PCCP

View Article Online

PAPER View Journal | View Issue

An ab initio anharmonic approach to study

vibrational spectra of small ammonia clusters†
Published on 06 October 2016. Downloaded by Academia Sinica - Taipei on 28/04/2018 02:46:10.

Cite this: Phys. Chem. Chem. Phys.,

2016, 18, 30498
Kun-Lin Ho,ab Lo-Yun Lee,a Marusu Katada,c Asuka Fujii*c and Jer-Lai Kuo*a

Received 9th August 2016,
Accepted 4th October 2016
DOI: 10.1039/c6cp05537k
www.rsc.org/pccp
Fermi resonance between the N–H stretching (n1 and n3) and the overtone of N–H bending (2n4) in ammonia
has hindered the interpretation and assignments of experimental spectra of small ammonia clusters. In this
work, we carried out anharmonic vibrational calculations using MP2/aug-cc-pVDZ to examine the vibrational
spectra of (NH3)n=1–5 with a focus on the size evolution. The enhancement of hydrogen bond strength due
to cooperative effects will cause n1 and n4 to red-shift and blue-shift, respectively, when the size of the cluster
increases. Our calculations show that the energy order of fundamental of n1 and overtone of n4 is reversed
between n = 3 and n = 4. Therefore, while the resultant mixed levels do not show remarkable shifts in their
peak positions, the main identity of these mixed levels changes and this causes significant re-distribution of
their intensities. Furthermore, our ab initio anharmonic calculation scheme can directly evaluate the coupling
strength between different N–H stretching and overtone of N–H bending without any experimental
parameters, thus leading us to a simpler picture to understand the Fermi resonance in (NH3)n.

1. Introduction
For X–H (X = O, N, and C) vibrations, Fermi resonance
frequently occurs between their stretching modes and over-
tones of bending modes. Though there have been many reports
on detailed theoretical analyses of the Fermi resonance of O–H
and C–H vibrations,1–8 such studies on N–H vibrations have been
scarce. A typical and important problem in Fermi resonance of
N–H vibrations has been known in neutral ammonia clusters,
(NH3)n. The band assignments of the N–H stretching vibrational
region of (NH3)n have been hindered by the existence of the
Fermi resonance.
The N–H stretching vibrational (3 mm) region of infrared (IR)
and Raman spectra of the ammonia monomer has been well
analyzed.9–11 The band origins of the symmetric (n1) and
asymmetric (n3) N–H stretching vibrations are 3336.1 and
3443.6 cm!1, respectively. In addition, overtones of the bending
mode (2n4) for angular momenta of l = 0 and l = 2 appear
a
Institute of Atomic and Molecular Sciences, Academia Sinica, Taipei 10617,
Taiwan. E-mail: jlkuo@pub.iams.sinica.edu.tw
b
Department of Physics, National Taiwan University, Taipei 10617, Taiwan
c
Department of Chemistry, Graduate School of Science, Tohoku University,
Sendai 980-8578, Japan. E-mail: asukafujii@m.tohoku.ac.jp
† Electronic supplementary information (ESI) available: (1) Cartesian coordinates
of the optimized cluster structures, (2) a table for the size of Hamiltonian
matrices, (3) tables for coupling matrix elements between all N–H stretching
modes and overtones (including combination modes) of the N–H bending modes,
(4) details of the calculated spectra (peak positions, intensities, and assignments)
and (5) component analysis of the simulated spectra of (NH3)n shown in Fig. 6.
See DOI: 10.1039/c6cp05537k
and they are attributed to the Fermi resonance with the n1
mode. The origins of the overtones are 3216 and 3238 cm!1,
respectively. Here we should note that the order of the energy
levels is 2n4, n1, and n3 (from low to high energy). On the other
hand, the vibrations of solid ammonia have also been clearly
assigned.12–17 In the solid phase, the band at B3200 cm!1 is
assigned to n1 and the weaker absorption at B3300 cm!1 is
attributed to 2n4. The n3 band appears at B3370 cm!1. The
energy order of n1 and 2n4 is reversed in the transition from the
monomer to the solid. This reversal of the order is reasonably
interpreted by hydrogen bond formation, which induces the
red-shift of the N–H stretching vibration and the blue-shift of the
N–H bending vibration. Therefore, (NH3)n clusters, which bridge
the gap between the gas phase monomer and the bulk, are expected
to show the reverse process of the energy order of the n1 and 2n4
levels with an increase of the cluster size. In spite of the previous
studies on (NH3)n,9–11,17–29 however, the critical size for the energy
order reversal has not been yet determined, and band assignments
in the N–H stretch region of the clusters have been ambiguous.
Among the previous infrared (IR) studies on (NH3)n, that on
the clusters in He droplets by Slipchenko et al.11 shows the most
definitely size selective spectra in the small size region. They have
demonstrated that with the size increase from the monomer to
the tetramer, the band intensity of the 3200 cm!1 region is
remarkably enhanced while only small and smooth shifts are
observed for the positions of both the 3200 and 3300 cm!1 region
bands. They have also reported the spectra of much larger
averaged-sized clusters up to hni = 10 000 in the He droplet.25
The peak shifts of the 3200 cm!1 and 3300 cm!1 region bands
are very small after hni = 5. In addition, moderately or average

30498 | Phys. Chem. Chem. Phys., 2016, 18, 30498--30506 This journal is © the Owner Societies 2016

Paper
size-selective IR spectroscopic studies on the molecular
beams22,28 have also shown that no obvious peak shift occurs
from the pentamer to hni = 989. Therefore, in the wide size
range, no sign of the crossing of the n1 and 2n4 bands has been
observed. Based on the empirical model potential with the
perturbation correction of the anharmonicity, the band cross-
ing between n1 and 2n4 has been predicted in the size range of
n = 10–100.22 But the experimental spectra obviously contradict
such band crossing in this size range. Because of the lack of
Published on 06 October 2016. Downloaded by Academia Sinica - Taipei on 28/04/2018 02:46:10.
the observation of the band crossing between n1 and 2n4, the
bands in the B3200 cm!1 and B3300 cm!1 regions in the clusters
have been tentatively assigned to 2n4 and n1, respectively, following
the assignments to the monomer. However, the band intensity
in the 3200 cm!1 region is larger than that in the 3300 cm!1
region with n Z 4.11,17,22,25,28 This is clearly inconsistent with
these band assignments since the intensity of the 2n4 band is
borrowed from the n1 band.
Though there have been many theoretical studies on vibra-
tional spectra of ammonia clusters,11,21–23,25,29 those on Fermi
resonance are more limited. Halonen has treated Fermi reso-
nance between the stretching and bending modes in ammonia
but it was limited to monomers only.30,31 In order to under-
stand the change in vibrational signatures from the monomer
to the solid, it is critical to develop a method that can be used to
calculate anharmonic vibrational spectra of larger clusters
using ab initio methods. Careful theoretical treatment of the
anharmonic coupling between n1 and 2n4 in the ammonia
cluster should provide new insight into Fermi resonance and
hopefully yield better band assignments in the puzzling
3200–3300 cm!1 region. In this paper, we build a high-dimensional
anharmonic Hamiltonian based on calculations at MP2/aug-cc-
pVDZ by including the cubic and quartic terms of the potential
energy surface and the linear term of the dipole moment. Direct
diagonalization of this Hamiltonian produces vibrational spectra
in satisfactory agreement with the experimental observations in
both peak positions and intensities. To assign the bands observed
in the IR spectra, components of anharmonic eigen-states in the
harmonic oscillator solution basis are analyzed in detail. Magnitudes
of anharmonic coupling among vibrational modes are also
examined. Further reduced dimension anharmonic vibrational
calculations are performed to show an essential and simple
picture of the mode coupling.
The rest of the paper is organized in the following manner.
In Section II, the details for the theoretical methods will be
given. In Section III, simple 2 " 2 and 3 " 3 models that catch
the essence of the vibrational couplings are discussed before
presenting the detailed analysis of the theoretical calculations
and discussion in comparison with experimental results. Finally,
we provide a brief conclusion in Section IV.
II. Calculation method
The structures of ammonia clusters have been extensively
studied before.11,20,21,23,24,26,29 For n = 3–4, the single-ring
structures are known to be the global minima predicted by
This journal is © the Owner Societies 2016
View Article Online
PCCP
Fig. 1 Structure of (NH3)n clusters optimized by MP2/aug-cc-pVDZ. Red
dashed lines are drawn to represent the hydrogen bonds. The global
minima of (NH3)3 and (NH3)4 are known to form ring structures. (NH3)5
has several low-energy minima and we only examine the ring structure in
this work to be compared with (NH3)4. Cartesian coordinates of these
clusters can be found in Table S1 of the ESI.†
the MP2/aug-cc-pVDZ level. Note that for the tetramer it is not
an exactly planar structure but a slightly bent boat structure.
According to the previous work,23,26,29 the global minimum
structures of size n = 2–4 have a large energy difference com-
pared to the second most stable minimum. Therefore, we only
consider the global minimum in each size. For the pentamer,
although there are several isomers having similar energy, we will
focus on the single-ring isomer for simplicity. Structures of these
isomers are shown in Fig. 1 and their Cartesian coordinates can
be found in Table S1 of the ESI.†
In this work, we include five vibrational modes of NH3
shown in Fig. 2. In addition to the symmetric n1 mode, there
are two degenerate bending modes (n4) and two degenerate
asymmetric stretching modes (n3). Umbrella motion (n2) is
excluded here, because its frequency is too low to be directly
involved in Fermi Resonance between 3000 and 3600 cm!1.
In ammonia clusters, the formation of hydrogen bonds will
break the C3v symmetry and thus n3 and n4 will no longer be
degenerate. To assist our discussion below, we partition n3 and
n4 into two kinds. The first kinds of modes, which involve
motion of the hydrogen-bonded N–H, are referred to as n3_HB
and n4_HB, and the second kinds of modes, which involve
mainly the ‘‘free’’ (non hydrogen-bonded) N–H, are referred to as
n3_free and n4_free. Schematic drawings of these modes are shown
on the right hand side of Fig. 2. Each group of the normal modes
in (NH3)n is color-coded differently using the following scheme to
assist our discussion (n4_free(black), n4_HB (blue), n1 (red), n3_free
(green) and n3_free (green)). We would like to emphasize here
that the main idea behind this naming scheme is the role of the
hydrogen bond in breaking the symmetry of the n3 and n4
modes. To be more precise, if we consider the symmetry with
respect to the mirror plane perpendicular to the page, we can
see that n4_free, n1 and n3_HB have a 0 symmetry and n4_HB and
n3_free have a00 symmetry. In the following discussions, we will
stick to this convention.
Phys. Chem. Chem. Phys., 2016, 18, 30498--30506 | 30499
 View Article Online

PCCP Paper

For a given structure of (NH3)n, we will construct a

5n-dimensional anharmonic vibrational Hamiltonian by including
all the cubic and part of the quartic terms in the potential
energy operator
X1 X1 X 1
V¼ fii qi2 þ fijk qi qj qk þ fijkl qi qj qk ql ;
i
2 i;j;k
6 i;j;k;l
24

where qi is the mass-weighted coordinate along the ith normal

mode. The coefficients are derivatives of the potential, shown
Published on 06 October 2016. Downloaded by Academia Sinica - Taipei on 28/04/2018 02:46:10.

as follows:
@2V @3V @4V
fii ¼ ; fijk ¼ ; fijkl ¼ :
@qi2 @qi @qj @qk @qi @qj @qk @ql

Second derivatives are computed by analytical expression in

quantum chemistry packages. The third and fourth derivatives
are calculated by numerical finite difference methods from
analytical second derivatives. Thus only the semi-diagonal
(up to three-body or three-different-modes) of the quartic terms
is computed.32–34
Since this anharmonic Hamiltonian is derived entirely by
ab initio methods (without any adjustable parameter), we refer
to this computational scheme as an ab initio anharmonic
approach. This form of the Hamiltonian is often used in many
anharmonic analyses.32,33 Except vibration-rotation coupling,
the Hamiltonian we use here is almost the same as the one in
second-order vibrational perturbation theory (VPT2) implemen-
ted in the Gaussian 09 program.34 For simplicity, we directly
use the third and fourth derivatives evaluated by the Gaussian
09 program based on MP2/aug-cc-pVDZ’s analytical second
Fig. 2 Five kinds of normal modes of NH3 are shown on the left-hand
side. Because of the C3v symmetry, two bending modes are degenerate
derivatives.35 The rotation-vibrational coupling is neglected
and two asymmetric stretching modes are also degenerate. In (NH3)n because of the small rotation constants of (NH3)n=3–5.11 Note
clusters shown in the right-hand side, the formation of hydrogen bonds that the fourth derivatives fijkl are neglected if i,j,k,l have totally
(shown as dashed lines) will break the C3v symmetry. The only symmetry different indices of the modes. In other words, we only consider
left is the mirror plane cutting through the hydrogen bond and perpendi- 3-body mode interaction in our calculation.
cular to the page. To assist our discussions below, we partition n3 and n4
into two kinds. The first kinds of modes, which involve the motion of the
Using the ladder operators (a and a†), the Hamiltonian can
hydrogen-bonded N–H, are referred to as n3_HB and n4_HB, and the second be recast as follows:
kinds of modes, which involve mainly the ‘‘free’’ (non hydrogen-bonded) ! " X sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
X 1 1 h3
!
N–H, are referred to as n3_free and n4_free. It is straightforward to see that y
H¼ hoi ai ai þ
! þ fijk
n4_free, n1 and n3_HB have a 0 symmetry and n4_HB and n3_free have a00 i
2 i;j;k
6 8oi oj ok
symmetry. Each group of the normal modes in (NH3)n is color-coded in
$ %$ %$ %
the following scheme to assist our discussions (n4_free(black), n4_HB (blue),
" ayi þ ai ayj þ aj ayk þ ak
n1 (red), n3_free (green) and n3_free (green)).
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
X 1 h4
!
þ fijkl
For donor ammonia in the dimer, we can classify its five 24 16oi oj ok ol
i;j;k;l
normal modes as shown in the right column of Fig. 2. But for $ %$ %$ %$ %
the acceptor, the use of ‘‘HB’’ can be slightly misleading as " ayi þ ai ayj þ aj ayk þ ak ayl þ al
there is only one hydrogen bond connecting the donor and the
acceptor, there are only ‘‘free’’ modes as none of the N–H Assuming |mji is the mth eigen-state of the jth harmonic
vibrations are hydrogen-bonded. Nevertheless, if one considers oscillator, we can construct the vector space of the
the symmetry of normal modes of two n3 modes in the acceptor, 5n-dimensional vibrational Hamiltonian by taking direct
one has a 0 symmetry and the other has a00 symmetry. In this products of the |mji to form the harmonic states (HSs)
work, we will use n3_HB and n3_free to refer to modes with a 0 and {|m1,m2,. . .m5ni} = {|m1i#|m2i#% % %#|m5ni}. All of the non-
a00 symmetry, respectively, even though none of them involves zero matrix elements can be easily calculated analytically by using
motion of hydrogen bonded NH. The same convention is applied ladder operators. The size of Hamiltonian matrices ranges from 456
to the two n4 modes in the acceptor (that is n4_HB and n4_free to in the monomer to more than 240 000 in the pentamer. A detailed
refer to the n4 modes with a00 and a 0 symmetry, respectively). list of the size of the matrices can be found in Table S2 of the ESI.†

30500 | Phys. Chem. Chem. Phys., 2016, 18, 30498--30506 This journal is © the Owner Societies 2016
 View Article Online

Paper PCCP

Since the increase of the matrix size is very rapid, the current Table 1 Maximum values (in cm!1) of the coupling matrix elements
approach can effectively treat up to 25 modes. Since these between bending overtones and stretching fundamental modes in (NH3)n.
For the monomer, there is no hydrogen bond, for simplicity of the table, we
matrices are sparse and we are interested mainly on the low-
still place n3 under the n3_HB mode. The full list of matrix elements can be
lying excited states, diagonalization can be done using ARPACK found in Table S3 of the ESI
in SciPy package and extract only first 1 + 5n + 25n2 eigen-values
and eigen-functions. After diagonalizing this Hamiltonian Maximum coupling value n1 n3_HB n3_free
matrix, we can track the eigen-functions (|Cki) back to the (NH3)1-2n4 38 35 0
original set of normal modes by examining the square of the (NH3)1-n4 + n4 0 0 49
(NH3)2-2n4_free 28 28 0
inner products between |Cki and |m1,m2, . . . mni. To calculate (NH3)2-2n4_HB 45 33 0
Published on 06 October 2016. Downloaded by Academia Sinica - Taipei on 28/04/2018 02:46:10.

the absorption intensities, we approximate dipole moment m (NH3)2-n4_free + n4_HB 0 0 32

P @m (NH3)3-2n4_free 19 35 0
with only the linear term (that is m ¼ m0 þ qi ). Then (NH3)3-2n4_HB 37 19 0
i @qi (NH3)3-n4_free + n4_HB 0 0 26
infrared intensities can be obtained by applying Fermi’s golden (NH3)4-2n4_free 14 31 3
rule. Since we neglect the electronic anharmonicity, the inten- (NH3)4-2n4_HB 34 14 3
(NH3)4-n4_free + n4_HB 5 4 22
sity of overtone modes is entirely borrowed from the oscillator (NH3)5-2n4_free 11 31 8
strength of the fundamental transition. (NH3)5-2n4_HB 35 11 6
(NH3)5-n4_free + n4_HB 11 11 27

III. Results and discussion

One of the advantages of the above-mention computational
scheme is that we can directly evaluate the magnitude of all of
the matrix elements. Before presenting the calculated spectra
and our assignments, we should explore a few general features
among the coupling between stretching and overtone of vibra-
tional modes in (NH3)n to build simple models to assist our
understanding of the vibrational coupling. Our calculated
spectra of (NH3)n=3–5 will be compared with experimental data
from our analysis and assignments.
As we mentioned earlier, the formation of hydrogen bonds
will break the C3v symmetry, but there still remains the sym-
metry with respect to the mirror plane. As seen in the normal
modes on the right-hand side of Fig. 2, the wave functions of
the overtone of 2n4_free and 2n4_HB are both a 0 (even) and the
wave function of combination bands (n4_free + n4_HB) is a00 (odd).
It is therefore straightforward to see that coupling between
2n4_free/2n4_HB and n1/n3_HB is allowed by symmetry. The combi-
nation bands (n4_free + n4_HB), however, can only couple to
n3_free. The full list of matrix elements is compiled in Table
S3 of the ESI.† In Table 1, we summarize the coupling matrix
elements by listing only the maximum values within each
group. In (NH3)2 and (NH3)3, this mirror symmetry is kept, so
the coupling matrix elements between combination bands
(n4_free + n4_HB) and n1/n3_HB are zero. In (NH3)4 and (NH3)5,
even though the planar symmetry is broken, their maximum
coupling strength remains relatively small (less than 10 cm!1).
It is interesting to note that the coupling between 2n4_HB and
n1 seems to be the strongest (45–35 cm!1) followed by the
coupling between 2n4_free and n3_HB (35–28 cm!1). The coupling
between 2n4_HB and n3_HB as well as the coupling between
2n4_free and n1 is smaller with the maximum value between
19 and 11 cm!1 for (NH3)n=3–5.
A. Simple models
In the following, we will present two simple modes to qualita-
tively explain the key features of the Fermi Resonance in (NH3)n.
It has been mentioned earlier that the magnitude of these
coupling terms is not too sensitive to the size of the clusters,
but frequencies of some of the vibrational modes are more
sensitive to the strength of the hydrogen bond which is enhanced
as the size of the cluster increases. This general phenomenon is
known as the cooperative effect that is found in many neutral
hydrogen bonded clusters. Among the five kinds of normal
modes, it is intuitive to assume that n1, n3_HB, and n4_HB are
more sensitive to the strength of the hydrogen bond. This notion
is confirmed by both the harmonic frequencies shown in Fig. 3(a)
and the anharmonic calculations shown in Fig. 3(b) and (c). We
shall come back to anharmonic results with more discussions
in the next section. Using the harmonic frequencies (Fig. 3(a))
as the reference, we can see that the n1 and n3_HB decrease by
B100 cm!1 and n4_HB increases by more than 20 cm!1 with the
increase of the size. Here we note that we use two times of
the harmonic frequencies of n4_free and n4_HB to represent
their overtone. Since there are too many combination modes
(n4_free + n4_HB) and they do not directly couple to n1 and n3_HB,
we do not show them in Fig. 3. However, all combinational
modes are included in our anharmonic calculations.
In Fig. 3(b), we show results of two reduced dimensional
anharmonic calculations. In the first one, we consider only the
coupling among 3n N–H stretching modes, and one can see
that the general trend of the size dependence is maintained but
the frequencies of all N–H stretching modes are lowered by
including the anharmonic coupling. In the second reduced
dimensional calculation, we include only 2n N–H bending. In
Fig. 3(b), the frequencies of the overtone of n4_free and n4_HB do
not deviate from those estimated from the harmonic frequen-
cies. This implies that the couplings among the bending modes
are not significant. The frequencies of n3_free and n3_HB are too
high to be directly involved in the Fermi resonance with n4_free
and n4_HB. However, it is important to note that they should not
be excluded in the calculation. As an example, we show in
Fig. 3(c) the frequencies 2n4_HB and 2n4_free after their couplings
with n3_free and n3_HB are explicitly included. One can see that

This journal is © the Owner Societies 2016 Phys. Chem. Chem. Phys., 2016, 18, 30498--30506 | 30501
 View Article Online

PCCP Paper

n1 and 2n4_HB become very close. We should note that in the data
shown in Fig. 3(c) the coupling between n1 and 2n4_HB/2n4_free is
excluded intentionally to ‘‘turn off’’ Fermi resonance. The results
of the full 5n-dimensional Hamiltonian will be shown in the next
section after introducing the simple models.
2 " 2 model. The simplest mode is to consider the coupling
between n1 and 2n4_HB, because (1) n1 has the largest oscillator
strength, (2) the energy matching between n1 and 2n4_HB is the
best, and (3) these two manifolds have the largest coupling
Published on 06 October 2016. Downloaded by Academia Sinica - Taipei on 28/04/2018 02:46:10.

matrix elements. In the dimer, the frequency of n1 is higher

Fig. 3 Frequencies of the N–H stretching and N–H bending modes calcu-
lated by different levels of anharmonicity. (a) Harmonic frequencies and two
times of the harmonic frequencies of n4_free and n4_HB are also used to
represent their overtones. The dashed lines represent the averaged frequency
of each kind of vibrational mode for a given size of the ammonia cluster.
(b) The results of two reduced dimensional anharmonic calculations. In the
first one, we consider only the coupling among 3n N–H stretching modes
(including n1, n3_HB and n3_free). In the other 2n dimensional calculation, we
only include n4_HB and n4_free. While the combination modes (n4_free + n4_HB)
are included in the calculations, they are not shown to keep this figure simple.
(c) the frequencies of n1, 2n4_HB and 2n4_free after their couplings with n3_free
and n3_HB are explicitly included. We should note here that coupling between
n1 and 2n4_HB/2n4_free is not included in (c). Spectra of full 5n-dimensional
Hamiltonian that includes Fermi resonance will be shown in Fig. 6.
the peak positions of 2n4_HB and 2n4_free are lowered in Fig. 3(c).
Thus, one can immediately expect that strong Fermi resonance
will begin to occur between n = 3 and n = 4 as the frequencies of
than that of 2n4_HB. Between the trimer and the tetramer, the
energetic order between these two modes is reversed. To assist
our discussions, a schematic diagram to summarize relative
positions of these vibrational states is shown in Fig. 4(a). To
further quantify this simple two-level model, we introduce a
parameter x into the 2 " 2 matrix shown in Fig. 5 to represent
the dependence of the frequencies of n1 and n4_HB.
Solution of this 2 " 2 model with x ranging between 0 and
160 cm!1 is shown in Fig. 5. Because the off-diagonal term is
35 cm!1, there always exists a gap of at least 70 cm!1 between
the high- and low-frequency mixed states. From the monomer to
the dimer, changes in the frequencies (x) are small (o40 cm!1),
and therefore the mixing of n1 and 2n4_HB is minor due to the
relatively large energy mismatch. As a result of the cooperative
effect, the hydrogen bonds in the trimer are strengthened, the
value of x reaches B80 cm!1, and the frequencies of 2n4_HB and
n1 states are very close to each other. Under such conditions,
the two states are heavily mixed consisting of nearly half 2n4_HB
and half n1. In the tetramer and the pentamer, the strength of
hydrogen bonds is saturated (with x B 130–160 cm!1). In these

Fig. 4 Two simple models to represent the key components of the Fermi resonance in (NH3)n. (a) Two-level models constructed to represent the
strongest coupling (45–35 cm!1) between n1 and 2n4_HB. The frequencies of n1 and 2n4_HB are most sensitive to the strength of hydrogen bonds in (NH3)n.
While the frequency of the latter increases by 40–50 cm!1 as the size of the cluster increases from the dimer to the pentamer, the former decreases by
B100 cm!1. (b) Three-level models constructed to illustrate the role of 2n4_free in (NH3)n=3–5. In our calculations, 2n4_free does not directly couple to
2n4_HB and its coupling with n1 is relatively weak (B15 cm!1). Because of the energy matching condition, it is reasonable to approximate that the main role
of 2n4_free is to split the low-energy states in the two-level models. In Fig. 5, we will show numerical solutions to their corresponding 2 " 2 and 3 " 3
matrices to quantify their size dependence.

30502 | Phys. Chem. Chem. Phys., 2016, 18, 30498--30506 This journal is © the Owner Societies 2016

Paper
Published on 06 October 2016. Downloaded by Academia Sinica - Taipei on 28/04/2018 02:46:10.
Fig. 5 Solutions of the two-level and three-level models shown in Fig. 4
(the same color-coding is applied). In the 2 " 2 and 3 " 3 matrices, a
parameter x is introduced to represent the shifts in the frequencies of n1
and 2n4_HB caused by the change in the strength of the hydrogen bonds as
the size of the cluster changes. In the top panels, the dashed lines depict
the dependence of the diagonal terms of the matrices. Dependence of the
eigen-values of the 2 " 2 and 3 " 3 matrices on x is shown as the solid
lines and they are colored according to their major components. In the
2 " 2 matrix, a gap larger than 70 cm!1 can be found. The high- and low-
frequency states are dominated by n1 and 2n4_HB when x is small and their
identities change completely when x becomes greater than 50 cm!1. In
the 3 " 3 matrix, the high-frequency state is also well separated with
character almost the same as their counterpart in the 2 " 2 matrix. The
middle- and low-frequency states are the mixture of 2n4_free and 2n4_HB.
cases, the assignment becomes clear again with the high-frequency
state dominated by 2n4_HB and the low-frequency state by n1.
The basic feature of such a two-level mode is the gap
between the high- and low-frequency mixed states. The magni-
tude of the gap is determined by the coupling constant.
This journal is © the Owner Societies 2016
View Article Online
PCCP
Because of this gap, even though the identity of these two
states is completely exchanged at x B 80 cm!1, the resultant
peak positions barely shift in this region. On the other hand,
the intensities (which are proportional to the square of the n1
component) are very sensitive to the change in x (i.e. magnitude
of the mixing). For x o 80 cm!1, the high-frequency state is the
main intensity carrier, and for x 4 80 cm!1, the low-frequency
state becomes more intense.
3 " 3 model. If we follow the frequencies of the reduced
dimensional anharmonic calculations shown in Fig. 3(c), it is
obvious that in the tetramer and the pentamer, the frequency of
the n1 state is lower than that of the 2n4_HB and it falls in the
vicinity of 2n4_free. One step forward to make this simple model
more realistic is to include the 2n4_free mode to build a three-
level model (shown in Fig. 4(b)).
As we mentioned earlier, the maximum strength of the
coupling between 2n4_free and n1 ranges is 11–19 cm!1. Then,
we use the averaged value of 15 cm!1 to construct a 3 " 3
matrix. The coupling between 2n4_free and 2n4_HB is found to be
essentially zero by checking the matrix elements in our calcula-
tions. Since the frequency of 2n4_free is less sensitive to the
cluster size than that of n1 and 2n4_HB, for simplicity, we assume
the frequency of 2n4_free to be a constant. Due to the relatively
large energy mismatch between 2n4_free and the high-frequency
state in the 2 " 2 model, it is reasonable to anticipate that
this high-frequency state is not significantly perturbed by the
addition of the 2n4_free mode. Thus it is qualitatively acceptable
to think that 2n4_HB and n1 mix first, and then the generated
low-frequency state will couple directly with the 2n4_free mode.
A schematic diagram to summarize relative positions of these
three vibrational states is shown in Fig. 4(b).
Solution of this 3 " 3 matrix with x ranging between 0 and
160 cm!1 is shown in Fig. 5. The three eigen-states are
separated with a larger gap between the highest frequency state
and the two low-energy states. As expected, the outcomes
in small and large values of x are similar to the results in the
2 " 2 model. Under the near resonance conditions (that is
x B 80 cm!1), the coupling among three states is a little bit
complicated hindering a clear assignment. A direct implication
is a complex vibrational spectrum with many intense peaks. In
the next section, we shall see that the ammonia trimer indeed
has such a complex vibrational spectrum.
B. Calculated spectra
The simulated IR spectra of (NH3)n, n = 1–5, based on the
5n-dimensional anharmonic Hamiltonian are shown in Fig. 6(a).
Experimental IR spectra in the He droplet measured by
Slipchenko et al.11 are reproduced in Fig. 6(b) for comparison.
The detailed peak positions, intensities, and assignments of these
spectra are compiled in Table S4 of the ESI.† In addition to this
table, we followed the spirit of the simple models presented in the
previous subsection and plotted the component analysis of these
complex spectra between 3100 and 3400 cm!1 in Fig. S1 of the
ESI.† The similarity between the full 5n-dimensional results in
Fig. S1 (ESI†) and those of the 2 " 2 and 3 " 3 models in Fig. 5
Phys. Chem. Chem. Phys., 2016, 18, 30498--30506 | 30503
 View Article Online

PCCP Paper
Published on 06 October 2016. Downloaded by Academia Sinica - Taipei on 28/04/2018 02:46:10.

Fig. 6 Comparison of the simulated (left panel) and experimental (right panel) IR spectra of (NH3)n, n = 1–5. Simulated spectra were obtained by solving
the full 5n-anharmonic Hamiltonian calculated at the MP2/aug-cc-pVDZ level. Experimental IR spectra of (NH3)n in He droplets were recorded by
Slipchenko et al. (ref. 11). The color-coding in the simulated spectra indicates the dominant component in each peak: green is N–H asymmetric
stretching (n3), red is N–H symmetric stretching (n1), and blue is N–H bending overtone (2n4). In the calculated spectrum of (NH3)1, the un-expected
splitting for the n3 doublet is due to numerical errors from the finite-difference method; the difference between two peaks is 4 cm!1. And since we use
only the linear part of the dipole moment, it may cause small intensities for 2n4 peaks. In (NH3)3, n1 and 2n4 couple heavily and it is hard to assign main
components; thus we color each peak in red and blue to indicate the weight of n1 and 2n4 modes, respectively. A full set of numerical data of these
spectra can be found in Table S4 of the ESI.† By comparing both the simulated and experimental spectra, we also attempt to assign the experimental
spectra using the same color-coding scheme. We can see for n = 4 and 5 that the main oscillator strength B3200 cm!1 is assigned to n1 modes. The
gap B3300 cm!1, found in both the simulated and experimental spectra, is direct evidence of the strong Fermi resonance between n1 and 2n4_HB.

justify that these simple models can capture the essence of the
Fermi resonance in (NH3)n.
For the monomer, the calculated peak positions for bending
overtone 2n4, symmetric stretching n1, and asymmetric stretch-
ing n3 are at 3165, 3351, and 3492 cm!1, respectively. These
peak positions of the calculated spectrum deviate from the
experimental values by B50 cm!1. We think that such a level of
agreement with experiment is satisfactory as we do not use
any fitting or experimental parameters and the anharmonic
Hamiltonian is truncated at the 4th order. Since the 3rd and
4th order terms were obtained by numerical finite differences,
numerical uncertainty can break symmetry, and in the monomer,
this uncertainty causes the doubly degenerate n3 mode to split
by 4 cm!1. Furthermore, we do not consider the rotational term,
so there is no splitting due to the different rotational angular
momentum and other couplings. The intensities of 2n4 are
relatively weak, less than 5% of the intensity of n1.
Because of the hydrogen bond in the dimer, some of the n1
and n3 modes involve movement of the bonded hydrogen and
thus their intensities are enhanced by more than 10 times and
the peak positions are red-shifted to 3312 cm!1 and 3450 cm!1,
respectively. The 2n4_HB (in donor) involving bonded hydrogen
is blue-shifted to 3211 cm!1. The intensity of 2n4_HB is also
enhanced due to the stronger mixing with n1, but it is still weak
about one-sixth of the intensity of n1. The other 2n4 modes
remain weak and they are not visible in the calculated spectra.
For the trimer, because the hydrogen bonds are further
strengthened by the formation of a ring structure (cooperative
effect),23 the peak positions of n3 are further red-shifted. The
2n4 modes are slightly blue-shifted. Between 2n4_HB and 2n4_free,
the former is more sensitive to the strength of the hydrogen
bond and thus we will see that the former has a more obvious
size dependence than the latter with n = 4 and 5. The calculated
spectrum of (NH3)3 shows that n3_HB modes are red-shifted to
3425 cm!1 and n3_free is less sensitive to the formation of the
hydrogen-bonded ring. Because of the heavy mixing between n1
and 2n4 due to the good energy matching for Fermi resonance,
clear and definite assignments of the peaks are not possible. To
illustrate this complex situation in Fig. 6(a), we use both red
and blue colors to represent the whole set of peaks with highly
mixed identities.
For the tetramer and the pentamer, the frequencies of the
2n4_HB modes are higher than those of the n1 modes. In the
calculated spectra, one can see a clean gap between the high-
frequency (blue) and low-frequency (red) peaks. In this figure,
we color-coded the former with blue to reflect that their main
components come from the 2n4_HB modes and the latter with
red to emphasize that their main identities are the n1 modes.
Thus, it is straightforward to see that the peaks at B3200 cm!1
are much more intense than those at B3300 cm!1. The overall
intensity in the tetramer is enhanced by about 6 times from the
trimer indicating that the cooperative effect still plays a role.

30504 | Phys. Chem. Chem. Phys., 2016, 18, 30498--30506 This journal is © the Owner Societies 2016

Paper
For the pentamer, there is no obvious shift in the spectrum
compared to the tetramer. Therefore, the hydrogen bond strength
in the cluster is almost saturated at this size. This is also
consistent with the previous work.20,21,23,25,28 Here, we should
also point out that the splitting among the peaks at B3200 cm!1
is caused by the coupling with 2n4_free as we demonstrated with
the three-level model.
Based on our ab initio anharmonic calculations, we shall
attempt to assign the experimental IR spectra of (NH3)n in the
Published on 06 October 2016. Downloaded by Academia Sinica - Taipei on 28/04/2018 02:46:10.
He droplet measured by Slipchenko et al.11 The assignment on
the cluster size of the observed bands has been done by
Slipchenko et al. For a given size of (NH3)n, we color-code their
experimental data by comparing with our simulations. One of the
amazing features in these spectra is that although the intensities are
dramatically re-distributed between the 3200 cm!1 and 3300 cm!1
regions with increasing size from n = 1–5, the peak positions show
only slight changes in the 3200 cm!1 and 3300 cm!1 regions
from the dimer to the pentamer. In other words, obvious shifts
in the peak positions are not seen with the increase of the cluster
size from the dimer to the pentamer, while the identities of these
bands are changed completely. Our simulations indicate that the
order of the frequencies of n1 and 2n4 is reversed at around n = 3,
and the strong Fermi resonance occurs.
We make a few more remarks on the comparison between
the experimental and simulated spectra in Fig. 6. In our
calculations on n = 4 and 5, the peak positions of n1 and n3
are lower and higher than the experimental values. The red-
shift of the position of n1 by 30 cm!1 will affect the intensity
distribution and that maybe responsible for missing of the
band B3250 cm!1 in the experimental data for n = 4+. Possible
origins of these mismatches are (1) ab initio methods better
than MP2/aug-cc-pVDZ required to precisely determine the
peak position, (2) high-order terms of the potential energy
surface not included and (3) possible influence from low-
frequency modes.
IV. Conclusions
We developed an ab initio anharmonic approach to compute
vibrational spectra of small-sized ammonia clusters, (NH3)n,
n = 1–5 with a focus on the Fermi resonance between the N–H
stretching and the overtone of bending modes. All of the
geometry optimizations and force constant (up to the 4th order)
calculations were performed at MP2/aug-cc-pVDZ and no fitting
and/or experimental parameters were used. The spectra calcu-
lated by our 5n-dimensional anharmonic Hamiltonian show
good qualitative agreements with the experimental spectra. We
found that the main players of the Fermi resonance are n1 and
2n4_HB, and the coupling with 2n4_free becomes more important
in the tetramer and the pentamer. A simple three-level model
was constructed to explain these key features. To conclude, we
believe that our computed spectra and analysis offer new perspec-
tives to the long-standing puzzle regarding the mismatch between
the band assignments of the monomer and the condensed phase
spectra of ammonia.
This journal is © the Owner Societies 2016
View Article Online
PCCP
Acknowledgements
We thank Professor Andrey F. Vilesov of University of Southern
California for kindly providing us their experimental IR spectra of
ammonia clusters in He droplets to enable direct comparisons.
This study was supported by the Ministry of Science and
Technology of Taiwan (MOST101-2113-M-001-023-MY3 and
MOST104-2113-M-001-017), Academia Sinica, and the JSPS Grant-
in-Aid for Scientific Research (project no. 26288002). Computational
resources were supported in part by the National Center for
High Performance Computing.
References
1 R. A. MacPhail, H. L. Strauss, R. G. Snyder and C. A. Elliger,
J. Phys. Chem., 1984, 88, 334–341.
2 R. G. Snyder, A. L. Aljibury, H. L. Strauss, H. L. Casal,
K. M. Gough and W. F. Murphy, J. Chem. Phys., 1984, 81,
5352.
3 E. G. Buchanan, J. C. Dean, T. S. Zwier and E. L. Sibert,
J. Chem. Phys., 2013, 138, 64308.
4 Y. Wang and J. M. Bowman, J. Phys. Chem. Lett., 2013, 4,
1104–1108.
5 K. E. Otto, Z. Xue, P. Zielke and M. A. Suhm, Phys. Chem.
Chem. Phys., 2014, 16, 9849.
6 E. L. Sibert, D. P. Tabor, N. M. Kidwell, J. C. Dean and
T. S. Zwier, J. Phys. Chem. A, 2014, 118, 11272–11281.
7 E. L. Sibert, N. M. Kidwell and T. S. Zwier, J. Phys. Chem. B,
2014, 118, 8236–8245.
8 D. P. Tabor, R. Kusaka, P. S. Walsh, E. L. Sibert and
T. S. Zwier, J. Phys. Chem. Lett., 2015, 6, 1989–1995.
9 I. Kleiner, L. R. Brown, G. Tarrago, Q.-L. Kou, N. Picqué,
G. Guelachvili, V. Dana and J.-Y. Mandin, J. Mol. Spectrosc.,
1999, 193, 46–71.
10 M. N. Slipchenko and A. F. Vilesov, Chem. Phys. Lett., 2005,
412, 176–183.
11 M. N. Slipchenko, B. G. Sartakov, A. F. Vilesov and
S. S. Xantheas, J. Phys. Chem. A, 2007, 111, 7460–7471.
12 F. P. Reding and D. F. Hornig, J. Chem. Phys., 1951, 19, 594.
13 H. Wolff, J. Chem. Phys., 1971, 55, 1373.
14 G. Sill, U. Fink and J. R. Ferraro, J. Opt. Soc. Am., 1980, 70,
724–739.
15 T. Kume, S. Sasaki and H. Shimizu, J. Raman Spectrosc.,
2001, 32, 383–387.
16 J. S. Holt, D. Sadoskas and C. J. Pursell, J. Chem. Phys., 2004,
120, 7153–7157.
17 M. Jetzki, A. Bonnamy and R. Signorell, J. Chem. Phys., 2004,
120, 11775–11784.
18 H. D. Barth and F. Huisken, J. Chem. Phys., 1987, 87, 2549.
19 S. Süzer and L. Andrews, J. Chem. Phys., 1987, 87, 5131.
20 T. A. Beu and U. Buck, J. Chem. Phys., 2001, 114, 7848.
21 T. A. Beu and U. Buck, J. Chem. Phys., 2001, 114, 7853.
22 C. Steinbach, U. Buck and T. A. Beu, J. Chem. Phys., 2006,
125, 133403.
23 P. E. Janeiro-Barral and M. Mella, J. Phys. Chem. A, 2006,
110, 11244–11251.
Phys. Chem. Chem. Phys., 2016, 18, 30498--30506 | 30505

PCCP
24 P. E. Janeiro-Barral, M. Mella and E. Curotto, J. Phys.
Chem. A, 2008, 112, 2888–2898.
25 M. N. Slipchenko, B. G. Sartakov and A. F. Vilesov, J. Chem.
Phys., 2008, 128, 134509.
26 L. Yu and Z. Z. Yang, J. Chem. Phys., 2010, 132, 174109.
27 Y. Matsumoto and K. Honma, Chem. Phys. Lett., 2010, 490,
9–13.
28 M. Katada, R. Shishido and A. Fujii, Phys. Chem. Chem.
Phys., 2014, 16, 7595–7601.
Published on 06 October 2016. Downloaded by Academia Sinica - Taipei on 28/04/2018 02:46:10.
29 A. Malloum, J. J. Fifen, Z. Dhaouadi, S. G. N. Engo and N.-E.
Jaidane, Phys. Chem. Chem. Phys., 2015, 17, 29226–29242.
30 T. Lukka, E. Kauppi and L. Halonen, J. Chem. Phys., 1995,
102, 5200.
31 E. Kauppi and L. Halonen, J. Chem. Phys., 1995, 103, 6861.
32 S. Califano, Vibrational States, John Wiley & Sons. Ltd., 1976.
33 W. H. Miller, R. Hernandez, N. C. Handy, D. Jayatilaka and
A. Willetts, Chem. Phys. Lett., 1990, 172, 62–68.
34 V. Barone, J. Chem. Phys., 2005, 122, 14108.
30506 | Phys. Chem. Chem. Phys., 2016, 18, 30498--30506
View Article Online
Paper
35 M. J. Frisch, G. W. Trucks, H. B. Schlegel, G. E. Scuseria,
M. A. Robb, J. R. Cheeseman, G. Scalmani, V. Barone,
B. Mennucci, G. A. Petersson, H. Nakatsuji, M. Caricato,
X. Li, H. P. Hratchian, A. F. Izmaylov, J. Bloino, G. Zheng,
J. L. Sonnenberg, M. Hada, M. Ehara, K. Toyota, R. Fukuda,
J. Hasegawa, M. Ishida, T. Nakajima, Y. Honda, O. Kitao,
H. Nakai, T. Vreven, J. A. Montgomery Jr., J. E. Peralta,
F. Ogliaro, M. Bearpark, J. J. Heyd, E. Brothers, K. N. Kudin,
V. N. Staroverov, R. Kobayashi, J. Normand, K. Raghavachari,
A. Rendell, J. C. Burant, S. S. Iyengar, J. Tomasi, M. Cossi,
N. Rega, J. M. Millam, M. Klene, J. E. Knox, J. B. Cross,
V. Bakken, C. Adamo, J. Jaramillo, R. Gomperts, R. E.
Stratmann, O. Yazyev, A. J. Austin, R. Cammi, C. Pomelli,
J. W. Ochterski, R. L. Martin, K. Morokuma, V. G. Zakrzewski,
G. A. Voth, P. Salvador, J. J. Dannenberg, S. Dapprich, A. D.
Daniels, Ö. Farkas, J. B. Foresman, J. V Ortiz, J. Cioslowski
and D. J. Fox, Gaussian 09, Revision D.01, Gaussian Inc.,
Wallingford, CT, 2009.
This journal is © the Owner Societies 2016