pubs.acs.

org/Langmuir
© 2009 American Chemical Society

Measurement of Multicomponent Solubility Parameters for Graphene

Facilitates Solvent Discovery
Yenny Hernandez, Mustafa Lotya, David Rickard, Shane D. Bergin, and Jonathan N. Coleman*
School of Physics and Centre for Research on Adaptive Nanostructures and Nanodevices (CRANN), Trinity
College Dublin, Dublin 2, Ireland

Received August 26, 2009. Revised Manuscript Received October 8, 2009

We have measured the dispersibility of graphene in 40 solvents, with 28 of them previously unreported. We have shown
that good solvents for graphene are characterized by a Hildebrand solubility parameter of δT ∼ 23 MPa1/2 and Hansen
solubility parameters of δD ∼ 18 MPa1/2, δP ∼ 9.3 MPa1/2, and δH ∼ 7.7 MPa1/2. The dispersibility is smaller for solvents
See https://pubs.acs.org/sharingguidelines for options on how to legitimately share published articles.

with Hansen parameters further from these values. We have used transmission electron microscopy (TEM) analysis to show
that the graphene is well exfoliated in all cases. Even in relatively poor solvents, >63% of observed flakes have <5 layers.
Downloaded via UNIV OF GOTHENBURG on August 1, 2018 at 05:09:03 (UTC).

Introduction oxidation of graphite renders it polar, allowing efficient exfoliation

Since the discovery of a method to isolate graphene monolayers
on insulating substrates,1 graphene has become one of the most
exciting of nanomaterials.2-5 It is most commonly prepared as
individual flakes by micromechanical cleavage of graphite. Such
samples are ideal for nanoscale electrical,1 spectroscopic,6-8 or
nanomechanical characterization.9 However, the yield and
throughput associated with this technique are extremely low.
This makes it completely unsuitable for any large scale applica-
tions. For example, graphene may be useful in thin-film form10,11
as a transparent conductor or as a filler in composites.12 For such
applications, the ability to exfoliate graphite in the liquid phase to
give graphene solutions or dispersions will be critical.
To this end, a number of groups have demonstrated the pre-
paration of dispersions of graphene oxide.12-18 Here, large-scale
*To whom correspondence should be addressed. E-mail: colemaj@tcd.ie.
(1) Novoselov, K. S.; Geim, A. K.; Morozov, S. V.; Jiang, D.; Zhang, Y.;
Dubonos, S. V.; Grigorieva, I. V.; Firsov, A. A. Science 2004, 306(5696), 666–669.
(2) Geim, A. K.; Novoselov, K. S. Nat. Mater. 2007, 6(3), 183–191.
(3) Geim, A. K. Science 2009, 324(5934), 1530–1534.
(4) Rao, C. N. R.; Biswas, K.; Subrahmanyam, K. S.; Govindaraj, A. J. Mater.
Chem. 2009, 19, 2459–2469.
(5) Rao, C. N. R.; Sood, A. K.; Subrahmanyam, K. S.; Govindaraj, A. Angew.
Chem., Int. Ed. 2009, 48, 7752–7777.
(6) Blake, P.; Hill, E. W.; Neto, A. H. C.; Novoselov, K. S.; Jiang, D.; Yang, R.;
Booth, T. J.; Geim, A. K. Appl. Phys. Lett. 2007, 91(6), 063124.
(7) Gaskell, P. E.; Skulason, H. S.; Rodenchuk, C.; Szkopek, T. Appl. Phys. Lett.
2009, 94(14), 143101.
(8) Wang, F.; Zhang, Y. B.; Tian, C. S.; Girit, C.; Zettl, A.; Crommie, M.; Shen,
Y. R. Science 2008, 320(5873), 206–209.
(9) Lee, C.; Wei, X. D.; Kysar, J. W.; Hone, J. Science 2008, 321(5887), 385–388.
(10) Blake, P.; Brimicombe, P. D.; Nair, R. R.; Booth, T. J.; Jiang, D.; Schedin,
F.; Ponomarenko, L. A.; Morozov, S. V.; Gleeson, H. F.; Hill, E. W.; Geim, A. K.;
Novoselov, K. S. Nano Lett. 2008, 8(6), 1704–1708.
(11) Wang, X.; Zhi, L. J.; Mullen, K. Nano Lett. 2008, 8(1), 323–327.
(12) Stankovich, S.; Dikin, D. A.; Dommett, G. H. B.; Kohlhaas, K. M.;
Zimney, E. J.; Stach, E. A.; Piner, R. D.; Nguyen, S. T.; Ruoff, R. S. Nature 2006,
442(7100), 282–286.
(13) Niyogi, S.; Bekyarova, E.; Itkis, M. E.; McWilliams, J. L.; Hamon, M. A.;
Haddon, R. C. J. Am. Chem. Soc. 2006, 128(24), 7720–7721.
(14) Stankovich, S.; Dikin, D. A.; Piner, R. D.; Kohlhaas, K. A.; Kleinhammes,
A.; Jia, Y.; Wu, Y.; Nguyen, S. T.; Ruoff, R. S. Carbon 2007, 45(7), 1558–1565.
(15) Eda, G.; Fanchini, G.; Chhowalla, M. Nat. Nanotechnol. 2008, 3(5), 270–
274.
(16) Park, S.; An, J. H.; Jung, I. W.; Piner, R. D.; An, S. J.; Li, X. S.;
Velamakanni, A.; Ruoff, R. S. Nano Lett. 2009, 9(4), 1593–1597.
(17) Li, D.; Muller, M. B.; Gilje, S.; Kaner, R. B.; Wallace, G. G. Nat.
Nanotechnol. 2008, 3(2), 101–105.
(18) Paredes, J. I.; Villar-Rodil, S.; Martinez-Alonso, A.; Tascon, J. M. D.
Langmuir 2008, 24(19), 10560–10564.
in water (and other solvents19). However, the oxidation process
extensively disrupts the bonding scheme in the basal plane, resulting
in a dramatic change in the electronic properties. In effect, the
exciting properties of graphene are destroyed. While chemical or
thermal reduction can remove the oxides,20 structural defects remain
and the original electronic properties are not recovered.14,15,21
To address this issue, our group22,23 and others10,24 have
demonstrated that graphite can be exfoliated to give graphene in
certain solvents or in aqueous surfactant solutions. The surfactant
dispersed graphene sheets22 are stabilized against reaggregation in
a manner similar to that described by Derjaguin-Landau-
Verwey-Overbeek (DLVO) theory.25 However, the solvent based
systems23 are more interesting. Here, it seems the solvent-gra-
phene interaction is strong enough to largely balance the intersheet
attractive forces. This process appears to involve only weak van
der Waals type bonding, specifically dispersion, dipolar, and
hydrogen-bonding interactions. Raman spectroscopy and X-ray
photoelectron spectroscopy show that no defects or oxides are
introduced as a result of the dispersion process.23 However, very
little is known about the details of the interactions between solvent
and graphene. In addition, few good solvents have been discovered
until now. We expect that a more detailed understanding of what
makes a good solvent would greatly facilitate solvent discovery,
thereby facilitating the exfoliation of graphite in a wider variety of
systems and opening a broader range of applications.
In this work, we measure the dispersibility of graphene in 40
different solvents. We use this data to identify the Hansen
solubility parameters of graphene. This allows the estimation of
the Flory-Huggins parameter, a number directly related to the
(19) Park, S.; Ruoff, R. S. Nat. Nanotechnol. 2009, 4(4), 217–224.
(20) Gao, W.; Alemany, L. B.; Ci, L.; Ajayan, P. M. Nat. Chem. 2009, 1, 403–
408.
(21) Yang, D.; Velamakanni, A.; Bozoklu, G.; Park, S.; Stoller, M.; Piner, R. D.;
Stankovich, S.; Jung, I.; Field, D. A.; Ventrice, C. A., Jr.; Ruoff, R. S. Carbon 2009,
47(1), 145–152.
(22) Lotya, M.; Hernandez, Y.; King, P. J.; Smith, R. J.; Nicolosi, V.; Karlsson,
L. S.; Blighe, F. M.; De, S.; Wang, Z.; McGovern, I. T.; Duesberg, G. S.; Coleman,
J. N. J. Am. Chem. Soc. 2009, 131(10), 3611–3620.
(23) Hernandez, Y.; Nicolosi, V.; Lotya, M.; Blighe, F. M.; Sun, Z.; De, S.;
McGovern, I. T.; Holland, B.; Byrne, M.; , Y. K.; Boland, J. J.; Niraj, P.; Duesberg,
G.; Krishnamurthy, S.; Goodhue, R.; Hutchison, J.; Scardaci, V.; Ferrari, A. C.;
Coleman, J. N. Nat. Nanotechnol. 2008, 3(9), 563–568.
(24) Bourlinos, A. B.; Georgakilas, V.; Zboril, R.; Steriotis, T. A.; Stubos, A. K.
Small 2009, 5(16), 1841–1845.
(25) Israelachvili, J. Intermolecular and Surface Forces; Academic Press: 1991.

3208 DOI: 10.1021/la903188a Published on Web 11/02/2009 Langmuir 2010, 26(5), 3208–3213
Hernandez et al.
energy cost of exfoliation. Finally, we use transmission electron
microscopy (TEM) to demonstrate that the graphene is heavily
exfoliated in these solvents.
Experimental Section
The graphite powder used in all experiments was purchased
from Sigma Aldrich (product number 332461) and sieved through
a 0.5 mm mesh to remove large particles. Dispersions of exfoliated
graphite were prepared, as previously reported,23 in a wide range
of solvents (all purchased from Sigma Aldrich, see Table S1 in the
Supporting Information). These solvents were identified in three
ways. Some (i.e., NMP,26,27 DMF26,28) were known from the
literature as nanotube solvents. Others were identified as close
structural relations to NMP. Still others were found by testing
solvents with surface energy close to that of graphene.23,29 In
addition, during this work, we made preliminary estimations of
the graphene Hansen parameters and used this information to
find new solvents. In all cases, an initial concentration of 0.1 mg/
mL of graphite in 10 mL of each solvent was subjected to
sonication in an ultrasonic bath (Branson 1510E-MT) for
30 min. To remove large aggregates, the dispersions were then
mildly centrifuged at 500 rpm for 90 min (Hettich Mikro 22R) and
the supernatant was retained for analysis. We define the graphene
dispersibility, CG, as the concentration of material remaining after
centrifugation (CF). We obtain this by measuring the optical
absorbance, A, over the wavelength range 200-1000 nm (Cary
6000i UV-vis NIR, l = 1 cm). We find spectra which are flat
and featureless as observed previously for dispersed graphene.22,23
We use the Lambert-Beer law to relate the absorbance to the
graphene concentration (A = RCGl), using the previously mea-
sured absorption coefficient (660 nm), R660 = 2460 L/g 3 m.23
Each dispersion was prepared a minimum of three times (nine
times for the best solvents), and the mean value of CG was used
with the standard deviation taken as an estimation of error. This
gives us a table of dispersibility values for graphene in 40 different
solvents (Table S1 in the Supporting Information).
Results and Discussion
The measured dispersibilities vary from 0.16 ( 0.05 μg/mL in
pentane to 8.5 ( 1.2 μg/mL in cyclopentanone (CPO). In some
cases, the dispersabilities of the previously reported23 solvents
have changed slightly. This is due to the fact that each of these
solvents has been measured nine times in this study, thus giving a
more accurate result. We note that the absolute values of these
dispersibilities are appropriate for the preparation conditions
used (i.e., sonication, centrifugation, etc). However, it is possible
that they could be improved by further optimization of the
preparation conditions. However, under those circumstances,
we would expect all dispersibilities to increase by the same factor,
thus retaining their order relative to each other. The stability of
the dispersed graphene in these solvents is reasonably good in
most cases. Shown in Figure 1 are sedimentation data for the best
solvent (CPO), a good solvent, N-methyl-2-pyrrolidone (NMP),
and a poor solvent, toluene. These were measured by tracking the
optical absorbance (after CF) over time using a specially built
apparatus.30 In each case, some sedimentation occurs; however,
(26) Furtado, C. A.; Kim, U. J.; Gutierrez, H. R.; Pan, L.; Dickey, E. C.;
Eklund, P. C. J. Am. Chem. Soc. 2004, 126(19), 6095–6105.
(27) Umek, P.; Vrbanic, D.; Remskar, M.; Mertelj, T.; Venturini, P.; Pejovnik,
S.; Mihailovic, D. Carbon 2002, 40(14), 2581–2585.
(28) Landi, B. J.; Ruf, H. J.; Worman, J. J.; Raffaelle, R. P. J. Phys. Chem. B
2004, 108(44), 17089–17095.
(29) Bergin, S. D.; Nicolosi, V.; Streich, P. V.; Giordani, S.; Sun, Z.; Windle,
A. H.; Ryan, P.; Peter, N.; Niraj, P.; Wang, Z.-T. T.; Carpenter, L.; Blau, W. J.;
Boland, J. J.; Hamilton, J. P.; Coleman, J. N. Adv. Mater. 2008, 20, 1876–81.
(30) Nicolosi, V.; Vrbanic, D.; Mrzel, A.; McCauley, J.; , S.; McGuinness, C.;
Compagnini, G.; Mihailovic, D.; Blau, W. J.; Coleman, J. N. J. Phys. Chem. B
2005, 109(15), 7124–7133.
Langmuir 2010, 26(5), 3208–3213
Article
Figure 1. Sedimentation data for graphene in toluene, NMP, and
CPO. Samples were prepared by 30 min sonication of 0.1 mg/mL
graphite followed by 500 rpm centrifugation for 90 min. The
optical absorbance (650 nm) of the supernatant was monitored
as a function of time. As shown in the inset, the dashed lines
represent exponential fits to the data. The fit constants are also
given in the inset.
for CHO and NMP, the sedimentation occurs over longer time
scales: >70% of the starting material remains dispersed after
250 h. However, toluene is less stable with 70% remaining
dispersed after 10 h. In all cases, the sedimentation curves can
be fit to single exponentials as expected for dispersions with one
sedimenting component.30 From the fits, we can estimate that
each dispersion has a stable (nonsedimenting) component con-
sisting of 75%, 72%, and 60% of the graphene mass for the NMP,
CPO, and toluene samples, respectively.
Previously, we showed that effective solvents for graphene had
surface tensions close to 40 mJ/m2.23 We developed a simple
model showing that, for two-dimensional nanomaterials, the
surface energy acts like a solubility parameter; good solvents
are those for which solvent and nanomaterial have the same
surface energy. Indeed, the data collected during this work fit well
to this model. As shown in Figure 2A, the graphene dispersibility
is maximized for solvents with surface tension close to 40 mJ/m2.
However, while we believe that the surface energy is an important
parameter for understanding the solvent-graphene interaction,
we also recognize it is a blunt tool which can only describe the
overall interaction. In most solvent-solute systems, we can divide
the intermolecular interactions into at least three types. The
simplest formulation divides them into dispersive (D), polar (P,)
and hydrogen-bonding (H) components.31,32 The most successful
solubility theories treat these interactions separately.32
Basic solubility theory states that, for molecular solutes, the
property most closely associated with solubility is the cohesive
energy density, EC,T/V, where EC,T is the total molar cohesive
energy and V is the molar volume of the solvent.32 The most
commonly used solubility parameter, the Hildebrand parameter, δT,
is just the square root of EC,T/V. It can be shown that the cohesive
energy density can be divided into D, P, and H components:
EC, T EC, D EC, P EC, H
¼ þ þ ð1Þ
V V V V
The square root of each of these components is a Hansen solubility
parameter δi (i = D, P, H).
δT2 ¼ δD2 þ δP2 þ δH2 ð2Þ
(31) Barton, A. F. M. CRC Handbook of Solubility Parameters and other
Cohesion Parameters; CRC Press Inc.: Boca Raton, FL, 1983.
(32) Hansen, C. M. Hansen Solubility Parameters - A User’s Handbook; CRC
Press: Boca Raton, FL, 2007.
DOI: 10.1021/la903188a 3209
Article
Figure 2. Graphene dispersibility, CG, as a function of (A) solvent
surface tension and (B) solvent Hildebrand parameter.
These solubility parameters are incredibly useful when it comes
to solvent selection for a given solute. Successful solvents are
governed by two criteria. First, the Hildebrand solubility para-
meters of solvent and solute must be very close to each other.33,34
This is usually sufficient to identify solvents for nonpolar solutes.
However, for polar solutes, a further criterion is often required:
good solvents are those where solvent and solute have similar
values of all three Hansen solubility parameters.31,32
However, graphene is not a molecular solute but a nanomater-
ial. Such materials are distinctive in that the interaction with the
solvent occurs at a well-defined surface. For this reason, the
surface energy based solubility parameters we have described
previously appear more appropriate.23,29 In fact, the surface
energy can also be divided up into dispersive, polar, and H-
bonding components.35-37 For nanomaterials, it would appear
more natural to use these components to develop Hansen-like
solubility parameters based on surface energy.38 However, the
components are only known for a small number of solvents. In
contrast, Hansen parameters have been published for 1200
solvents.39 This makes them much more useful even though
surface energy may be a more intuitive parameter for nanostruc-
tured systems such as nanotubes and graphene. Thus, we analyze
the dispersibility of graphene using Hildebrand and Hansen
solubility parameters.
We can test the first solvent criterion by plotting the measured
graphene dispersibility, CG, as a function of δT as shown in
Figure 2B. A well-defined peak in the data is observed with the
best solvents having δT ∼ 23 MPa1/2. This is in good agreement
with previous work on solvent-assisted dispersions of carbon
nanotubes.38,40 However, some of the solvents with δT very close
to 23 MPa1/2 display poor dispersibility. This means that the
(33) Hildebrand, J. H.; Prausnitz, J. M.; Scott, R. L. Regular and related
solutions, 1st ed.; Van Nostrand Reinhold Company: New York, 1970; p 228.
(34) Rubinstein, M.; Colby, R. H. Polymer Physics, 1st ed.; Oxford University
Press: Oxford, 2003; p 440.
(35) Beerbower, A. J. Colloid Interface Sci. 1971, 35, 126–132.
(36) Brandrup, J.; Immergut, E. H.; Grulke, E. A.; Akihiro, A.; Bloch, D. R.
Polymer Handbook, 4th ed.; 1999.
(37) Koenhen, D. M.; Smolders, C. A. J. Appl. Polym. Sci. 1975, 19, 1163–1179.
(38) Bergin, S. D.; Sun, Z. Y.; Rickard, D.; Streich, P. V.; Hamilton, J. P.;
Coleman, J. N. ACS Nano 2009, 3(8), 2340–2350.
(39) Hansen, C. M.; Abbott, S. HsPiP software, www.hansen-solubility.com.
(40) Detriche, S.; Zorzini, G.; Colomer, J. F.; Fonseca, A.; Nagy, J. B.
J. Nanosci. Nanotechnol. 2007, 8(11), 6082–6092.
3210 DOI: 10.1021/la903188a
Hernandez et al.
Figure 3. Dispersibility for all solvents tested as a function of (A)
dispersive Hansen solubility parameter, (B) polar Hansen solubi-
lity parameter, and (C) hydrogen-bonding Hansen solubility para-
meter.
second criterion (that solute and solvent must have similar
Hansen parameters) must be fulfilled.
To test this, we plot CG as a function of δD, δP, and δH, as
shown in Figure 3. The Hansen solubility parameters (and molar
volumes used later) of the solvents were taken from HSPiP
software (www.hansen-solubility.com) and are given in Table
S1 in the Supporting Information. This software has a database of
values from literature32 but also provides algorithms to calculate
unknown Hansen parameters.41,42 Figure 3A shows a well-
defined dependence of CG on the dispersive Hansen parameter
with a peak around 18 MPa1/2. Clearly, good solvents for
graphene have 15 MPa1/2 < δD < 21 MPa1/2. These results for
the dispersive Hansen parameter are in good agreement with
recent work on single walled carbon nanotube dispersions.38 As
shown in Figure 3B and C, when plotted versus polar and
H-bonding Hansen parameters, CG also displays broad peaks
centered at δP ∼ 12 MPa1/2 and δH ∼ 9 MPa1/2, respectively. The
most surprising thing about this data is that good solvents have
nonzero values of δP and δH at all. The nonpolar nature of
graphene would suggest that the Hildebrand parameter or the
dispersive Hansen parameter alone would determine dispersibil-
ity. For example, we note that the solubility of C60 is very well
described by the Hildebrand parameter alone.43 X-ray photoelec-
tron spectroscopy measurements have confirmed that no oxides
or other polar groups are attached to the graphene either before or
after dispersion in solvents such as NMP.23 Park et al. have shown
that highly reduced graphene oxide can be dispersed in solvents
defined by 10 < (δH+δH) < 30.44 These results agree well with
our data for defect free graphene, suggesting a fundamental
requirement for a well defined degree of polarity in successful
(41) Stefanis, E.; Panayiotou, C. Int. J. Thermophys. 2008, 29(2), 568–585.
(42) Isu, Y.; Nagashima, U.; Aoyama, T.; Hosoya, H. J. Chem. Inf. Comput. Sci.
1996, 36(2), 286–293.
(43) Ruoff, R. S.; Tse, D. S.; Malhotra, R.; Lorents, D. C. J. Phys. Chem. 1993,
97, 3379–3383.
(44) Park, S.; An, J.; Jung, I.; Piner, R. D.; An, S. J.; Li, X.; Velamakanni, A.;
Ruoff, R. S. Nano Lett. 2009, 9, 1593–1597.
Langmuir 2010, 26(5), 3208–3213
Hernandez et al.
graphene solvents. However, the exact reason why nonzero values
of δP and δH are required to disperse graphene remains a mystery.
While Figure 3 supports the validity of the use of Hansen
parameters, much scatter is present. This is likely due to the fact
that each panel in Figure 3 treats the Hansen parameters one by
one. Thus, a solvent with only one matching solubility parameter
will appear under the peak in one panel but with lower than
expected CG. To address this, we need to test the suitability of all
three Hansen parameters simultaneously.
To do this, we first estimate the Hansen parameters of
the graphene itself. The simplest way to do this is by solvent
screening, that is, by associating the solubility parameters of the
solute with those of the most successful solvents.31-33,36 Here, we
estimate the graphene Hansen parameters from the weighted
averages of the Hansen parameters of the solvents. We suggest
using the quantitative measure of the solvent quality, here the
dispersibility, as the weighting factor. The three Hansen para-
meters are then given by
P
CG δi, sol
P
Æδi æ ¼ solvent ð3Þ
CG
solvent
where i = D, P, or H (or T). Here, CG is the graphene
dispersibility in a given solvent and δi,sol is the ith Hansen
parameter in a given solvent. The summation is taken over all
solvents studied. The advantage of this approach is that solvents
contribute to the final result in proportion to their quality. For the
solvents studied in this work, we estimate the Hansen parameters
for graphene to be ÆδDæ ≈ 18.0 MPa1/2, ÆδPæ ≈ 9.3 MPa1/2, and
ÆδHæ ≈ 7.7 MPa1/2. These values agree reasonably well with the
Hansen parameters for carbon nanotubes measured by us (ÆδDæ =
17.8 MPa1/2, ÆδPæ = 7.5 MPa1/2, and ÆδHæ = 7.6 MPa1/2)38 and
others.40
To test the validity of this methodology, we need a parameter
which represents the potential of a given solvent. The simplest
parameter is the distance in Hansen parameter space from solvent
to solute Hansen parameters. This distance is just the length, R, of
the vector from the point in Hansen space representing the
solute, A (δD,A,δP,A,δH,A), to the point representing the solvent,
B (δD,B,δP,B,δH,B):
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
R ¼ ðδD, A - δD, B Þ2 þ ðδP, A - δP, B Þ2 þ ðδH, A - δH, B Þ2 ð4Þ
However, rather than using R, we borrow an alternative metric
from Flory-Huggins theory. The Flory-Huggins parameter,34
χ, is a measure of the cost of mixing solvent and solute and can be
related to the Hansen solubility parameters by32
v0 2 v0 h
χ≈ R ¼ ðδD, A - δD, B Þ2 þ ðδP, A - δP, B Þ2
kT kT
i taken with a Jeol 2100 instrument operating at 200 kV. This was
þ ðδH, A - δH, B Þ2 ð5Þ
where v0 is the solvent molecular volume. Hansen has written this
expression with a prefactor of 0.25 before the second and third
terms on the right-hand side.32 Incorporation of this prefactor is
extremely common, and its usage is supported empirically by
significant quantities of data from many areas of solubility
research.32 However, it is not supported by a rigorous theoretical
foundation. Thus, we use eq 5 in its more intuitive form. However,
we note that we have also carried out all the analysis described
below using eq 5 but including the factor of 0.25. We find the same
trends in both cases.
Langmuir 2010, 26(5), 3208–3213
Article
Figure 4. Dispersibility, CG, as a function of Flory-Huggins
parameter, χ.
According to eq 5, the lower the value of χ, the lower the
energetic cost of dispersing (and exfoliating) the graphene. How-
ever, this expression is approximate as it does not allow for
negative values of χ. Such values are of course possible and occur
when solvent-solute interactions are particularly strong. Equa-
tion 5 also confirms that dispersion and exfoliation is favored
when both solute (graphene) and solvent have similar values of all
three Hansen parameters (D, P, and H).
In the simplest case, χ gets smaller as solvent and solute Hansen
parameters converge. Thus, if the concept of Hansen parameters
is appropriate to graphene dispersions and if our estimate of the
graphene Hansen parameters is realistic, CG should decrease with
increasing χ. We use eq 5 to calculate the Flory-Huggins
parameter for each solvent-graphene dispersion. Figure 4 plots
CG as a function of χ and clearly shows the expected behavior. We
observe a clear improvement in graphene dispersibility as the
estimated χ values tend toward zero. This graph confirms that our
estimated Hansen parameters are close to the true values. We
expect that identification of the exact values would reduce
the scatter in Figure 4 further. In addition, it clearly shows that
the dispersion of graphene is at least partly controlled by the
energetic cost of exfoliation.
These results mean that good solvents for graphene should have
Hansen parameters close to δD = 18.0 MPa1/2, δP = 9.3 MPa1/2,
and δH = 7.7 MPa1/2. In fact, during this work, we made a
preliminary estimation of ÆδDæ, ÆδPæ, and ÆδHæ. We then tested a
number of solvents with Hansen parameters in this region. This
resulted in the discovery of our best two solvents, cyclopentanone
(δD = 17.9 MPa1/2, δP = 11.9 MPa1/2, δH = 5.2 MPa1/2) and
cyclohexanone (δD = 17.8 MPa1/2, δP = 8.4 MPa1/2, δH = 5.1
MPa1/2). We feel this strongly validates our results and our method.
Of course, probing dispersibility through optical absorbance
measurements does not confirm the presence of graphene in the
dispersions. In order to prove exfoliation of graphene from the
starting graphite, we conducted TEM analysis. Samples for TEM
were prepared by pipetting a few milliliters of dispersion onto
holey carbon grids (400 mesh). Bright field TEM images were
done for solvents with the highest CG, a selection of intermediate
CG values, and a low value of CG (the square bracketed numbers
give the solvent rankings from Table S1 in the Supporting
Information): cyclopentanone[1] (CPO, CG = 8.5 ( 1.1 μg/
mL), 1,3-dimethyl-2-imidazolidinone[5] (DMEU, CG = 5.2 (
1.3 μg/mL), N-ethyl-2-pyrrolidone[13] (NEP, CG = 4.0 ( 0.7 μg/
mL), N-dodecyl-2-pyrrolidinone[28] (N12P, CG = 2.1 ( 1.1 μg/
mL), and acetone[34] (CG = 1.2 ( 0.4 μg/mL). These samples
were compared to previously studied23 N-methyl-2-pyrrolidone-
[8] (NMP, CG = 4.7 ( 1.9 μg/mL). Figure 5 shows selected TEM
images for monolayer and multilayered graphene flakes derived
from these solvents. All the solvents produce some monolayer
DOI: 10.1021/la903188a 3211
Article
Figure 5. Selected monolayer and few-layer graphene flakes from
various solvents: (A-C) monolayers and (D-F) multilayers.
Figure 6. Histograms for numbers of layers per flake from TEM
analysis of six solvents.
graphene with no visible evidence of large scale defects. Many of
the flakes observed under TEM are composed of few-layer
graphene. In order to investigate the quality of the dispersions
further, we conducted detailed statistical analysis of the TEM
images. We counted the numbers of graphene layers per flake by
examining the edges of the flakes. In TEM images of graphene
multilayers, the edges of the individual flakes are almost always
distinguishable. Thus, by carefully counting the flake edges, it is
possible to measure the number of layers per flake.23 (We note
that most researchers use atomic force microscopy to measure
flake thickness. However, this is not possible here. Due to the high
boiling point of most successful graphene solvents, flakes tend to
aggregate during deposition onto surfaces, rendering quantitative
AFM analysis impossible). A minimum of 64 flakes per sample
were analyzed; the data are summarized in Figure 6. It is clear that
all the solvents examined produce distributions of flake types that
are weighted toward few-layer graphene (see Table 1). All the
solvents produce a significant population of monolayer material.
As shown in Table 1, the best solvent was NMP with a mean value
of 2.5 layers per flake and the highest number fraction of
monolayer graphene (29%). What is most interesting is that all
the solvents, good and poor, have >63% of flakes with 1-5
layers. This suggests that even solvents with poor dispersibility,
such as N12P and acetone, can produce dispersions of high flake
quality.
Finally, we note that the mechanism of graphene dispersion in
these solvents is still not known. We expect the χ values estimated
for the vast majority of solvents are simply too big to result in
thermodynamic solubility of graphene sheets. The reason for this
3212 DOI: 10.1021/la903188a
Hernandez et al.
Table 1. Mean Numbers of Graphene Layers per Flake, ÆNæ, Number
Fraction of Monolayer Graphene, N1/NT, and Number Fraction of
Few-Layer Graphene (1-5 layer), N1-5/NT, from TEM Analysis of
Six Solvents
N1/NT N1-5/NT
ÆNæ (%) (%)
NMP 2.5 29 97
DMEU 4.5 11 70
cyclopentanone 4.8 5 69
NEP 4.2 6 65
N12P 5.2 5 64
acetone 4.3 7 74
is that such large structures would be expected to have very low
entropy of mixing.38 Thus, only systems with extremely small
values of χ could be solutions (i.e., have negative free energy of
mixing). Here, some of the solvents have enormous χ values, up to
∼10. To put this in context, in polymer systems, χ values >0.5 are
considered large. In addition, the observed sedimentation sug-
gests these systems are dispersions rather than solutions. How-
ever, if graphene is not soluble, it must be somehow stabilized
against aggregation. The means for this stabilization remain
unknown. However, the mechanism appears to be different from
that of nanotubes in solvents. The best solvents for nanotubes
tend to be amide solvents,29,38 suggesting a molecule dependent,
specific interaction. In addition, a number of solvents with the
correct Hansen parameters do not disperse nanotubes.38 Neither
of these phenomena applies to graphene: Successful solvents are
not limited to amides, and all solvents with Hansen parameters in
the correct region tested by us have dispersed graphene to some
degree. Another mystery is that fact that NMP is the eighth best
solvent in terms of dispersibility. However, it is the best solvent in
terms of exfoliation and stability. Future work is required to
answer these outstanding questions.
Conclusion
In conclusion, we have demonstrated the dispersion of gra-
phene in 40 solvents of which 28 are new graphene solvents,
attaining dispersibilities up to 8 μg/mL. We confirmed that
successful solvents are characterized by surface tensions close to
40 mJ/m2 as described previously.23 However, we also showed that
good solvents are characterized by a Hildebrand solubility para-
meter close to 23 MPa1/2. As both surface tension and Hildeb-
rand parameter are related to the overall solvent-graphene
interaction, we have also investigated the dispersive, polar, and
hydrogen-bonding components of the interaction. We do this
by showing that successful solvents have well-defined values of
the Hansen solubility parameters. Surprisingly, nonzero values
of the polar and H-bonding Hansen parameters are required
for a solvent to disperse graphene. This is unexpected for such a
nonpolar material. In addition, we have shown that the
graphene is extensively exfoliated in solvents irrespective of
the graphene dispersibility in that solvent.
We feel this work is important for two reasons. First, we have
demonstrated 28 new solvents for graphene. Many of these
solvents have distinct advantages over previously known solvents
such as NMP. For example, NMP is handicapped by its high
boiling point of 205 C which makes it very difficult to remove
solvent in applications such as deposition of flakes on surfaces. In
contrast, chloroform[20] and isopropanol[19] have boiling points
of 61 and 82 C, respectively. In addition, access to such a wide
range of solvents means a broad choice of polymers are available
for liquid phase composite formation. Second, knowledge of the
Hansen parameters for any solvent allows the discovery of new
Langmuir 2010, 26(5), 3208–3213
Hernandez et al. Article

solvents and facilitates the formulation of solvent blends. Thus,
this work opens the field of liquid phase graphene exfoliation. We
anticipate advances in areas such as film formation and composite
processing will stem from these results.
Acknowledgment. J.N.C. acknowledgments SFI funding
under the PI award scheme, Contract Number
07/IN.1/I1772. M.L. acknowledges financial support from
IRCSET.
Supporting Information Available: Full list of solvents
studied including Hansen parameters and dispersibility va-
lues. This material is available free of charge via the Internet
at http://pubs.acs.org.

Langmuir 2010, 26(5), 3208–3213 DOI: 10.1021/la903188a 3213