Published on 04 November 2015. Downloaded by Max-Planck-Institut für Dynamik und Selbstorganisation on 01/08/2016 15:07:32.

Faraday Discussions
Cite this: Faraday Discuss., 2016, 186, 229

View Article Online

PAPER View Journal | View Issue

How do polydisperse repulsive colloids

crystallize?
Robert Botet,*a Bernard Cabane,b Lucas Goehring,c Joaquim Lic
and Franck Artznerd

Received 14th October 2015, Accepted 4th November 2015

DOI: 10.1039/c5fd00145e

A modified version of the Gibbs-ensemble Monte-Carlo method reveals how polydisperse

charged colloidal particles can build complex colloidal crystals. It provides general rules
that are applicable to this fractionated crystallization that stems from size segregation. It
explains the spontaneous formation of complex crystals with very large unit-cells in
suspensions of nanoparticles with a broad size distribution.

1 Introduction
Fundamental physics requires experiments to observe, and theory to explain. In
principle, nothing more is needed for a complete understanding of phenomena.
However, in some situations, this sharing of the work is not sufficient, either
because the experimental data are counter-intuitive (disagreement with the
theory) or because the theory itself is impracticable. If one of these cases arises,
numerical simulations turn out to be the essential tool to complement our ideas
through direct ‘observation’ of the theory or through articial changes of the
microscopic rules (as in a Maxwell's demon-like approach to analyze cause-to-
effect mechanisms).
In the present article, we discuss results of numerical simulations dedicated to
the comprehension of recent experiments in which a dispersion of polydisperse
repulsive nanoparticles was compressed through osmosis. Details of the experi-
ments will be published elsewhere.1 A suspension of ordinary colloidal silica
particles (Ludox, polydispersity 14%, average radius 8 nm), was compressed
through dialysis. This caused a spontaneous segregation of the particles, frac-
tionated crystallization and multiple-phase coexistence (involving various crystal
structures). For instance, the low-compression part of the phase sequence was
seen to be a colloidal liquid at the lowest volume fractions, then nucleation of the

a
Laboratoire de Physique des Solides, CNRS UMR8502, Univ. Paris-Sud, Université Paris-Saclay, 91405 Orsay
Cedex, France. E-mail: robert.botet@u-psud.fr; Fax: +33 1 6915 6086; Tel: +33 1 6915 6925
b
LCMD, CNRS UMR 8231, ESPCI, 10 rue Vauquelin, 75231 Paris Cedex 05, France
c
Max Planck Institute for Dynamics and Self-Organization (MPIDS), 37077 Göttingen, Germany
d
Institut de Physique, CNRS UMR 6626, Univ. Rennes, 35042 Rennes, France

This journal is © The Royal Society of Chemistry 2016 Faraday Discuss., 2016, 186, 229–240 | 229
Published on 04 November 2015. Downloaded by Max-Planck-Institut für Dynamik und Selbstorganisation on 01/08/2016 15:07:32. View Article Online
Faraday Discussions Paper
BCC crystalline phase occurred at a silica volume fraction f > 0.19, and then there
was coexistence of the BCC + Laves AB2 crystal phases for f > 0.22. The complete
phase diagram was found to be much richer than expected according to current
theories.
It is important to note that we do not describe here ordinary fractionated
crystallization caused by different solubilities of coexisting substances, but a new
kind of fractionation that organizes populations of particles with different sizes,
and makes the best use of available space. Without any theory at hand to explain
the formation of such uncommon structures as the Laves AB2 crystals,2 we used
the Gibbs-ensemble Monte-Carlo method3 to understand the mechanisms used
by the system to sort the various particle sizes and to put them in the correct
places.
Actually, we encountered many unanswered questions from the experimental
data. The numerical simulations helped us to answer some of them. We replicate
hereaer this particular state of mind, dividing our comprehension process in
a series of short questions and discussing the answers given by the numerical
study. The reader interested in a quantitative comparison between the experi-
mental data and the numerical simulations should refer to ref. 1.
To close this Introduction, let us give a warning which might be important
when comparing experimental data with numerical simulations: the regular
Monte-Carlo method does not consider real kinetic processes, but is an efficient
tool to search for the equilibrium state using the physical ground of free energy
minimization. Therefore, the results presented below are linked to the equilib-
rium state. If this is not the case (in experiments, in kinetic numerical simula-
tions), other intermediate crystalline structures may appear for some time as
unstable phases.

2 The system under study

We consider a dispersion of spherical hard particles. The surface of each particle
is electrically charged with 0.4e per nm2. The particle radius is denoted: a, and its
value is Gaussian-distributed with the average value hai ¼ 8 nm and standard
deviation s, truncated to the range: [4 nm; 12 nm] (that is: no radius is smaller
than 4 nm or larger than 12 nm). The polydispersity, d, of this population is
dened as: d2 ¼ ha2i/hai2  1. Then, the value of d coincides with s/hai for the
pffiffiffiffiffi
small values of s, and has the limiting value 1= 12x29% for the innite value of
s (rectangular distribution). Throughout the article, we shall take the value s ¼
1.2 realizing the polydispersity d ¼ 15% of the population.
If large polydispersity (say d > 25%) has to be considered, the inuence of
moments of order larger than 2 should be studied, that is to investigate the
various distributions – such as log-normal or Schulz distributions – with the same
values of hai and d. In related problems, the shape of the radius-distribution is
known to be unessential.4
We consider here the DLVO theory,5 in which the interaction energy between
two charged particles, labelled i and j, is generally of the Yukawa (screened-
Coulomb) form:
 lB
Uij kB T ¼ Zi Zj ekrij ; for rij $ ai þ aj (1)
rij

230 | Faraday Discuss., 2016, 186, 229–240 This journal is © The Royal Society of Chemistry 2016
Published on 04 November 2015. Downloaded by Max-Planck-Institut für Dynamik und Selbstorganisation on 01/08/2016 15:07:32. View Article Online
Paper Faraday Discussions
where Zi is the effective charge of the particle i, and rij the distance between their
centers. lB ¼ 0.7 nm is the Bjerrum length, and k ¼ 1/lD is the screening
parameter, with lD ¼ 3 nm the Debye screening length in our experimental
conditions. The values of lB and lD are expected to be constant throughout the
system.
Two denitions have to be stated:
 The energy at contact (i.e. using (1) for rij ¼ ai + aj), traditionally written: b3, is
a central parameter for discussion. Its averaged value is b3 ¼ 21.6 for the present
set of parameters.
 The potential energy per particle at equilibrium is here denoted generically:
E. It is dened as the sum of the pair-potentials (1) between all the couples of
particles forming the equilibrated system, divided by the total number of parti-
cles, aer having minimized the system free-energy using the Monte-Carlo
method.
At small or intermediate volume fractions – namely: f # 0.41 for d ¼ 15% – the
Wigner–Seitz radius is larger than hai + lD. In these conditions, the attractive van
der Waals interaction between silica spheres in water is smaller than the thermal
energy, kBT, and thus can be neglected. In the same way, we do not discuss here
controversial additional weak long-ranged attractive pairwise potentials such as
those appearing in the Sogami–Ise theory.6 Within our approach, such interac-
tions would result in the same results as in the case of the purely repulsive DLVO
potential, with an effective value of the osmotic pressure slightly larger than the
experimental pressure.7

3 Q1: which of the FCC or BCC crystalline

structures should appear for a system made of
a polydisperse distribution of repulsive particles at
the thermodynamic equilibrium?
A1: systems of monodisperse particles interacting through repulsive Yukawa
potentials, form either FCC or BCC crystal structures in the equilibrium solid
phase.8 The BCC phase appears when f < 0.5 and the Debye screening length lD is
larger than a threshold depending on f. In all other cases, the FCC phase is the
solid structure at equilibrium.
In the case of polydisperse particles, the question is more complicated. Let us
suppose rst that the solid phase can only be FCC or BCC, without coexistence
(note that, for the parameters as chosen, only the FCC crystalline phase can
appear in the monodisperse case, since the value of lD/hai 0.38 is below the
liquid–BCC–FCC triple critical point). We performed random exchanges of
couples of particles in the system composed of either a single FCC phase or
a single BCC phase, according to the standard Metropolis Monte-Carlo algorithm,
until thermodynamic equilibrium was reached. The evolution of the difference of
energy at equilibrium per particle, (EFCC  EBCC)/kBT, versus the volume fraction is
plotted in Fig. 1, and the monodisperse case is also plotted for comparison.
Unlike the monodisperse case, the polydisperse population is seen to undergo
a clear change in the stability of the BCC/FCC structure: when f > 0.3, the BCC
crystal is energetically more likely to appear than the FCC structure. Moreover this
trend increases with the volume fraction.

This journal is © The Royal Society of Chemistry 2016 Faraday Discuss., 2016, 186, 229–240 | 231
Published on 04 November 2015. Downloaded by Max-Planck-Institut für Dynamik und Selbstorganisation on 01/08/2016 15:07:32. View Article Online
Faraday Discussions Paper

Fig. 1 Differences between the energies per particle of the FCC crystal and of the BCC
crystal, at equilibrium, versus the system volume fraction, f, for particles of average radius
hai ¼ 8 nm, Debye length lD ¼ 3 nm, density of charge ¼ 0.4 nm2. Black circles are the
exact calculation for the monodisperse case (d ¼ 0); red squares are from the Metropolis
Monte-Carlo method, 500 000 random particle-exchanges for each of the 50 realizations
of the random radius-distribution with polydispersity d ¼ 0.15, on the 2662 sites of the BCC
crystal, or the 2916 sites of the FCC crystal. Positive values of the energy difference
indicate that the BCC structure should be the stable phase at equilibrium. Then, poly-
dispersity favors the BCC structure for the high volume fractions.

This is a sign of the correlated organization of the charges in the system,

decreasing the energy of the BCC crystal more efficiently than for the FCC structure.

4 Q2: how do the polydisperse particles

distribute in a system with the BCC crystal phase
coexisting with FCC crystal phase?
A2: to understand why BCC becomes the stable phase when polydispersity is high
enough, we consider the problem of the 2-phase system (coexistence): let us then
suppose that the population of the polydisperse particles is arranged in part as
a BCC lattice and in part as a FCC lattice, with proportions rBCC and rFCC ¼ 1 
rBCC, respectively.
To know the thermodynamic equilibrium of the system, we use a variant of the
Gibbs ensemble Metropolis random exchanges of particles, introduced to study
the thermodynamic evolution of different phases coexisting in a liquid.9 Particles
are exchanged randomly between the phases according to the usual Metropolis
Monte-Carlo scheme, keeping constant throughout the total number of particles,
the total volume of the system and the temperature. In this method, the homo-
geneous coexisting phases are simulated in separated boxes without interfaces
between the phases, and each box has periodic boundary conditions. Particle
exchanges occur either inside a single phase, or from different phases.
To make the approach more efficient, we sampled the conguration space into
a nite number of congurations, for example taking the 19 congurations

232 | Faraday Discuss., 2016, 186, 229–240 This journal is © The Royal Society of Chemistry 2016
Published on 04 November 2015. Downloaded by Max-Planck-Institut für Dynamik und Selbstorganisation on 01/08/2016 15:07:32. View Article Online
Paper Faraday Discussions
corresponding to the coexistence of one cubic BCC crystal and one cubic FCC
crystal, the total sum of the sites being between 3500 and 4500 (in order to
compare systems of about the same total sizes). When thermodynamic equilib-
rium is reached in every conguration, the total energy, E, of the system is eval-
uated for each conguration. The program then scanned, little by little, the energy
landscape of (a sampling of) the conguration space, and the approximate
X
partition function Z ¼ expðE=kB TÞ gives access to all thermodynamic
conf
quantities at equilibrium. The average value of the proportion of the BCC phase at
equilibrium is estimated through:
1 X
hrBCC i ¼ r eE=kB T (2)
Z conf BCC

and similarly for hrFCCi.

Let us discuss the results for the value f ¼ 0.3 realizing the equality of the pure
BCC and pure FCC energies, as seen in Fig. 1. As could be expected, hrBCCi ¼
hrFCCi ¼ 0.5 is obtained in this case because of the choice EBCC ¼ EFCC for the pure
phases. However, the actual value of the energy per particle for BCC and FCC
crystals of the same size, is found to be E ¼ 22.2 kBT, signicantly lower than the
value for the single phase EBCC ¼ EFCC ¼ 23.0 kBT. This means that the system put
different particle distributions in the different phases. This occurrence is exem-
plied in Fig. 2 where the at-equilibrium radius distributions in the two phases
are compared for the conguration corresponding to a BCC crystal of 2000 sites,
and a FCC crystal of 2048 sites. The radius-distribution in the FCC phase has
approximately a Gaussian shape with a polydispersity of 6.5%, that is much
smaller then the polydispersity (15%) of the total distribution including all the

Fig. 2 At-equilibrium radius-distributions of the particles located in the phases of

a system made of a FCC finite lattice of 2048 sites and of a BCC finite lattice of 2000 sites,
both with periodic boundary conditions. The Gibbs Monte-Carlo scheme was applied with
300 000 random particle exchanges and 50 random realizations of the total radius-
distribution (plotted as the dashed curve) with polydispersity d ¼ 0.15, for the system
volume fraction f ¼ 0.3, all other parameters as listed in the caption of Fig. 1. The FCC
phase takes all the particles with radii close to the average value hai, letting the wings of the
distribution be managed by the BCC lattice. The polydispersity of the population of
particles making the FCC lattice is 0.065, much smaller than the polydispersity of the total
distribution, d ¼ 0.15.

This journal is © The Royal Society of Chemistry 2016 Faraday Discuss., 2016, 186, 229–240 | 233
Published on 04 November 2015. Downloaded by Max-Planck-Institut für Dynamik und Selbstorganisation on 01/08/2016 15:07:32. View Article Online
Faraday Discussions Paper
particles in the system. Consequently, both wings of the initial Gaussian distri-
bution distribute in the BCC phase.

5 Q3: why does the BCC crystal phase accept

handling two modes of particle radius while FCC
sticks to one mode?
A3: the detailed management of the initial radius-distribution by the system results
from the different abilities of the phases to accept particles of different radii (that is
here, of different electric charges). Following this idea, one can dene a parameter of
tolerance as the maximum relative decrease of the system energy when changing the
charges of a couple of particles, keeping the total charge unchanged. More precisely,
let us dene the tolerance, s, by the expression:
 
DE
s ¼ min  (3)
Uo

in which DE is the variation of the system energy when the charges of two particles
are changed from the initial values Z to: 2Z and 0, respectively, and Uo ¼ Z2klB is
a reference energy for the system. The minimum value in (3) is taken over all the
possible positions of the couple of particles with modied charge. This concept
supposes that the system is able to detect the best (from the point of view of the
energy) relative positions of the two concerned particles. Then the repulsive inter-
actions are expected to be long-ranged – as the Yukawa potential is – even if the
characteristic lengths are nite. For a regular lattice with only one sort of site, the
value of s is given by the expression:
ekro
s¼ (4)
kro
where ro is the value of the distance between the nearest neighbours.
 pffiffiffiffiffi 1=3
For the BCC lattice of volume fraction f, one has ro =a ¼ p 27=3f , and for
 pffiffiffiffiffi 1=3
the FCC lattice at the same volume fraction: ro =a ¼ p 32=3f . Then,
sBCC > sFCC for all values of f (for lD ¼ 3 nm and a ¼ 8 nm considered here:
sBCC/sFCC ¼ 1.3, for f ¼ 0.3). This indicates that the FCC crystal phase is less
tolerant than BCC, or, equivalently, that changing the charges of two particles as
follows: Z / 2Z, and Z / 0, results in a larger decrease of the BCC lattice energy
than in the FCC case. This is the reason why, from the point of view of the energy,
the system prefers to ll the FCC lattice with particles of almost the same charges,
and to arrange the rest of the particles in the BCC lattice.
As a conclusion, we propose here a ‘tolerance rule’ of the form: in the case of
coexistence of several phases, the less tolerant phase prevails over all the other
phases in the distribution of the radii.

6 Q4: what information can the method give

about experiments of compression of a system of
charged polydisperse particles?
A4: as explained in the Introduction, we recently explored the colloidal crystalli-
zation of polydisperse Ludox HS silica nanoparticles dispersed in water through

234 | Faraday Discuss., 2016, 186, 229–240 This journal is © The Royal Society of Chemistry 2016
Published on 04 November 2015. Downloaded by Max-Planck-Institut für Dynamik und Selbstorganisation on 01/08/2016 15:07:32. View Article Online
Paper Faraday Discussions
near-equilibrium dialysis to achieve very slow compression of the polydisperse
suspension.1 We demonstrated that, in charge-stabilized colloidal dispersions at
intermediate compressions, the system exhibits two crystalline phases, namely:
a BCC phase and a Laves AB2 phase, in coexistence with the liquid. Moreover, the
FCC phase is missing. The Laves AB2 crystalline system is known to be relevant for
10 11
the opal structure, and
ffi more generally for packing of bidisperse particles with
pffiffiffiffiffiffiffi
a radius ratio of 2=3x0:82. The tolerance of this phase, as dened in (3), is such
that: if sAB2/sBCC ¼ 3.55, in the experimental conditions, then the Laves lattice is
much more tolerant than BCC to accepting uneven particles. Therefore, an
argument similar to the one developed in Section B, leads to the conclusion that
the BCC phase should collect most of the particles close to the mode of the radius-
distribution, while the wings of the distribution should ll the AB2 phase.
To be more quantitative, the Gibbs ensemble Monte-Carlo method was used
for the system made of a BCC phase in coexistence with an AB2 phase. As previ-
ously, the ensemble average of each crystalline phase proportion is evaluated aer
sampling the conguration space. This results in the ensemble-average values:
hrBCCi ¼ 0.52; hrAB2i ¼ 0.48 corresponding to the respective volume proportions of
the phases at equilibrium. Then, considering a system close to these values, that
is: a BCC crystal of 9  9  9 unit cells, and a Laves AB2 crystal of 5  5  5 unit
cells, the resulting respective radius-distributions are plotted in Fig. 3.
Here too, we nd that the less tolerant phase (the BCC phase in this case)
collects most of the particles with radii close to the average radius. The nal
polydispersity of the BCC phase particles is 7.5% in this example, emphasizing
the role of the repulsive interactions in the process, since this polydispersity
signicantly exceeds the terminal value 5.7% of the polydispersity of the crystal
phase made of hard sphere colloidal particles precipitating from a uid phase at
equilibrium.12,13 Moreover, the AB2 phase is pretty well lled with particles with
a particle-radius ratio of about 0.8, the bigger particles being located in the
octahedral sites, and the smaller ones in the tetrahedral sites.
This suggests a simple “rule of thumb” for the respective proportions of the
two phases: suppose that the radii of the particles lling the less tolerant phase
are distributed in the Gaussian shape with polydispersity smaller than the overall
polydispersity d, and almost all the particles of radius hai belong to this phase.
Calculating the two modes of the remaining radius-distribution of the particles in
the high-tolerant phase is then a simple
pffiffiffiffiffiffiffi
ffi exercise. If we constrain the ratio between
the two radius modes to be 2=3 for the system to efficiently t the AB2 phase,†
then one nds the relation:
d*
rB x (5)
d
with rB the proportion of the less-tolerant phase at equilibrium, and
pffiffiffiffiffi
d* ¼ 5  24x10%. In (5), we approximated: 2 ln rB/(1  1/rB2) x rB.
Interpretation of eqn (5) is as follows: when d < d* only the less-tolerant phase
(here: BCC) can exist; when d > d*, the proportion of the BCC phase is a decreasing

† In principle, one should consider the ratio of effective radii based on the energy density. This does not
change eqn (5) but the value of the limit of polydispersity d* which generally depends on the system
volume fraction.

This journal is © The Royal Society of Chemistry 2016 Faraday Discuss., 2016, 186, 229–240 | 235
Published on 04 November 2015. Downloaded by Max-Planck-Institut für Dynamik und Selbstorganisation on 01/08/2016 15:07:32. View Article Online
Faraday Discussions Paper

Fig. 3 At-equilibrium radius-distributions of the particles located in the phases of

a system made of a BCC crystal of 1458 sites and of a Laves AB2 crystal of 1500 sites, both
with periodic boundary conditions. The Gibbs Monte-Carlo scheme was applied with
300 000 random particle exchanges and 50 random realizations of the total Gaussian
radius-distribution with hai ¼ 8 nm, polydispersity d ¼ 0.15, for the system volume fraction
f ¼ 0.3, all other parameters were as listed in the caption of Fig. 1. The BCC phase takes
most of the particles with radii close to the average value hai, letting the wings of the
distribution be managed by the AB2 lattice. The polydispersity of the population of particles
making the BCC lattice is 7.5% compared with the polydispersity of the total distribution,
d ¼ 15%. The two modes of the AB2 radius-distribution are the vertical red segments,
respectively at 6.9 nm for the tetrahedral sites, and at 9.3 nm for the octahedral sites. The
ratio between these two values is 0.74, close to the value 0.82 for the AB2 packing of hard
spheres. Compared with Fig. 2, one can easily understand that the system prefers the
present configuration for which all the modes are nearly at the correct place (unlike the
BCC phase in Fig. 2, which had to be filled with a bi-disperse distribution).

function of the polydispersity, d, of the system, that is: polydispersity favors the
search for more tolerant phases.
It is worth noting that the appearance of the Laves phase in this context can be
related to a work on bidisperse particles.14 Indeed, such complex crystal structures
were found to be spontaneously stabilized in the case of mixtures of spherical
particles with two different diameters, interacting via simple repulsive potentials.
This was shown by numerical simulations using the thermodynamic integration
technique. However, our case is a priori much more surprising since there is not,
at the beginning, any ‘magic’ radii ratio.

7 Q5: how long-ranged should the interaction

be between the particles?
A5: the physical parameters that we have considered up to now are rather typical
of short-range interactions, namely: an exponential decrease in (1) with a char-
acteristic length lD ¼ 3 nm, much smaller than the nearest-neighbor distances,
which are of the order 20 nm. Then, it is natural to ask the question whether only
the nearest-neighbors are important or if the farther neighbors play a role too.
A quick answer is given considering the monodisperse case for the BCC or FCC
lattices. In these cases, one knows the analytical formula for the energy per

236 | Faraday Discuss., 2016, 186, 229–240 This journal is © The Royal Society of Chemistry 2016
Published on 04 November 2015. Downloaded by Max-Planck-Institut für Dynamik und Selbstorganisation on 01/08/2016 15:07:32. View Article Online
Paper Faraday Discussions

Fig. 4 Energies at equilibrium per particle of the system of the BCC crystal phase in the
proportion 1  rAB2 coexisting with a Laves AB2 crystal phase in the proportion rAB2.
Equilibrium is reached after 300 000 random particle exchanges and 50 random reali-
zations of the Gaussian radius-distribution with polydispersity d ¼ 0.15, for the system
volume fraction f ¼ 0.3 and total number of sites between 2500 and 3500, all other
parameters as listed in the caption of Fig. 1. The open circles are the total Yukawa inter-
action, while the red triangles are the Yukawa interaction truncated at the nearest-
neighbors (the respective black and red continuous line are guides for the eyes). The black
(respectively red) dashed line is the energy of the pure BCC phase considering all the
neighbors (respectively only the first-neighbors). The statistical error bars are of the order
0.1kBT. When all the neighbors are included, the system energy reaches the minimum
value for rAB2 x 0.42 (to be compared to the average value hrAB2i x 0.48 found in Section
6 using the partition function). On the other hand, all the results including only the nearest-
neighbors are found to be above the energy of the pure BCC lattice, indicating that the
system should not consider the AB2 phase if the Yukawa interaction was truncated. As
a matter of fact, the difference between the nearest-neighbors/all-neighbors is essentially
important only for the less tolerant crystalline structure in this case (i.e. the BCC phase).

particle as a series in the order of the neighbors, then it is easy to compare the
energy when only the nearest-neighbors are considered. For the same parameters
as in the caption of Fig. 1 and f ¼ 0.3, only the rst-neighbors give the results:
BCC/kBT ¼ 24.98 < EFCC/kBT ¼ 29.79 which seems to show that the BCC crystal is
E(1) (1)

more stable than the FCC crystal. But including the second-neighbors turns
round that conclusion, as: E(2) BCC/kBT ¼ 30.47 > EFCC/kBT ¼ 30.32. Taking into
(2)

account all the neighbors conrms the latter conclusion: E(N)BCC/kBT ¼ 30.82 > EFCC/
(N)

kBT ¼ 30.51. Actually, this discussion was about the point d ¼ 0; f ¼ 0.3; (EFCC 
EBCC)/kBT ¼ 0.31 of Fig. 1. Then, even if lD/ro  1, one must consider more than
only the nearest-neighbors to get the correct conclusion about the stability of the
crystals.
To check the relevance of this ‘long-range versus short-range’ question in the
case of the coexistence of the BCC and AB2 phases of polydisperse particles, we cut
the Yukawa potential (1) right aer the rst nearest-neighbors (three kinds of sites
are to be considered: one for the BCC lattice, the tetrahedral site and the octa-
hedral site of the AB2 lattice). Then, the second- and farther neighbors of a given
particle do not interact with it any more. The main results are conveniently
visualized in Fig. 4, aer plotting the energies per particle of the system in phase
coexistence versus the composition of the system (characterized by: rBCC ¼ 1 
rAB2 and rAB2 as the respective volume proportions).

This journal is © The Royal Society of Chemistry 2016 Faraday Discuss., 2016, 186, 229–240 | 237
Published on 04 November 2015. Downloaded by Max-Planck-Institut für Dynamik und Selbstorganisation on 01/08/2016 15:07:32. View Article Online
Faraday Discussions Paper
The conclusion is that the number of neighbors considered is relevant in the
energy prole versus the composition of the system. In the example in Fig. 4, the
system is probably a pure BCC crystal phase if the Yukawa interaction is cut at the
rst-neighbors, while coexistence of the BCC and AB2 phases is clear when all the
neighbors are taken into account.

8 Conclusions – what did you learn from the

method?
In the case of the fractionated crystallization, the Gibbs ensemble Monte-Carlo
method does not predict the nature of the phases at equilibrium. However, it can
provide essential help for a quantitative description of the crystals in coexistence
(such as: the respective volumes of the phases, the particle size-distributions in
each phase, etc.) as well as the fundamental mechanisms leading to the nal
composition at equilibrium, within the framework of the DLVO theory. In the two
cases presented here (the FCC + BCC and the BCC + AB2 systems), spontaneous
self-assembly processes sort out particles from a broad polydisperse population
and build different sets of colloidal crystals with structures that are adapted to
a best use of this population. One process selects the most popular particles to
build a rst set of crystals (the less tolerant crystal) and “burns” a hole in the
original population of particle sizes. The residual particle population then
resembles a bi-disperse mixture, and a second process makes use of the
remaining sub-populations on either side of the “hole” to construct other crystals
(more tolerant). This process of sieving is probably general and leads naturally to
a sequence of fractionated colloid crystallization governed by the tolerance rule.
On the basis of what we know about binary nanoparticle systems,11,15 we can
expect still more complex superlattices to be built spontaneously that way for still
larger polydispersities.
This description of the mechanisms is close to the idea by Pusey16 in a different
context – an idea which has been validated much later through numerical
simulations17 and theoretical developments13 – that suspensions of highly poly-
disperse hard spheres might crystallize by splitting the broad overall distribution
into a number of narrower distributions, each of which being accommodated
within a single crystalline phase.
We shall close this presentation with three discussions about the limitations of
the numerical method.
 The method does not consider the real kinetics of the mechanisms, then
reliability of the method for large volume fractions may be questionable since
movements of the particles can be hindered. However, for systems of spherical
particles, repulsive interactions act as lubrication between the particles and
generate ergodicity. In other words, local rearrangement of the particles (related
to the system capability to reach thermodynamic equilibrium) is facilitated by the
repulsive interactions, and more precisely by the value of the energy at contact.
This is not the case in the opposite situation of attractive interactions, resulting
generally in non-equilibrium patterns.18
 In principle, the method cannot predict the phases at equilibrium, since it
supposes to have a priori information – or to guess – about the phases in coex-
istence. However, one should note that if large unit crystal cells are being formed

238 | Faraday Discuss., 2016, 186, 229–240 This journal is © The Royal Society of Chemistry 2016
Published on 04 November 2015. Downloaded by Max-Planck-Institut für Dynamik und Selbstorganisation on 01/08/2016 15:07:32. View Article Online
Paper Faraday Discussions
in the system, no numerical kinetic method (e.g. molecular dynamics) can
nowadays predict the phases, due to essential computing limitations.
 A last but fundamental question is the role of the liquid phase in coexistence
with the crystals. In principle, the same Gibbs ensemble Monte-Carlo method for
a system of two crystal phases plus a liquid phase can be used if a fast-computing
free energy method is at hand for the liquid phase. The optimized random-phase
approximation (ORPA) is a good candidate for a systematic study. This approach
could help to solve the similar problem with large polydispersities. Indeed, in this
case, the wings of the charge-distribution function are probably too wide to be
incorporated in the only Laves AB2 phase, and the system should choose the
liquid phase as a reservoir of untting particles (too large or too small), expressing
the high tolerance of the liquid phase. However, experimental data for such
systems with large polydispersity are scarce, and it is not clear whether such
a system will choose intensive use of the liquid phase (i.e. melting the phases) or if
it will nd another crystalline structure with a still larger crystal unit cell.

Acknowledgements
Part of this work was realized in the framework of the GDR AMC2 “Approches
Multi-Physiques pour les Colloı̈des Concentrés”, CNRS. We thank Christophe
Labbez and Guillaume Bareigts for interesting comments about the Monte-Carlo
numerical method used in this work.

References
1 B. Cabane, J. Li, F. Artzner, R. Botet, C. Labbez, G. Bareigts, M. Sztucki and
L. Goehring, Hiding in plain view: Colloidal self-assembly from polydisperse
populations, 2015, ArXiv: 1511.02493v1
2 F. Laves, Crystal Structure and Atomic Size, in Theory of Alloy Phases, The
American Society of Metals, Cleveland, OH, 1956, pp. 124–198.
3 A. Z. Panagiotopoulos, in Observation, prediction and simulation of phase
transitions in complex uids, ed. M. Baus, L. R. Rull and J. P. Ryckaert,
Kluwer Academic Publishers, Dordrecht, The Netherlands, NATO ASI Series
C, 1995, vol. 460, pp. 463–501.
4 E. Zaccarelli, S. M. Liddle and W. C. K. Poon, So Matter, 2015, 11, 324–330.
5 E. J. Verwey and J. T. G. Overbeek, Theory of the Stability of Lyophobic Colloids,
Elsevier, Amsterdam, The Netherlands, 1948.
6 I. Sogami and N. Ise, J. Chem. Phys., 1984, 81, 6320–6332.
7 Ordering and Phase Transitions in Charged Colloids, ed. A. K. Arora and B. Tata,
Wiley-VCH, New York, 1996.
8 M. O. Robbins, K. Kremer and G. S. Grest, J. Chem. Phys., 1988, 88, 3286–3312.
9 The Monte-Carlo Method in Condensed Matter Physics, Topics in Applied
Physics, ed. K. Binder, Springer-Verlag Berlin Heidelberg, New York, 1992,
vol. 71.
10 M. J. Murray and J. V. Sanders, Philos. Mag. A, 1980, 42, 721–740.
11 Z. Chen and S. O'Brien, ACS Nano, 2008, 2, 1219–1229.
12 D. A. Koe and P. G. Bolhuis, Phys. Rev. E: Stat. Phys., Plasmas, Fluids, Relat.
Interdiscip. Top., 1999, 59, 618–622.
13 M. Fasolo and P. Sollich, Phys. Rev. Lett., 2003, 91, 068301.

This journal is © The Royal Society of Chemistry 2016 Faraday Discuss., 2016, 186, 229–240 | 239
Published on 04 November 2015. Downloaded by Max-Planck-Institut für Dynamik und Selbstorganisation on 01/08/2016 15:07:32. View Article Online
Faraday Discussions Paper
14 M. D. Eldridge, P. A. Madden and D. Frenkel, Nature, 1993, 365, 35–37.
15 L. Filion and M. Dijkstra, Phys. Rev. E: Stat., Nonlinear, So Matter Phys., 2009,
79, 046714.
16 P. N. Pusey, J. Phys., 1987, 48, 709–712.
17 P. Bartlett, J. Chem. Phys., 1998, 109, 10970–10975.
18 G. Foffi, C. de Michele, F. Sciortino and P. Tartaglia, J. Chem. Phys., 2005, 122,
224903.

240 | Faraday Discuss., 2016, 186, 229–240 This journal is © The Royal Society of Chemistry 2016