Chemical Science

Cite this: Chem. Sci., 2012, 3, 541

Published on 04 November 2011. Downloaded by Universidade Federal de Goias on 29/05/2015 01:34:41.

www.rsc.org/chemicalscience

View Article Online / Journal Homepage / Table of Contents for this issue

Dynamic Article Links <

EDGE ARTICLE

Molecular mechanisms of salt effects on carbon nanotube dispersions in an

organic solvent (N-methyl-2-pyrrolidone)
Andrey I. Frolov,a Raz N. Arif,b Martin Kolar,bc Anastasia O. Romanova,a Maxim V. Fedorov*ad
and Aleksey G. Rozhin*b
Received 9th April 2011, Accepted 18th October 2011
DOI: 10.1039/c1sc00232e
We consider the effects of salt (sodium iodide) on pristine carbon nanotube (CNT) dispersions in an
organic solvent, N-methyl-2-pyrrolidone (NMP). We investigate the molecular-scale mechanisms of
ion interactions with the nanotube surface and we show how the microscopic ionsurface interactions
affect the stability of CNT dispersions in NMP. In our study we use a combination of fully atomistic
Molecular Dynamics simulations of sodium and iodide ions at the CNTNMP interface with direct
experiments on the CNT dispersions. In the experiments we analyze the effects of salt on the stability of
the dispersions by photoluminescence (PL) and optical absorption spectroscopy of the samples as well
as by visual inspection. By fully atomistic Molecular Dynamics simulations we investigate the
molecular-scale mechanisms of sodium and iodide ion interactions with the nanotube surface. Our
simulations reveal that both ions are depleted from the CNT surface in the CNTNMP dispersions
mainly due to the two reasons: (1) there is a high energy penalty for the ion partial desolvation at the
CNT surface; (2) NMP molecules form a dense solvation layer at the CNT surface that prevents ions to
come close to the CNT surface. As a result, an increase of the salt concentration increases the osmotic
stress in the CNTNMP system and, thus, decreases the stability of the CNT dispersions in NMP.
Direct experiments confirm the simulation results: addition of NaI salt into the NMP dispersions of
pristine CNTs leads to precipitation of CNTs (bundle formation) even at very small salt concentration
(
103 mol L1). In line with the simulation predictions, the effect increases with the increase of the salt
concentration. Overall, our results show that dissolved salt ions have strong effects on the stability of
CNT dispersions. Therefore, it is possible to stimulate the bundle formation in the CNTNMP
dispersions and regulate the overall concentration of nanotubes in the dispersions by changing the NaI
concentration in the solvent.

Introduction
Liquid dispersions of carbon nanomaterials (CNM), and in
particular, CNTs, is a subject of intensive research these days due
to their wide areas of applications.15 However, the inert nature
of the CNT surface makes them extremely solvophobic to the
a
Max Planck Institute for Mathematics in the Sciences, D 04103 Leipzig,
Germany. E-mail: fedorov@mis.mpg.de
b
School of Engineering & Applied Science, Aston University, Aston
Triangle, Birmingham, UK B4 7ET. E-mail: a.rozhin@aston.ac.uk
c
Faculty of Mechanical Engineering, Brno University of Technology, 616
69 Brno, Czech Republic Address
d
Department of Physics, Scottish Universities Physics Alliance (SUPA),
University of Strathclyde, John Anderson Building, 107 Rottenrow East,
Glasgow, UK G4 0NG. E-mail: maxim.fedorov@strath.ac.uk
Electronic supplementary information (ESI) available: Simulation
details, details on the calculation of the preferential interaction
coefficients, comparison of water and NMP radial density profiles at
the CNT surface, details on the experimental results, results of the
simulations with the 0.01 M NaI concentration and examples of the
simulation input files. See DOI: 10.1039/c1sc00232e

This journal is The Royal Society of Chemistry 2012

most of commonly used solvents.15 That poses a great challenge

for chemists and nanoscientists working on CNT applications.
As a consequence, there have been spent many efforts to find
efficient ways of dispersing CNTs in liquid phase either through
chemical modification of their surface13 or through adding
dispersing agents (e.g. surfactants) into solution to stabilise the
CNT dispersions.58
Although water as a solvent is unavoidable in many biorelated applications of CNTs,6,9,10 this is actually not a good
solvent for making stable CNT dispersions.11,12 Indeed, it is
practically impossible to disperse any noticeable amount of
pristine CNTs in bulk water, where in the absence of dispersants,
pristine carbon materials instantly precipitate.5,11 However, it has
been recently shown that several organic solvents, such as e.g.
N-methyl-2-pyrrolidone (NMP), are much more efficient for
dispersing CNTs and other CNMs than water.1114 In several
studies it has been demonstrated that it is possible to disperse
considerable amounts of pristine CNT and graphene materials in
bulk NMP (i.e. without adding any dispersing agents).14,15 That
Chem. Sci., 2012, 3, 541548 | 541

Published on 04 November 2011. Downloaded by Universidade Federal de Goias on 29/05/2015 01:34:41.

View Article Online

explains why these days non-aqueous (organic) solvents attract

considerable attention of experimental nanoscientists as well as
theoreticians.12,14,16,17
Recently, the mechanisms of interactions between CNM
species in organic solvents were investigated by molecular
dynamics simulations. In the work of MacKernan and Blau17 it
was found that a single layer of NMP raises a big barrier between
interacting carbon nanotubes in CNTNMP dispersions, preventing their aggregation into bundles. Similar effects were
observed for graphene sheets by Shih et al.14 They showed that
the origin of the barrier is the strong van der Waals interactions
of the confined NMP molecules with the surface of the graphene
sheets. In contrast, as the results of ref. 14 show, there is almost
no such barrier in aqueous solution, because of the weak water
molecule interactions with the inert surface of graphene.14 We
note, however, that the discussed works were mainly focused on
the mechanisms of CNM aggregation, rather than on molecular
effects of solvent environment associated with dispersion mechanisms of a single carbon nanotube or graphene sheet in NMP.
Still, there is some lack of information about the molecular-scale
effects at the CNMorganic solvent interface. The role of other
dispersion components (ions, cosolvents, cosolutes) is also
not clear.
The effects of ions may be of particular interest here because,
as is known from biochemical and colloidal sciences, ions may
have profound effects on the stability of biomolecular1820 and
colloidal dispersions,2124 interacting in different ways with
biomolecules and dispersed particles.18,2528 Recently, ion effects
have attracted attention of scientists working on carbon nanotubes. For instance, there have been shown that different ions
can cause various effects on the PL spectra of single-walled
carbon nanotubes (SWNT) in aqueous dispersions stabilized by
surfactants.2933 However, we are not aware of a detailed study of
ion effects on CNTs in their organic solvent dispersions.
We recently experimentally observed that small amounts of
NaI salt (
0.1 mM) lead to interesting effects on bundle
formation in CNTNMP dispersions with a possibility to control
the bundle size. CNT bundle engineering is an essential element
for realization of CNT-based ultrafast photonics devices,
because the absorption recovery time in CNT bundles is typically
much faster (
fs time scale) than in isolated CNT.34 This will
define the operation speed of the CNT based optical switching
devices in lasers and telecom applications.34 However, the
fundamental issue is that the bundle size has to be significantly
smaller than the working wavelength in photonic devices to
minimize scattering losses in the system.34 The preliminary results
on CNT bundle engineering were reported in a brief communication.35 However, neither the molecular mechanisms of these
effects nor the effects of salts of higher concentrations were
explored in ref. 35.
In this work we study the salt (NaI) effects on CNTNMP
dispersions by a combination of fully atomistic molecular
simulations with photoluminescence (PL) spectroscopy and
optical absorption spectroscopy experiments. We have chosen
NMP solvent because this is one of the most promising solvents
for making dispersions of pristine CNMs.13,14 Indeed, it is
possible to obtain at least 0.02 mg mL1 concentration of CNTs
in the bulk NMP solution without any additional dispersing
agents and the dispersion may stay stable for weeks.15 To study
542 | Chem. Sci., 2012, 3, 541548

the ion effects we use sodium iodide because in contrast to

sodium chloride and many other inorganic salts, NaI is soluble in
NMP at least up to 0.2 M.36,37
We note that an addition of surfactants (e.g. polyvinylpyrrolidone, Triton X-100 etc.) can increase the concentration of the CNTs in the dispersion much further than 0.02 mg
mL1.38 However, we focus on the ion effects in bulk NMPCNT
dispersions to avoid any interference of ion effects on CNTs with
possible ion effects on the surfactant molecules.
The paper is organised as follows. In the Models and methods
section we describe our simulations and experimental methods.
The Results and discussion section contains two parts, (i)
Molecular Modelling results where we describe our findings on
molecular mechanisms of ion and NMP interactions with the
CNT surface; (ii) Experimental results where we show that the
proposed mechanisms of ion interactions with the CNT surface
correlate with experimental observations of changes in CNT PL
and optical absorption spectra upon addition of different
concentration of salts. In the Conclusions section we summarize
our results and discuss possible avenues for future research.

Models and methods

Molecular modelling methods
We performed molecular dynamics (MD) simulations of a singlewall CNT with (8,6) and (6,5) chiralities dissolved in 0.15 M NaI
NMP solution to reveal the basic molecular mechanisms of ion
interactions with the carbon nanotube surfaces. Salt concentrations higher than in the experiments were used to collect good
simulation statistics at reasonable computational cost (we were
able to put many ion pairs into a relatively small simulation box).
However, to prove that the trends obtained for higher concentrations can be extrapolated to lower salt concentrations, we also
investigated a CNT with (8,6) chirality dissolved in 0.01 M NaI
solution in NMP. In Section 1.3 and ESI we show that the main
trends and conclusions remain the same even at this concentration. In simulations we could not explore concentrations lower
than 10 mM, because this would require much larger simulation
boxes that would lead to impractically large computational
expenses. To reveal ion distribution at the CNT(8,6) surface at
0.01 M NaI concentration we performed a series of 304 small
simulations, which are described in detail in Part 5 of ESI.
Below we present only the computational details for the systems
with CNT(8,6) and CNT(6,5) dissolved in 0.15 M NaI solutions.
For Fig. 2 and 3 we present the results of the simulations performed for (8,6) chirality and 0.15 M solution, because the use of
higher salt concentrations allows one to obtain smoother ion
density profiles at the CNT surfaces and to resolve the density
profiles for larger separations from the CNT surface. We used
the Gromacs 4.5 software.39 Segments of CNTs were placed in
rectangular boxes (7.50  7.50  5.19 nm3) and (7.25  7.25 
4.07 nm3) in the case of (8,6) and (6,5) CNT chiralities, correspondingly. The CNTs were oriented along the Z axis. Then,
1600 (1200) NMP molecules, 27 (19) Na+ ions and 27 (19) I ions
were placed inside the box with the help of Packmol program in
the case of CNT(8,6) (CNT(6,5)).40 The initial configuration was
first optimized by energy minimization,39 and then the density of
the system was equilibrated during a 0.2 ns simulation in the
This journal is The Royal Society of Chemistry 2012

Published on 04 November 2011. Downloaded by Universidade Federal de Goias on 29/05/2015 01:34:41.

View Article Online

NPT-ensemble. Then we performed simulations for 60 and 50 ns

in the NVT-ensemble at 300 K for CNT(8,6) and CNT(6,5)
correspondingly. Next, to enhance statistics and to estimate the
errors in the calculated preferential interaction coefficients and
the free energy changes, we performed 10 replica simulations for
each CNT starting from different initial coordinates. The initial
configurations for the replica runs were collected by taking
coordinates of the system each 3 ns and 5 ns from the initial
simulations starting from 30 and 0 ns for the systems with CNT
(8,6) and CNT(6,5), respectively. Each replica was first simulated
for 0.2 ns at elevated temperature (450 K) and then annealed to
300 K during 0.1 ns of simulation time in the NVT ensemble. The
production simulation times for each replica were 15.9 and 20.7
ns for systems with CNT(8,6) and CNT(6,5), respectively.
Coordinates of the system were sampled each 0.3 ps (for the
initial runs) and 0.4 ps (for replica simulations) for further
analysis. We employed the fully atomistic OPLS-AA force
field,4143 which has been thoroughly tested for NMP and similar
organic solvents in refs. 14, 44 and 45. For sodium iodide, we
used the recent set of ion parameters developed consistently with
the general framework of the OPLS force field.43 We used the
Particle-Mesh Ewald method to evaluate the electrostatic interactions.46 To analyze molecular mechanisms of ion solvation in
NMP, we performed an additional simulation of NaI solution in
bulk NMP (without the CNT) using similar simulation setup to
the one described above. Further details of the molecular
modeling are given in ESI.
Experimental methods
The CNTs (SWeNt CG100, Lot # 000-0012) were dispersed in
neat N-methyl-2-pyrrolidone (spectrophotometric grade, $99%,
Sigma-Aldrich) via ultrasonication (20 kHz, 200 W, 1 h) with
a Nanoruptor sonicator (Diagenode). The dispersion was subjected to ultracentrifugation (2.5 h; 17  C; 47 000 rpm) using
a MLS 50 swinging bucket rotor in the Optima Max XP ultracentrifuge (Beckman Coulter). The resulting CNT dispersion was
divided into five samples which we number as samples 0, 1, 2, 3
and 4 correspondingly. We used sample 0, as a control sample
(without salt). We put four different salt concentrations (0.1, 0.5,
1 and 10 mM) into the other samples. We prepared the final
concentrations of the salt in the samples as follows. A predefined
amount of concentrated salt (NaI; ACS reagent, $99.5; SigmaAldrich) solution in NMP was added to each sample with
a calibrated micropipette. All samples were exposed to shaking at
300 rpm for 10 min. Due to the small volume of the added drops
of the concentrated salt solution we did not take the dilution into
account. The final salt concentrations were 0.1, 0.5, 1 and 10 mM
for samples 1, 2, 3, and 4.
We investigated the changes in the CNT dispersions in
response to addition of different amounts of salt by using photoluminescence (PL) spectroscopy and optical absorption spectroscopy as well as by visual inspection of the samples
(see ESIfor further details).
We use PL spectroscopy as a tool for monitoring the presence
of isolated SWNTs and their small bundles in the dispersions.34,47
We assume that the concentration of nanotubes within the
sample correlates with the intensity of the PL bands and optical
absorbance for the corresponding nanotube chiralities. That
This journal is The Royal Society of Chemistry 2012

allows us to make quantitative analysis of the changes in the

dispersions upon additions of salt.

Results and discussion

1.

Molecular modelling results

1.1 Ion solvation in the bulk NaI NMP solution. In our

simulations we observe that ions dissolved in NMP have distinct
solvation shells (see Fig. 1). That was expectable, because NMP
is a very polar solvent (the dipole moment is about 4.1 Debye)
and, therefore, the NMP molecules strongly interact with the
dissolved ions due to the electrostatic chargedipole interactions.
Both of the ionNMP radial distribution functions show
a peak (at r 0.42 and r 0.55 nm for Na+ and I, respectively)
followed by a hollow (at r 0.50 and r 0.70 nm for Na+ and I,
respectively), indicating formation of the first solvation shells
around the ions.48 The boundaries of the ion solvation shell were
estimated as the region with non-zero ionNMP radial distribution function ending at the first distinct minimum on the
corresponding function. The structures of the solvation shells of
the ions are different: the negatively charged NMP oxygen atoms
are strongly attracted to the positively charged sodium ions (see
the snapshot illustrating typical configuration of NMP molecules
around this ion in Fig. 1A); from another side, the oxygen atoms
are placed outwards from the negatively charged iodide ions (see
the molecular snapshot on Fig. 1B). In general, similar to the
mechanisms of ion solvation in water20,49,50 there is a strong
asymmetry in sodium and iodide solvation in NMP. As illustrated by the high peak on Na+NMP g(r) (Fig. 1), the sodium
ion solvation shell is very dense because the ion is relatively small
and, consequently, has a large surface charge density.48,51
Therefore, the Na+ ions strongly coordinate polar solvent
molecules around them, that resulted in the height of the first
peak on Na+NMP radial distribution function to be about 8.0.
On the other hand, the iodide ion solvation shell is much more
diffuse because of the larger size of the I ion and, consequently,
its smaller surface charge density than that of the Na+ ion.48,51 As
a result, the height of the first peak on INMP radial distribution function is much smaller than that on the Na+NMP g(r) (it
is slightly above 3.0). The position of the first maximum on the
INMP g(r) is also shifted compared to the Na+NMP g(r)
(from 0.42 to 0.55 nm). This is because of two main reasons: (i)
I (r 0.220 nm52) has a larger radius compared to Na+
(r 0.102 nm52); (ii) as discussed above, the INMP interactions are weaker than the Na+-NMP interactions.
1.2. Ions behavior at the CNT surface in CNTNMP dispersion. Results of our MD simulations indicate two major effects
which take place during ion interactions with the CNT surface:
(1) First, ions have to become partially desolvated to make
direct contacts with the CNT surface (see Fig. 2). The partial
desolvation of ions at the CNT surface happens because of the
steric restraints caused by the surface. The MD simulations show
that upon approaching the CNT surface the sodium ions have to
release one NMP molecule and the iodide ions have to release two
NMP molecules from their solvation shells (there is a significant
decrease of the solvation number of ions at the CNT surface see,
Fig. 2A). The partial desolvation of ions also means that the
Chem. Sci., 2012, 3, 541548 | 543

Published on 04 November 2011. Downloaded by Universidade Federal de Goias on 29/05/2015 01:34:41.

View Article Online

Fig. 1 IonNMP (center of mass) radial distribution functions, g(r), in NaINMP solution combined with corresponding simulation snapshots: (A)
sodium ion (shown as a blue sphere), (B) iodide ion (shown as a magenta sphere). The highly noticeable peaks on the g(r) functions indicate distinct solvation
shells around the ions. On the simulation snapshots the NMP molecules in the first solvation shell around the ions are represented by thick sticks. The oxygen
atoms in these molecules are colored by red, nitrogens by blue, carbons by cyan and hydrogens by white. Other molecules are shown by thin cyan lines.

strong ionNMP dipole interactions (discussed above) are

substituted by the much weaker van der Waals interactions of
ions with the non-polar CNT surface. That leads to large energy
costs for desolvation of ions (so-called desolvation penalty53)
and, overall, makes the direct contacts of ions with the CNT
surface to be energetically unfavorable.
(2) Secondly, to approach the CNT surface, ions have to
squeeze through a very dense layer of NMP molecules in the first

Fig. 2 (A) Solvation number of ions (the number of NMP molecules in

the first solvation shell of the ions) as a function of distance from the
CNT with (8,6) chirality. (B) Schematic representation of the partial
desolvation of the ions upon the direct contact with the CNT surface.
Because the iodide ion is larger than the sodium ion it consequently has
a larger solvation number.

544 | Chem. Sci., 2012, 3, 541548

CNT solvation shell (see Fig. 3). Contrary to water, NMP is

known to interact strongly with the surface of carbon nanomaterials.14,54 The dense CNT solvation shell in NMP corresponds to a broad region on the CNTNMP radial density

Fig. 3 Radial density profiles of ions and NMP molecules (center of

mass) at the CNT with (8,6) chirality. There is a distinct salt depletion
area at the CNT surface (marked by the beige color): a region with
enhanced concentration of NMP molecules but with much lower
concentration of ions than in the bulk. The simulation snapshot represents the NMP molecules in the first solvation shell of CNT(8,6) (shown
by thick sticks) preferentially having two different orientations, flat and
perpendicular to the surface. These two preferential orientations correspond to the two distinct peaks on the NMP radial density profile. The
NMP molecules outside the first solvation shell are represented as thin
lines. The colored circles schematically show ions (Na+ by blue and I by
magenta) in the solution. A similar depletion ion area is formed at the
CNT with (6,5) chirality.

This journal is The Royal Society of Chemistry 2012

Published on 04 November 2011. Downloaded by Universidade Federal de Goias on 29/05/2015 01:34:41.

View Article Online

profile with two distinct high peaks in the vicinity of the CNT
surface (see Fig. 3). This NMP enrichment area is followed by
a deep hollow at about r 1.1 nm (see Fig. 3). Thus, during the
formation of ionCNT direct contacts, not only the ions have to
be partially desolvated (as shown on Fig. 2), but also the CNT
has to release some NMP molecules from its solvation shell. This
further increases the energetic barrier for the formation of ion
CNT direct contacts.
To conclude, the energetic penalties in the processes of partial
desolvation of ions and partial desolvation of CNT can not be
compensated by the weak van der Waals interactions of the ions
with the non-polar CNT surface. As a result, ions prefer to stay in
the bulk of the NMP solution rather than close to the CNT
surface. This leads to a formation of ion depletion area around the
non-polar CNT surface, where the concentration of ions is
significantly lower than in the bulk solution (Fig. 3). Geometrically, the depletion area roughly corresponds to the CNT solvation shell. The depletion area thickness is about 0.35 nm, which is
about the width of one NMP molecule. Ions can attach to the
CNT solvation shell beyond the depletion area, where there is even
enhanced concentration of iodides nearby the boundary of the
CNT solvation shell (Fig. 3, area around r
1.2 nm). The sodium
ions can occasionally enter the CNT solvation shell (there is small
but non-zero peak on the Na+CNT radial density profile at
r
1.0 nm, see Fig. 3), while in the case of the iodide ions it
happens very rarely (ICNT radial density profile is almost zero
at shorter distances to the CNT than r
1.1 nm Fig. 3).
To understand the main mechanisms of changes in Na+ and I
solvation upon approaching the CNT surfaces we calculated the
solvation number (the average number of NMP molecules in the
first solvation shell of the ions) as a function of distance between
the corresponding ion and the CNT surface (Fig. 2). As illustrated by Fig. 2, the difference in the solvation strength of Na+
and I results in different mechanisms of their penetration into
the CNT solvation shell. Na+ is strongly solvated, and therefore,
it keeps its solvation number unchanged until it comes very close
to the CNT surface (see Fig. 2A). The Na+ ions become
(abruptly) partially desolvated only at very close vicinity to the
CNT surface (Fig. 2A, r < 1.0 nm) when this is absolutely
unavoidable because of the steric constraints posed by the CNT
surface. On the other hand, the iodide ion is much more weakly
solvated in NMP compared to Na+. As a result, the I ions
gradually loose NMP molecules from their solvation shell
(starting from r
1.5 nm) diving into the CNT solvation shell
(see Fig. 2A).
Here we also would like to briefly compare the proposed
mechanisms of ion interactions with the CNT surface in NMP
solution to the previously reported results on molecular simulations of ionCNT interactions in aqueous solutions54 as well as ion
interactions with other non-polar surfaces.55 It is shown in ref. 54
that the probability of a direct ion contact with the CNT surface
strongly correlates with the strength of ion solvation. The less
solvated the ion is, the higher its concentration at the CNT
interface.54 For instance, the simulations in ref. 54 show that the
concentration of iodide at the CNT surface in aqueous solution is
about 40% of its bulk concentration, while the interface concentration of sodium ions is practically zero. This is apparently not
the case for NMP solutions, despite the fact that the strength of the
ion solvation in NMP as well as the strength of ion solvation in
This journal is The Royal Society of Chemistry 2012

aqueous solutions decreases from sodium to iodide. There are

almost negligible concentrations of both Na+ and I ions in the
ionCNT interface shell (see Fig. 3). We attribute this to the fact
that the CNT surface can be much more easily dehydrated in water
than desolvated in NMP (see also part 2 of ESI).
1.3. Thermodynamics of ion depletion at the CNT surface. To
quantify the effect of the salt depletion area on the solvation
thermodynamics of CNTs in CNTNMP dispersions, we employ
the GibbsDuhem relation (as described in ref. 55 and 56) and an
approach based on the KirkwoodBuff theory of solutions5760 as
discussed in ref. 61. In general, within the framework of the
GibbsDuhem theory of solution, the change in the chemical
potential of a solute molecule (S) dissolved in a solvent (V), upon
addition of a cosolvent (X) is written as:55
DmS z GSXDmX,

(1)

where DmS is the change in the chemical potential of solute, DmX

is change of chemical potential of the co-solvent, GSX is the
soluteco-solvent preferential interaction coefficient (deficit or
excess of the number of co-solvent molecules around a solute
molecule, compared to the same volume of the bulk solution).
We note that this is an approximate relationship (that is why we
use here the z sign) because we approximated the differentials
by the finite differences. Here we consider the salt as a co-solvent.
In our case, the soluteco-solvent preferential interaction
coefficient can be calculated via the radial density profiles of the
corresponding species in the solution (see ESI):
N


GSX

r0X
0


rSX r rSV r
2pr$dr;

r0X
r0V

(2)

where {r0X, r0V} are the number density of particles X and V

correspondingly in the bulk solution, {rSX(r), rSV(r)} are the
densities of particles X and V as a function of the distance r from
the axis of cylindrical symmetry of CNT (denoted by S).
Following the works (ref. 55 and 61) we applied the KirkwoodBuff theory to estimate the preferential interaction coefficient of NaI at the CNT surface. Kusalik and Patey proved (see
ref. 61 and 62) that within the KirkwoodBuff theory one should
consider dissolved cations and anions not as independent
components, but rather as one component in the solution. Thus,
in this study we considered the dissolved Na+ and I ions as a cosolvent (the component X) to NMP solvent (the component V)
(see additional details in ESI). The calculated CNTion preferential interaction coefficients (GSX) are presented in Table 1.
For all the systems GSX is negative, indicating that there is a lack
Table 1 The preferential interaction coefficient (eqn (2)) normalized to
the surface area of the CNTs and the estimated increase of the CNT free
energy in the CNTNMP dispersions upon salt addition (DmS, eqn (1))
for the three considered systems with different CNT chiralities and two
salt concentrations. See Table S2 (ESI) for more details
System

GSX (ions nm2)

DmS (kBT nm2)

CNT(8,6), C(NaI) 0.15 M

CNT(6,5), C(NaI) 0.15 M
CNT(8,6), C(NaI) 0.01 M

0.12  0.03
0.13  0.03
0.007  0.003

0.23  0.06
0.26  0.06
0.013  0.005

Chem. Sci., 2012, 3, 541548 | 545

View Article Online

of ions around the CNT compared to bulk NaI NMP solution

because of the formation of the salt depletion area (as shown on
Fig. 3). The value of DmX in eqn (1) was estimated as:

Published on 04 November 2011. Downloaded by Universidade Federal de Goias on 29/05/2015 01:34:41.

DmX z 

xV
k T lnxV
xX B

(3)

based on the GibbsDuhem relation between changes in the cosolvent and solvent chemical potentials and some approximations (where {xX, xV} are the molar fraction of particles V and X,
kB is the Boltzmann constant, T is the temperature)55 (see
ESIfor further details).
The calculations show that the addition of the NaI salt into
CNTNMP dispersion increases the CNT surface free energy
(see Table 1). The positive values of DmS indicate an increase of
the CNT solvophobicity upon the salt addition. The values of
DmS (kBT nm2) of the CNT surface are very similar for the CNTs
with both (8,6) and (6,5) chiralities at 0.15 M NaI concentration.
This indicates that the salt effect is proportional to the CNT
surface area, and does not depend considerably on a specific
order of carbon atoms in the CNT (CNT chirality). Increase of
the salt concentration from 0.01 to 0.15 M significantly increases
DmS, and thus decreases the thermodynamic stability of the
CNTNMP dispersions (see Table 1). This can be rationalized in
the following way. In the supporting information (ESI) we
showed that the particle radial density profiles do not change
much with the increase of the salt concentration from 0.01 to
0.15 M. Thus, we may assume that the integral in eqn (2) is
a constant for the range of low salt concentrations considered in
the study (see Table S2, ESI). Consequently (see eqn (2)), the
preferential interaction coefficient is proportional to the salt
concentration: GSX z CxX, where the C is a negative (C < 0)
constant (see above), and xX is the mole fraction of salt in NMP.
For small salt concentrations used in our experiments (110 mM)
we also may assume ln(1  xX) z xX and (1  xX) z 1. The
combination of the above mentioned formulas gives the
following relation of DmS and the salt concentration xX:
DmS z CkBT(xX)

(4)

Minding the negative constant C, we may conclude that with

the increase of the salt concentration, DmS increases (becomes

more positive), and, therefore, the CNT becomes more

solvophobic.
The results of our simulations and simple analysis of thermodynamic relations reveal that an increase of the salt concentration increases the CNT solvophobicity in CNTNMP
dispersions. This originates from the formation of the ion
depletion area around CNTs, and that in turn results in an
increase of the free energy of CNTs in NMP dispersions. These
thermodynamic changes should make the CNTNMP dispersions less stable with the increase of salt concentration. That
means that CNTs can more easily interact with each other
forming CNT aggregates and bundles, decreasing the total CNT
surface exposed to the salt NMP solutions. We refer to this effect
as salting out of CNTs from CNTNMP dispersions.
We think that the mechanism of salting out of CNTs from
NMP dispersions resulting from the analysis of the simulation
data in the current work has a similar origin to the well known
salting out effects in aqueous solutions, which are widely used
in biotechnology for coagulation and separation of biomolecules
from aqueous solutions as well as for protein
crystallization.18,63,64

2.

Experimental results

To verify the general trends predicted by the simulations, we

performed a series of experiments on the salting out of CNTs
from CNTNMP dispersions. We added different quantities of
inorganic salt (NaI) to CNT dispersions in NMP. After 15 min
upon the salt addition we observed formation of CNT bundles
(see Fig. 4B, S4, ESI). The quantity of the bundles increases
with the increase of the salt concentration. The formed bundles
stay insoluble and can be removed from the dispersion.
The PL spectra of the control CNTNMP dispersion is presented in Fig. 4A. Detailed PL spectra of all samples are presented in ESI (Table S1). A significant drop in the PL intensity
upon the salt addition indicates a decrease of the CNTs
concentration in dispersions34,38 (Fig. 4C). At salt concentrations
of 0.1 and 0.5 mM a certain concentration of CNTs remains in
the dispersion. However, concentrations higher than 1 mM
apparently lead to complete precipitation of all CNTs in the
dispersion. We note that in the case of the control no salt

Fig. 4 (A) PL map of the control sample; (B) the control sample and samples with salt (NaI) addition aged for 5 h. 0 control (no salt), 1 0.1 mM;
2 0.5 mM; 3 1 mM; 4 10 mM of NaI; (C) PL spectra of the control sample and samples with salt (NaI) addition at excitation wavelength 570 nm.

546 | Chem. Sci., 2012, 3, 541548

This journal is The Royal Society of Chemistry 2012

Published on 04 November 2011. Downloaded by Universidade Federal de Goias on 29/05/2015 01:34:41.

View Article Online

Fig. 5 Absorption spectra of the CNTNMP dispersions with different

salt concentrations after an additional centrifugation. The centrifugation
removes big CNT bundles from dispersion. For NaI concentrations
above 0.5 mM the spectra show practical absence of CNTs in the
dispersions after the additional centrifugation that corresponds to
complete salting out of CNTs from dispersions (note: the high increase
of the intensity of the 10 mM spectra in the area of 400700 nm is caused
by the high concentration of NaI, not by CNTs).

sample (0) the additional centrifugation almost does not affect

the PL intensity (see ESI). This is a clear illustration that the
decrease in the PL intensity for other samples is caused by the
addition of salt, rather than aging of the dispersions.
The experiments show that the concentration of the CNTs in
the dispersion can be regulated by the quantity of added salts.
Indeed, there is a monotonic drop in the PL intensity with the
increase of the salt concentration (Fig. 4C).
Absorption spectroscopy allows determining the overall
concentration of CNT in the dispersion. We used this method to
support our PL data. Absorption spectroscopy measurements
show the same general trend: the intensity drops with the increase
of the salt concentration (see Fig. 5 and S5, ESI).
The experimental data clearly show that it is possible to salt
out carbon nanotubes from their NMP dispersions. The
salting out effect is observed already at very small concentrations of NaI salt (mM range). The degree of CNT aggregation
monotonically increases with the increase of salt concentration,
indicating an enhancement of the CNT solvophobicity in the
CNTNMP dispersions. Thus, the conclusions coming from the
computer simulation and general physico-chemical principles of
salt effects discussed in the paper are confirmed by direct
experiments on the CNT precipitation by salts.

Conclusions
The analysis of the simulation and experimental data on sodium
iodide salt effects on CNT dispersions in NMP results in the
following conclusions:
(1) To make the direct contacts with the CNT surface in CNT
NMP dispersions, ions have to become partially desolvated. The
partial desolvation is energetically unfavorable because of the
strong ionsolvent interactions in highly polar NMP solution.
(2) The CNT itself has a very distinct and dense solvation shell
in the NMP solution, and that adds an additional barrier for ions
to come close to the CNT surface.
This journal is The Royal Society of Chemistry 2012

(3) As a result of the unfavorable interactions of ions with the

CNT surface, a salt depletion area is formed around the CNT,
where the concentration of ions is much less than in the bulk
solution. The width of the salt depletion area corresponds to the
width of the first solvation shell of CNT in NMP.
(4) The preferential depletion of ions from the CNTNMP
interface results in increase of solvophobic CNTNMP interactions with the increase of salt concentration. As a consequence,
the CNTNMP dispersions become thermodynamically less
stable at higher concentrations of salts.
(5) The increase of the CNT solvophobicity upon the salt
addition was confirmed by direct experiments. We added
different amount of NaI salt to prepared CNT dispersions in neat
NMP and observed precipitation of the CNTs. The degree of the
precipitation increased with the increase of salt concentration,
which is in line with the results of the theoretical and simulation
results of the study.
We believe that the revealed mechanisms of the ion interaction
with the CNT surface in non-aqueous (NMP) dispersions should
bring new insights on salt effects on nanotubes in non-aqueous
solutions.35 We show that, to make stable CNT dispersions in the
NMP organic solvent one has to keep the solvent free of even
small salt impurities/additives. Also, we show that the salting
out effect leads to an efficient safe and inexpensive method of
regulating the CNT concentration in non-aqueous dispersions.
In addition, our findings should be useful for environmental
applications. Indeed, there have been already raised serious
concerns about safety issues in nanoengineering. Our findings
provide an inexpensive and easy route to completely precipitate
carbon nanoparticles from their dispersions. The macro-bundles
formed upon precipitation can then be easily utilized. We hope
that the results of the paper can be useful for rational development of new methods for processing of liquid dispersions of
carbon nanomaterials based on the salting out effect in nonaqueous media.

Acknowledgements
We acknowledge Volodymyr Sergiievskyi and David Palmer for
their critical reading of the manuscript. We acknowledge the
supercomputing support from the John von Neumann-Institut
f
ur Computing (NIC), Juelich Supercomputing Centre (JSC),
Forschungszentrum Juelich GmbH, Germany. Project ID:
HLZ16. The work has been performed under the HPCEUROPA2 project (project number: 228398) with the support of
the European Commission - Capacities Area - Research Infrastructures. The simulation part of this work made use of the
facilities of HECToR, the UKs national high-performance
computing service, which is provided by UoE HPCx Ltd at the
University of Edinburgh, Cray Inc and NAG Ltd, and funded by
the Office of Science and Technology through EPSRCs High
End Computing Programme. A. O. R. is supported by Deutsche
Forschungsgemeinschaft within the Graduate School BuildMoNa. We used VMD software for visualization of simulation
results.65 A. I. F. is very thankful to the financial support of the
International Max Planck Research School Mathematics in the
Sciences (supported by the Klaus Tschira Stiftung, Germany).
We acknowledge the partial financial support from the MarieCurie International Exchange Scheme TelaSens, Research
Chem. Sci., 2012, 3, 541548 | 547

View Article Online

Executive Agency Grant No 269271, Programme: FP7PEOPLE-2010-IRSES.

Published on 04 November 2011. Downloaded by Universidade Federal de Goias on 29/05/2015 01:34:41.

Notes and references

1 D. Tasis, N. Tagmatarchis, A. Bianco and M. Prato, Chem. Rev.,
2006, 106, 11051136.
2 T. J. Park, S. Banerjee, T. Hemraj-Benny and S. S. Wong, J. Mater.
Chem., 2006, 16, 141154.
3 A. R. Boccaccini, J. Cho, J. A. Roether, B. J. Thomas, E. Jane Minay
and M. S. Shaffer, Carbon, 2006, 44, 31493160.
4 D. Mattia and Y. Gogotsi, Microfluid. Nanofluid., 2008, 5, 289305.
5 L. Vaisman, H. D. Wagner and G. Marom, Adv. Colloid Interface
Sci., 2006, 128130, 3746.
6 S. C. Lin and D. Blankschtein, J. Phys. Chem. B, 2010, 114, 15616
15625.
7 S. D. Bergin, V. Nicolosi, H. Cathcart, M. Lotya, D. Rickard,
Z. Y. Sun, W. J. Blau and J. N. Coleman, J. Phys. Chem. C, 2008,
112, 972977.
8 N. R. Tummala and A. Striolo, ACS Nano, 2009, 3, 595602.
9 A. Bianco, K. Kostarelos, C. D. Partidos and M. Prato, Chem.
Commun., 2005, 571577.
10 Y. Gogotsi, N. Naguib and J. A. Libera, Chem. Phys. Lett., 2002, 365,
354360.
11 K. D. Ausman, R. Piner, O. Lourie, R. S. Ruoff and M. Korobov, J.
Phys. Chem. B, 2000, 104, 89118915.
12 (a) S. D. Bergin, Z. Sun, P. Streich, J. Hamilton and J. N. Coleman, J.
Phys. Chem. C, 2010, 114, 231237; (b) S. D. Bergin, Z. Sun,
D. Rickard, P. Streich, J. Hamilton and J. N. Coleman, ACS Nano,
2009, 3, 23402350.
13 S. Giordani, S. Bergin, V. Nicolosi, S. Lebedkin, W. J. Blau and
J. N. Coleman, Phys. Status Solidi B, 2006, 243, 30583062.
14 C. Shih, S. Lin, M. S. Strano and D. Blankschtein, J. Am. Chem. Soc.,
2010, 132, 1463814648.
15 S. Giordani, S. D. Bergin, V. Nicolosi, S. Lebedkin, M. M. Kappes,
W. J. Blau and J. N. Coleman, J. Phys. Chem. B, 2006, 110, 15708
15718.
16 J. Wang, D. Fruchtl, Z. Y. Sun, J. N. Coleman and W. J. Blau, J.
Phys. Chem. C, 2010, 114, 61486156.
17 D. Mac Kernan and W. J. Blau, Europhys. Lett., 2008, 83, 66009.
18 K. D. Collins and M. W. Washabaugh, Q. Rev. Biophys., 2009, 18,
323422.
19 K. D. Collins, Methods, 2004, 34, 300311.
20 K. D. Collins, Biophys. Chem., 2006, 119, 271281.
21 D. V. Schur, B. P. Tarasov, S. Y. Zaginaichenko, V. K. Pishuk,
T. N. Veziroglu, Y. M. Shulga, A. G. Dubovoi, N. S. Anikina,
A. P. Pomytkin and A. D. Zolotarenko, Int. J. Hydrogen Energy,
2002, 27, 10631069.

22 R. Hidalgo-Alvarez,
A. Martn, A. Fernandez, D. Bastos,
F. Martnez and F. J. de las Nieves, Adv. Colloid Interface Sci.,
1996, 67, 1118.

23 J. L. Ortega-Vinuesa, A. Martn-Rodrguez and R. Hidalgo-Alvarez,
J. Colloid Interface Sci., 1996, 184, 259267.
24 S. V. Solomatin, T. K. Bronich, A. Eisenberg, V. A. Kabanov and
A. V. Kabanov, Langmuir, 2004, 20, 20662068.
25 M. V. Fedorov, J. M. Goodman and S. Schumm, J. Am. Chem. Soc.,
2009, 131, 1085410856.
26 M. Lund, P. Jungwirth and C. E. Woodward, Phys. Rev. Lett., 2008,
100, 258105.
27 M. V. Fedorov, J. M. Goodman and S. Schumm, Chem. Commun.,
2009, 896898.
28 I. V. Terekhova, A. O. Romanova, R. S. Kumeev and M. V. Fedorov,
J. Phys. Chem. B, 2010, 114, 1260712613.
29 S. Niyogi, S. Boukhalfa, S. B. Chikkannanavar, T. J. McDonald,
M. J. Heben and S. K. Doorn, J. Am. Chem. Soc., 2007, 129, 1898
1899.

548 | Chem. Sci., 2012, 3, 541548

30 S. Niyogi, C. G. Densmore and S. K. Doom, J. Am. Chem. Soc., 2009,

131, 11441153.
31 J. J. Brege, C. Gallaway and A. R. Barron, J. Phys. Chem. C, 2007,
111, 1781217820.
32 J. J. Brege, C. Gallaway and A. R. Barron, J. Phys. Chem. C, 2009,
113, 42704276.
33 S. Ju, J. Doll, I. Sharma and F. Papadimitrakopoulos, Nat.
Nanotechnol., 2008, 3, 356362.
34 T. Hasan, Z. P. Sun, F. Q. Wang, F. Bonaccorso, P. H. Tan,
A. G. Rozhin and A. C. Ferrari, Adv. Mater., 2009, 21, 3874
3899.
35 M. V. Fedorov, R. N. Arif, A. I. Frolov, M. Kolar, A. O. Romanova
and A. G. Rozhin, Phys. Chem. Chem. Phys., 2011, 13, 12399.
36 GafCorporation, M-Pyrol N-methyl-2-pyrrolidone Handbook, GAF
Corpoation, 1972.
37 (a) A. Domard, M. Rinaudo and C. Terrassin, Int. J. Biol. Macromol.,
1986, 8, 105107; (b) W. E. Willy, D. R. Mckean and B. A. Garcia,
Bull. Chem. Soc. Jpn., 1976, 49, 19891995.
38 T. Hasan, V. Scardaci, P. Tan, A. G. Rozhin, W. I. Milne and
A. C. Ferrari, J. Phys. Chem. C, 2007, 111, 1259412602.
39 B. Hess, C. Kutzner, D. van der Spoel and E. Lindahl, J. Chem.
Theory Comput., 2008, 4, 435447.
40 L. Martnez, R. Andrade, E. G. Birgin and J. M. Martnez, J. Comput.
Chem., 2009, 30, 21572164.
41 W. L. Jorgensen, D. S. Maxwell and J. Tirado-Rives, J. Am. Chem.
Soc., 1996, 118, 1122511236.
42 M. L. P. Price, D. Ostrovsky and W. L. Jorgensen, J. Comput. Chem.,
2001, 22, 13401352.
43 K. P. Jensen and W. L. Jorgensen, J. Chem. Theory Comput., 2006, 2,
14991509.
44 S. Aparicio, R. Alcalde, M. J. Davila, B. Garca and J. M. Leal, J.
Phys. Chem. B, 2008, 112, 1136111373.
45 S. Aparicio and R. Alcalde, Phys. Chem. Chem. Phys., 2009, 11,
6455.
46 U. Essmann, L. Perera, M. L. Berkowitz, T. Darden, H. Lee and
L. G. Pedersen, J. Chem. Phys., 1995, 103, 85778593.
47 P. H. Tan, A. G. Rozhin, T. Hasan, P. Hu, V. Scardaci, W. I. Milne
and A. C. Ferrari, Phys. Rev. Lett., 2007, 99, 137402.
48 Y. Marcus, Ion Solvation, John Wiley & Sons, Chichester, UK, 1985.
49 M. V. Fedorov, J. M. Goodman and S. Schumm, Phys. Chem. Chem.
Phys., 2007, 9, 54235435.
50 M. V. Fedorov and A. A. Kornyshev, Mol. Phys., 2007, 105, 116.
51 K. Collins, Biophys. J., 1997, 72, 6576.
52 R. D. Shannon, Acta Crystallogr., Sect. A: Cryst. Phys., Diffr., Theor.
Gen. Crystallogr., 1976, 32, 751767.
53 R. M. Levy, L. Y. Zhang, E. Gallicchio and A. K. Felts, J. Am. Chem.
Soc., 2003, 125, 95239530.
54 A. I. Frolov, A. G. Rozhin and M. V. Fedorov, ChemPhysChem,
2010, 11, 26122616.
55 S. Pal and F. M
uller-Plathe, J. Phys. Chem. B, 2005, 109, 64056415.
56 V. A. Parsegian, R. P. Rand and D. C. Rau, Proc. Natl. Acad. Sci.
U. S. A., 2000, 97, 39873992.
57 J. G. Kirkwood and F. P. Buff, J. Chem. Phys., 1951, 19, 774777.
58 S. Shimizu, W. M. McLaren and N. Matubayasi, J. Chem. Phys.,
2006, 124, 234905.
59 J. M. Schurr, D. P. Rangel and S. R. Aragon, Biophys. J., 2005, 89,
22582276.
60 I. Shulgin and E. Ruckenstein, Biophys. J., 2006, 90, 704707.
61 V. Pierce, M. Kang, M. Aburi, S. Weerasinghe and P. E. Smith, Cell
Biochem. Biophys., 2007, 50, 122.
62 P. G. Kusalik and G. N. Patey, J. Chem. Phys., 1987, 86, 5110.
63 J. Porath, L. Sundberg, N. Fornstedt and I. Olsson, Nature, 1973, 245,
465466.
64 S. A. Miller, D. D. Dykes and H. F. Polesky, Nucleic Acids Res., 1988,
16, 1215.
65 W. Humphrey, A. Dalke and K. Schulten, J. Mol. Graphics, 1996, 14,
3338.

This journal is The Royal Society of Chemistry 2012