I. Phys.: Condens. Matter 7 (1995) 4775-4785. Printed in the UK
A neutron BriUouin scattering study of Mg70Zn30
C J Benmaret, B J Oliver$T, J-B Sucks, R A Robinson$ and P A Egelstafft
t Depamnent of Physics, University of Guelph Guelph, Ontaio NIG 2W1.Canada
t Los Alamos National Laboratory. LANSCE. Los Alamos, NM 87545. USA
lmtitut Laue-Langevin, BP156 F-38M2 Grenoble C6dex 9. France
Received 3 FebNW 1995
Abstract. Inelastic neutron scattering measurements have been made with momentum uansfers
within the first pseudoBrillourn zone ofa magnesiumlzinc glass af 100 K and 297 K.We describe
the applieaaon of a new spectrometer (PHAROS) for these studies. Because the longitudinal
velocity of sound is about 4300 m s-', a high incident neutron energy (0.187 eV) was required.
Data have been obtained down to Q z 8 nm-' and h o = 25 meV well inside the first pseudoBrillouin zone of the glass. Evidence for two predominmtly longitudinal excitations in the
Mg,oZn,o glass is presented. Their positions agree appmximately with theoretical predictions.
1. Introduction
By analogy with the Brillouin scattering of light, neutron Brillouin scattering (NBS)describes
neutron inelastic scattering measurements made at momentum transfers hQ within the
first Brillouin zone. Whereas conventional inelastic scattering experiments on crystals are
generally performed in higher-order Brillouin zones to gain intensity and the results are
presented after reduction to the first zone, the investigation of collective excitations (such as
vibrations or magnetic excitations for example) in disordered matter is considerably more
difficult, as no reciprocal lattice exists. However, provided these systems have short- and/or
medium-range order, low-order pseudo-Brillouin zones can be defined close to the origin
of reciprocal space. Here, the first maximum Qp in the static structure factor S ( Q ) can be
regarded as the first, broadened reciprocal lattice point, such that the first pseudo-Brillouin
zone extends up to Qp/2 in reciprocal space [I].
An important role of NBS experiments is the investigation of the collective dynamics,
by determining the dynamic structure factor S(Q. w ) , for liquids and amorphous materials
and the extraction of the single-excitation part S , ( Q , w ) of S ( Q , w ) , where hw is the
experimentally observed energy transfer. This procedure is useful because the peaks in
Sl(Q. w). which contain information on the collective dynamical modes of the system,
are more easily accessible to theory and more reliably interpreted. As Q increases these
peaks become less pronounced, and at Q > 2Qp the structure of S(Q, o)is dominated
by the vibrational density of states, describing single-particle motion [1,21. A unique
advantage of NBS experiments is their ability to distinguish between longitudinal and
transverse excitations. as whilst neutrons couple directly to longitudinal vibrations they
only couple to transverse vibrations via Umklapp-type processes outside the first Brillouin
zone.
7 Present address: Department of Physics. Oklahoma State University.
0953-8984/95N4775t11$l9.50
@ 1995 IOP Publishing Ltd
Stillwater, OK 74078, USA.
4175
4116
C J Benmore et a1
Such experiments place stringent kinematic constraints on the spectrometer, since they
require a large w range at the smallest Q values. A full picture of the dispersion curve
necessitates the use of high neutron energies at scattering angles close to the straight-through
beam, and there is a large amount of associated background and multiple scattering with
few single scattering events [3].
The metallic glass Mg7oZnso can be melt-spun in continuous ribbons, from which a
suitable neutron scattering specimen may be made. Since the scattering amplitudes for Mg
and Zn are similar (5.38and 5.69 fm respectively) and the coherent scattering is about 98%
of the total, it is a suitable sample for the investigation of collective excitations. We may
represent, therefore, the scattering cross-section by an approximate formula
where E is the average scattering amplitude for Mg and Zn. Moreover, this glass has been
the subject of investigation in previous experimental studies down to momentum transfers of
12 nm-' at energy transfers of 28 meV. just inside the boundary of the fvst pseudo-Brillouin
zone (Qp/2 = 13 nm-I) [4,5].In addition, there are extensive computer simulations using
apriori pair potentials and other theoretical work [6].
Our previous experimental study of Mg7021130 on the HET spectrometer at the Rutherford
Appleton Laboratory [4,5]used 0.18 eV neutrons, and an angular range of 3.3" to 7". The
sample was enclosed in a multicellular cadmium grid, in which each cell was 6 mm square
and 2 cm deep. This arrangement enhances the first scattering compared with multiple
scattering processes. Data were taken near the boundary of the first pseudo-Brillouin zone.
In order to extend the data further into this zone it is necessary to use an instrument that
offers significantly lower scattering angles, and also to make an improved sample geometry.
These were the technical objectives of the new experiment described here: in addition, we
have begun a study of the temperature variations.
The intention of this paper is to present out NBS results on glassy Mg7oZn30 in which the
accessible Q-range has been extended down to momentum transfers of 8.0 nm-'. The ncw
experimental conditions could be met by the new instrument [71, PHAROS,at the Los Alamos
pulsed, spallation source facility (LANSCE). We employed a relatively high incident energy
(0.187 eV or a neutron velocity of 6000 m s-' compared with the velocity of longitudinal
sound in the sample of 4300 m s-'), and scattered angles down to 1'. (However, due
to the low counting rates, we averaged several detectors giving a practical minimum of
2.W.) Moreover, we have used a more sophisticated analysis of the data than was possible
in the earlier experiments, since the majority of measurements were taken well inside the
first Brillouin zone. These new data are compared to existing theoretical predictions [6].
A further goal of this experiment is to examine the variation of the spectral shape with
temperature: for example, at room temperature the neutron scattering cross-section is higher
than at low temperature (e.g. LOO K) where the excitation peaks may become sharper and
better defined.
In section 2, the theoretical background will be presented briefly, and in section 3, the
experimental samples and procedures will be described. This will be followed by sections
on the data analysis and interpretation, and their comparison with published theoretical
predictions.
2. Theoretical remarks
The structure (peaks and shoulders) of the single excitation part &(e,w) of the dynamic
structure factor S ( Q , w ) contains the information about the collective dynamics of the
A neutron Brillouin scattering study of Mg7,,Znlo
4777
atoms. Thus we separate this latter function into three parts as:
S(Q.
U)
=SdQ.
+ Si(Q, + S d Q ,
0)
0)
W)
(2)
where So represents elastic scattering and S,,,represents the multi-excitation component.
The (smallest) Q values which can be reached in an NBS experiment at a given o depend
on the scattering angle 0 and the incident energy EO or incident (k,) and scattered (k)
wavevectors necessary for the excitation of the modes. These quantities are related by the
conditions for conservation of momentum and energy, namely,
Q2=kz+k2-2kk,cosB
'
h
o=-(k:-k2)
2m
(3)
where m is the neutron mass. They may be combined into the equation:
h Q = [ Z m ( E o - h w ) - 2 ~ ~ ~ 0 ~ 0 ] " ~ .
(4)
In this section we assume that these equations yield values in the first pseudo-Brillouin zone
of the Mg702nw glass.
We start from SI(Q, w ) for an elementary polycrystal with isotropic force constanls and
an atomic mass M. and scattering in the first Brillouin zone [SI i.e.
2 W hwJZksT
SI(Q,4 = e
e
[~6 ( Q - W'(o- o q ) p ( o q ) ]
(5)
where
A=
BhQ2
4Mosin h(liwf2k~T)
and e-2Wis the Debye-Waller factor, (q,oq)are the wavenumber and frequency of a
phonon in the first Brillouin zone, p(o,) is the density of states normalized to unit area
and B is a dimensional factor. For a linear branch, with a sound velocity c, we would have
U,, = cq. It is clear that for low values of Q the amplitude factor A is small and the density
of states (which varies initially as 0,")
is also small. However, in this limit the product
Ap(w) is independent of o.Because these modes in a crystal are expected to be long-lived,
they are represented as 6 functions. For damped modes in an amorphous material, they
might be represented as Lorentzian functions, with a peak width r proportional to Qz in
the (small-Q) hydrodynamic limit. Provided that they have sufficient intensity, these peaks
are therefore easiest to detect, and their width l' is most reliably measured at the smallest
momentum transfers. But in practice, the present experimental limitations have restricted
our work to the peak positions only. To do this we need to subtract the multiphonon and
multiple scattering intensities, by methods we shall describe in section 4.
3. Experimental details
The magnesidzinc samples were fabricated from high-purity elements by melt-spinning.
The sample container comprised a flat aluminium alloy frame container of volume 8.0
(height) x 5.5 (width) x 2.0 (depth) cm3 intersected by a 6 mm x 6 nun square grid of
0.5 mm thick cadmium blades. Strips of the Mg70%3o melt-spun glass (of mass = 55.4 g
4778
C J Benmore et a1
and mean density = 1.72 g ~ m - were
~ ) pressed into each of the 96 cells in the grid. Once
loaded, the sample container was flushed for 30 minutes with helium gas before sealing, to
avoid any additional background scattering from air left in the container,
Two experiments were performed on the Mg70Zn3o glass, at temperatures of 100 K
and 297 K, using epithermal neutrons. An incident beam energy of 0.187 eV (close to the
absorption maximum for neutrons in Cd) was chosen to provide a neutron velocity greater
than the velocity of longitudinal sound, and to maximize the absorption of the grid and
thereby reduce the large background of multiply-scattered neutrons. An identical grid was
also used for the empty sample container run, and two halves of the same grid construction
were placed either side of a 3 mm vanadium plate at room temperature for the calibration
run. All sets of measurements were carried out under identical experimental conditions and
temperature fluctuations were kept to within 0.5" between runs. The fast Fermi chopper
(see [7]) for its characteristics) was run at 240 Hz for these experiments, giving a measured
resolution of 10.5 meV FWHM at the elastic line, which was constant over the entire angular
range. This frequency is slower than the optimum, giving broader than optimum resolution,
a choice that was made to increase the observed intensity. We note that 5 meV corresponds
to a timescale, given by % / w , of 0.1 ps.
PHAROS [7] is a direct-geometry inelastic chopper spectrometer, purpose-built for NBS
experiments requiring high-o-resolution, low-angle work (in another mode it can also
be used for high-angle, high-resolution work). It has a water moderator poisoned with
gadolinium places near the spallation target, and a 20 m incident flight path (see figure 1).
Two mechanical choppers are installed in the incident beam. At 14 m from the moderator,
a double-bladed chopper, rotating at 120 Hz (the fundamental frequency of the linear
accelerator) is employed to suppress the high-energy neutron background. Downstream
(at 18 m) the high-speed Fermi chopper (referred to above) is phased with the accelerator
pulse to select the required incident neutron energy.
A boron carbide mask defined the incident beam of 7.5 cm x 5.0 cm at the sample
position, and the scattered neutrons travelled a further distance of 5.75 m along the secondary
path before reaching the detector plane. The detector grid comprised five banks of linear
position-sensitive 3He detectors, each consisting of eight, one inch diameter and 36 inch
long detectors mounted vertically. For convenience, this arrangement was asymmetric in the
horizontal plane with two banks located on each side of the incident beam (covering angles
from 1.3" to 5.4" and 3.5" to 6.9", separately), with an additional single bank covering the
angles between 5.8" to 8.8" on one side. In total, this gave an angular range of 1.3" to
8.8" in the forward scattering direction and with the vertical position sensitivity gave an
x-y distribution of data points. The detector distribution for the experiment is illustrated
in figure 2. Detector calibration measurements performed before and after the experiment
showed a high degree of detector stability throughout the experiment. Due to the relatively
low flux at high incident energies, low scattering angles and attenuation of the scattered
beam by the cadmium grid, sample run times of 9 days at 100 K and 4.5 days at 297 K
were required to obtain useful statistics on the summed data. (A longer run at 297 K would
have been desirable but was not possible.)
The pixels of the x-y detector were grouped into four rings centred at zero degrees. The
profiles for these constant-@ rings are shown in figure 3, together with theoretical
curves for the two longitudinal modes in Mg&kro glass (see Hafner [6], figure 6). From
an experimental viewpoint the situation is very favourable as the U-Q rings intersect with
the theoretical dispersion curves at an angle close to 90". In addition, the width of each
ring is only spread over a 4 meV range when intersecting either of the two dispersion
(we)
A neutron Brillouin scattering study of Mg70Znao
4779
Bulk shield
Building wall
Figure 1. A schematic diagnm of the initial sac-up for [he PHAROS spectrometer at Los Alamos.
Full details wn be found in [TI.
0
A
2
Ll
0
4
6
10
Figure 2. The angular distnbution of the detector pixels on PHAROS for this eaperiment. The
decreae beyond 5' is due to the limited size of the reclangular detector a m y .
curves shown in figure 3. The ringed data sets were converted from time-of-flight onto an
4780
C I Benmore et al
60
40
-40
-60
0
5
10
15
20
25
Q b-')
Figure 3. The four cuts in u-Q space along which x-y data were collected into constantangle data (solid lines). The dashed lines conespond to the theoretical c u ~ y e sfor the two
longitudinal modes (seetext) calculatfd by Hafner (1-51 figure 6) for the metallic glass Mg.loZnw
(u=longitudinal acoustic, V)=longiludinal optic).
energy scale, the backgrounds subtracted and analysed using a two-stage process.
4. Analysis of data
In the first stage of the analysis, the ringed data were rebinned in energy onto a linear scale
using 2.0 meV bins. A correction for neutron capture efficiency in cylindrical detectors was
made 191, and the data sets normalized to the ringed vanadium spectra using the number
of counts under the vanadium elastic peak. An energy-independent correction of the type
described in [IO] was applied to the data and empty container scattering using the equation
where I,,, is the corrected intensity. Is+c and IC denote the measured intensities of the
sample plus container and smoothed empty container runs respectively. As.sc, Ac.sc and
Ac.c are the corresponding absorption correction factors calculated for Bat-plate geometry,
and M is the monitor count ratio of the two runs.
For the second stage, a two-channel maximum entropy method [I I] was used to separate
out the relatively sharp, single-excitation scattering events from the large multiply-scattered
and multi-excitation background. The maximum entropy algorithm creates two channels;
one is designed to accept only sharp features i.e. the peaks (see equation 2) corresponding
to the required So(@, U ) S l ( 0 , o),whilst the other accepts only broad structure i.e.
S,,,(0, U ) . Our experimental data will be presented as constant0 results, and the relationship
+
A neutron Brillouin scarrering study of Mg,oZnso
4781
energy transfer (mev)
Figure 4. Step in data handling for Q = 5.35‘ at temperatures of 100 K ( ( a ) to ( d ) ) and
297 K ((e) to (k)). respectively. (U)and (e): initial spectm corrected for detector efficiency and
normalized to vanadium. (b)and
rebinned energy spectra for sample plus can and smoothed
empty can rum. (c) and ( 8 ) :sampleonly spectra and broad maxtmum entropy mmponent. ( d )
and (h): SI(0. o)t SI (Q, w ) for the Mg.ioZnle glass (error bars) and So@, 0 ) (line) obtained
from the vanadium s p t m
v):
between SI(0,
o)and SI(Q,
o)for example, may be obtained readily from equations such
as (4) and (5). It was expected that the multiple-scattering and multi-excitation contributions
(which include both multiple and single scattered multiphonon terms) would dominate the
broad part of the spectra. To aid in the analysis, the following constraints were imposed
on the two channels: the broad component should be smoothly varying with the ends of
the spectra decreasing to a constant value, and the excitation peak area in the sharp spectra
should increase as Q2 (see section 2). Thus the lowest angular spectra taken at an average
angle of 0 = 2.87” were fitted assuming virtually all the inelastic scattering was due to
multiple scattering. The width of the angular rings ranged from A Q = 1.5 nm-I for the
middle two angular rings to about 3.9 nm-’ for the highest angular ring. At each angle,
a comparison of the broad components at both temperatures showed them to be similar in
shape, although the T = 297 K data are greater in intensity on the neutron energy gain side
4782
C J Benmore et a1
energy transfer (meV)
+
Figure 5. SdQ, U ) &(e. U ) (see equation 2 ) spectra for MnoZnlo ilt ( 0 ) 100 K and ( b )
297 K derived by wllecling x-y data into angular goups and interpreting by the maximum
enmpy method. The points (with mor bur) represents he experimental data while the solid
line represents the maximum envopy fit to the data (see [IO]).
as expected. Moreover, both temperature runs showed a strong angular dependence of the
broad component, a possible effect of the cadmium grid in reducing the multiple scattering
below the level of the multi-excitation processes. These broad spectra were subtracted from
+
the original four data sets to give the function SO(@,w ) SI(@, w ) (measured at fixed
scattering angle 0).
Figure 4 shows this process at both temperatures; the original data, the rebinned data
and the separation into broad and structured components are illustrated. Figure 5 shows
the SO(@, w ) SI(@. w ) data points with the most likely smooth line through the data.
Differences taken between detector rings (at the same and different temperatures) showed
little structure above the statistical noise, which suggested the presence of a plateau in
the dispersion curve which is evident when the ( W . Q) relationship for the observed peak
positions is plotted in figure 6.
An alternative interpretation may be made by calculating the density of states function
+
A neutron Brillouin scattering study of MgmZnso
4783
Figure 6. An +Q plot for the peaks in Sr(0, U ) for the melallic glass Mg$na.
The results
of the present investigalion are shown as solid diamonds and s q u m for the 291 K run and
100 K run, respectively. The solid circle denates the lower beak position obtained hom the
fi (0.w ) function shown in figure 1. The open symbols correspond to the previous results
on the same metallic glass made on the IN4, IN6 and HET spemomelers 14.51. The top line
corresponds to the theoretical pzak positions in S,(B. w). calculated by J-B Suck (see Egelsraff
[I21 figure 3). The bottom line corresponds 10 b e dispersion relation for the 'longitudinal
acoustic' mode calculated Q by Hafner (see [6] figure 6). The error bars shown on our data
are a combination of the vatialion in Q xising from the finite width of the angulv ring and the
shape of the resalved peak in SI(@, w ) .
f ' ( Q , w ) as in [6]. If we assume there is no incoherent contribution, then by using
equation (3) of 141, this quantity may be written, where ci and M, are the concentration and
mass of component i . as:
where
and SA6 is the single-excitation partial dynamic structure factor f ' ( Q , w ) will denote a
w ) partials, and ft(Q, o)will denote the similar weighted sum
weighted sum of f,s(Q,
using constant angle data in place of constant Q data. These Functions are shown in
figure I. For OUI case, the Debye-Waller factor was calculated to be very close to unity
(using equation (3.63) in [8]). The advantages of this presentation are that it emphasizes
the energy transfer intensity distribution, and that a theoretical predication is available in
[6] (figure 4 of [6]). In order to separate ft(6', w), the elastic peak contribution to the
spectra, S&3, w ) needed to be subtracted from the spectra shown in figure 5 . &(e, w ) was
C J Benmore et a1
4784
obtained by fitting the elastic peak of the vanadium spectra with a Gaussian distribution
(see figures 4(d) and 4(h)). To improve our statistics, we have added together the fl(0,o)
spectra for both the 4.29" and 5.35" rings (these angles were chosen as they have the best
8 resolutions) and both temperatures. Finally, the comparison of the experimental and
theoretical functions is shown at figure 7.
-
h
3
8
g
6
ru-
ru
0
0
0
10
20
30
40
energy (mev)
50
0
IO
20
30
40
50
energy (mev)
Figure 7. A comparison of the /'(e. w ) function calculated from the sum of individual
/.s(b', 01 functions t&en hoin [61 and shown in 7(b).and the function f,(s,o)(full line
in 7(a)) derived from our d3ta at anaes of 4 . 3 ' . 5.39 and both temperatures combined.
The individual partial dynamic stmciure factors xe represented by /M&Q,
0 ) (dotled line).
h(Q.
w ) (dashed line), / M ~ ~ ( Q U)
, (cham line) shown in (b) and the solid line in (a) is
a fit to the experimental data. The two CUNS in 7(n) (from experiment) and ( b ) (from theory)
have not been drawn to the same scale.
5. Discussion and conclusions
It can be seen in figures 5(a) and (b) that SI(e. w ) exhibits a single-excitation peak even
at the lowest angle in both the 100 K and 297 K data sets. Furthermore, despite the shorter
run time, the greater count rate associated with the single-excitation peaks in the room
temperature experiment is clearly illustrated, especially at the largest angles. The ratio of
the areas of the measured single-excitation peak at the two temperatures is also in reasonable
agreement with those values calculated from an isotropic Einstein model of the glass given
by equation (5). The high signalhackground ratio obtained on PHAROS and an improved data
treatment process have yielded overall better data compared with our previous experiments
[4,5], although the energy resolution is worse.
From the data in figure 5, the peak positions (in w and Q)have been extracted and are
plotted in figure 6. Also shown in this figure are our earlier data identified as a longitudinal
and it can be seen that the new and old data are in good
mode in the MgToZn30 glass [4,5],
agreement. The new data also extend significantly further into the first zone. The solid lines
in this figure are based on the calculation by Hafner [6]of the two longitudinal modes: as
in our earlier work, we estimated the theoretical peak positions in SI(8, w ) from Hafner's
results. Only at the lowest momentum transfers (calculated to be Q 6 3.1 nm-] in 161)
A neutron Brillouin scattering study of Mg70Znao
4785
can the acoustic and optic modes (as defined by Hafner) be clearly identified. For larger
Q values, towards the zone boundary ( Q 6.2 om-' in [e]) it has been predicted that the
acoustic optic modes will tend to be dominated by the vibrations of the Zn and Mg atoms,
respectively.
The main phonon intensity measured in this study is therefore expected to be due to
the longitudinal 'optic' mode which mainly corresponds to the vibrations of the lighter Mg
atoms. There will also be a less intense phonon peak at lower energy transfers, close to
the elastic peak, primarily due to the Zn atomic vibrations (i.e. the longitudinal 'acoustic'
mode) e.g. see figure 7 ( a ) . This explanation is the simplest interpretation of the present
results i.e. that two peaks are found in the spectra at around 18 meV and 30 meV. Due
to the experimental set-up, there is a finite band of Q values over which the phonon peak
energies were averaged: however, this phenomenon does not alter the overall results of this
experiment. The broader peaks observed at higher energy transfers in figure 5 ( E 40meV)
are probably due to residual multiphonon effects; a full calculation of the expected S ( Q , o)
intensity is required for a detailed description of this w region.
The comparison shown in figures 7 ( a ) and ( b ) suggests that a larger contribution from
the Zn-Zn correlations is seen experimentally than is predicted by theory. In addition, the
peak positions are not in full agreement, which suggests that the small discrepancy between
theory and experiment in figure 6 may be real.
It is concluded that a modern NBS instrument can extend the data on high velocity of
sound materials (e.g. Mg,oZn,o) into the first Brillouin zone. In this way, the nature of the
modes can be identified positively and excellent comparisons with theoretical predictions
(based on fundamental information) may be attempted.
>
>
Acknowledgments
We would like to acknowledge the assistance of P S Lysaght, E Larson and J P Sandoval
during the c o m e of our experiments, and the advice of Dr D S Sivia during the maximum
entropy processing. We are also grateful to the NATO Collaborative Research Grants
Program for a travel grant and to the NSERC of Canada for their financial support of
the Canadian participants. This work was funded in part by the division of Basic Energy
Sciences of the US Department of Energy.
References
Suck J-B, Rudin H. Giintherodt H-J and Beck H 1983 Phys. Rev. Lett. 50 49
Suck 1-B, Rudin H, Giintherodt H-J and Beck H I980 3. Phys. C:SolidState Phys. 13 LlC45. 1981 14 2305
Robinson R A 1989 Advanced Neutron Sources 1988 OOP Con! Ser. Topics in Applied Physics) p 31 I
Suck J-B, Egelstaff P A, Robinson R A. Sivia D S and Taylor A D 1992 Eumphys. Len. 19 207
Suck J-B. Egelstaff P A, Robinson R A. Sivin D S and Taylor A D 1992 J. N m . C f y t . Solids 150 245
Hafner J 1983 3. Phys. C: SolidStote Phys. 16 5773
[7] Robinson R A. Nuuer M. Rickets R L. Larson E. Sandoval J P, Lrsaght P S and OIivier B J 1993 Pmc.
[l]
121
[3]
[4]
151
[6]
ICANS-XI/, Ruthc,j+vd Appleton Luboratmy Repon RAL 94025 YOI
I, p 44
[SI Squires G L 1978 Introduction to the Theory of Thermal Neutron Scattering (Cambridge University Press)
[9] Windsor C G 1981 Pulsed Neutmn Scattering (London: Taylor Francis)
[lo] S o p A K and EgelstafF P A 1980 Nucl. l n s t r m Methods 178 415
[ I I] Sivia D S 1990 Maximum Entropy end Bayesian method^ ed P Fougere (Netherlands: Kluwer) p 195
1121 Egelstaff P A 1992 3. Phys. C:SolidSlate Phys. 156-158 1 4
View publication stats