JOURNAL OF NANOSCIENCE AND NANOTECHNOLOGY

The Role of Surfactant Adsorption during

Ultrasonication in the Dispersion
of Single-Walled Carbon Nanotubes
Michael S. Strano, Valerie C. Moore, Michael K. Miller, Mathew J. Allen, Erik H. Haroz,
Carter Kittrell, Robert H. Hauge, and R. E. Smalley
Center for Nanoscale Science and Technology, Rice University, Houston, Texas, USA

The ionic surfactant-assisted dispersion of single-walled carbon nanotubes in aqueous solution has
been studied by Raman and uorescent spectroscopy during ultrasonic processing. During the
process, an equilibrium is established between free individuals and aggregates or bundles that limits
the concentration of the former that is possible. This equilibrium is a function of free sodium dodecyl
sulfate concentration. At surfactant concentrations below this value, uorescence is shifted to a
lower energy due to an increase in micropolarity from water association at the nanotube surface.
The mechanism of dispersion is postulated as the formation of gaps or spaces at the bundle ends
in the high shear environment of the ultrasonicated solution. Surfactant adsorption and diffusion
then propagate this space along the bundle length, thereby separating the individual nanotube. The
former is found to be controlling, with the use of a derived kinetic model for the dispersion process
and extraction of the characteristic rate of nanotube isolation.
Keywords: Single-Walled Carbon Nanotubes, Surfactant Adsorption, Raman Spectroscopy,
Fluorescent Spectroscopy,
Delivered by Publishing Ultrasonic
TechnologyProcess.
to: Imperial College London
IP: 117.253.100.176 On: Wed, 10 Feb 2016 20:17:05
Copyright: American Scientific Publishers

1. INTRODUCTION
The solution phase dispersion of carbon nanotubes is
important for the development of bulk processing tech-
niques and makes possible the exploration of established
reaction chemistries. The problem is not straightforward,
however; carbon nanotubes invariably exist after synthesis
as aligned aggregates or bundles that are tightly bound by
an estimated 500 eV/m of tube length.11 2 Although ultra-
sonication of carbon nanotubes in organic solvents35 and
aqueous surfactant solutions21 61 7 is widely used, the mech-
anism of this dispersion and the factors that optimize its
ef ciency are poorly understood. Many researchers have
screened various surfactants61 7 and polymer adsorbates2
to modify the hydrophobic surface of carbon nanotubes
to permit water solubilization. Few techniques exist for
monitoring dispersion in the solution phase, and past work
has ignored the aggregation tenancy of nanotubes and
has instead relied upon the solution mass density of car-
bon or optical density as a measure of solubility and
dispersion.21 61 8
The discovery of nanotube uorescence7 makes pos-
sible a more precise method of detecting individ-
ual nanotube dispersion and separation from a bundle.
Author to whom correspondence should be addressed.
RESEARCH ARTICLE
A nanotube in an aligned bundle does not emit because
of energy transfer to neighboring tubes, particularly those
that are metallic, that then allow nonirradiative depopu-
lation of the excited state. In this paper, the dispersion
process is monitored by examining transient uorescent
emission and Raman scattering at 785-nm excitation as
a function of ultrasonic processing in sodium dodecyl
sulfate.
2. EXPERIMENTAL DETAILS
The dispersion process was monitored with a ow-
through apparatus where high-shear homogenization,
ultrasonication, and Raman and uorescence spectro-
scopies were performed continuously on a 1-liter car-
bon nanotube/water slurry (not previously sonicated) at
200 mg carbon/liter recycled from a central reservoir.
Solution was pumped at a ow rate of 420 ml/min from
a sealed, N2 -blanketed ask with a diaphragm pump
(Cole-Palmer) into a ow-through high-shear homoge-
nizer (Cole-Palmer.) This solution was then fed to a
ow-through ultrasonication unit (Sonics and Materials,
Inc., Newtown, CT) with temperature regulation at 32 C
through a cooling jacket on the device itself. The intensity
of the unit was calibrated by measuring the temperature

J. Nanosci. Nanotech. 2003, Vol. 3, No. 1/2 2003 by American Scienti c Publishers 1533-4880/2003/01/081/006/$17.00+.25 doi:10.1166/jnn.2003.194 81
 Strano et al./Surfactant Adsorption in Ultrasonication of Nanotubes J. Nanosci. Nanotech. 2003, 3, 8186

rise of pure water fed to the device and performing an 16

energy balance (230 W). 14 (6,5) (7,5)
The outlet of the sonicator passed through a ow- 12 (8,3)

Relative Intensity
through cuvette (Sterna), where Raman and uorescent 10
spectroscopies were performed at 1-min intervals with a 8
Kaiser Raman process spectrometer (Ann Arbor, MI) at
6
785-nm excitation. The CCD camera of the spectrometer b)
4
allows for detection out to 1050 nm, where it captures a (6,4)
(9,1)
2
portion of the nanotube uorescence spectrum at 785-nm a)
excitation. In this way, both Raman and uorescent spec- 0
0 1000 2000 3000
troscopies were captured in real time during processing.
Raman Shift (1/cm)
The observed intensity I of a spectral feature was related
to the concentration of emitting and scattering carbon Fig. 1. (a) Raman spectrum at the start of the sonication process
nanotubes, c, with the use of a point-scattering model91 10 from carbon nanotubes in solution in the absence of surfactant (3).
where (b) Raman/ uorescence spectrum after sonication in 34 mM SDS. The
lattice index of a particular uorescent feature is labeled.
I
ln D ln4Jc5 kc41 1 C 2 2 5
Io is evident at the start of the experiment, where sonication
Here Io is the incident excitation intensity, J is the scat- proceeds in the absence of surfactant. After each addition
tering coef cient, k is log4e5 D 20303, and i and i are of surfactant, the uorescence intensity rises and stabi-
the absorptivities and effective path lengths of the incident lizes to a steady-state value over a period of about 20 min
(1) and scattered (2) radiation. A calibration plot gen- to 1 h, depending on the amount of added surfactant. An
D
erated in the experimental setup yields J 60609 L/mg increased sonication time did not result in an increase in
C D
and (1 1 2 2 5 2003 L/mg and suggest that attenua- the Raman D-band, indicating that few defects if any were
tion due to reabsorption and scattering were minor below introduced by the processing. The effect of oxygen was
50 mg/liter of completely sonicated 1% sodium dodecyl explored by rst degassing the reservoir for 2 h at 100 C
sulfate (SDS) suspendedDelivered by Publishing Technology to:and
nanotubes. then conducting
Imperial the processing in a N2 shelter gas
College London
IP: 117.253.100.176 On: Wed, 10and Feb 2016
again in 20:17:05
a stream of owing O2 at 1 atm. The change
During the actual ow-through Copyright:
experiment,American
SDS (98%Scientific Publishers
pure; Aldrich) was increasingly added to the solution in purge gas had no apparent effect.
Also during the process, immediately after the addition
RESEARCH ARTICLE

reservoir. Dilution experiments were also conducted off-

line, and processed solution was subsequently diluted by of surfactant, the orescent features shift systematically
the addition of deionized water to produce the desired to higher energy. Figure 3 shows the peak centers of the
SDS concentration. The nanotubes were synthesized by (8, 3) (a) and (7, 5) (b) nanotubes as a function of time.
high-pressure CO disproportionation (HiPco) as described The shifts occur immediately after each addition, and,
previously 11
and were generated as reactor run 106.1 at when compared on the energy scale, they do so by nearly
Rice University. The nanotubes were used without further the same magnitude (8 meV).
puri cation. Raman peaks in the region from 200 to 400 cm1
arise from the scattering of distinct nanotubes with the
frequency as an inverse function of the diameter of the
3. RESULTS AND DISCUSSION
Figure 1 shows the Raman/ uorescence spectrum of the 16 Experimental
nanotube solution at the start of the experiment at 0 mM Kinetic Model
SDS concentration 435 and after 5 h of ultrasonic pro- I II III
12
Relative Intensity

cessing with progressive SDS addition terminating at IV

V (8,3)
34 mM. Ultrasonic processing increases the Raman tan- 8
gential mode at 1593 1/cm by a factor of 3.6 as individual
nanotubes are dispersed but does not change the Raman
4
shift of this peak.
During the process, the evolution of uorescent emis-
sion is monitored as a function of time. Figure 1 labels 0
the emission from the (6, 4), (9, 1), (8, 3), (6, 5), and 0 2 4 6

(7, 5) nanotubes as indicated.12 Following the growth of Time (hrs)

these uorescent features as a function of time, the inten-
sity from the sample appears to rise and saturate after the
addition of an aliquot of surfactant. Figure 2 shows the
growth of the (8, 3) feature as an example. No emission
Time (hrs)
Fig. 2. Fluorescence intensity as a function of sonication process time
for the (8, 3) feature. Following each surfactant addition (at marked
times), the intensity saturates to a steady-state value. Surfactant concen-
trations are: (I) 0.5, (II) 2.7, (III) 4, (IV) 6.2, and (V) 6.3 mM.

82
J. Nanosci. Nanotech. 2003, 3, 8186 Strano et al./Surfactant Adsorption in Ultrasonication of Nanotubes

(a) (8,3) 3.1. Increases in Resonance Enhancement

1304 with Debundling
1302 II III
Peak Center (meV)
When in aligned bundles, carbon nanotubes are known
1300
to undergo signi cant changes to their composite elec-
1298 tronic structure, including the formation of a pseudo-gap
1296 for metallic nanotubes and an orthogonal electronic dis-
1294 IV V persion in the otherwise 1-D electronic structure.14 Exper-
1292 imentally, this electronic dispersion signi cantly broadens
0 2 4 6 photon absorption, shifts it to lower energy, and reduces
Time (hrs) the overall intensity.7 When freed from bundles, isolated
(b) (7,5) individuals demonstrate sharper, more intense absorption
1214 features corresponding to singularities in their 1-D elec-
1212 II III tron density of states. Raman scattering can couple to and
Peak Center (meV)

1210 become resonantly enhanced with these transitions and

1208
increase in scattering intensity.13
In Figure 1, this is why the tangential mode at
1206
1593 cm1 increases in intensity during the sonication
1204 IV V process. At lower frequencies, the radial breathing modes
1202 undergo a unique reversal in relative intensities. The
0 2 4 6 266 cm1 feature corresponds to the (10, 2) nanotube with
Time (hrs) a v2 ! c2 optical transition at 734 nm.12 In a bundle, the
Fig. 3. Peak centers for the (a) (8, 3) and (b) (7, 5) uorescence peaks interband transitions for this feature shift to lower energy
as a function of time during sonication. Shifts in energy coincide with the and become partially resonant with the excitation laser at
addition of surfactant. The SDS concentrations are labeled per Figure 2. 785 nm. The 233 cm1 feature belongs to the (11, 3) with
a v2 ! c2 at (792 nm). When isolated, this nanotube
13 Delivered 1 by Publishing Technologyshows
tube. Peaks at 233 and 266 cm undergo enhancement to: Imperial Collegeenhancement
a considerable London in Raman scattering.
IP: 117.253.100.176 On: Wed,Similarly, 10 Feb 2016the 20:17:05
features at 225 cm 1
(10, 5) with v2 ! c2
and diminishment, respectively, as theCopyright: sonication proceedsAmerican Scientific Publishers 1
and additional surfactant is added. Figure 4a shows the at 786 nm and 215 cm (9, 7) at 790 nm also increase in
intensity with increasing individual dispersion. The fea-

RESEARCH ARTICLE
changes to the relative peak intensities, and Figure 4b
1
traces these two particular peaks in time during the exper- ture at 203 cm (13, 3) has a transition at a lower wave-
iment. It is clear that the dispersion of individual nano- length (760 nm) and does not increase notably. The width
tubes from the decay of bundles results in a signi cant of the resonance window in energy necessarily increases
increase in the 233 cm1 mode and a moderate decrease with aggregation as the absorption peaks broaden. The
1
in this 266 cm mode for Raman scattering at 785-nm (13, 3) feature is both enhanced by the shifting of reso-
excitation. nance and diminished by the broadening of the transition
and remains unchanged.
Fluorescence becomes observable during the sonication
(b)
0.6
experiment only with the addition of surfactant, indicat-
267 1/cm V
II III IV
ing that the adsorbed phase plays a critical role in the
Relative Intensity

(a) 233 1/cm

18 0.4
debundling process. The mechanism of isolating individ-
233 cm -1
16 6.3 mM 0.2 uals is explored in more detail below, but bundles are
14 267 cm postulated not to uoresce because of energy transfer to
-1

0.0
neighboring nanotubes in the bundle. Energy transfer to
Relative Intensity

12 4 mM 0 1 2 3 4

10 Time (hrs) metallic nanotubes particularly quenches the emission by

8 2.7 mM providing a nonradiative decay pathway.
6
0 mM
4 3.2. Solvatochromic Shifts Observed
2 with Surfactant Adsorption
0
150 200 250 300 350 The systematic shifting of uorescent emission to higher
Raman Shift (1/cm) energy follows the addition of surfactant during the son-
ication process. This behavior was found to occur as
Fig. 4. (a) Low-frequency Raman shift region of the spectrum para-
metric in surfactant concentration during sonication. (b) Peak intensities
an effect that is separate from ultrasonic processing.
for the 233 cm1 and 266 cm1 modes as a function of time during To examine this process in more detail, sonicated solu-
sonication. tion was centrifuged at 30,000 rpm to remove large

83
 Strano et al./Surfactant Adsorption in Ultrasonication of Nanotubes J. Nanosci. Nanotech. 2003, 3, 8186

6
1.2
5 11.3 mM
Normalized Intensity 1
4.25 mM
4

DE/DEmax
1.89 mM 0.8 1.31 eV (8,3)
3 1.42 mM 1.43 eV (6,4)
0.6
0.63 mM 1.36 eV (9,1)
2 0.4 1.27 eV (6,5)
1 0.2 Langmuir

0 0
850 900 950 1000
0 10 20 30 40
Wav e le ngth (nm)
SDS Concentration (mM)
Fig. 5. Fluorescence spectra with increasing dilution of surfactant.
Fig. 6. Peak centers in energy scaled by the maximum observable shift
for four uorescent features. The smooth curve is the trend, assuming
aggregates.7 This solution was then diluted by the addition that the shift is directly proportion to surfactant surface coverage and
of deionized water and transferred to a cuvette. Raman that the latter demonstrates a simple Langmuir monolayer adsorption
equilibrium (Eq. (1)).
and uorescence spectra were generated as a function of
solution SDS concentration. No change in the peak posi-
3.3. Insights from Modeling the Transient
tions of the uorescence is observed at concentrations
Dispersion Process
above the critical micelle concentration (CMC) of SDS
in water at 298 K (8.1 mM). Below this concentration, The concentration of uorescent individuals appears to
peaks shift to lower energy (Fig. 5), and the addition of come to a stable equilibrium at a particular surfactant con-
SDS to reconcentrate reverses this shift. These changes centration. This implies that the separation of the nano-
were metastable over a period of hours. Equilibration of tube from the bundle is reversible, leading to a dynamic
the samples overnightDelivered
resulted inbyirreversible
Publishingroping
Technology
and to:equilibrium
Imperial College
that is aLondon
function of both carbon and surfac-
IP: 117.253.100.176 On: Wed, 10tant Febconcentration.
2016 20:17:05 Modeling this process, we write
hence elimination of uorescence.Copyright: American Scientific Publishers
When normalized with respect to the maximum dCI k
D kz Cb CSDS z CI
RESEARCH ARTICLE

shift observed in a solution of carbon nanotubes in (2)

dt Keq
1% wt SDS/H2 O, this ratio traces a Langmuir adsorption
isotherm for SDS on the carbon surface: where CI and Cb are the concentrations of individual
(I) and bundled (b) nanotubes in solution, and the SDS
E KSDS CSDS
D (1) monomer concentration in the bulk is CSDS . The constant
Emax KSDS CSDS C 1 kz is the bundle unzippering rate, and Keq is the equilib-
rium constant between individual nanotubes and bundles.
Here E is the shifted peak position relative to that in
By de nition the total nanotube mass concentration CT is
a 1% SDS/H2 O solution, and Emax is the maximum constant:
shift observed (by extrapolation to zero concentration).
The trend is overlaid in Figure 6. The adsorption equilib- CT D Cb 4t5 C CI 4t5 (3)
rium constant, Kads , is 0.98 liter/mol and similar to SDS
adsorption from solution on graphitic surfaces.15 Adsorp- Integration of Eqs. (2) and (3) yields
tion of the surfactant necessarily excludes water from the Keq CT CSDS 44t5 15 C CI 405CSDS 41 C Keq CSDS 5
nanotube surface. This decrease in micropolarity16 with CI 4t5 D
4t541 C Keq CSDS 5
decreasing water association on the chromophore surface (4)
shifts the absorption and emission spectrum to higher
energy and increases the uorescence intensity through with
the elimination of nonradiative energy transfer processes. 4t5 D ekz 4CSDS C1=Keqr 5t
The rapid reversibility of the shift with SDS concentration Equation (4) accurately models many aspects of nan-
indicates that the adsorbed phase on the nanotube sur- otube dispersion in solution during sonication, including
face is in dynamic equilibrium with the bulk, and hence the observation that production of uorescent nanotubes
surfactant removal below the CSDS D CMC threshold is saturates for a given SDS concentration. This limiting
expected to ultimately destabilize suspended nanotubes. concentration is given by
This was observed after overnight equilibration of sam-
Keq CSDS
ples prepared at 34 mM SDS and then diluted to below CI1 max D C (5)
K eq CSDS
C1 T
8.1 mM.

84
J. Nanosci. Nanotech. 2003, 3, 8186 Strano et al./Surfactant Adsorption in Ultrasonication of Nanotubes

for all portions of the transient response is 1/(0.75 h), and,

hence, the initial activation via sonication must instead
be rate limiting. The overall process rate re ects the
probability that the collapse of a microbubble during the
ultrasonic processing actually initiates the unzippering of
the individual from the bundle and is therefore intensity
dependent. The latter process, once initiated, proceeds rel-
+
atively quickly.

(i) (ii) (iii) (iv)

4. CONCLUSIONS
Fig. 7. Proposed mechanism of nanotube isolation from bundles (i).
Ultrasonic processing frays the bundle end (ii), which then becomes
During ultrasonic processing of carbon nanotubes in aque-
a site for additional surfactant adsorption. This latter process continues
ous solution with SDS, an equilibrium is established
in an unzippering fashion (iii) that terminates with the release of an
isolated, surfactant-coated nanotube in solution (iv). between free individuals and aggregates or bundles that
limits the concentration of the former that is possible.
Figure 2 shows the agreement between the experimental This equilibrium is a function of free SDS concentration.
response to surfactant addition and (2). The quantity kz is For surfactant concentrations below the CMC, uorescent
found to be 0.12 liter mol1 h1 . emission is shifted to lower energy as the surface cover-
age falls below saturation. The micropolarity increase due
3.4. Implications for the Dispersion Mechanism to water association at the nanotube surface accounts for
the broadening and shifting of the spectrum in this way.
Figure 7 outlines the proposed mechanism of dispersion The mechanism of dispersion is postulated as consisting
from nanotube ropes. The role of ultrasonic processing is of the formation of gaps or spaces at the bundle ends in
likely to provide high local shear, particularly to the nano- the high-shear environment of the ultrasonicated solution.
tube bundle end (ii). Once frayed in this way, spaces
Delivered by Publishing TechnologySurfactant
to: Imperialadsorption
Collegeand diffusion then propagate these
London
between the bundle and individual nanotubes dangling
IP: 117.253.100.176 On: Wed,spaces
10 Feb along
2016 the20:17:05
bundle length, ultimately separating the
in solution become sites for surfactant adsorption.
Copyright: In the Scientific
American Publishers
individual nanotube. The former step is reasoned to be
absence of surfactant or an adsorbing polymer, the indi- rate determining, using the results of a derived kinetic

vidual is unstable to complete reattachment through van
der Waals attraction. The surfactant adsorbs and acts to
keep the two connected entities partially separated. As the
individual moves relative to the bundle through Brownian
motion, surfactant continues to progress along the nan-
otube length (iii) until separation (iv). Evidence for this
unzippering type of mechanism has been captured by
cryo-transmission electron microscopy (cryo-TEM) in the
work of Bandyopadhyaya and co-workers.8
The characteristic time for nanotube isolation, Keq =kz ,
is 0.75 h. Even correcting for the residence time within
the ultrasonicator, this time is much longer than the
unzippering rate. Brownian motion is rapid with respect
to surfactant adsorption. The characteristic time z of step
(iii) can be estimated as
s
L 1. A. Thess, R. Lee, P. Nikolaev, H. J. Dai, P. Petit, J. Robert, C. H.
z D a
2 ap
Here a is the characteristic time for surfactant adsorption
and can be conservatively estimated as the SDS micelle
decomposition rate (a 00001 s) (Refs. 17, 18) and ap is
the SDS head-group effective area on the nanotube sur-
face (ap 0044 nm2 5. The parameter L is the average
length of nanotubes in our sample. Through cryo-TEM
imaging of processed solutions, this length is typically
greater than 1 m. Hence z 1 s. The value of kz =Keq
RESEARCH ARTICLE
model for the dispersion process and extracting the char-
acteristic rate of nanotube isolation.
Acknowledgments: Financial support was provided by
the National Science Foundation (NSF) Focused Research
Group on Fullerene Nanotube Chemistry (DMR-
0073046), the NSF Center for Biological and Environ-
mental Nanotechnology (EEC-0118007), and the Robert
A. Welch Foundation (C-0689). Support from NASA
(NCC9-77) for the development of the HiPco method is
also gratefully acknowledged.
References and Notes
Xu, Y. H. Lee, S. G. Kim, A. G. Rinzler, D. T. Colbert, G. E.
Scuseria, D. Tomanek, J. E. Fischer, and R. E. Smalley, Science
273, 483 (1996).
2. M. J. OConnell, P. Boul, L. M. Ericson, C. Huffman, Y. H. Wang,
E. Haroz, C. Kuper, J. Tour, K. D. Ausman, and R. E. Smalley,
Chem. Phys. Lett. 342, 265 (2001).
3. A. Koshio, M. Yudasaka, M. Zhang, and S. Iijima, NanoLetters 1,
361 (2001).
4. P. X. Hou, C. Liu, Y. Tong, S. T. Xu, M. Liu, and H. M. Cheng,
J. Mater. Res. 16, 2526 (2001).
5. P. X. Hou, S. Bai, Q. H. Yang, C. Liu, and H. M. Cheng, Carbon
40, 81 (2002).

85
 Strano et al./Surfactant Adsorption in Ultrasonication of Nanotubes J. Nanosci. Nanotech. 2003, 3, 8186

6. N. Nakashima, Y. Tomonari, and H. Murakami, Chem. Lett. 638
(2002).
7. M. J. OConnell, S. M. Bachilo, C. B. Huffman, V. C. Moore,
M. S. Strano, E. H. Haroz, K. L. Rialon, P. J. Boul, W. H. Noon,
C. Kittrell, J. P. Ma, R. H. Hauge, R. B. Weisman, and R. E.
Smalley, Science 297, 593 (2002).
8. R. E. N.-R. Bandyopadhyaya and O. R. Y.-R. Regev, NanoLetters
2, 25 (2002).
9. R. L. McCreery, Photometric Standards for Raman Spectroscopy,
Wiley, Chichester (2002).
10. T. C. Strekas, D. H. Adams, A. Packer, and T. G. Spiro, Appl.
Spectrosc. 28, 324 (1974).
11. M. J. Bronikowski, P. A. Willis, D. T. Colbert, K. A. Smith, and
R. E. Smalley, J. Vac. Sci. Technol. (2001).
12. S. M. Bachilo, M. S. Strano, C. Kittrell, R. H. Hauge, R. E. Smalley,
and R. B. Weisman, Science 298, 2361 (2002).
13. M. S. Dresselhaus, G. Dresselhaus, and P. C. Eklund, Science of
Fullerenes and Carbon Nanotubes, Academic Press, San Diego
(1996).
14. S. Reich, C. Thomsen, and P. Ordejon, Phys. Rev. B 65, 155411(2002).
15. J. Djuve, L. M. Grant, J. Sjblom, T. P. Goloub, and R. J. Pugh,
Langmuir 18, 2673 (2002).
16. L. Q. Amaral, M. E. M. Helene, D. R. Bittencourt, and R. Itri,
J. Phys. Chem. 91, 5949 (1987).
17. E. Lessner, M. Teubner, and M. Kahlwet, J. Phys. Chem. 85, 1529
(1981).
18. E. Lessner, M. Teubner, and M. Kahlwelt, J. Phys. Chem. 85, 3167
(1981).

Received: 11 October 2002. Revised/Accepted: 31 January 2003.

Delivered by Publishing Technology to: Imperial College London

IP: 117.253.100.176 On: Wed, 10 Feb 2016 20:17:05
Copyright: American Scientific Publishers
RESEARCH ARTICLE

86