to derive a force field and establish the force field
harmonic. When the anharmonicity is neglected, one may parameters. For that purpose the so-called Wilson GF matrix
expect a difference between calculated and experimental formalism had been developed [17]. The so-called molecular
vibrational frequencies. For empirical methods such as the mechanics method, commonly abbreviated to MM, works the
force field method the anharmonicity might be hidden in other way around: given a set of force constants the vibrational
some parameters, though indirect, and its transferability to spectrum may be calculated. Up to the present day, both
somewhat different molecular structures is, therefore, not schemes are used in practice. The MM approach, however,
guaranteed. For quantum mechanical ab initio calculations differs from the original force field in a number of other
this difference is well-known. For the Hartree–Fock method it respects. MM may be applied to a large variety of problems. In
is of the order of 10–15% of the vibrational frequency, a MM force field the force field is usually parametrized to
although this is not a priori only due to the neglect of reproduce the molecular geometries and relative conforma-
anharmonicity. In order to obtain a better match with tional energies. Using functional minimisation strategies,
experimental frequencies the so-called scaling factors have molecular mechanics can be employed to obtain minimum
been introduced. These will be discussed separately in energy structures of molecules. Therefore, when using a force
Section 4. field approach, one always has to bear in mind for what purpose
the particular force field was devised and therefore whether it is
3.2. The force field approach an appropriate choice for the calculations of vibrational spectra.
In the past there were both generic, widely applicable (large
The force field approach starts from a potential energy set of varying molecules), force fields as well as more dedicated
expression comprising the so-called force constants. There is specific ones. The latter ones have become obsolete for a while
a different force constant for each different pair of atoms, i.e. [18], except for very specific examples involving large systems.
the force constant for the C–C bond stretch is different Furthermore, currently the force field approach is out-dated
from the force constant for C–N, etc. For a molecule with respect of calculating the spectra of small molecules. Here
comprising more than just two atoms, additional potential the quantum mechanics (QM) based approach has successfully
energy terms arise, i.e. bond angle and torsional energy taken over in recent years. For large systems, comprising
terms, and finally non-bonded van der Waals and Coulombic hundreds to thousands of atoms, a force field method is still the
interaction terms. This leads to a potential energy expression only feasible tool. However, when the larger system can be
of the form: easily decomposed in smaller entities for which spectra can be
X X calculated with QM methods that might be the preferred way to
Epot ¼ K 2 ðb  b0 Þ2 þ H 2 ðu  u0 Þ2 obtain independent and reliable spectroscopic information.
b u
X When explicit dynamical effects of a large system, e.g. a
þ fV 1 ½1  cosðf  f1 0 Þ þ V 2 ½1  cosð2f  f02 Þ molecule in a solvent or another condensed state system, are
f involved, the force field method can be a useful tool. On the
X
þ V 3 ½1  cosð3f  f3 0 Þg þ K x x2 other hand recent developments have led to methods treating
x the core of a large system by QM, whereas the environment is
X qi q j X treated by a force field approach, and which are known as
þ þ ei j ½2ðrij =r i j Þ12  3ðrij =r i j Þ6  (7)
er
i> j r ij i> j
hybrid methods. Experience with respect to calculating
vibrational spectra is still extremely limited, however.
The successive terms are the bond stretching term with force
constants K2, the valence angle term with force constants H2, a 3.3. Quantum mechanics based calculations
three-term Fourier expansion of the torsional potential (force
constants V1, V2, V3), an out-of-plane term with force constant The alternative is to employ a quantum mechanics based
Kx, a Coulomb term describing the interaction between approach to obtain the potential energy surface (the second
charged atoms, and finally a Lennard–Jones 12–6 potential for derivatives, see above) for a molecule. Using ab initio quantum
describing the van der Waals interactions. Various force fields mechanics, all properties of a molecular system can, at least in
have been designed, varying in the number and precise type of principle, be calculated for a molecular system starting from the
individual terms in Eq. (7). Apart from the above-mentioned Schrödinger equation, which requires no experimental input.
central force field other examples are the valence force field Therefore such methods are also known as ab initio (from the
and the Urey–Bradley force field. For a discussion see the beginning) or first principles. The only input is initial positions
specialized but easy-to-read book by Burkert and Allinger for the nuclei, whereas the electronic wavefunctions are
[16]. described by a set of mathematical functions. A calculation
30 R.J. Meier / Vibrational Spectroscopy 43 (2007) 26–37

involves the optimisation of the electronic wavefunction to find Semi-empirical methods are not that often used anymore. Ab
the minimum energy state. In addition, an optimisation of the initio calculations have really become feasible for small (to
structure (position of the nuclei) is performed for it is necessary medium) systems, whereas large sized systems still need (and
to bring the molecule to a stationary point (a proper minimum will always need) force field methods. Nevertheless, semi-
energy structure for which the first derivatives, Eq. (3), are zero) empirical methods may have their benefits and are certainly not
before a frequency calculation is attempted. The larger the always worse than ab initio calculations, though application is
number of mathematical functions (larger basis set) to describe mainly limited to organic molecules. A more recent example
the electronic wavefunction, the lower the energy and therefore illustrating the interpretation of thus far misinterpreted data is
in principle the better the result. Because of the nature of the discussed in Ref. [12].
method, any molecule can be calculated and one is not limited
to available force fields. Early developments of ab initio 3.4. Vibrational frequencies from molecular dynamics
calculation of vibrational spectra go back to the 1970’s [19,20]. simulations
It is primarily due to the significant increase in computing
power over the last 10 years that QM based calculations have It is also possible to evaluate both the vibrational frequencies
become possible for relevant experimental systems, i.e. not just and the intensities from molecular dynamics simulations.
limited to the smallest molecules only. Imagine a diatomic molecule such as N2. The interatomic
Various QM based approaches are possible. The oldest distance as a function of time is a sine function. When we
method is the ab initio Hartree–Fock method, which however, measure the oscillatory period, or in a mathematical sense when
largely neglects the correlation (mutual interaction) between we Fourier transform this to the time-domain, we obtain the
electrons. That problem may be overcome by resorting to the vibrational frequency (see e.g., Ref. [21] for illustration). This
so-called post-HF methods inclduing configuration interaction procedure can be extended to vibrational analysis of molecules
(CI), multi configuration self consistent field (MCSF) and and extended (3D-periodic) systems, following meanwhile
Møller Plesset perturbation theory methods of order n (MPn). well-known algorithms [22].
These methods are, however, computational very intensive and The advantage of MD simulation over static normal mode
hardly feasible for the calculation of vibrational frequencies for analysis as described in previous sections is that it is not
any realistically sized molecular species as they scale with the necessary to impose the harmonic approximation. This,
number of electrons to the power of 5–7. More recently, however, depends how the actual analysis is performed. One
however, density functional theory (DFT) based methods have of the early first principle MD simulations revealing vibrational
become very popular in computational chemistry as they are data on ferrocene was reported by Margl et al. [23]. In that
computationally as efficient as HF calculations but also take paper reference is made to various methods to accomplish
account of a significant portion of electron correlation. vibrational analysis from MD simulation, some do impose the
To conclude this paragraph, the following additional harmonic approximation. Using a MD approach also allows for
information is important. As just said, the optimisation of the study of finite temperature effects. This approach is,
molecular structure and wavefunction aims for finding the obviously, applicable to both the force field as well as to
minimum energy state of the molecule. In practice we will never quantum mechanics methods. An example illustrating the
use an infinite basis set and 100% coverage of electron difference between static calculations and molecular dynamics
correlation. The better the basis set, however, the lower the simulation calculations (both with the same type of density
energy, but not a priori the better the vibrational frequencies and functional theory first principles calculations) is presented in
intensities. Only for perfect calculation, covering all electron Fig. 1 [24]. Here experimental inelastic neutron scattering data,
correlation and an infinite basis set will also supply the best revealing the full vibrational spectum of the sample and not just
frequencies and intensities. So moderate quantum calculations IR or Raman active modes, are compared with static normal
might actually provide very acceptable vibrational data. This will mode analysis and the spectrum from MD simulation. A main
be illustrated in some examples provided later on in this article. discrepancy appears near 800 cm1, where we see, however,
Semi-empirical methods, such as those implemented in the that the MD calculated band position is 40 cm1 more towards
well-known MOPAC program, simplify the Schrödinger the experimental value (800 cm1) compared to the static
equation considerably by simplifying certain integrals in the normal mode analysis (890 cm1) which is due to the fact that
calculation, but then compensate for this by parameterisation of the MD simulation takes into account the anharmonicity. The
some of them so that the calculations reproduce experimental remaining difference was attributed to the performance of the
information on, for example, the heat of formation. Once the density functional applied.
various approximations are made, the molecular properties to
which the parameters are fitted, and the molecules used in the 3.5. First principles periodic calculations (3D structures)
fitting, define a model Hamiltonian, of which the most
commonly used are the AM1 and the PM3 Hamiltonians A major breakthrough in quantum simulations of molecular
found in MOPAC. A major advantage of semi-empirical systems was the introduction of what is now known as the Car-
methods is the speed of computation compared to full ab initio Parrinello quantum molecular dynamics method [25]. This
calculations. Molecules of several hundred atoms can be treated method using periodic boundary conditions to allow for the
using a semi-empirical method. study of extended systems, though isolated molecules may also
 R.J. Meier / Vibrational Spectroscopy 43 (2007) 26–37 31

comprising up to a few hundred atoms while still retaining full

first principle (ab initio character). This and the fact that periodic
systems provides us with a method that can finally lead to
conclusions which could not be obtained previously using
isolated molecules or the cluster approach (piece of solid cut out
from the solid and subsequently treated as an object in vacuo) has
opened the possibility to study more realistic chemical species of
real practical relevance. A very appealing example is that of
methanol adsorption on zeolites that will be discussed in the
applications section.

3.6. Size limitations

Obviously, even though capabilities have substantially

increased over the last decade due to the fastly growing speed
and size in terms of memory and disk space of modern
computers, there are still severe limitations as to the size of
molecules that can be treated when calculating vibrational
Fig. 1. Comparison of measured (top) and calculated (middle: molecular
spectra. Force field analysis can handle thousands of atoms, but
dynamics and bottom: normal modes analysis) inelastic neutron scattering has its limits in application. Semi-empirical methods can
spectra of urea–phosphoric acid. (Reproduced with permission from Ref. [24].) handle up to hundreds of atoms, but also suffer from the same
problem as force field methods in that it heavily depends on the
be studied when sufficient vacuum is created around the empirical fitting parameters once employed while setting up the
molecule so that interactions with species in the neighbouring method whether good results will be obtained. Moreover,
box are negligible as illustrated in Fig. 2. The method, for a whereas in principle IR intensities can be obtained from both
review on applications in chemistry see Ref. [26], is methods, they are hardly ever reported and are likely to be
comparatively very efficient and allows for the study of systems pretty unreliable.

Fig. 2. Illustration of periodic boundary conditions as applied in a plane-wave based method. The molecules are put in a unit cell, as if we are dealing with a
crystalline structure. For a homogeneous catalyst, as shown, some vacuum is present around the catalyst to minimize interactions between images in neighbouring
cells.
32 R.J. Meier / Vibrational Spectroscopy 43 (2007) 26–37

Ab initio methods are the most versatile, but also Hartree–Fock method with a moderate basis set. To provide a
computationally a lot more demanding. High-level calculations better match between calculated and experimental vibrational
beyond the Hartree–Fock approach with a large basis set are frequencies the so-called scaling parameters were introduced
only feasible for really small systems. However, as we have long ago [27,28]. Actually there are two different scaling
argued elsewhere in this paper, one can often do with much less, approaches in use. One is due to Pulay [29] (also called ‘‘SQM
i.e. Hartree–Fock or DFT methods with a moderate basis set are force field method’’) and the other is the simple ‘‘frequency
appropriate to tackle many problems of interpretation. The scaling’’ method. The Pulay-type scaling is done through the
traditional isolated molecule approach can handle say a few force field for which the force field has to be transformed first to
dozen of atoms at maximum, for Raman intensities the demand internal coordinates and the secular equation following Eq. (4)
is so high that the limits may be reached a little earlier. Periodic has to be solved again after scaling. The other method involves
calculations of the Car–Parrinello type, which can reveal the direct application of a scaling factor to the calculated
vibrational spectra from MD simulation, are computational a frequencies and there is no need to repeat any calculation. The
lot more efficient, and quite a few dozen of atoms can still be results from both types of scaling are different, but both have
handled. Moreover, the latter approach allows to overcome the the same intention, i.e. to improve agreement between
traditional problem of recovering the harmonic frequencies calculated and experimental vibrational frequencies.
only. Computing anharmonic contributions using standard HF Effective frequency scaling factors lie typically in between
or DFT methods on the other hand quickly becomes prohibitive 0.85 and very slightly above 1.00 depending on the method of
for reason of computational demand. calculation. Values for the scaling parameters have been
These statements give a summary of general experience. Of tabulated [30] and have also been presented in many papers
course this all depends on the system of interest and the [31–33]. Pulay et al. [34] have for developed a set of
available computational resources. transferable scaling factors for DFT force fields which can be
applied to a variety of molecules. They recommend the use of a
4. The scaling parameter in QM calculations selective scaling procedure to achieve a better reproduction of
the frequencies. Although frequencies are becoming better
We dedicate a separate section to this issue because of some represented, there is an increasing factor of empiricism coming
relevant developments reported recently. As explained in the in this approach. Nevertheless, in general the Pulay method is
above, there is generally a difference between experimental and the more sophisticated one and leads to better quality results.
calculated frequencies, amounting to some 10–15% for the Whether this is sufficiently accurate must be investigated case

Fig. 3. Top: average absolute error w.r.t. CCSD(T)/cc-pVTZ results for different DFT harmonic frequencies (cm1) of H2CO, H2CS, H2CNH and C2H4. Bottom:
same for DFT anharmonicities. (Peproduced with permission from Ref. [36], and adapted from Ref. [36] with thanks to Dr. Philippe Carbonniere.)
 R.J. Meier / Vibrational Spectroscopy 43 (2007) 26–37 33

by case. In particular for larger molecules, which unfortunately rather than force field methods. Therefore we will focus here on
not that often reported in literature, with a high number of the former, though starting with an example of the latter.
vibrational bands in the fingerprint region, a highly reliable Unfortunately, most papers contain an interesting discussion on
method is required to arrive at an unambiguous and correct the topic, but few illustrations with respect to comparing
assignment. The reader is referred to the aforementioned calculated and experimental spectra.
literature for more information on methods and quality of
results. 5.1. Force field example: vibrational spectra of methane
Factors that contribute to the difference between calculated clathrate hydrates from molecular dynamics simulation
and experimental frequencies include the neglect of electron
correlation, the (limited) size of the basis set, and the neglect of An application using force fields provided information on the
anharmonic contributions. Larger basis sets do tend to yield differences between methane in a smaller and in a large clathrate
better results, and similarly do calculations including electron cage [41]. Clathrate hydrates have the general formula XnH2O
correlation (DFT or post-Hartree–Fock). It seems then the in which X is the guest molecule within a water cage, and n is the
calculations needs to be improved with respect to all these three hydration number per guest molecule. The unit cell of the
factors to obtain good agreement between calculated and common methane gas hydrate consist of two small and six large
experimental vibrational frequencies. This, however, is too cages. A MD simulation involving eight of these units was
quick a conclusion. For various combinations of method and a reported. Next to structural information, e.g. thermal expansion
small basis set the scaling factor is very close to 1.0000 [30], in the range 0–300 K, vibrational spectra were recovered and
whereas more advanced combinations such as Configuration discussed. Simulated spectra are shown in Fig. 4. Peaks for
Interaction with a large basis set (i.e. cc-pVDZ) may lead to
scaling parameters in the range 0.92–0.96 [30]. In addition, low
frequency vibrations are usually very well reproduced by
calculations within the harmonic approximation (see e.g., Refs.
[32,35]).
Very interestingly, Barone et al. recently reported calcula-
tions including anharmonic corrections [36–38]. The conclu-
sions are nicely summarized in Fig. 3 which we have adapted
from Ref. [36]. These results imply that most of the errors are
related to the harmonic approximation, i.e. to the neglect of
anharmonic contributions. This is further illustrated by the
observation that when excluding the C–H stretching modes,
which are known to contain a significant anharmonic
contribution, the average error is generally reduced (values
in parenthesis in the top graph of Fig. 3). It is also interesting to
compare two recent studies involving the 2-thiouracil molecule
to appreciate the impact of anharmonic contributions [38,39].
Furthermore, in the above we argued that in quantum
mechanical calculations a larger basis set, although providing a
lower total energy, does not a priori provide better frequencies.
This is also what we observe from the data in Fig. 3, e.g., the
B3LYP with the smaller 6-31+(d,p) basis set performs clearly
better than in combination with the larger cc-pVTZ and TZ2P
basis sets. For other functionals, e.g. B1B95, this may be the
other way around.
Thus, in conclusion, recent work has shown that anharmonic
contributions should be included to arrive at good absolute
vibrational frequencies without the need for scaling parameters.
Including anharmonicity also allows for the calculations of
overtones and combination bands [40]. Despite the larger
computational effort, this can be expected to be a significant
step forwards to the unambiguous interpretation of vibrational
spectra. Fig. 4. Top: simulated power spectra of methane atoms in methane hydrate. The
solid line represents methane H atoms, and the dashed lines represents methane
5. Applications C atoms in large and small cages. The C atom peaks at 1300 and 3000 cm1
were multiplied by a factor of 50 for ease of viewing. Contributions from water
molecules are not shown for clarity. Bottom: simulated power spectra of
The overwhelming majority of papers published on calculated methane H atoms in hydrate (solid lines) and gas (dashed line) phases, focussing
vibrational frequencies involves quantum mechanical methods on the C–H stretch region. (Reproduced with permission from Ref. [41].)
34 R.J. Meier / Vibrational Spectroscopy 43 (2007) 26–37

methane in the smaller and the larger cage are separated by is of the same order of magnitude as the mass of the halogen.
14 cm1, which agrees well with the value of 10 cm1 as This causes the effects observed as is explained in more detail
obtained from experimental Raman spectra. The paper is a nice in Ref. [12].
illustration on how powerful force field simulation can be when
an appropriate problem and force field are at hand. 5.3. Ab initio examples: I. Alkanes and polyethylene
Raman spectra
5.2. Semi-empirical example: LAM-1 Raman frequencies
in a-bromoalkanes and a,v-dibromoalkanes A problem in the spectra of alkanes and polyethylenes is that
it is hard if at all possible to obtain certain spectroscopic
The longitudinal acoustic or accordion mode (LAM) is a information from experimental data. In addition, calculations
well-known characteristic vibrational mode in linear segments may be helpful in rationalizing certain experimental data. In a
of polymeric chains. In general, for long chain length, the series of papers [14,15,42,43] we have reported calculations on
frequency is inversely proportional to this chain length. In Raman spectra of these species.
Fig. 5 the LAM frequencies are shown for the alkanes and Firstly, it is not possible to obtain spectroscopic information
bromoalkanes as taken from Mizushima and Shimanouchi on individual, specific, conformations except for the all-trans
(alkanes), and Viras and Booth and coworkers for the state, which is present in the crystalline form. The applicability
bromoalkanes (for the full references see Ref. [12]). Viras of ab initio calculated Raman spectra using octane, dodecane
et al. have adopted a model to describe the LAM frequencies in and hexadecane as models to describe parts of the Raman
which the chain groups (e.g. CH2) as well as the chain ends spectrum of PE has been reported [14,42]. The effect of
taken as point masses with appropriate interaction potential. conformation on the spectral intensities was studied. The
Whereas the gross features concerning the chain length dependence on chain conformation of the integrated intensity in
dependence of the LAM frequency in the alkanes and the 0–600 cm1 and the 700–1000 cm1 range of the Raman
bromoalkanes could be described by this model, some of the spectrum of alkanes and PE could be rationalized. The practical
essential details could not. We particularly mention the insensitivity towards conformation of the intensity in the
relatively strong odd–even effect in the dibromoalkanes and 1300 cm1 was recovered. The two all-trans C–C stretching
the jump in the LAM frequency for the monobromoalkanes bands were found to react differently with respect to
when going from C13Br to C14Br. The latter experimental conformational changes. These findings could be directly
observation was tentatively explained assuming a phase compared to experimental data, and therefore provide a further
transition between different crystal structures similar to the rationalization of the experimentally observed features. The
triclinic to monoclinic transition in the n-alkanes at C24/C26. feasibility of this type of calculations make that several of the
When, however, a straightforward fully atomic approach is early interpretations presented by Snyder and others can now be
taken, semi-empirical calculations fully recovered [12] the yet confirmed on basis of ab initio calculated data, whereas
unexplained named experimental observations. Whereas for previous studies involving interpretation have often relied on
infinite (or a least very long) chain length the additional mass of force fields which were parametrised using the vibrational
the halogen is negligible, and therefore all data fall nicely on frequencies (i.e. the Wilson GF matrix formalism referred to in
one curve as is the case for the pure alkanes (Fig. 5), for the Section 3.2). Here, the ab initio methodology provides
shorter substituted alkanes the halogen at the end of the chain independent and in this respect unbiased information. More-
causes an unbalance when the mass of the alkyl part of the chain over, the ab initio calculations provide the partly required
vibrational intensities too. These data also suggested that ab
initio calculated Raman data on relatively short oligomers
provide valuable information regarding the interpretation of
polymer Raman spectra, in particular concerning issues where
interpretation based on experimental verification is not
possible.
A long-standing problem is the issue how long a trans
sequence should be in order to contribute to the ‘crystalline’, or
better the all-trans, Raman bands at 1060 and 1130 cm1 in the
experimental spectrum. On basis of ab initio Hartree–Fock
calculations on conformers of hexadecane it was found [15] that
very short trans sequences, down to a length of 2 or even a
single trans bond, adds intensity to the frequency corresponding
to the ‘infinite’ all-trans chain. The intensity of such peaks
drops very quickly, however, much quicker than the all-trans
Fig. 5. AM1 calculated and experimental LAM frequencies vs. the number of segment length shortens. When we look at all vibrational modes
carbon atoms in the alkane for the pure alkanes (&, &), the monobromoalkanes
(^, ^) and the a, v-dibromoalkanes (~, ~). Note that the calculated and calculated within the spectral range also comprising the all-
experimental data point for the C23 dibromoalkane practically coincide (taken trans range ‘‘1060’’ and ‘‘1130’’ cm1 bands, for trans
from Ref. [12]). sequences shorter than about 10 the intensity distribution is
 R.J. Meier / Vibrational Spectroscopy 43 (2007) 26–37 35

results show that far-IR spectra may be accurately predicted for

medium sized molecules using HF or DFT type calculations with
moderate basis set size.

5.5. Ab initio examples: III. Methanol adsorption on

zeolites

Apart from studying chemical reactions within zeolites, an

important activity is the characterisation of zeolites. A popular
method is to use a small species such as methanol to
characterize acidity. However, in particular the adsorption of
water and methanol in zeolites and whether these species are
protonated by the zeolites SiOH groups has been a topic with a
Fig. 6. Experimental (dashed curve) and computed far-infrared spectra of the long history. Only about a decade ago Nusterer et al. [45] were
most stable configuration of the Al2Me2Cl4 dimer. Additional spectral features
observed in the experimental spectrum can be attributed to other, minority,
able to finally settle the case by performing periodic DFT
species, in particularly the monomer AlMeCl2. calculations allowing for the study of more than a single probe
molecule at the same time. That study revealed that whereas at
low methanol coverage methanol is unprotonated, at higher
broadly spread, contributing to the signal representing the coverage the proton is transferred from the zeolite to the
amorphous phase rather than adding to the ‘crystalline’ bands at methanol. In addition, the authors evaluated the infrared spectra
experimental Raman shifts 1060 and 1130 cm1. Only trans for this 3D structure, and found good agreement with
sequences longer than about 10 are found to add more or less experiment (see Section 3.4 in Ref. [45]).
selectively to the same C–C stretching Raman bands as
characteristic for crystalline polyethylene. 5.6. Ab initio examples: IV. CO adsorption on Co and
Both examples just discussed show that it is crucial to NiMO sulfide catalysts
evaluate vibrational intensities in these cases.
Molybdenum based sulfide catalysts are very much used in
5.4. Ab initio examples II. Aluminium-alkyl cocatalyst far- processes aiming at the reduction of heteroatom content (S, N and
infrared spectra metals) of petroleum feedstocks. Like in the previous example,
also here a small molecule, CO, is used as a probe molecule and
The structure of aluminiummethyl(chloride) complexes was the infrared spectra are used to characterize the interaction
investigated by experimental far-infrared spectroscopy and by between CO and the catalyst. As in many systems of catalytic
Hartree–Fock [44] and by DFT based calculations [35]. The nature, the interpretation of the spectra is a matter of long debate.
incentive for these studies was to combine modelling with Travert et al. [46] have performed 3D periodic DFT calculations
experimental spectroscopy in order to arrive at the best possible to model potential interaction geometries and calculated the
suggestion for the structure of a catalyst. More explicitly, by vibrational frequencies for such structures. To illustrate, two
comparing the experimental spectrum with a series of calculated adsorption geometries of surfaces considered are shown in Fig. 7.
spectra on assumed likely molecular structures it is anticipated The work by Travert et al., in which experimental IR
that the correct geometry is found. For the aluminiummethyl spectroscopy was combined with modelling, allowed for
(chloride) complexes both the vibrational frequencies and making realistic interpretation on what is going on at the
relative intensities were reproduced very well using the surface when CO adsorbs. More particularly, a CO band could
B3LYP based density functional method (DFT), viz. Fig. 6. be assigned characteristic to six-fold coordinated Mo centers,
The computed spectra allow for a unique identification of the whereas another band could be assigned to Mo centers of the
configuration of the aluminiummethyl(chloride) complexes. The S-edge of MoS2. Another band was attributed to CO

Fig. 7. Adsorption geometries of CO molecule on the 100% Ni-substituted NiMo/Al2O3 catalyst: (A) the so-called M-edge; (B) the S-edge. Yellow circles, S atoms;
light blue circles, Mo atoms; dark circles, Ni atoms; grey circles, C atoms; red circles, O atoms. (Reproduced with permission from Travert et al. [46].) (For
interpretation of the references to colour in this figure legend, the reader is referred to the web version of the article.)
36 R.J. Meier / Vibrational Spectroscopy 43 (2007) 26–37

and molecular modelling the surface and calculating the

vibrational frequencies of suggested surface geometries. g-
alumina is an important material either as a catalyst and even
much more so as a catalyst support, but there are many open
questions as to the structure at the surface. Digne et al. [47]
studied the hydroxyl groups at the g-alumina surface and
arrived at the conclusion that a popular interpretation of the
IR spectra due to Knözinger (1978) needs significant
revision. The key factor describing the OH frequency for
bridging OH groups was found to be the existence of
hydrogen bonds rather than the OH coordination. Again, this
information could only be obtained by combining experi-
mental spectra and modern 3D periodic DFT simulation
methods. Furthermore, adsorption of CO and pyridine at the
alumina surfaces was studied. Fig. 8 shows a correlation
between calculated CO frequency and adsorption energy on
Fig. 8. Correlation between the CO adsorption energy (in absolute value) and the one hand, and for different adsorption configurations (Al
the stretching frequency for the (1 0 0) surface (full circles) and (1 1 0) surface sites on (1 0 0) and (1 1 0) surfaces, respectively) on the other
(full triangles). For the two surfaces, the lowest hydration state is represented by hand.
diamonds, the highest hydration by empty circles, and intermediate hydration
by empty squares. (Reproduced with permission from Digne et al. [47].)
5.8. Ab initio examples: VI. Polyhedral vanadium oxide
cages
coordinated Mo sites adjacent to Co. This kind of information
cannot be obtained from a single spectroscopic technique, but The size dependent properties of transition metal oxide
by comparing experimental spectra and spectra calculated on clusters are of contemporary interest for their role as
realistically assumed surface geometries unique information (potential) building blocks in nanostructured materials.
can be gathered. For further interesting information that was Sauer et al. [48] studied vanadium oxide clusters both
obtained in the study we refer to the original paper [46]. experimentally (infrared multiple photon dissociation)
and computationally (DFT calculations). The study rev-
5.7. Ab initio examples: V. Acid base properties of g- ealed striking differences between different DFT func-
alumina surfaces tionals (BLYP, B3LYP, BHLYP), with only the B3LYP
reproducing the clustersize-depending change from delo-
Like the former two examples, this is a case of surface calized to localized d-electron states, as is illustrated in
characterisation by combining experimental IR spectroscopy Fig. 9.

Fig. 9. Experimental and simulated vibrational spectra of vanadium oxide cluster anions in the region of the V–O and V O bond stretch modes. The calculated stick
spectra were convoluted for better comparison with the experiment. Gray shaded peaks indicate localization of the unpaired d-electron (Reproduced with permission
from Ref. [48].)
 R.J. Meier / Vibrational Spectroscopy 43 (2007) 26–37 37

6. Summary and outlook [16] U. Burkert, N.L. Allinger, Molecular Mechanics, ACS Monograph, vol.
177, American Chemical Society, 1982.
[17] E. Bright Wilson Jr., J.C. Decius, P.C. Cross, Molecular Vibrations, The
Force fields have been a powerful tool in the history of Theory of Infrared and Raman Vibrational Spectra, McGraw-Hill, New
interpreting vibrational spectra. However, with the advent of York, 1955;
time, quantum mechanics based calculations have been taking E. Bright Wilson Jr., J.C. Decius, P.C. Cross, Molecular Vibrations, The
over, though practical limitations still apply. In general Theory of Infrared and Raman Vibrational Spectra, Dover Publications,
molecular modelling studies involve the treatment of the New York, 1980.
[18] R.J. Meier, Vib. Spectrosc. 10 (1996) 319–323.
molecular system in vacuum, i.e. no solvent or any other type of [19] P. Pulay, Mol. Phys. 21 (1971) 329.
environment (as in the solid state) is taken into account. In [20] H.B. Schlegel, S. Wolfe, F. Bernardi, J. Chem. Phys. 63 (1975) 3632.
many cases this suffices, but one has to be aware of the [21] V. Wathelet, B. Champagne, D.H. Mosley, J.-M. Andre, S. Massida,
limitations. Force field calculations are very fast, and may be Chem. Phys. Lett. 275 (1997) 506–512.
[22] D. Frenkel, B. Smit, Understanding Molecular Simulation: From Algo-
applied to a system comprising thousands of atoms. So, when a
rithms to Applications, 2nd. ed., Academic Press, San Diego, 2002.
sufficiently accurate force field is available, amorphous [23] P. Margl, K. Schwarz, P.E. Blöchl, J. Chem. Phys. 100 (1994) 8194–8203.
polymer systems or a molecule in a solvent may be treated [24] F. Fontaine-Vive, M.R. Johnson, G.J. Kearly, J.A.K. Howard, S.F. Parker,
explicitly. On the other hand, when no appropriate force field is J. Am. Chem. Soc. 128 (2006) 2863–2869.
available, the quantum mechanics approach is the way to go. [25] R. Car, M. Parrinello, Phys. Rev. Lett. 55 (1985) 2471–2474.
Recent developments have shown that it is primarily the [26] V. VanSpeybroeck, R.J. Meier, Chem. Soc. Rev. 32 (2003) 151–157.
[27] P. Pulay, G. Fogarasi, G. Pongor, J.E. Boggs, A. Vargha, J. Am. Chem. Soc.
neglect of anharmonicity that is responsible for the difference 105 (1983) 7037–7047.
between calculated and experimental vibrational frequencies. [28] G. Fogarasi, P. Pulay, Ab initio force fields and vibrational spectra, in: J.R.
Experience with vibrational intensities, although considered Durig (Ed.), Vibrational Spectra and Structure, vol. 14, Elsevier,
highly relevant for the interpretation of medium sized molecular Amsterdam, 1985, pp. 125–219.
species, is still limited. Nevertheless, the examples discussed [29] J. Baker, A.A. Jarzecki, P. Pulay, J. Phys. Chem. A 102 (1998) 1412–
1424.
reveal that combining experiment and theory may lead to unique [30] http://srdata/nist/gov/cccbdb/vsf.asp.
information, which cannot be obtained otherwise. [31] A.P. Scott, L. Radom, J. Phys. Chem. 100 (1996) 16502–16513.
With the current capabilities at hand, one does not need to [32] M.P. Andersson, P. Uvdal, J. Phys. Chem. A 109 (2005) 2937–2941.
await further developments in theory or computational power to [33] Y. Tantirungrotechai, K. Phanasant, S. Roddecha, P. Suarawatanawong, V.
Sutthikhum, J. Limtrakul, J. Mol. Struct.: Theochem. 760 (2006) 189–192.
benefit significantly of the calculation of vibrational spectra.
[34] G. Rauhut, P. Pulay, J. Phys. Chem. 99 (1995) 3093–3100.
This includes the simulation of spectra via Molecular Dynamics [35] E. Koglin, D. Koglin, R.J. Meier, S. van Heel, Chem. Phys. Lett. 290
simulation. However, in many cases the involvement of a (1998) 99–104.
theoretician still seems required in most practical exercises. [36] P. Carbonniere, V. Barone, Chem. Phys. Lett. 399 (2004) 226–229.
[37] V. Barone, Chem. Phys. Lett. 383 (2004) 528–532.
[38] V. Barone, G. Festa, A. Grandi, N. Rega, N. Sanna, Chem. Phys. Lett. 388
References (2004) 279–283.
[39] M. Alcolea Palafox, V.K. Rastogi, R.P. Tanwar, L. Mittal, Spectrochim.
[1] J.M. Chalmers, P.R. Griffiths (Eds.), Handbook of Vibrational Spectro- Acta Part A 59 (2003) 2473–2486.
scopy, John Wiley & Sons, 2002. [40] D. Begue, P. Carbonniere, V. Barone, C. Pouchan, Chem. Phys. Lett. 415
[2] R.J. Meier, Chem. Soc. Rev. 34 (2005) 743–752. (2005) 25–29.
[3] B. Galabov, Y. Yamaguchi, R.B. Remington, H.F. Schaefer III, J. Phys. [41] J.A. Greathouse, R.T. Cygan, B.A. Simmons, J. Phys. Chem. B 110 (2006)
Chem. A 106 (2002) 819–832. 6428–6431.
[4] C.F. Kettering, L.W. Shutts, D.H. Andrews, Phys. Rev. 36 (1930) 531. [42] A. Tarazona, E. Koglin, B. Coussens, R.J. Meier, Vib. Spectrosc. 14 (1997)
[5] J. Dechant, Ultrarotspektroskopische Untersuchungen an Polymeren, 159–170;
Akademie-Verlag, Berlin, 1972. A. Tarazona, E. Koglin, B. Coussens, R.J. Meier, Vib. Spectrosc. 15 (1997)
[6] T. Simanouti, S.-I. Mizushima, J. Chem. Phys. 17 (1949) 1102. 147.
[7] J.H. Schachtschneider, R.G. Snyder, Spectrochim. Acta 19 (1963) 117. [43] R.J. Meier, E. Koglin, Vib. Spectrosc. 20 (1999) 151–153.
[8] R.G. Snyder, J. Chem. Phys. 47 (1967) 1316. [44] A. Tarazona, E. Koglin, F. Buda, B.B. Coussens, J. Renkema, S. van Heel,
[9] R.G. Snyder, J. Mol. Spectrosc. 23 (1967) 224. R.J. Meier, J. Phys. Chem. 101B (1997) 4370–4378.
[10] P.C. Painter, M.M. Coleman, J.L. Koenig, The Theory of Vibrational [45] E. Nusterer, P.E. Blöchl, K. Schwarz, Chem. Phys. Lett. 253 (1996) 448–
Spectroscopy and its Application to Polymeric Materials, John Wiley & 455.
Sons, New York, 1982. [46] A. Travert, C. Dujardin, F. Maugé, E. Veilly, S. Cristol, J.-F. Paul, E.
[11] E. Koglin, E.G. Witte, R.J. Meier, Vib. Spectrosc. 33 (2003) 49–61. Payen, J. Phys. Chem. 110 (2006) 1261–1270.
[12] R.J. Meier, Vib. Spectrosc. 24 (2000) 165–170. [47] M. Digne, P. Sautet, P. Raybaud, P. Euzen, H. Toulhoat, J. Catal. 226
[13] N. Gohaud, D. Begue, C. Pouchan, Chem. Phys. Lett. 420 (2006) 453–458. (2004) 54–68.
[14] E. Koglin, R.J. Meier, Comput. Theor. Polym. Sci. 9 (1999) 327–333. [48] K.R. Asmis, G. Santambrogio, M. Brümmer, J. Sauer, Angew. Chem. Int.
[15] R.J. Meier, Polymer 43 (2002) 517–522. Ed. 44 (2005) 3122–3125.