Colloidal Clusters in Spherical Confinement

Kolloid-Cluster in Beschränkten Räumen

Der Technischen Fakultät der

Friedrich-Alexander-Universität Erlangen-Nürnberg
zur Erlangung des Doktorgrades
DOKTOR–INGENIEUR

vorgelegt von
Junwei Wang
aus Hunan, China
Als Dissertation genehmigt
von der Technischen Fakultät
der Friedrich-Alexander-Universität Erlangen-Nürnberg

Tag der mündlichen Prüfung: 27.09.2021

Vorsitzender des Promotionsorgans: Prof. Dr.-ing. Knut Graichen

Gutachter: Prof. Dr. Nicolas Vogel

Prof. Dr. Michael Engel
_______________________________

Kindness is our only way out

_______________________________
THIS PAGE IS LFFT INTENTIONALLY BLANK
 Abstract
Structure formation in finite confined space is fundamentally intriguing and technologically
relevant. In this thesis, we investigate the self-assembly of spherical colloidal particles in
spherical confinement. We first develop methods to equilibrate the crystalline colloidal
clusters in droplet confinement. We then establish a framework of magic numbers to
reveal the number – structure – property relationship for cluster sizes ranging from a few
hundred to a few hundred thousand. We discuss the emergence of magic numbers
associated with energy minimum cluster structures, the off-magic number structures with
defect accumulations and the breakdown of magic numbers due to geometric constraints.
In addition, we discuss formation pathways of colloidal clusters in icosahedral, decahedral
and octahedral symmetry. We demonstrate the anisotropic structural color and its use in
monitoring the dynamics of colloidal clusters in situ. At last, we investigate the mechanical
properties of colloidal supraparticles under compression, preparing for future applications
of such self-assembled particle-based materials.
 Zusammenfassung
Strukturbildung im confinement beschränkter Räume ist grundsätzlich faszinierend und
technologisch relevant. In dieser Arbeit wird die Selbstorganisation kugelförmiger
kolloidaler Suprapartikel (Kolloid-Cluster) in Emulsionstropfen als typisches Beispiel
eines solchen Strukturbildungsprozesses in räumlicher Beschränkung untersucht. Zuerst
werden Methoden entwickelt, um ein thermodynamisches Gleichgewicht in diesem
Selbstorganisationsprozess zu erreichen. Darauf aufbauend wird ausgehend von
Nummer – Struktur – Eigenschaftsbeziehungen von Kolloidclustern zwischen 100 und
100.000 Primärpartikeln gezeigt, dass solche Systeme einen magic-number Effekt
aufweisen. Dabei entstehen für bestimmte Systemgrößen Strukturen mit niedriger
minimaler freier Energie. Systeme zwischen diesen thermodynamisch-bevorzugten
Strukturen, sogenannte Off-Magic-Number-Kluster weisen definierte Defektstrukturen auf,
die im Folgenden untersucht werden. Schließlich wird gezeigt, dass die Bildung von
magic-number Strukturen ab einer kritischen Größe geometrisch nicht mehr möglich ist.
Neben thermodynamischen Gesichtspunkten wird die Kinetik dieser Strukturbildung
untersucht, deren Verständnis essentiell ist, um die Entstehung von Kolloidclustern mit
ikosaedrischer, dekaedrischer und oktaedrischer Symmetrie erklären zu können.
Experimentell wird die Beobachtung der anisotropen Strukturfarben solcher Kolloidcluster
als neue Methode zur Untersuchung der internen Struktur und des
Selbstorganisationsprozesses entwickelt. Schließlich werden auch mechanische
Eigenschaften von kolloidalen Suprapartikeln unter Kompression beleuchtet und
sozukünftige Anwendungen solcher selbstorganisierten partikelbasierten Materialien
vorbereitet.
Table of Contents
CHAPTER 1 ............................................................................................................................................................. 1

INTRODUCTION..............................................................................................................................................................1
Colloid crystallization ...........................................................................................................................................2
Structural color of colloidal crystal .......................................................................................................................6
Colloid crystallization in spherical confinements .................................................................................................8
Motivation ..........................................................................................................................................................13

CHAPTER 2 ........................................................................................................................................................... 15

EXPERIMENTAL METHODS .............................................................................................................................................15

Particles and surfactants ....................................................................................................................................15
Microfluidic fabrication ......................................................................................................................................15
Fabrication, storage and drying of emulsion droplets .......................................................................................16
Geometric modelling ..........................................................................................................................................17
Simulation of colloidal clusters...........................................................................................................................17
Characterization of colloidal clusters and supraparticles...................................................................................17
Indentation of colloidal supraparticles ...............................................................................................................19

CHAPTER 3 ........................................................................................................................................................... 20

FABRICATION OF COLLOIDAL CLUSTERS IN SINGLE AND DOUBLE EMULSIONS ...........................................................................20

Colloidal clusters in single emulsion droplets .....................................................................................................22
Colloidal clusters in double emulsion droplets ...................................................................................................27

CHAPTER 4 ........................................................................................................................................................... 33

MAGIC NUMBERS IN MINIMUM ENERGY COLLOIDAL CLUSTERS ............................................................................................33

Icosahedral colloidal clusters .............................................................................................................................33
Magic number colloidal clusters ........................................................................................................................41

CHAPTER 5 ........................................................................................................................................................... 47

A DETAILED INVESTIGATION OF FREE ENERGY LANDSCAPE OF COLLOIDAL CLUSTERS .................................................................47

Magic number region .........................................................................................................................................50
Off-magic number region ...................................................................................................................................54

CHAPTER 6 ........................................................................................................................................................... 62

THE BREAKDOWN OF MAGIC NUMBERS IN COLLOIDAL CLUSTERS ..........................................................................................62

Colloidal clusters with non-close shells ..............................................................................................................63
 Geometric requirement for shell closure ............................................................................................................64
Breakdown of magic numbers ...........................................................................................................................71

CHAPTER 7 ........................................................................................................................................................... 74

STRUCTURAL COLOR AS A TOOL TO INVESTIGATE STRUCTURES AND DYNAMICS ........................................................................74

Colloidal clusters with three types of symmetries ..............................................................................................75
Structural color of fcc clusters ............................................................................................................................78
Structural color of decahedral clusters...............................................................................................................80
Structural color of icosahedral clusters ..............................................................................................................81
Dynamics of colloidal cluster monitored by structural color ..............................................................................83

CHAPTER 8 ........................................................................................................................................................... 88

COMPETING MAGIC NUMBERS: ICOSAHEDRON VS DECAHEDRON ..........................................................................................88

Decahedral colloidal clusters ..............................................................................................................................89
Magic numbers in icosahedral and decahedral clusters ....................................................................................91
Formation pathways ..........................................................................................................................................93

CHAPTER 9 ......................................................................................................................................................... 100

MECHANICS OF COLLOIDAL SUPRAPARTICLES UNDER COMPRESSION ....................................................................................100

The fracture of colloidal supraparticles under compression ............................................................................102
The scaling between fracture and supraparticle geometry .............................................................................106
Influence of cohesive forces between primary particles ..................................................................................111
Influence of primary particle material..............................................................................................................115

SUMMARY ......................................................................................................................................................... 119

REFERENCE ........................................................................................................................................................ 123

LIST OF PUBLICATIONS ....................................................................................................................................... 135

ACKNOWLEDGEMENT ........................................................................................................................................ 138

 Chapter 1
Introduction
Understanding the structure of matter and its formation is a central theme in science. It
not only satisfies our curiosity but also improves our quality of life – because the structure
of matter determines its property, which we utilize to our benefits. From the early attempt
to probe the local structure of liquids by shaking ion balls in a basket, to studying
dislocation movement in crystalline solids using bubble rafts floating on water, we are
now able to look into the crystalline structure transformation of a nanoparticle with atomic
resolution, as well as investigate amorphous structures in films as thin as one layer of
atoms. Among many branches of research towards the understanding of structure and its
formation in both two- and three-dimensions, one class is particularly intriguing – the self-
organization of matter in confinement.

After the discovery of electrons, Thomson proposed the plum pudding model to describe
atoms based on what was known at that time – electrons carry negative charges and
atoms have no net charges. He postulated that “atoms consist of a number of negative
electrified corpuscles (his word for electrons) enclosed in a sphere of uniform positive
electrification”.1 This naturally led to the problem of determining the minimum energy
configuration of N repulsive spheres constrained at the surface of larger sphere. After the
discovery of positively charged nucleus, the plum pudding model was abandoned, but the
Thomson problem was generalized and remained.2,3 It has been solved for small number
of N, larger numbers are still unknown.

Another example is atoms condensing from gas to solid structures. At a certain point the
ensemble of atoms has an interface that separates an atom-rich phase and its
surrounding deprived of atoms. Now the atoms must self-organize under the confinement
of their own surface. The minimum energy structure is a balance of the energy
contribution from interior crystalline structure and the contribution from the surface.
Depending on the interatomic potential, number of atoms and the kinetics of condensation,
diverse structures and symmetries are found, which often differ from their bulk crystalline
structures.4

1
The self-organization of matter in confinement is also observed in natural minerals such
as framboids.5 These structures are quasi-spherical aggregates of up to 250 μm,
consisting of typically a few thousand polyhedral crystallites that exhibit order interior and
sometimes even icosahedral structures6. Framboids are commonly pyrites, the most
abundant sulfide mineral, found in sedimentary or volcanic rocks, dated back to as far as
the Pleistoncene Epoch about 2 millions years ago on earth.7 Magnetite framboids,
consisting of a few thousand particles, are found in meteorite formed 4 billions years ago,8
with indication of a vestigial water droplet confinement.9 However, attempts to synthesize
framboids with interior order in the lab were not successful up to date, and their formation
mechanism remains unclear.6

From atoms to minerals, to even biology such as cellular packing in embryos,10 or the
ribonucleoprotein assembly in the SARS-CoV-2 virus,11 structure formation in
confinement fascinates researchers from a wide range of fields. A major reason for such
diverse phenomena is the versatile interaction potentials of the self-assembling building
blocks – atoms interact in short range and strongly, mineral crystallites in longer range
and weaker, interaction in biological matter are anisotropic and the building blocks can
deform. To obtain a general insight of structure and its formation in confinement, it
perhaps helps to abstract the problem into simpler form. The simplest representation of
all the above-mentioned building block is a sphere; the simplest confinement is also a
sphere. Our question becomes – what is the behavior of a number of small spheres inside
a large sphere? This thesis tries to provide some answers.

Colloid crystallization

Structure formation, or in a broader sense, the transition between states of matter, is

governed by thermodynamics. The laws of thermodynamics tell us that a spontaneous
process is driven by the reduction of Helmholtz free energy:

H = U − TS

where H is the Helmholtz free energy, U , T and S are internal energy, absolute
temperature and entropy of the physical system, respectively. The internal energy U is
determined by the potential of the interacting entities. The magnitude, range and shape

2
of the interaction potential takes many different forms in atoms, molecules, polymers and
particles, hence creating rich variety of phenomena and the wonderful world we live in.

The simplest interaction potential is the potential of hard spheres, which are impenetrable
spheres that cannot overlap in space.

0, |𝑟1 − 𝑟2 | ≥ 𝜎
𝑉(𝑟1 , 𝑟2 ) = {
∞, |𝑟1 − 𝑟2 | < 𝜎

where the pairwise potential 𝑉 depends on the position 𝑟1 and 𝑟2 of the two spheres –
zero when their center to center distance is larger than sphere diameter 𝜎, and infinitely
large when smaller. When these non-deformable frictionless spheres are given inertia to
move freely in space, they collide and change moving directions to conserve momentum.
Since they cannot overlap, the internal energy U for hard spheres is zero. In an ensemble
of hard spheres, the packing fraction is defined as the total volume of the spheres divided
by the total volume of the space that the spheres can occupy. At low packing fraction,
hard spheres are randomly distributed in space. When the packing fraction increases to
a critical value, the spheres face two scenarios – to remain disorder (as sketched in
Figure 1a, left box) or to organize themselves into orderly in a crystalline lattice (as
sketched in Figure 1a, right box). Since hexagonal close packing is the densest packing
of spheres, the latter utilize space more efficiently, and have on average more space
around each sphere compared to the former case of randomly distributed sphere. Recall
that entropy is defined as:

S = k 𝐵 𝑙𝑜𝑔Ω

where k 𝐵 is the Boltzmann constant, and Ω is the number of microstates. For spheres
arranged orderly in a crystalline lattice, each sphere has more space for local movement,
where each tiny movement counts as one microstate, it has more microstates and hence
higher entropy than the ensemble of disorder spheres. Since internal energy U is zero, at
constant temperature, an increase of entropy S results in reduction of Helmholtz free
energy H. This means that for at this packing fraction, hard spheres transition from
disordered state to an order crystalline state spontaneously, demonstrated in one of the
first computer simulations in the 1950s, after which simulations start to take increasingly
important role in science.12–14 This phase transition is driven purely by entropy, since no

3
other forces are present. This hard sphere crystallization is illustrated in a computer
simulation where disorder spherical particles spontaneously organize into hexagonally
close packed lattice at 54% packing fraction (Figure 1a, left to right, particles with non-
crystalline surround is marked in red, with crystalline surround marked green, credit of M.
Engel).

The simplistic hard sphere model provides remarkable insights in the behavior of matter,
particularly for those whose repulsion becomes extremely large at close distance, such
as atoms and colloidal particles.

Colloidal particles in an aqueous medium have more complex interaction potentials than
the simplistic hard sphere potential. In the simplest case, spherical solid colloidal particles
can be made with accurate sizes and monodispersity via solution-based synthesis.15,16
Common materials involve silica,17 polystyrene,18 PMMA,19 which can be fluorescently
labeled and provide advantages of matching density20 to the continuous medium for long
time experiments or matching refractive index for optical imaging experiments. At close
distance, colloidal particles always experience attractive forces due to the omnipresent
van der Waals interactions.21 This attraction causes agglomeration or aggregation when
no repulsive forces are present. Since many solid surfaces are charged in water, the
electrostatic repulsion between colloidal particles provides a primary source for
stabilization. The DLVO theory21,22 describes the interaction of colloidal particles, taking
accounts of only the attractive van der Waals and repulsive electrostatic forces.

According to DLVO theory, the approximated energy between two particles (radius of 𝑅)
separated by a gap (x) follows:

𝐴 (x)
64𝜋𝑘𝐵 𝑇𝑅𝑐0 𝛼 2 −𝜅𝑥 𝐴𝐻 𝑅
𝑉 = ∙𝑒 −
𝜅2 12𝑥

where 𝑘𝐵 is the Boltzmann constant, 𝑇 is the temperature, 𝑐0 is the ion density, 𝜅 −1 is the
Debye length, 𝐴𝐻 is the Hamaker constant. 𝛼 depends on the potential of the charged
surface, for low surface potential:

𝛼 = e𝜓0 /4𝑘𝐵 𝑇

where e is unit electron charge, 𝜓0 is potential at the charged surface.

4
The Debye length 𝜅 −1 is given as:

2𝑐0 𝑒 2
𝜅=√
𝜀𝜀0 𝑘𝐵 𝑇

where 𝜀 is the relative permittivity, 𝜀0 is the vacuum permittivity.

It can be seen that the effective potential 𝑉 𝐴 (x) is a variable of separation distance and
consists of two parts. The first part is long ranged, which comes from electrostatic
repulsion and decreases exponentially with the distance to Debye length ratio. The
second part is short ranged, which comes from van der Waals attraction and increases
rapidly at small distance. The interparticle potential takes the form as illustrated in Figure
1b (bottom). The range and shape of the potential can be modulated to approximate the
hard sphere potential (Figure 1b, top) by influencing the electrostatics and matching the
refractive index to minimize van der Waals attraction.21

Better hard sphere approximation can be achieved by steric-stabilized colloidal particles.

In a pioneering experiment23 (Figure 1c), PMMA polymer particles (diameter of 300 nm)
were coated with a polymer stabilizing layer, whose thickness is about 5% of the particle
diameter. The particles are dispersed in non-aqueous solvents to reduce the long range
the electrostatic repulsion. The repulsive force is short ranged, provided by entropy
penalty by the intertwined polymer brushes at small particle distance. 24 After times
ranging from minutes to hours, concentrated stock dispersion stored in a thin glass
capillary started to show iridescent color when illuminated by a broad beam of white light.
This coloration is an indication of order and formed by constructive interference of light
beams that are reflected by the periodic lattices of the formed colloidal crystal, termed
Bragg-reflecting crystallite at the time (Chapter 2).

Since this early work, colloidal particles have been used as model for hard sphere to
investigate diverse phase behaviors such as crystallization, 25 structure transformation26
and glass transition27 (Figure 1d), providing insights also for statistical physics28 and
condensed matter.29

5
Figure 1, crystallization of colloidal particles. a, hard spheres crystallization driven by entropy. The order ensemble
(right) has higher entropy than the disordered (left). Spheres spontaneously transition from liquid (red) to crystalline
solid (green). b, sketch of hard sphere potential and potential of colloidal particles in dispersion described by the DLVO
theory. c, sterically-stabilized colloidal particle with near-hard-sphere potential exhibit constructive interference by
reflection from the crystal planes of the colloidal assembly when illuminated by broad white light. From [23]. Reprinted
with permission from Springer Nature. d, phase behavior of hard spheres. From [29]. Reprinted with permission from
Springer Nature.

Structural color of colloidal crystal

The above-mentioned Bragg-reflecting crystallite refers to crystalline arrangement of

colloidal particles that fulfill the Bragg’s law for constructive interference of light. As
illustrated in Figure 2a, when two beams of incident light of the same wavelength and
phase (dashed line) approaches a solid consisting of a number of building blocks at an

6
angle θ, they are reflected at each scattering bodies (in this case, colloidal particles,
represented by grey dots). When the Bragg condition:

2d sin 𝜃 = 𝑛λ

is fulfilled, constructive interface occurs, increasing the magnitude of the reflected light.
Here, the light path difference is determined by the spacing between two crystalplanes d
and incident angle θ to be 2d sin 𝜃. This length should be the integer multiple of the
incident light wavelength λ. When a broad white light consisting of different wavelengths
approaches the crystalline solid, only the wavelengths that constructively interfere are
observed. For colloidal crystals whose constituent particle size is around a few hundred
nanometer, the constructive interference occurs in the visible spectral range.30 When the
incident angle changes (or the viewing angle), the color of the colloidal crystal also
changes. This is known as iridescence, and the color arising from the interference by the
scattering of nanostructure is known as structural color. Iridescence has been observed
in nature31 in minerals such as opals,32 as well as biological material such as the scales
of butterflies33 and mimicked in the lab, often using colloids.30,34–36 Figure 2b shows a
typical colloidal crystal fabricated by spin-coating a dense colloidal dispersion on a
wafer.37 The crystallinity is evident by the periodicity of colloidal particles in cross-section
of the crystal revealed in the SEM image. Macroscopically, the colloidal crystal shows
intense iridescent color as illustrated in Figure 2c.

Since colloidal particles can be made with different materials with well-controlled shape
and sizes in large scale; their interaction tunable in liquid medium in precision, various
techniques have been developed to assemble colloids into two- and three-dimensional
crystals38 with applications in various fields.39 However, typical assembly methods such
as interface-templated methods,40 vertical deposition,41 or simply drying a dispersion on

7
glass slide produce bulk colloidal crystals where at least one dimension can be
considered infinitely large compared to the size of the colloidal particle.

Figure 2, Structural color of colloidal crystals. a, conditions for constructive interference in Bragg diffraction of light
reflected by a colloidal crystal. b, SEM image of the cross-section of a colloidal crystal. c, iridescence of the colloidal
crystal on a water. Reprinted with permission from [37]. Copyright [2004] American Chemical Society.

Colloid crystallization in spherical confinements

The crystallization of colloids is different in confinement, whose geometry limits the space
that colloidal particles can explore. On two-dimensional flat surface, spherical colloidal
particles self-organize into hexagonally close-packed monolayers, each particle having
six neighboring particles. A similar monolayer, where all particles are coordinated by six
neighbors can form on a cylindrical surface, as its zero Gaussian curvature allows it to be
spread into a flat surface. However, on a spherical surface, it is impossible for all particles
to have six closest neighbors due to the positive curvature. Each particle can be
considered as a vertex of a polyhedron and the connection between neighboring particles
can be imagined as the edge of the polyhedron. Euler’s convex polyhedral formula reads:

V+F−E=2

where V is the number of vertices, F is the number of faces, E is the number of edges.
Consider 60 particles assembled on a spherical surface (V = 60). If each particle has 6
neighbor (hence 6 edges) and an edge is shared by two particles, the number of edges
is E = 180. Similarly, if each particle is connected to 6 faces and each face is connected
to three particles, the number of faces is F = 120. Now since the sum of vertices and faces
subtracted by edges is zero, such a polyhedron does not exist according to Eulers formula.
This geometric constraint forbids all particles to have 6 neighbors. Instead, some must
have 5 or 7 neighbors. As demonstrated in Figure 3a, on a spherical surface, the particles

8
with 6 coordination number form grains with hexagonally close-packed structure, and
particles with 5 or 7 coordination number line up into “scars” at the boundary of these
grains.42 As required by topology, at least twelve of such disclination is required to fulfill
the Euler rule of polyhedral. While such defects for crystals at flat surface can be zero,
they are inevitable on spherical surface, whose size scales linearly with system size.42
The elastic strain caused by distortion of the hexagonal lattice prefers these disclinations
to maximize their distance.43 For stronger interacting particles on a spherical surface
(Figure 3b), defect islands are observed to form and arrange icosahedrally, and the
ordered particles can be mapped to an icosahedral surface to exhibit long-range
orientational order.44 In bold words, as hexagonal close packing is the ground state for
spheres at flat two-dimensional surface, the ground state at spherical surface has to
associate with icosahedral symmetry. When colloidal particles have strong tendency to
form hexagonal crystalline lattice, for example in the presence of small microgel particles
causing depletion attraction (Figure 3c), they form ribbon-like regions at the curved
spherical surface. Having topological defects or deviating from regular hexagonal lattice
similar to the case in Figure 3a and Figure 3b induces extra elastic deformation energy
induced by the curved spherical surface. Such elastic stress limits the growth of isotropic
domain to a maximum size, resulting in the ribbon-like domains that minimize the
curvature-induced elasticity.

Figure 3, Confinement influences the crystallization of colloidal particles in two-dimension. a, grain boundary
scars of colloidal particles at spherical surface. From [42]. Reprinted with permission from AAAS. b, spherical crystal
of icosahedral symmetry at spherical surface. From [44]. Reprinted with permission from Springer Nature. c, ribbon-
like crystalline domains minimizing curvature induced elastic energy. From [43]. Reprinted with permission from AAAS.

9
Even more interesting phenomena emerges when confining colloidal particles inside a
sphere. This is usually achieved by encapsulating colloids in a droplet, where the particles
can explore equilibrium structures as the droplet shrinks. For small numbers of particles
(N<10, 1 micrometer in diameter), the cluster structure is uniquely determined by the
number of particles and the wetting of particles at the droplet interface (Figure 4a).45,46
For example, four particles always form a tetrahedral cluster, five particles always a
triangular bipyramid, six particles an octahedron and seven a pentagonal bipyramid. For
hundreds to a few thousand nanoparticles (Figure 4b, 6 nanometer in diameter), clusters
with Lennard Jones geometries are found.47 For finite number of particles interacting with
such potentials, theory and simulation often predicts icosahedral or decahedral cluster
structure, which are indeed observed.

For even larger system sizes, such as a few thousand and even more particles, they
organize into crystalline clusters in droplets.48 As shown in Figure 4c, the clear five-fold
patterns on the surface (top and middle) indicates that they form clusters with icosahedral
symmetry. Hard sphere simulations in spherical confinement with sufficient equilibration
shows that the icosahedral symmetry can arise solely due to entropy. Calculation
indicates that for thousands of hard spheres, the icosahedral arrangement has lower free
energy than a single grain of face center cubic lattice in spherical confinement.
Tomography studies reveal that such a cluster has twenty twinned domains of tetrahedral
grains with face center cubic lattice. As the number of particles increases, face center
cubic clusters eventual prevails (Figure 4c, bottom). The global symmetry of such
crystalline colloidal cluster is influenced by the system size, as indicated in the diagram
in Figure 4d. For up to 100,000 of particles, icosahedral clusters are found, followed by
a gradual transition to fcc clusters.

However, clusters formed by sub-micron colloidal particles (instead of tens of nanometer

nanoparticles) in droplets often exhibit less crystalline order and a global spherical
symmetry. For more than thousands of hard sphere-like colloidal particles, spherical
clusters are often observed (Figure 4e).49,50 They exhibit a surface similar to the sphere
packing at spherical surface (Figure 3a), showing large domains of particles with 6
coordination number and additional grain boundary scars. Cross-section of such spherical
reveals that while regions near the surface consists of highly ordered concentric layers of

10
particles, the interior consist of disordered assembly of particles (Figure 4f). It is clear
that equilibrium has not been reached – these spherical clusters are kinetically arrested.
However, similar to sphere packing at spherical surface, crystallinity is not compatible
with spherical symmetry. Imagine stacking concentric shells of hexagonally close-packed
monolayer inside a sphere. Even though monolayers near the surface can conform to the
spherical curvature by introducing defects and tolerating strains, this curvature increases
significantly towards the center. In addition, the translational order across the shells is
missing. It can be inferred that given sufficient equilibration time, crystalline colloidal
clusters can form.

Figure 4, Confinement influences the crystallization of colloidal particles in three-dimension. a, small clusters
of unique symmetries determined by the number of particles. From [45]. Reprinted with permission from AAAS. b,
nanoparticle clusters with Lennard Jones geometry. Reprinted with permission from [47]. Copyright [2012] American
Chemical Society. c, icosahedral, rhombicosidodecahedral, and fcc nanoparticle clusters at different system sizes. d,
structural transition from icosahedral to fcc cluster as system size increases. From [48]. Reprinted with permission from
Springer Nature. e, kinetically-arrested spherical colloidal clusters. f, ordered surface region and disordered core region
in spherical colloidal clusters. From [49]. Copyright (2015) National Academy of Sciences.

The Mackay icosahedron model of identical sphere packing, as well as the anti-Mackay
derivative structure, provides an important basis to understand the crystalline cluster. 51,52
As illustrated in Figure 5, the simplest Mackay icosahedron construction consists of 12
spheres around a center sphere at the icosahedral vertices, which forms the first shell.
Building on top of the one-shell Mackay icosahedron, larger shells can be added
sequentially, following a stacking sequence. In Mackay icosahedron model, all shells are

11
stacked in ABCABC sequence, as in a fcc lattice, hence the twenty grains in the
icosahedron follows closely to the fcc lattice with slight deformation. This is illustrated in
Figure 5b and Figure 5c. Each shell requires a specific number of spheres to be complete.
The spheres on the Mackay icosahedron surface can be classified as three types: on the
vertices, on the edges and on the faces, each having a certain neighboring environment.
However, based on any existing Mackay icosahedron, larger shells can be stacked
without following the ABCABC sequence. This produces anti-Mackay icosahedron. As
illustrated in Figure 5d, the second shell is added to the existing one-shell Mackay
icosahedron (Figure 5a) in a ABA sequence, hence requiring different numbers of
spheres to complete such shell and exhibits different surface topology compared to the
case of ABC sequence in Figure 5b. Similarly, in Figure 5e, the third shell is stacked on
a two-shell Mackay icosahedron (Figure 5b) in a ABCB sequence, instead of a ABCA
sequence shown in Figure 5c. As a result, the third shell in anti-Mackay icosahedron only
requires 72 spheres (60 B type and 12 A type), compared to the 92 spheres (A type) in
the third shell of Mackay icosahedron. The anti-Mackay icosahedron in Figure 5e shows
a more complex surface topology and is in fact more spherical than the Mackay
icosahedron. Historically, anti-Mackay type stacking is only considered as the most outer
shell of the icosahedral cluster, but theoretically more shells can be stacked on top of an
anti-Mackay shell in ABCABC sequence, which will result in two twinned fcc domains.
The arrangement of spheres in icosahedral clusters deviates little from concentric
spherical shells of spheres as seen in Figure 4f but ensures that all spheres have local
environment closest to perfect crystalline lattice at the energy cost of thirty twinned faces.
Under the curved confinement, icosahedral symmetry is a good compromise of
crystallinity and sphericity. However, the geometry of icosahedral dictates that the edge
distance is 1.05 times the distance between the vertex to center, hence even though
icosahedral shells can be added indefinitely over a smaller structure, the strain at the
surface becomes increasingly unfavorable at large sizes.

12
Figure 5, Mackay and anti-Mackay icosahedral sphere packing model. a, Mackay icosahedral sphere packing of
one shell consisting of a central sphere (denoted as A in the ABC sequence of fcc lattice) and twelve spheres (denoted
as B in the ABC sequence of fcc lattice). b, Mackay icosahedral sphere packing of two shells. Additional 42 spheres
(denoted as C in the ABC sequence of fcc lattice) are added on top of a. c, Mackay icosahedral sphere packing of three
shells, the third shell consists of 92 A-type spheres. d, anti-Mackay icosahedral sphere packing of two shells. The
second shell does not follow the ABC stacking sequence, instead, it is stacked in ABA manner. Hence the second shell
consists of 32 spheres and exhibit different surface structures. e, anti-Mackay icosahedral sphere packing of three
shells, the third shell is a stacking fault on top of structure b, making it ABCB sequence, hence consisting of 72 spheres
instead of 92 spheres in the case of c. From [52]. Reprinted with permission from Springer Nature.

Motivation
The roles of geometry, packing and entropy are now widely acknowledged for colloidal
matter.53 However, when confinement is involved, an implicit theme is hidden in plain
sights – the numbers. For a bulk system, particle numbers can be considered infinite and
adding a few thousand more particles to the system would not change its properties.
However, confinement not only limits the space in which colloidal particles can explore,
but also sets a finite number of building blocks to self-organize. Despite our current
understanding of colloids in spherical confinement. Can we understand the cluster
structure in depth, apart from its global symmetry? How do flat facets and sharp vertices
of crystalline clusters adapt to the curvature of the confinement? How do crystalline
clusters tolerate defects? Are all crystalline clusters structurally the same? Must
crystalline clusters have icosahedral symmetry? What is the role of kinetics during
structure formation? Experiments of small clusters (Figure 4a) already showcase that

13
number dictates structure formation in small confined systems but understanding the
number – structure – property relationship in detail for larger systems (Figure 4) face
tremendous difficulty. To start with, the number of permutations increases rapidly with
system size, making it impractical to numerate possible arrangements for even a hundred
particles, nor does it carry much meaning since there may be many competing metastable
states with little energy difference. It also carries little significance to investigate any
particular number – the principle of the number effect for colloids is important, the
behavior of a particular number is not. Here comes the dilemma – how to understand the
whole without understanding the parts. A new framework is needed.

In this thesis, I will introduce the concept of magic numbers to colloidal clusters in
spherical confinement, with a focus on geometric sphere packing, to explain their number
– structure – property relationship.

The thesis will be organized as followed. Chapter 2 describe the experiments and
methods. Chapter 3 discuss the requirements for successful fabrication of colloidal
clusters. Chapter 4 discuss the structure of icosahedral colloidal clusters in detail and the
emergence of magic numbers in finite system with purely entropic interaction. Chapter 5
discuss the free energy landscape of colloidal clusters across a wide size range size via
magic and off-magic numbers, including the influence of defects. Chapter 6 discuss the
breakdown of magic numbers in colloidal clusters as the system size approaches the bulk
limit. Chapter 7 discuss the use of structural color to identify clusters of other competing
symmetries and monitor their formation dynamics. Chapter 8 discuss the dominant role
of kinetics over thermodynamics in the formation of colloidal clusters to bias against non-
icosahedral symmetries. Chapter 9 investigates the mechanical stability of spherical
colloidal clusters under compression and relates pivotal mechanical properties to their
geometric features.

14
 Chapter 2
Experimental Methods

Particles and surfactants

Chemicals were purchased from Sigma Aldrich and used as received, including styrene,
acrylic acid, and ammonium peroxodisulfate. Surfactant free emulsion polymerization
was used to synthesized polystyrene colloidal particles of various sizes using acrylic acid
as comonomer and ammonium peroxodisulfate as initiator in following literature
protocols.40 After synthesis, the particle dispersion was either centrifuged by ethanol
water mixture (1:1) several times or kept in dialysis tube for cleaning. A range of particles
were used, the diameter from 150 nm to 1000 nm. They were kindly provided by Houda
Ichanti, Denise Niederkorn, Salvatore Chiera, Umair Sultan, Gudrun Bleyer and Giulia
Magnabosco from our group. Fluorescently labeled polystyrene particles (500 nm, 750
nm) were purchased from PolySciences and used as received. Several fluoro-surfactants
were used. Krytox 157 FSH (Perfluoropolyether carboxylic acid) was purchased from
Costennoble (Dupont). PFPE-PEG-PFPE (perfluoropolypropyleneglycol-block-
polyethyleneglycol-block-perfluoropolypropyleneglycol) was purchased from Creative
PEGWorks. Surf-O (PFPE-Jaffamine-PFPE, perfluoropolypropyleneglycol-block-
poly(propylene glycol)-block-poly(ethylene glycol)-block-poly(propylene glycol)-
perfluoropolypropyleneglycol-block) was kindly provided by a friend. Surf-W (same as
PFPE-Jeffamine-PFPE) was synthesized by Tobias Salbaum and Yaraset Galvan and
purified by me following recipes in the literature.54,55

Microfluidic fabrication

PDMS microfluidics devices were produced following typical soft lithography methods as
described in literature.49 In short, a silicon wafer was spin-coated with SU-8 negative
photoresist, a pattern mask was used to create microstructures on the coated water
surface through UV light. The microstructures were then hardened and coated with an
anti-sticking layer to produce the master wafer. Polydimethylsiloxane (Sylgard 184 PDMS
from Dow Corning) was mixed with curing agent at 10:1 weight ratio and degassed before

15
pouring onto the master wafer for approximately 1 cm thickness. The PDMS was cured
in 80°C oven overnight. The cured PDMS chip was cut off by blazer and peeled off from
the wafer. Biopsy punch (1.0 mm in diameter, Kai Group) was used to create the inlets
and outlets in the PDMS chip. Chips with 2-dimenionsal flow focusing of 15, 25, 50, 100
and 120 μm were used. A scotch tape was used to remove debris and dust on the PDMS
chip surface. The chip was washed by ethanol and water (Milli-Q) and dried with
compressed air. After cleaning, the PDMS chip and a clean glass slide was plasma
treated for 30 seconds in oxygen environment at 30 W power (Diener electronic, Femto).
After surface activation, the PDMS chip was bonded to the glass slide and put in oven for
30 minutes to enhance bonding. Afterwards, the microfluidics channels were immediately
flushed with Aquapel (PPG industry) or 2 wt% fluorosilane dissolved in HFE 7500 oil (3M)
by injection through a 1 mL syringe of a flat-head suiting needle (diameter slightly larger
than 1mm). After 1 hour, the channels were flushed by compressed gas to remove the
liquid and kept in oven to dry.

Fabrication, storage and drying of emulsion droplets

The fabrication of monodisperse droplets uses the PDMS microfluidic chip with two inlets
and one inlet. Inlets were connected to the syringe tube via medical grade polyethylene
tube. The sizes of the PDMS inlet, tube, needle, tube diameter needs to be chosen
accordingly. Typically, the inlet has 1.0 mm diameter, the tube has 1.2 mm outer diameter
and 0.4 mm diameter. The needle needs to be slightly larger than the tube to both fit in
and not slide out during usage. The syringe was placed and pushed by precision pump
(Cronus) to ensure constant flow rate. For chips channels of 20, 50 and 100 μm, typically
100 uL/hr to 400 uL/hr flow rate for water phase is used. For oil phase, typically 100 uL/hr
to 1000 uL/hr flow rate was used. Colloidal particles are dispersed in water phase (usually
1 wt%). Typically, 0.1 wt% to 2 wt% fluoro-surfactant (Figure 6)were dissolved in HFE
novec 7500 oil (3M). Emulsion droplets were produced at the cross-junction in the
microfluidic channel (Figure 6) and collected at the outlet, either through a pipette tip or
through a tube into a glass vial. Typically, 1.5 mL glass vial was used to store the water
in oil droplets. Drying of droplet is controlled by either closing the vial cap, or closing the
vial opening with parafilm with holes punched by needle. The glass vial is stored in room

16
temperature or in the fridge at 5°C. For the double emulsions, the droplets were stored in
5 mL, 10 mL glass vials or 10 mL, 50 mL centrifuge tubes with closed cap.

Geometric modelling

The coordinates in the geometric sphere packing model proposed by me and generated
in a Python code. Illustrations were drawn in GeoGebra Classic 6.

Simulation of colloidal clusters

Simulations were performed by Chrameh Fru Mbah in Michael Engel’s group. The cluster
structures were examined in a Java program written by Michael Engel by Chrameh and
me. Event driven molecular dynamics was implemented in C++ for hard spheres in rigid
spherical confinement with in-house code details described in literature.56,57 In this step,
the particles in simulation does not have interparticle interaction, nor any potential with
the non-deformable spherical confinement. The droplet shrinking process in the
experiment is mimicked by slowing reducing the confinement radius, which increases the
packing fraction of hard particles until crystallization. The confinement radius continues
to be reduced further to reduce interparticle distance in the crystalline cluster. Clusters of
icosahedral, decahedral, and octahedral symmetries are found. In the second step,
numerical quenching was realized by fast inertial relaxation engine (FIRE)58 implemented
in HOOMD-blue59,60 which effectively mimicked capillary forces that consolidate colloidal
clusters in the final stage of droplet drying. This step is only necessary to obtain the
faceted cluster surface as observed in the experiments but does not alter the overall
symmetry and barely change the crystallinity of the structure obtained from the first step.
Absolute free energy of a colloidal cluster at packing fraction ϕ was calculated with the
Einstein crystal method61 applied to hard spheres in hard spherical confinement
implemented in a Monte Carlo simulation using harmonic springs.

Characterization of colloidal clusters and supraparticles

Oil droplets containing colloidal supraparticles were drop-casted on a silicon wafer and
examined under scanning electron microscopy (Zeiss Gemini Ultra 50 SEM, 1 kV, 5 mm
working distance). The sample transfer was performed by Thomas Przybilla with my
assistance. Tomography study was performed by Thomas Przybilla and Silvan English

17
with the help of Benjamin Apeleo Zubiri in Erdmann Spiecker’s group.56,57,62,63 For
electron tomography, the colloidal suparparitcles were deposited onto standard 200 mesh
Lacey carbon copper grid. After drying of solvent, solid supraparticles remained on the
grid. Typically, the grid together with deposited samples were examined in SEM to identify
suitable samples. The grid with selected was mounted onto the ultrathin single-tilt
tomography holder (Fischione model 2020) and transferred to the transmission electron
microscope (TEM, FEI Titan3 Themis 60-300). Scanning transmission electron
microscopy (STEM) tomography was performed using a dual probe- and image-side
aberration-corrected TEM at an acceleration voltage of 300 kV in high-angle annular dark
field (HAADF) STEM imaging mode at a camera length of 91 mm. The tilt series was
acquired using FEI Tomography 4.0 software in a tilt angle range from -76° to 62° with 1°
tilt increment, continuous and linear tilting scheme. Tilt series alignment was performed
using FEI Inspect 3D software (cross-correlation technique). Reconstructed volumes
were visualized with VSG Avizo 8.1 for FEI systems software. For 360° tomography, the
selected colloidal supraparticles were transferred onto a pre-prepared FIB-cut micro pillar
by stamping transfer technique, which was later mounted on Fischione Model 2050 On-
Axis Rotation Tomography holder (E.A. Fischione Instruments, Inc.) before imaging in the
TEM. The tilt series has a full tilt angle range of 180° with 1° tilt increment. Continuous
and linear tilting scheme and auto focus and tracking was used before acquisition. X-ray
tomography was performed on a ZEISS Xradia 810 Ultra X-ray microscope with a 5.4 keV
rotating anode Cr-source and a Zernike phase ring for phase contrast imaging of down to
50 nm optical resolution. Several clusters of one ensemble were prepared on the edge of
a sticky carbon pad to find one the rare decahedron samples by transmission imaging in
the microscope. The chosen sample was transferred on to a needle tip in the pre-
alignment light microscope of the X-ray microscope for 360° tomography without missing
wedge. Some gold sphere particles on the top of the cluster improve image alignment.
For the datasets, a tilt series with a total number of 901 transmission images was recorded
with an acquisition time of 400 s/frame. The image series is aligned along the rotational
axis by manual gold sphere tracking and the complete 2D dataset is relocated to fit the
reconstruction geometry. Acquisition and alignment were performed in the native ZEISS
microscope software (XMController and Scout&Scan) 3D reconstruction was performed

18
employing the Simultaneous Iterative Reconstruction Technique (SIRT) algorithm
implemented via an in-house Python script based on the Astra-Toolbox. Arivis Vision4D
and InViewR were used for visualization, segmentation and quantitative 3D analysis.

Indentation of colloidal supraparticles

Experiments were performed and initially processed by Andreas Ströbel in the group of
Matthias Göken, with the help of Patrick Feldner under the supervision of Benoit Merle.
The in-situ measurements in SEM were performed by Jan Schwenger, Patrick Herre
under the supervision of Stefan Romeis from Wolfgang Peukert’s group. I further
processed and analyzed the data. Nanoindentation on the colloidal supraparticles were
performed in Agilent Nanoindenter G200 with 0.01 nm depth resolution and 50 nN load
resolution in displacement-controlled. The flat diamond indenter tip has a diameter of 90
μm (Synton MDP). The diameters of the colloidal supraparticles were measured in SEM.
For the indentation measurements, the samples were deposited onto a piece of silicon
wafer directly from the fluorinated oil phase. The silicon wafer was glued to the sample
holder with a minimum amount of superglue (Ultra gel, Pattex). After fixing the wafer, the
sample were left for 24 hours. After mounting the sample holder on the indenter stage, at
least one-hour equilibrium time was allowed after closing the indenter chamber. After
calibration of the tip to the built-in microscope, a test indent was performed on bare silicon
wafer to position the tip to the substrate. In a measurement, the tip advanced towards the
sample at maximum speed until 2 μm above the sample and changed to fixed speed. The
loading and unloading speed were fixed at 50 nm/s, maximum displacement at 60% strain,
peak holding time for 10 s. A double-sided tape was attached to the sample holder to
facilitate cleaning of the indenter tip and remove possible particle residues after every
measurement. Measurements in vacuum were performed with a customized setup inside
an SEM with a nanomanipulator and a loading cell, using a mixture of force- and
displacement-controlled mode. The detail of the device is described elsewhere. 64 The
force-displacement curve was analyzed in a Python code written by Andreas Ströbel.

19
 Chapter 3
Fabrication of Colloidal Clusters in Single and Double
Emulsions
Manuscript based on parts of the results described in this chapter will be submitted as:
Junwei Wang, Simon Hahn, Esther Amstad, and Nicolas Vogel
Thin Shell Double Emulsion by Centrifugal Shearing Atomization (in preparation)

I performed the experiments, S. Hahn assited the experiments, conventional double

emulsion microfluidic chips and fabrication technique are provided by E. Amstad, N. Vogel
supervised the work.

We produce colloidal clusters using water in oil emulsion droplets as templates. To control
the number of particles in the cluster, in other words, the cluster size, the droplet size
must be uniform. Therefore, we resort to PDMS microfluidics which fabricates
monodisperse droplets.65,66 As illustrated in Figure 6a, an aqueous colloidal dispersion
of certain concentration, usually 1 wt%, is injected into the inlet (blue), whose flow rate is
controlled by a precision pump. The perfluorinated oil phase (HFE 7500) is injected from
another inlet (white) from the sides which breaks up the water phase into dispersed
droplets and carries them towards the outlet, where we collect the fluid sample in a pipette
and transfer to glass vial for storage and drying. As the viscosity and the interfacial tension
between the dispersed and continuous phase is constant, the droplet sizes are controlled
conveniently by flow rate of the water and oil phase, as well as the channel dimension.
Perfluorinated oil is chosen due to its compatibility with PDMS chip, as hydrocarbon oil
molecules enters the porous polymer network, which swell the PDMS and eventually alter
the channel dimension, or even delaminate the chip from the substrate. 67 To stabilize the
droplet against coalescence, a fluorosurfactant is dissolved in the oil phase which adsorb
at the droplet interface to provide steric repulsion. We use two types of surfactants. One
is an anionic surfactant54 (PFPE-COOH, commercially known as Krytox, Figure 6c), the
other is a non-ionic surfactant PFPE-PPO-PEO-PPO-PFPE (Figure 6d). Both surfactants
have the identical fluorophilic polymer block (note that the non-ionic surfactant is a triblock

20
co-polymer with two fluorophilic blocks). The hydrophilic part of the former surfactant
consists of a single carboxyl group, which provides electrostatic repulsion against anionic
spices, the latter by the water-soluble polymer which provides steric repulsion. We note
that the non-ionic surfactant is commonly synthesized using the anionic surfactant as the
source of the fluorophilic block.55 Due to incomplete reaction, chemical equilibrium or
purification limits, the non-ionic surfactant typically contains some traces of the anionic
surfactant. Both surfactants are efficient in stabilizing the water droplets against
coalescence at 0.1 wt% concentration, during the microfluidic droplet fabrication, the
subsequent handling and storage of the sample. However, they have different effect on
the formation of colloidal clusters.

Since the fluorinated oil has higher density than water, the water droplets assemble at the
top of the continuous phase. The oil phase containing particle-laden water droplets is
usually kept in glass vials where evaporation of both water and oil take place
simultaneously. With sufficient volume of oil, water droplets dry out completely before the
continuous phase dries out. As sketched in Figure 6b, the droplet now contains a certain
number of colloidal particles. The numbers of particles can be controlled by tuning the
droplet size or the dispersion concentration. However, we note that the exact number of
particles are not identical in each droplet due to fluctuation of particle concentration in
space, as well as slight variation of droplet sizes. The volume of droplet continuously
decreases during drying, eventually consolidating the encapsulated colloidal particles into
a solid colloidal cluster suspended in the oil.49

21
Figure 6, Water in oil droplets as templates for colloidal clusters. a, flow-focusing PDMS microfluidic device
produces water in perfluorinated oil droplets in uniform sizes. b, the number of encapsulated colloidal particles can be
controlled by droplet size or dispersion concentration. Evaporation of the droplet increases the packing fraction of
colloidal particles and eventually consolidate them into a colloidal cluster. The droplet can be stabilized by either an
anionic surfactant PFPE-COOH (c), or a non-ionic copolymer surfactant PFPE-PPO-PEO-PPO-PFPE (d). The
fluorophilic block is marked in green, the hydrophilic block blue.

Colloidal clusters in single emulsion droplets

One prerequisite of the successful formation of colloidal clusters is that the colloidal
particles should stay inside the droplet - not adsorbing to the droplet interface or migrating
from the droplet to the continuous phase. As shown in Figure 7a, we encapsulate
negatively charged PS colloidal particles (carboxyl groups on surface, d = 250 nm) in
droplets encapsulated by the non-ionic surfactant at 5 wt%, an unusual high concentration
to probe the extremity. Under light microscope in transmission mode, the particle scatters
light randomly in water and dims the intensity arriving at the objective, thus droplets
containing particles appear less transparent than pure water droplets. As marked by the
blue circle, particles (as black dots) are rapidly leaving the droplet. After 30 seconds, the
droplet becomes more transparent as most particles have migrated from water to
continuous oil phase. This particle migration is unexpected because the particles are
more hydrophilic than fluorophilic, and large hydrophilic block in the non-ionic fluoro
surfactant should provide sufficient steric repulsion at 5 wt% (more than the maximum
droplet interface adsorption coverage).55 We rule out the contribution of electrostatics

22
when similar particle migration is observed for positively charged PS particles (amino
groups on surface, d = 600 nm, credit of S. Ciera) in the same condition (Figure 7c). As
a reference, we pipette a 2 μL colloidal dispersion drop in the oil without any surfactant
(only one droplet per sample is possible, as more droplets coalescence upon contact),
which show no such particle migration.

Figure 7, Migration of positively and negatively charged colloidal particles across water-oil interface stabilized
at high non-ionic surfactant concentration. a, negatively charged PS colloidal particles migrate into the continuous
oil phase through droplets interface stabilized by 5 wt% PFPE-PPO-PEO-PPO-PFPE non-ionic surfactant (indicated
by blue circle). b, the droplet appears more transparent after 30 seconds, as the encapsulated particles have left. c,
same particle migration observed for positively charged PS colloidal particles.

We reduce the non-ionic surfactant concentration significantly to 0.1 wt% and observe no
particle migration across the droplet interface. However, after droplet drying, the
consolidated colloidal assembly appears like a crumpled sheet of paper (Figure 8a). This
morphology is likely caused by strong and irreversible particle adsorption at the droplet
interface, which forms a dense monolayer undergoing thin-sheet buckling when the
droplet continuous to shrink.68–70 As our colloidal particles are not cross-linked polymers,
after storing the dispersions for more than a few months, we suspect that short chain
polymer or oligomer formed during the radical polymerization synthesis may remain in the
particle and diffuse into water phase slowly over time. 71 These ionic spices contain a
hydrophobic PS block and a hydrophilic charged group, 71 which may also be surface-
active and aggravate the particle adsorption to the droplet interface in synergy with the
fluorosurfactant already present at the droplet interface. Indeed, after centrifuging the

23
particle dispersion to remove traces of such spices (water ethanol mixture at 1:1 volume
ratio, particles are sedimented in the centrifuge tube and re-dispersed in deionized water
after supernatant removal), the colloidal assembly formed after droplet drying becomes
more spherical (Figure 8b), even though the surface is rough and does not show any
ordering. This suggests that the tendency for particles adsorption at droplet interface is
weakened, forming no rigid monolayer at early stages of the drying process at large
droplet size. However, particle adsorption seems to occur at a later stage of drying, when
the droplet has shrunk to a smaller size. It is likely that in this stage, the impurities have
concentrated, and the particles collide more frequently with the interface. After cleaning
the dispersion by three or more centrifugation-redispersion steps, the assembly not only
has a spherical outline (Figure 8c), but also exhibits a smooth ordered surface consisting
of domains in hexagonal close-packed lattice separated by grain boundary scars, similar
to sphere packing at two-dimensional surface (Figure 3a).

Figure 8, Impurity in colloidal dispersion controls the colloidal cluster morphology. a, buckling of colloidal
clusters with a crumpled sheet of paper morphology due to strong particle adsorption to the droplet interface at 0.1 wt%
non-ionic surfactant concentration. b, colloidal clusters with rough surface and deformed spherical morphology due to
weakened particle adsorption to the droplet interface after purifying the dispersion once using the centrifuge. c,
spherical colloidal clusters with smooth surface of hexagonal close packing and grain boundary scars after purifying
the dispersion three times using the centrifuge. This morphology indicates that particle adsorption to the droplet
interface is prevented.

We find a simple solution to avoid particle adsorption to water-oil droplet interface by

adding anionic surfactant Krytox to the continuous oil phase. A mixture of 20% of anionic
surfactant prevents particle adsorption at the droplet interface completely when the
particle is negatively charged. Higher mixture ratio or pure anionic Krytox surfactant

24
sufficiently prevents negatively charged colloidal particles from adsorbing to the droplet
interface due to electrostatic repulsion. Under these conditions, colloidal clusters always
appear spherical and show highly ordered surface, without additional purification steps.
The morphology of these clusters is identical to the morphology shown in (Figure 8c),
which we consider successful spherical colloidal cluster.

The above experiments indicate that particle tend to adsorb to the droplet interface,
facilitated by both the non-ionic surfactant in the continuous oil phase and possibly the
amphiphilic impurities in the dispersed water phase. At high non-ionic surfactant
concentration, particle migrate towards the oil phase; at low concentration, particles
adsorb irreversibly at the interface. This correlation suggests that it is not the typical
scenario of particle adsorption at water hydrocarbon oil interface without surfactants. 72,73
There are evidences that such fluorosurfactant can transport cargos in the form of
micelles.74 Yet in our experiments, even 1 μm particles have been observed to migrate
through the droplet interface, a size much too large to be explained by a micellar transport.
We postulate that hydrophobic interactions may favor the PPO-PEO-PPO block of the
surfactant to adsorb at the PS particle surface (as both PPO and PS are hydrophobic),75–
77 which renders the particle at least partially fluorophilic in the oil phase. However, we
also observe particle adsorption at droplet interfaces stabilized by a commercial PFPE-
PEG-PFPE surfactant, which has a decreased hydrophobicity in the segments
penetrating into the water phase. Nevertheless, electrostatic repulsion of the anionic
surfactant proves an efficient way to prevent the negatively charged colloids to adsorb at
the water-oil droplet interface. The mechanisms of particle adsorption to interface in the
small volume of a droplet, as well as its synergy with impurity and surfactant at the
interface is an interesting question that requires more detailed investigation in the future.

The second prerequisite for successful formation of colloidal cluster is sufficient

equilibration time, determined by the time to complete droplet drying. We observe four
distinct cluster morphologies at different droplet drying rates (Figure 9a–d). With fastest
evaporation, buckled clusters form as the droplet interface moves faster than colloidal
particles can consolidate (Figure 9a). When the evaporation rate is lowered, spherical
clusters dominate (Figure 9b). Spherical clusters exhibit a uniformly curved surface with
only weak crystalline order. Partial crystalline clusters with incompletely developed five-

25
fold symmetry pattern at the surface (Figure 9c) form with further decreasing evaporation
rate. Very slow evaporation provides sufficient time for the colloidal particles to arrange
into crystalline clusters (Figure 9d), which are characterized by a fully developed pattern
of five-fold axes at the surface. The crystalline structure will be discussed in detail in
Chapter 4. The uniformity of the colloidal clusters enables statistically evaluation of the
evaporation rate-dependent cluster formation: the fraction of crystalline clusters increases
from almost zero (Figure 9e) up to 75% (Figure 9f) as evaporation is slowed down
(Figure 9g). Importantly, at the lowest evaporation rate, the dominant species of the
observed clusters evolve from buckled to spherical to crystalline with increasing assembly
time (Figure 9h). Note that the unusually-long 40 days of droplet dying does not mean
that colloidal particles in one droplet reaches equilibrium in 40 days, depending on the
number of particles, full crystallization can be completed in a few days. However, even
though the droplets are uniform in size and kept in the same condition in a 1.5 mL glass
vial, their environment for drying is not homogenous. During storage and drying, the
droplets are pushed towards the oil air interface due to density difference between water
and fluorinated oil and jammed under the meniscus near the glass wall. Droplets closest
to the oil surface dry the fastest, but even small vibration rearranges the droplets. As a
result, each droplet experiences a different and random drying history. To ensure
sufficient equilibration, the fastest dried droplet has to be dried slow enough. In addition,
the drying speed of droplets only plays an important role when the packing fraction of
colloids is close to the critical value of crystallization, which is only a small fraction of the
total droplet drying time at the end. However, to avoid misinterpretation and ensure
erliable experiments, we choose to dry the droplets in 40 days in Figure 9h.

26
Figure 9, Colloidal clusters from confined self-assembled in water-in-oil emulsion droplets generated by
microfluidics. Four distinct cluster morphologies with increasing degree of ordering are observed: a, buckled clusters
are large in comparison and partially collapse upon evaporation into a non-spherical shape; b, spherical clusters exhibit
only local order; c, partial icosahedral clusters show one or more five-fold symmetry axes and incomplete faceting
(dotted blue boxes); e, icosahedral clusters have clearly defined facets, edges, and vertices and complete icosahedral
symmetry. e,f, Low magnification SEM images highlight the uniformity in size and structure of the prepared clusters.
Spherical and icosahedral clusters dominate in the limit of fast (e) and slow (f) evaporation, respectively. g,h, Statistical
evaluation of the observed morphologies as a function of the evaporation rate (g) and as an evolution over time for the
slowest evaporation rate (h) showing the progression from spherical to icosahedral (Ih). Scale bars, 2 µm

Colloidal clusters in double emulsion droplets

The drying of single emulsion suffers from inhomogeneities in droplets local environment
as the shrinkage is caused by the gradient in water content in the continuous phase. This
prevents the uniformity of crystallinity of colloidal clusters even in one sample stored in
one glass vial. To circumvent this problem, we device a different strategy for droplet drying.
Osmotic pressure can also drive droplets to shrink if there the osmotic pressure in the
continuous phase is higher than the disperse phase. This requires the water droplet to be
separated by a thin liquid film to a continuous water phase. For this effect to occur, we
need to encapsulate colloids in water-in-oil-in-water (w/o/w) double emulsions.78 The
osmotic pressure in the continuous phase can be easily controlled by tuning the
concentration of osmotic species such as salt, dissolved polymer or surfactant. To reach
equilibrium, the volume of water droplet will decrease if the osmotic pressure in the

27
continuous phase is higher or increase if the osmotic pressure in the continuous phase is
lower. This provides a precise and dynamic control of the water droplet volume.

Up to date, double emulsions with accurate size and shell thickness rely on complex
PDMS microfluidic devices or glass capillary microfluidics.79 The former usually requires
3D microfluidic channels and tedious, locally controlled surface treatment of the walls at
different positions within two consecutive droplet-forming junctions.55 The latter requires
skillful craft of aligning two glass capillaries with a few-dozen micrometer openings,80 and
the produced droplets are usually around 100 µm or larger, too large for our purpose.81–
83 To produce thin oil shells for efficient water diffusion under osmotic pressure, both
techniques needs to operate at a very small window of flow rates. In this case, the droplet
size is fixed by the dimension of the PDMS channel or glass capillary opening – one
device can only produce droplets of one size. If double emulsions are produced with thick
oil shell, a new device is required to reduce the shell thickness below 5% to qualify as
thin-shells.84,85 We therefore developed a fast and simple method to produce double
emulsion with single water core and thin oil shell to mitigate these inconveniences. The
method involves sequential vortexing of liquids in conventional, macroscopic centrifuge
tubes (made of polypropylene from Rotilabo, always works for glass centrifuge tubes or
glass vials of various sizes). It involves either only a conventional microfluidic chip to
fabricate the inner phase (if homogeneous droplets are required), or no microfluidics at
all, in which case, polydisperse droplets with thin shells are produced. As sketched in
Figure 10a, an aqueous colloid dispersion is pipetted into the perfluorinated oil containing
afluorosurfactant (1 wt% anionic Krytox or non-ionic PFPE-PEG-PFPE) into a centrifuge
tube. This mixture is vortexed at 2500 rpm for 30 seconds, which creates water droplets
of inhomogeneous sizes. The upper part of the oil, which contains the lighter water
droplets, are pipetted into another centrifuge tube containing an aqueous solution of
Sodium dodecyl sulfate (SDS,1 wt%) solution as the continuous phase. After vortexing
for 60 seconds, single core w/o/w double emulsion is formed (Figure 10b). The double
emulsion sinks to the bottom of the container as its overall density is slightly higher than
that of water due to the fluorinated oil shell. It appears brown under the optical microscope
due to the scattering properties of the colloidal dispersion. More than 95% of droplets
contains a single water core (the rest contains satellite droplets around the size of the oil

28
shell) and the oil shell is too thin to be visible at this magnification. This method involves
vortexing a centrifuge tube twice and can create high quality double emulsion in milliliter
quantities in a few minutes without any microfluidic devices. It therefore significantly
reduces the effort required for our colloidal cluster experiments.

When uniform w/o/w double emulsion droplet are desired, the first vortex step can be
replaced by the simple water in oil emulsion PDMS microfluidics as illustrated in Figure
6a. Using uniform water droplets also helps us to understand the mechanism of this
simple but surprisingly successful method. Before vortexing, the oil droplet contains large
number of water droplets, as shown in Figure 10c. Slight shaking of the centrifuge tube,
or any other container of the liquid sample, breaks up the oil droplets into smaller droplets
containing less water cores (Figure 10d). After some vortexing time, the oil droplets are
discretized into smaller droplets, each containing only a few water cores, and eventually
shears the emulsion into single water core double emulsion with a thin oil shell (Figure
10f). The size of the double emulsion depends on the core water droplets produced in the
microfluidics and can be precisely adjusted from 30 to 90 µm (Figure 10g). The shell
remains between 2 to 5 µm.

We investigate the limitations of this method, which also sheds light into the physical
principles underpinning the emulsification process. Droplets above 100 µm show large
size deviation (Figure 10g), presumably from coalescence of multiple core water droplets
(Figure 10e). Droplets below 30 µm are not uniformly single core double emulsion – many
contains multiple cores, similar to Figure 10e, and do not shear off into single core double
emulsion despite extended vortex time. This suggests a critical core droplet size, under
which the small curvature resists the breaking-up of the oil droplet at the given interfacial
tension between the fluorinated oil phase and the continuous water phase. The method
works with other concentration and types of surfactants as well, such as TritonX-100 or
PVA in the continuous water phase, resulting in different boundaries of available droplet
size range. Increasing the viscosity in the continuous phase increases the smallest double
emulsion size, while increasing the surfactant concentration in the oil phase increases the
largest double emulsion size. Oil-in-water in oil double emulsion can also be produced by
this method, although the water shell is thicker and less uniform (in the range of 20 µm)
and contains a lot more satellite oil droplets. This suggests that the atomization process

29
(for water-in-oil-in-water double emulsion) involves not only the break-up of the oil droplet
by friction against the continuous water phase, but also the removal of the oil shell around
a water core by the friction against the centrifuge tube wall, enhanced by the density
difference between the two liquids. This hypothesis is supported by the many small oil
droplets containing no water core in Figure 10b. The atomization of multicore double
emulsion into single core is possible by turbulent flow or high shear in the continuous
phase, but a uniform and thin shell around the core requires a wall of high friction.

Figure 10, Simple and fast fabrication of double emulsion with thin shell. a, sequential vortexing to produce water
in oil in water double emulsion with inhomogenous sizes. b, the double emulsion consists of single water core (colloid
dispersion) and thin oil shell. c, the same protocol applied to simple water in oil emulsion of homogenous size produced
by microfluidics. Before vortexing, an oil droplet contains multiple water droplets. d, e, f, vortexing breaks up the oil
droplet containing less water droplets, eventually discretize it into single water core double emulsion with thin oil shell
(inset in f). g, the size of homogenous double emulsion droplet ranges from 30 μm to 90 μm with uniform thin oil shell.

30
Fabricating colloidal clusters in double emulsions provides several benefits over simple
water in oil emulsions. The droplets sediment to the bottom of the container due to density
different and reside there stationary, insensitive to small vibrations, making it convenient
for optical characterization (Figure 11, optical properties will be discussed in Chapter 7).
The droplet shrinkage rate can be conveniently controlled by tuning the osmotic-active
spices in the continuous water phase by salt or surfactant concentration. As their
concentration is homogenous in space, the droplet shrinkage rate is the same for all
droplets. Compared to 75% crystalline colloidal clusters after 40 days of droplet drying,
the portion of crystalline clusters achieves 100% within a few days, evident by various
structural color pattern as shown in Figure 11b (details in Chapter 7). Importantly, the
volume of the water core in double emulsion can be increased by reducing the osmotic
pressure in the continuous phase. The expansion of the droplet confinement melts the
crystalline colloidal cluster (Figure 11c, i to v as time proceeds). As interparticle distance
increases, their structural color red shifts due to larger lattice spacing (Chapter 1) and
eventually the cluster melts. The melting of colloids is an important but less explored topic
in soft condensed matter. It is often achieved by changing the range of interparticle
potential,26 as experimental setups to slowly increase packing fraction is difficult. Double
emulsion offers a unique opportunity to control the packing fraction at will, providing an
ideal platform to investigate the kinetics in particle structure formation. However, this
thesis focuses on the crystallization of colloidal clusters, their melting will not be discussed
and should be subject to future studies.

When colloidal particles are prevented from adsorbing to droplet interface, and when
sufficient equilibration time is allowed, they self-assemble into colloidal clusters with
complex structure and intriguing symmetry. Now we are ready to study them.

31
Figure 11, Colloidal clusters in double emulsion. a, optical image of colloidal clusters in double emulsion after
osmotic annealing. The volume of water droplet encapsulated in an oil droplet shrink significantly, leading to
crystallization of colloidal particles. b, optical image of crystalline colloidal clusters showing structural color and diverse
color patterns. c, osmotic melting of colloidal clusters as their confinement volume increases due to water diffusion
from the continuous phase to the core water droplet (credit of S. Hahn).

32
 Chapter 4
Magic Numbers in Minimum Energy Colloidal Clusters
The results described in this chapter were published in:
Junwei Wang, Chrameh F. Mbah, Thomas Przybilla, Benjamin Apeleo Zubiri, Erdmann
Spiecker, Michael Engel, Nicolas Vogel
Magic number colloidal clusters as minimum free energy structures.
Nature Communications 9, 5259 (2018)

I performed the colloidal cluster experiments, proposed and wrote the code for the model.
C.F. Mbah performed the simulation and free energy calculation. T. Przybilla and B.A.
Zubiri performed the electron tomography. E. Spiecker, M. Engel and N. Vogel supervised
the work.

The previous chapter described the two requirements for the colloidal particles to reach
equilibrium and form order structures in droplet. Starting from this chapter, we discuss
their thermodynamics– what are the energy minimum structures for colloidal particles in
droplet confinement? We will see that for such a confined system, the number of particles
is a determining factor, a feature not seen in the bulk but unique in finite systems. To
make distinctions with the commonly observed spherical assembly of colloidal
particles,49,50 widely acknowledged as colloidal supraparticles, we term the completely
crystalline assembly described in this thesis as colloidal clusters to highlight their
crystalline nature, in analogue to atomic clusters.

Icosahedral colloidal clusters

We produce monodispersed droplets of an aqueous dispersion of polystyrene (PS, 244

nm in diameter) colloidal particles in a continuous oil phase by PDMS microfluidics. We
ensure sufficiently slow water evaporation of the water in oil emulsion, as described in
Chapter 3, for colloidal particles to equilibrate into their minimum energy structure.

We find a wide range of well-formed icosahedral clusters with distinct surface features.
The surface is tiled with rectangles and truncated triangles, which alternate around five-
fold axes. Together, these structural elements form a closed shell spanning the surface

33
of the cluster (Figure 12a) In analogy to atomic clusters with complete outer shells86,87,
and for the simplicity of discussion, we now term such clusters as magic number colloidal
clusters (MCCs). The word “magic number” refers to the number of particles that can form
clusters in with minimum energy configurations, which will be discussed later. To
rationalize the appearance of MCCs, we recall that twelve identical spheres readily
arrange into an icosahedron around a central sphere. Subsequent Mackay shells can be
added concentrically88. Figure 12b exemplarily shows a ten-shell Mackay icosahedron
as the starting point for a model to describe the MCC structure. It consists of 20 slightly
deformed tetrahedra with FCC structure sharing a central sphere, each twinned with three
neighboring tetrahedra. We expand the Mackay core by adding anti-Mackay shells89–93
that consist of 20 additional tetrahedra over the icosahedron faces, filling their gaps with
30 tetrahedra over the icosahedron edges, and finishing the construction with 60
tetrahedra over the icosahedron vertices. The resulting Pentakis dodecahedron model 94
maintains icosahedral symmetry and consists of 130 twinned tetrahedra. Finally, we apply
spherical truncation to remove spheres in the model whose distances to the center is
larger than the truncation radius. This mimics the effect of confinement of the confined
colloids, which are forced into a spherical shape by the geometry of the emulsion droplet.

Although complex at first glance, MCCs are determined by only two parameters: the
number of shells of the Mackay core m and the number of anti-Mackay shells a, which is
controlled by the truncation radius Figure 12c). We introduce a notation to classify the
MCCs as (m+a)a types, specifying the total number of shells, m + a, and the number of
anti-Mackay shells a. The distinct surface patterns of the MCCs are directly correlated
with their crystal structure and can be used to deduce the number of Mackay and anti-
Mackay shells. From the model, we identify that the rectangular surface features are the
characteristic structural element associated with the anti-Mackay shells (Figure 12b).
They result from truncation of the 20 twinned tetrahedra added to the facets of the
icosahedral structure and the additional 90 tetrahedra filling the open voids of the model
structure. In the absence of an anti-Mackay shell, a single line of colloids marks the facet
of the icosahedron. A single anti-Mackay shell produces a rectangle with a width of two
particles, two anti-Mackay shells a width of three particles (as in the example structure)
and so on. An equation of the relationship between surface and interior shell structure

34
can be given. We use a Python code to generate MCC models following the procedure in
Figure 12b. By comparison with the model, we can now identify the MCC in Figure 12a
as of 122 type, i.e., consisting of 10 Mackay and 2 anti-Mackay shells.

The geometry of the Pentakis dodecahedron necessitates deformation on all tetrahedral

grains in the model (Figure 12d). Grains in the Mackay core and over anti-Mackay faces
are identical and least deformed, grains over anti-Mackay edges and vertices are more
deformed. The deformation analysis of our model suggests particles along the width of
the rectangles in the cluster surface are separated by 1.13 times their diameter, compared
to those along the length of the rectangle by 1.05, which can be observed experimentally
(Figure 12a). It also suggests extra particles are prone to accumulate in the vertices
regions where tetrahedral grains have the highest degree of deformation. This geometric
feature is unique to icosahedral sphere packing under spherical confinement and
provides mechanisms to accommodate structural defects in the colloidal cluster (to be
discussed in the next chapter). It also implies that structural impurities, such as a few
larger or many smaller particles may be entropically favorable to reside in these vertices
regions. The new Pentakis dodecahedron model extends the well-established Mackay
icosahedron52, which may help the understanding of binary clusters where slightly larger
component resides in more deformed regions94–96.

35
Figure 12, Model of magic number colloidal clusters. a, SEM image of a magic number colloidal cluster (MCC). The
surface is tiled with 5 × 3 rectangles (along two-fold symmetry axes, ‘2f’), truncated triangles (‘3f’), and concentric rings
around five-fold axes (‘5f’). Scale bar, 2 µm. b, Model of a MCC. A Mackay icosahedron core is expanded by adding
twinned tetrahedral grains over the core icosahedral faces, edges and vertices, forming anti-Mackay shells. The outer
geometry of the complete model is a Pentakis dodecahedron. The effect of droplet confinement is mimicked by applying
spherical truncation to remove some spheres in the model. The resulting structure accurately reproduces the
experimentally observed MCC in (a). c, Building block of the model with 10 Mackay and 2 anti-Mackay shells. MCCs
are denoted as of (m+a)a type, where m is the number of Mackay shells and a the number of anti-Mackay shells. d,
Tetrahedral grains in the different parts of the model with marked scale factors compared to a perfect FCC grain. All
grains deviate from regular tetrahedra as a consequence of strains in the icosahedral structure. Grains in anti-Mackay
edges and vertices are deformed more strongly.

To confirm the accuracy of the model, we deposit a colloidal cluster on a TEM grid, and
Thomas Przybilla from the Spiecker Group performed electron tomography study to
reveal the three-dimensional structure of this 72 type MCC. The cluster structure and
model agree quantitatively, as seen by the similarity of bright field scanning transmission
electron microscopy (STEM) images and the semi-transparent model images in
projections along the symmetry axes (Figure 13a-c,f-h). This agreement confirms that
the MCC indeed is crystalline and that the cluster type deduced from the surface pattern
is consistent with its internal structure. The recorded tilt series underlines the icosahedral

36
nature of the cluster. During rotation, views along two-, three-, and five-fold symmetry
axes gradually transition from one to another. In the tomographic reconstruction of the
data, characteristic surface features (Figure 13d,i) and details of the interior (Figure 13e,j)
coincide between experiment and model. In the cross-section of the reconstructed cluster,
two triangles of 21 hexagonally closed-packed particles sharing a single particle at the
center can be clearly distinguished (Figure 13e,j). These particles form the triangular
faces of two tetrahedra in a five-shell Mackay icosahedron. Four particles at the surface
belong to the second anti-Mackay shell (indicated by a green line), which enables us to
identify the cluster as a 72 type.

Figure 13, Electron microscopy tomography confirmation of model. a-c, Scanning transmission electron
microscopy (STEM) bright-field images of a magic number colloidal cluster viewing along two-fold, five-fold, and three-
fold axes, characteristic for icosahedral symmetry. Two staggered 10-spike rings in the center along the five-fold axis
are clearly seen in (b). d, STEM tomography reconstruction provides a three-dimensional visualization of characteristic
surface features of triangles with four particles at side, rectangles of three and four particles at width and length, and
pentagon with three particles at side (indicated in blue). e, Cross-sectional views through the reconstruction at the
central plane reveals the characteristic internal structure. A section of the Mackay core is marked by two blue triangles.
The vertex region is marked by its characteristic pentagon. The second anti-Mackay shell is indicated by a green line
segment. Scale bars, 1 µm. f-j, The experimental images show excellent agreement with 72 MCC type model
reconstructions using semi-transparent spheres mimicking the STEM imaging process (f-h), external (i) and cross-
sectional views (j).

Having established by tomography study that our geometric model accurately describes
the colloidal cluster structure, C. Mbah from the Engel Group performed Event Driven
Molecular Dynamics simulation (EDMD) to mimic the experimental process.56,57 More

37
details of the simulation and kinetics of crystallization will be discussed later. In short,
hard spheres are programmed to follow the Newtonian laws of motion to change direction
based on momentum reservation, without interparticle potential, elastic or plastic
deformation, or friction upon contact. The particles are confined in a rigid and hard
spherical confinement, whose wall has no potential with the encapsulated particles. Upon
critical packing fraction, the crystallization begins spontaneously due to entropy
maximization. After crystalline clusters are formed in the confinement, a second
simulation step is utilized to mimic the last stage of droplet shrinking in the experiment
where droplet interface is deformed, and the cluster is faceted. As seen in Figure 14, this
two-step simulation scheme produces clusters that accurately reproduce structures
observed in experiments, tomography and model. The characteristic surface rectangular
region with a width of three particles is a result of two anti-Mackay shells in the sphere
packing model as described in Figure 12. The precise agreement between experiment,
theory and simulation confirms our understanding of the colloidal cluster structure from a
few selected samples and allows us to extend the investigation to a larger range of system
size with confidence. Although the hard sphere simulation ignores hydrodynamics, the
slight deformability of droplet interface, as well as the slight repulsion in colloidal particles,
it accurately captures the structure formation of colloidal clusters, which highlights the
dominant role of entropy in the formation process.

Figure 14, Comparison of experiment, tomography, model and simulation. a, an experimental colloidal cluster
imaged in SEM, the surface showing characteristic rectangular region with a width of three particles (marked by red
lines). b, an electron tomography reconstructed cluster obtained from the abovementioned tilt series of TEM images.
The surface renders the unique rectangular region marked by three red lines. c, a theoretical model cluster from
spherical truncation of icosahedral sphere packing with anti-Mackay shells, showing agreement with a and b. d, cluster
obtained from hard sphere simulation in rigid spherical confinement, followed by a relaxation scheme, showing accurate
structural details that agrees with all experiment, tomography reconstruction and model.

38
From the analysis of hundreds of clusters of sizes ranging from 100 to 10000 colloidal
particles, we generate a library of experimentally observed icosahedral MCCs (Figure
15). All these clusters exhibit complete, well-defined outer shells and can be assigned a
magic number cluster type of the form (𝑚 + 𝑎)𝑎 , deduced from their surface features. In
all cases, experimental observation and model coincide. The model predicts that the
number of colloids per cluster follows the approximate relationship 𝑁 = 10
3
(𝑚 + 𝑎)3
derived for Mackay clusters88. Furthermore, the number of colloidal particles per cluster
decreases from type (m+a)a to type (m+a)a+1, i.e., when a Mackay shell converts to an
anti-Mackay shell (Figure 15j-l), as a result of lower local density in anti-Mackay shells.
Figure 15m summarizes all observations of MCCs in experiment and simulation. The

39
number of anti-Mackay shells increases with cluster size as a result of the improved
sphericity but is limited by 𝑎 < 𝑚/2.

Figure 15, Library of magic number colloidal clusters and comparison to model. Rich variety of MCCs are
observed with increasing number of particles. a-c, MCCs without anti-Mackay shells (m0 type) correspond to truncated
Mackay icosahedra. d-f, MCCs with one anti-Mackay shell ((m+1)1 type) are characterized by a two-particle wide
rectangular region and a varying number of Mackay shells. g-i, Similarly, two anti-Mackay shell clusters ((m+2)2 type)
feature a width of the rectangular region of three particles. j-l, MCCs with a fixed total number of thirteen shells but a
varying number of anti-Mackay shells (13a type). In each example, SEM images (left) are compared to the
corresponding model (right). Scale bars, 1 µm. m, Summary of all MCCs observed in experiment or simulation
organized by type. Clusters with the same total number of shells m+a fall on a diagonal. Such structures are close in
the total number of particles. As the cluster size increases, more anti-Mackay shells can typically be accommodated
around a larger Mackay core.

40
Magic number colloidal clusters

We observe the MCC formation in computer simulations to reveal details of the formation
mechanism, as a confocal study to track all particle positions in-situ is difficult due to small
particle size and fast diffusion. As described before, our simulation scheme involves two
steps (credit of C. Mbah). The first step (‘self-assembly’) mimics colloids in a shrinking
droplet during evaporation by hard spheres in spherical confinement with decreasing
radius. This computational approach follows recent work97 and ignores hydrodynamic
interactions, which affect crystallization speed98,99 and colloidal aggregation far from
equilibrium100,101 but are expected to have a weak influence on the equilibrium cluster
structure and near-equilibrium structure formation. We record the equations of state for
six system sizes (Figure 17a,b) and generally observe a sudden pressure drop indicative
of a first-order transition from a disordered fluid to an ordered cluster. The second step
(‘quenching’) uses numerical relaxation58 to mimic the capillary forces that consolidate
the colloidal cluster in the final stage of droplet drying where the droplet interface no
longer remains spherical due to loss of water volume 102. In this step, all spheres in the
structure obtained from the first step are assigned a Morse pair potential and let to relax
into local energy minimum structure. The Morse potential was chosen due to its tunable
range and form of interaction. It is important to note that capillary forces and attractive
interactions in the quenching step are not essential for the crystallization of hard spheres
in confinement. They occur only after the actual self-assembly process and merely push
the particles into their final facetted structure but do not change neighbors.

To understand the structure of simulated clusters, we utilized a bond orientational order

diagram (BOD) as illustrated in Figure 16. The BOD is visualized in a java program written
by Michael Engel. In short, a vector can be formed to connect any two particles in space
with a direction and length. The BOD is generated by selecting all such vectors in the
cluster that have a certain length (this can select, for example, all vectors connecting
nearest neighboring particles), and projecting these vectors from the center to a spherical
surface. Since every vector yields a projected dot in the spherical surface, the pattern of
the superimposed dots represents the overall spatial arrangement of the selected
particles. Due to translational and orientational symmetry in crystal lattice, particles in a

41
crystalline state give rise to overlapped, hence, bright and sharp spots in the BOD, while
particles in amorphous state does not. As shown in Figure 16a, the tetrahedral grain of
slightly deformed fcc lattice give rise to one central spot and four spots in the ring of the
BOD (viewed from this angle). More spots are visible when additional grains are added
(Figure 16b-e). Due to the icosahedral arrangement of the grains, their BOD shows ten
spots in the ring with a five-fold pattern, when viewed along the five-fold axis. We show
the development of the BOD as more grains are added until the completion of an
icosahedron with twenty grains (Figure 16f-i). The BOD informs us of the global symmetry,
crystal type, grain orientation and arrangement, and even distribution of defects in the
colloidal cluster (to be discussed in detail in the next chapter).

Figure 16, Bond orientational order diagram of colloidal clusters. a-e, spots in the bond orientational order diagram
(BOD) as a result of one, two, three, four and five tetrahedral grains. Each sets of spots is rotated 72 degrees due to
the icosahedral arrangement of the grains, which yields a final five-fold symmetric pattern in the BOD. f, five grains
viewed along its five-fold axis, g, ten grains viewed along its two-fold axis, h, ten grains viewed along its five-fold axis,
i, the complete icosahedron with twenty grains viewed along its two-fold axis.

42
With the help of BOD, we can analyze in the simulation how particles in disorder fluid
state transformed into icosahedral clusters (Figure 17c-f). Quenching the ordered
clusters of Figure 17a demonstrates high propensity for the development of surface
features characteristic for MCCs. However, this ordering process depends sensitively on
cluster size. The hysteresis loop becomes wider and pressure in the ordered cluster lower
for one cluster size (N = 1905 in Figure 17b) than slightly smaller and larger cluster sizes
(N = 1774 and N = 2166), suggesting size-dependent thermodynamic stability. This will
be further elaborated below.

We investigate the kinetics of cluster formation during the assembly process. Particle
displacements are recorded over time during EDMD when the packing fraction is
increased from 0.48 (fluid phase) to 0.55 (solid phase) and subsequently decreased. Due
to the presence of hysteresis, the fluid and solid phase occur at the same packing fraction
𝜙 = 0.51 during the compression stage and the expansion stage, respectively. In both
cases, particles form concentric shells with occasional migration between neighboring
shells. To reveal how kinetics varies throughout the cluster, we measure mean square
displacement separately for particles starting in different shells. In the fluid phase (Figure
17g, top), particles near the center diffuse significantly more than particles near the
surface, which appear to be hindered in their mobility by the confining wall. In contrast, in
the icosahedral cluster (Figure 17g, bottom), particles in outer shells show an increased
mobility while particles in the interior are nearly arrested. This abrupt change in the
kinetics is key for understanding MCC formation and highlights the importance of the
outermost shells for the ordering process. Crystalline patches develop early in the outer
shells and rapidly grow towards the interior97,103,104. Importantly, particles near the surface
effectively retain mobility through the phase transition. This mobility is necessary to heal
surface defects efficiently and helps forming anti-Mackay surface shells. The enhanced
mobility after crystallization of the cluster’s interior is essential to equilibrate to the energy
minimum structure under the curved confinement, as the effect of confinement curvature
is most severe near the surface. These few surface layers play an important role in colloid
crystallization under confinement –particles here comply to the geometry of the
confinement to nucleate the crystallization that grows towards the core, after the core

43
becomes crystalline, they serve as an interface that cushions the structural distortion
between the lattice and the incommensurable confinement curvature.

The predominant formation of closed-shell clusters poses the fundamental question if

MCCs are thermodynamically favored. In the absence of particle interactions, we
determine free energies of clusters (credit of C. Mbah) with between 100 and 8000 hard
spheres by calculating entropy using the Einstein crystal method. 61 As an important
modification of the conventional Einstein crystal method we include swap moves to
sample diffusion efficiently near the ordering transition105 and subtract the bulk
contribution. The calculated free energy shows a series of distinct minima as the cluster
size is varied (Figure 17h). This demonstrates that icosahedral order is not only favored
over FCC for hard spheres in confinement97 but is realized as thermodynamically stable
MCCs with highly defined shell structures. Our analysis reveals that the depth of the free
energy wells per particle ranges from 0.6 kBT for MCCs with 3 shells to 0.018 kBT for
MCCs with 12 shells. The absolute free energy gain for forming a complete shell is
constant in the order of 100 kBT per shell and increases with packing fraction. In both
experiments and simulations, we observe more anti-Mackay shells as colloidal clusters
increases in size (Figure 15m).

44
Figure 17, Kinetics and stability of magic number colloidal clusters. a-b, Equations of states of hard spheres in
spherical confinement. Dimensionless pressure P* is recorded for sets of 100 simulations. Packing fraction 𝜙 is first
increased and then decreased to test for hysteresis. c-f, Snapshots obtained by quenching at the four packing fractions
indicated in (a). The disordered cluster (c) preorders at the surface (d) and then rapidly transitions into an icosahedral
cluster (e). The remaining defects heal given sufficient time (f). Particles are colored by the number of neighbors. The
bond orientational order diagram (insets) is a global order parameter for the transition from the fluid to the ordered state.
g, Mean square displacement (MSD) of particles for a 9-shell cluster at coexistence packing fraction (𝜙 = 0.51). The
dependence of the diffusivity on the shell number reverses at the transition from the fluid (top) to the icosahedral cluster
(bottom). The shell number corresponds to the shell the particle was part of at the start of the MSD measurement.
h, Free energy of colloidal clusters computed at 𝜙 = 0.52 as a function of the radius of the confining sphere R
normalized by subtraction of the free energy of the bulk corrected for surface effects F 0 (see Supplementary Information).
Free energy minima correspond to clusters with complete shells. Data of clusters with face-centered cubic (FCC) crystal
structure is included for comparison for N > 1000 where FCC clusters are metastable. This figure is contributed by
Chrameh Fru Mbah from Engel’s group.

The closed-shell magic number cluster structure requires particles to efficiently utilize
space in spherical confinement to maximize entropy, particularly in the vicinity of the
curved confinement. Icosahedral symmetry is favored over FCC97 because the highly
facetted closed-shell structure formed by the latter (Wulff shape between octahedron and
cube) creates large mismatch and gaps between facets and the spherical confinement.

45
Icosahedral arrangement reduces this mismatch via more but smaller facets, improving
local dense packing near the curved interface106. Our simulations reveal that there is
always a preference for a certain anti-Mackay and Mackay clusters type. The introduction
of a few anti-Mackay shells makes the cluster more spherical and compact compared to
the truncated Mackay icosahedron. This means that the anti-Mackay shells provide
means for the cluster to more closely follow the geometry of the confinement and thus
use the available space more efficiently. Because the packing in Mackay and anti-Mackay
shells differs, both types of shells accommodate different number of particles. Therefore,
we speculate that the confined system can adopt different Mackay/anti-Mackay
combinations to find the optimal closed-shell magic number arrangements for a given
specific numbers of particles.

Magic number configurations are a well-known concept in atomic nuclei and atomic
clusters,4,86,107–110 where interaction between building blocks are strong. Similar effect
was found in systems with other types of interactions such as micelles111, fullerene
assemblies112, protein aggregates113, supramolecules of ionic liquid114, inorganic
clusters115, and natural framboidal pyrite6,7. In this chapter, we give evidence that strong
interaction is not necessary to form minimum energy structures in finite systems, purely
entropic interaction suffice to give rise to magic number effect in colloids in confinement.

46
 Chapter 5
A Detailed Investigation of Free Energy Landscape of
Colloidal Clusters
The results described in this chapter were published in:
Junwei Wang, Chrameh F. Mbah, Thomas Przybilla, Silvan Englisch, Erdmann
Spiecker, Michael Engel, Nicholas Vogel
Free energy landscape of colloidal clusters in spherical confinement.
ACS Nano 13, 9005-9015 (2019)

I performed the colloidal cluster experiments, analyzed the simulation data, extract the
defect fractions and proposed the defect model. C.F. Mbah performed the simulation and
free energy calculation. T. Przybilla and S. Englisch performed the electron tomography.
E. Spiecker, M. Engel, and N. Vogel supervised the work.

As we showed in Chapter 4, magic number colloidal clusters are close shell structures of
sphere packing. Naturally, the larger the shells, the more spheres it has. The number of
spheres in a close shell increases approximately quadratically to the radius of the shell.
Therefore, magic numbers are not only discrete numbers, but are dispersed increasingly
sparsely at larger system sizes. In reality, the numbers of colloidal particles encapsulated
in droplets are determined by the concentration of the dispersion and the size of the
droplets, along with some fluctuations. Once encapsulated, the numbers of colloidal
particles in the confinement does not change. In this chapter, we discuss colloidal
particles’ effort towards lower energy when the number of particles is not an exact magic
number. There are two scenarios – when the number is close to a magic number, and
when the number is far off from a magic number.

Recall that the shell structure of colloidal clusters can be described by the (m+a)a notation,
where m is the number of Mackay shells and a is the number of anti-Mackay shells at the
cluster surface (Figure 12). The anti-Mackay shells, stacked onto an icosahedral core,
are responsible for the characteristic surface rectangle region in the colloidal clusters.

47
For recapitulation, we show in Figure 18a (left) a cluster of 8 Mackay shells (m) and one
anti-Mackay shells (a), hence the notation 91. Figure 18a (right) shows a cluster with 8
Mackay shells and 3 anti-Mackay shells, hence the notation 113.

C. Mbah calculate the free energy of simulated clusters in spherical confinement for a
range of system sizes (details see Chapter 2 and Chapter 4). The resulting free energy
landscape of colloidal clusters is shown in Figure 18b (52% packing fraction in spherical
confinement and normalized to energy per particle). The free energy landscape of
colloidal clusters shows distinct minima. These minima can be assigned to regions where
the number of constituent particles in the colloidal cluster affords closed shells, and
subsequent minima have increasing numbers of total shells. The free energy minima of
simulated clusters coincide with the particle numbers required in the ideal model,
suggesting that clusters with concentric icosahedral shells, as described by the model,
are thermodynamically favored.

Noteworthily, the free energy curves of colloidal clusters from simulation do not show
sharp peaks but rather broadened regions of low free energy (termed magic number
regions, green in Figure 18b), followed by plateaus with higher free energy (termed off-
magic number regions, red in Figure 18b). The appearance of broad magic number
regions is surprising at first because varying the spherical truncation radius of the model
to include more particles generates clusters with isolated particles at the cluster surface.
Such particles lower the density and prevent efficient packing in spherical confinement,
which increases the free energy. Apparently, the simple picture of sparsely located,
discrete magic number common in the study of atomic clusters is not applicable to
colloidal clusters.

We use bond orientational order diagrams (BODs) to assess the structural order in
colloidal clusters. As a reminder (Chapter 4, Figure 16), BODs provide information about
the symmetry and arrangement of particles by projecting all bonds between neighboring
particles to a spherical surface. In colloidal clusters, crystalline arrangements of particles
generate BODs with clear and bright spots where the intensity of a spot depends on the
number of aligned bonds.

48
Figure 18c shows BODs of several simulated colloidal clusters from about 3000 to 4000
particles in simulation following the transition in free energy between two high free energy
plateaus through a minimum. Within the free energy minimum (particle numbers from
3402 to 3567), colloidal clusters exhibit near-perfect icosahedral symmetry (middle row
in Figure 18c). Their BODs show precise and clear spots with five-fold symmetric pattern.
In addition, five triplets of subspots are found between the major spots with a decreasing
intensity as cluster size increases. The structural origin of the subspots will be discussed
below. Colloidal clusters outside of the free energy minimum (off-magic number) show
broken icosahedral symmetry where spots in the BOD are split into doublets (particle
numbers of 3010 to 3375 and 3652 to 3974; first and third row in Figure 18c).

Figure 18, Magic and off-magic number regions in the free energy landscape of colloidal clusters. a, Two
exemplary icosahedral magic number colloidal clusters from experiments (left) and simulation (right). The cluster
surface is tiled with alternating triangles and rectangles around five-fold axes. The 91 type cluster consists of nine
icosahedral concentric closed shells among which there is one anti-Mackay shell at the surface. The 113 type cluster
consists of a total of eleven shells, including three anti-Mackay surface shells. Scale bars: 2 μm. b, Free energy
landscape of clusters from between 2500 to 4500 colloids at packing fraction 52% in simulations. Line is a guide to the
eyes. As the number of particles in the cluster increases, two free energy minima are observed that correspond to nine
and ten closed shells, respectively. Off-magic and magic number regions are marked in red and green, respectively. c,
Bond orientational order diagrams (BODs) of selected simulated colloidal clusters. The free energy of the chosen
clusters transitions from a plateau of high free energy through a minimum to another plateau. For the off-magic number
region (3010 to 3375, red, top), individual spots in BODs split into duplets. For the magic number region (3402 to 3567,
green, middle), BODs show clear and bright spots with five-fold patterns indicative of high-quality icosahedral ordering.

49
When transitioning out of the free energy minimum into the next off-magic number region (3652 to 3974, red, bottom),
BODs again show splitting of spots.

Magic number region

We first discuss three mechanisms that broaden minima in the free energy landscape
and explain how icosahedral clusters can form throughout an extended region of magic
numbers and do not exist as individual, discrete magic numbers only. This deals with the
first scenario – when the number is close to a magic number.

The first mechanism relates to anti-Mackay shells. The number of anti-Mackay shells sets
upper and lower bounds for magic number regions. Figure 19a shows a typical magic
number colloidal cluster found in experiment together with the corresponding 10 2 cluster
model (Figure 19b). As discussed in Chapter 4, the number of anti-Mackay shells can
be identified from the structure of the cluster surface. The appearance of rectangular
patterns with three rows in Figure 19a,b indicates that the cluster has two anti-Mackay
shells.

Maintaining ten closed shells in the cluster, the number of anti-Mackay shells can vary
(Figure 19b e). Because rectangular regions in anti-Mackay shells accommodate fewer
particles compared to Mackay shells, their presence decreases the packing density.
Indeed, the number of particles decreases from 3607 in the cluster with no anti-Mackay
shells (100 type cluster, Figure 19c) to 3247 in the cluster with three anti-Mackay shells
(103 type cluster, Figure 19e). For well-formed icosahedral clusters, we always observe
42 bright spots in BODs in the directions of the vertices of a pentakis icosidodecahedron.
These spots originate from 20 (slightly deformed) fcc grains in icosahedral arrangement
(Figure 19c). Anti-Mackay shells, which are twinned with Mackay shells, generate 20
triplets of subspots (Figure 19b,d,e). These subspots become brighter with increasing
number of anti-Mackay shells (Figure 19b-e), which indicates that they are indeed caused
by the anti-Mackay structure.

The intensity of the subspots in BODs of simulated clusters within the magic number
region (second row in Figure 18c) gradually diminishes with increasing particle number.
This indicates a reduction of the number of anti-Mackay shells with cluster size and
corroborates with the possibility to form different structural arrangements within a single

50
free energy minimum. However, different combinations of Mackay shells and anti-Mackay
shells correspond to distinct magic numbers within a single magic number region and
would be expected to lead to small and well-defined subfeatures. However, the free
energy landscape does not resolve these sub-features within the free energy minimum,
indicating additional mechanisms to accommodate variations in particle number.

a
a b

=
2
8
=
m
102
N = 3427

c d e

100 101 103

N = 3607 N = 3577 N = 3247

Figure 19, Ten-shell colloidal clusters with variable number of anti-Mackay shells. a,b, Scanning electron
microscope (SEM) image of experimentally observed colloidal cluster (a) and corresponding model (b). Scale bar: 2
μm. The cluster is identified as a 102 type with 3427 colloids. One of the 20 tetrahedral grains and the BOD of the
cluster are shown as inset. c, The BOD of the 100 cluster consists of 42 bright spots along the vertices of a pentakis
icosidodecahedron. d,e, BODs of the 101 cluster (d) and the 103 cluster (e) have additional triplets of less intense
subspots. These subspots result from the anti-Mackay surface shells. With increasing number of anti-Mackay shells
from (c) via (d) and (b) to (e) the intensity of subspots increases proportional to the total number of particles in all anti-
Mackay shells.

The second mechanism relates to partial anti-Mackay shell. Additional types of ordered
colloidal clusters are possible by adopting a partial anti-Mackay shell. Figure 20a shows
a simulated cluster in the magic number region. Its surface shows the characteristic
rectangular patterns of the anti-Mackay shells. Noteworthily, rectangles with a width of
both three and four particles can be seen. The joint appearance of these two rectangular
patterns suggests that grains with two and three anti-Mackay shells coexist. Indeed, the
BOD shows asymmetric subspot triplets with uneven intensity, corroborating an

51
anisotropic thickness of anti-Mackay shells at the cluster surface. Figure 20b shows a
side-view of a grain extracted from the cluster shown in Figure 20a with seven Mackay
shells in the core and three anti-Mackay shells at the surface. Note that our notation
considers the central sphere in the cluster as the zeroth Mackay shell. Figure 20c shows
another grain from the same cluster with eight Mackay shells and two anti-Mackay shells.
Partial anti-Mackay shells are also frequently observed in experiment (Figure 20d,e).

Recall that for a fixed total number of shells, grains with a higher number of anti-Mackay
shells consist of fewer particles (Figure 19). Given there are now two types of grains that
coexist within a cluster (Figure 20b,c), it becomes clear that partial anti-Mackay shells
increase the types of ordered clusters that comprise the magic number region. Note that
in a strict sense icosahedral symmetry is disrupted by adapting a partial anti-Mackay shell.
However, because the underlying core structure remains unaffected (evident by the clear
five-fold patterns in the bond order diagram), such clusters have similar free energies as
magic number clusters with complete icosahedral symmetry. The formation of partial anti-
Mackay shells smears out the discrete free energy minima arising from Mackay and anti-
Mackay clusters. The ability to accommodate a certain number of defects at the cluster
surface while maintaining the global icosahedral symmetry is a consequence from the
enhanced mobility of particles near the confinement interface, which facilitates finding the
entropically favorable arrangement for the anti-Mackay surface, as discussed in Chapter
4.

52
Figure 20, Colloidal clusters with partial anti-Mackay shell. a, Simulated cluster within the magic number region
(3402 spheres) that exhibits a surface structure with rectangular patterns of a width of both three and four particles,
indicating the presence of different numbers of anti-Mackay shells within the same cluster. The uneven intensity of
triplet subspots in the BOD (inset) results from an anisotropic thickness of anti-Mackay shells. b,c, Side-view of two
grains extracted from (a). Anti-Mackay twinning is indicated by blue lines. Spheres are drawn slightly smaller for clarity.
d,e, Experimental observation of two colloidal clusters with a partial anti-Mackay shell. Scale bar: 2 μm.

The third mechanism relates to disorder near cluster vertices. The areas near the vertices
pointing along five-fold symmetry axes are usually more disordered than the otherwise
regularly tiled surface of magic number clusters, as a result of higher geometric
deformation of the constituent grain, inevitable when twinning tetrahedra around a central
point, as illustrated in the previous chapter in Figure 12. Figure 21a shows a typical
colloidal cluster with a surface exhibiting rectangular patterns with a width of four particles
and a side-length of seven particles (marked in blue). The corresponding model structure,
composed of 9063 spheres, identifies the cluster as 14 3 type (Figure 21b). Figure 21c
shows a different colloidal cluster in the experiment along with its model (Figure 21d). As
evidenced from the width of the surface rectangular patterns of four particles, this cluster
also consists of three anti-Mackay layer and therefore is equally classified as 143 type.
However, two lines of these rectangular patterns are removed at both sides leading to a
total side-length of five particles. The similarity of the two clusters is obvious from their
BODs, which are indistinguishable (Figure 21b,d, insets). Subtle differences in the vertex
region are not captured by our cluster classification. In the model, both clusters can be

53
generated by different truncation radii. The vicinity of the vertices are typically the most
deformed parts of icosahedral clusters. Defects near the vertices therefore have the
lowest free energy cost. In both experiments and simulations, these regions typically
exhibit several extra or missing particles, while the overall structure remains unaffected.
In the example of Figure 21, such excess particles near the vertices amount to 360
particles, about 4% of the total number of particles.

Figure 21, Excess or deficient particles near the cluster vertices. a,c, Magic number colloidal clusters with two
different side-length of rectangular surface patterns but equal width. Both clusters can be assigned a 14 3 type. Scale
bars: 1 μm. b,d, Corresponding geometric models and BODs (inset). The model and the indistinguishable BODs
suggest that the two clusters only differ near the vertices but are identical in the core region. Both BODs are rotated to
their five-fold axes for clarity.

Off-magic number region

All three mechanisms jointly contribute to the broadening of minima in the free energy
landscape of colloidal clusters. Whenever possible, self-assembly in spherical
confinement attempts to maintain icosahedral symmetry with closed, concentric shell
structure. System size fluctuations up to several hundred particles can be accommodated
efficiently via changes in the number of anti-Mackay layers, the introduction of partial anti-
Mackay layers, and excess particles in the vicinity of the cluster vertices. However, these
mechanisms are not sufficient to explain the plateau regions of higher free energy in the
off-Mackay regions. BODs of colloidal clusters in these regions exhibit a broken
icosahedral symmetry, often with smeared out spots suggesting the presence of defects
within the core region (Figure 18c).

Here, we are dealing with the second scenario – when the number is far off from a magic
number. The key question underpinning such clusters is how they incorporate defects
particles and distributed them within the cluster to minimize their detrimental effects to

54
free energy. Figure 22a shows a magic number cluster obtained from simulation using a
number of particles near the center of a free energy minimum (N = 3402 in Figure 18).
The cluster is viewed along three different symmetry axes. The cluster surface appears
well ordered, and the BODs exhibit clear and bright spots. All characteristic surface
features predicted by the anti-Mackay model are present. In contrast, Figure 22b shows
an off-magic number cluster with only a few hundred more particles such that the number
of particles is now outside of the broadened free energy minimum (N = 3738 in Figure
18). Again, the cluster surface appears well ordered in most orientations without clear
visual distinction from the magic number cluster above. Still, close inspection reveals
signs of disorder when viewed along one of the two-fold axes (Figure 22b, middle).
Differences are much more pronounced in the BODs. Recall in Figure 16 that the sharp
BOD spots are superimposed from dots contributed by several grains, and a perfect
icosahedral arrangement of twenty identical grains give rise to spots with icosahedral
symmetry. Deviation from such position will result in splitting into less intense, or blurry
spots. Here in Figure 22, the splitting of BOD spots, typically into doublets, indicate
breaking of icosahedral symmetry in the off-magic number cluster. Apparently, the relative
orientation of tetrahedral grains is disturbed. From this example it is obvious that
crystalline order of colloidal clusters should not solely be evaluated from analysis of the
cluster surface.

We investigate the distribution of defects in the simulated clusters. After removing surface
particles and particles in anti-Mackay shells that have reduced numbers of neighbors, we
identify particles with less than 12 nearest neighbors as defects. The coloring is
performed by C. Mbah by counting nearest neighbors of all particles in a Python code.
While the magic number cluster of Figure 22a contains no defects in this analysis (not
shown), the off-magic number cluster accumulates defects in a narrow wedge (Figure
22b, red particles in inset). We find that this wedge takes up the volume of between three
and four tetrahedron grains. The accumulation of defects in the wedge leaves non-
affected grains in the cluster defect-free.

We model the orientation and contact of tetrahedron grains in a magic number cluster
and an off-magic number cluster. The magic number (Mackay) cluster has 20 large grains
with icosahedral symmetry (Figure 22c). Recall from Figure 12 that the distance between

55
icosahedron vertices in the model is 1.05 times larger than the distance from the center
to the vertices. Internal stress in the Mackay cluster is inevitable because tetrahedron
grains with dihedral angle 𝑎𝑐𝑜𝑠(1⁄3) = 70.53° do not quite fit pentagonal symmetry with
rotation angle 360 °⁄5 = 72° . Magic number clusters accept the free energy penalty
associated with this stress because it is more than compensated by the highly efficient
surface packing that adopts well to spherical confinement.

The situation is different for off-magic number clusters (Figure 22d). We construct a
defected icosahedron model starting from a pentagonal bipyramid whose 15 edges are
of equal length. Regular tetrahedra are subsequently added until overlap becomes
inevitable and a large opening (marked in red) is left. The opening covers a volume of
approximately three more tetrahedra. The opening, which accommodates the defects,
has a wedge-like shape in which two twinned tetrahedra connect to a third one through a
narrow gap.

The defected icosahedron model accurately reproduces the split spots in BODs of off-
magic number clusters. The defect wedge forces the cluster into two halves, each slightly
tilted from its original position in the Mackay cluster. According to the defect model and in
agreement with observations in simulated clusters, there remains no five-fold symmetry
axis in the structure but only mirror planes. The sacrifice of icosahedral symmetry in the
defect model relaxes internal stress by leaving the remaining grains less deformed. Due
to the geometry of icosahedron, the constituent tetrahedra is not regular but deformed,
which causes increasing strain at increasing system size, and is believed to hinder the
occurrence of large icosahedral assembly in nature.51,52 For a regular tetrahedral grain
in fcc lattice, the averaged nearest neighbor distance is one. This distance is slightly larger
than one for a deformed tetrahedral grain, as some of its edges are slightly longer than
others (Figure 12). This geometric stress is relaxed after the icosahedral symmetry is
broken, which is directly visible in the radial distribution function (Figure 22e). In this plot,
the first peak, corresponding to nearest neighbor distances, is much narrower for off-
magic number clusters (red) than for magic number clusters (green), which means that
on average, particles in the off-magic number clusters are closer. Although our
calculations suggest that the transition from magic number region to off-magic number
region is gradual (Figure 18b), the amount of peak splitting is always very similar. This

56
suggests that the opening angle of the wedge in the defected icosahedron model is
unique and likely determined by geometry – as soon as the system size reaches a number
where excess defects cannot be tolerated, maintaining the perfect icosahedral symmetry
(which involves the geometric strain) becomes entropically less favorable compared to
breaking the symmetry (which relaxes the strain).

Figure 22, Defects accumulate in a wedge for simulated off-magic number cluster. a, An exemplary simulated
magic number cluster (3402 spheres) viewed along two two-fold and one five-fold axis. BODs (insets) show sharp,
bright spots with near-perfect icosahedral symmetry. b, A simulated off-magic number cluster (3738 spheres) has
similar order on the surface but BODs now show duplets of split spots indicating broken icosahedral symmetry. Top
right insets indicate the distribution of defect particles (marked red) inside the cluster (semi-transparent). Defects
accumulate in a wedge perpendicular to a two-fold axis (left), parallel to a two-fold axis (middle) and parallel the five-
fold axis (right). Defect particles near the surface and the different geometry of the anti-Mackay shells were ignored by
analyzing only particles located within 75% of the cluster radius. c, Perfect icosahedron model for the Mackay (and

57
anti-Mackay) cluster. d, Defected icosahedron model obtained by minimizing the stress between tetrahedron grains.
An inevitable gap (approximate size three tetrahedra, boundary marked in red) accommodates the accumulated defects.
e, First peak of the radial distribution function of magic number clusters (green colors) and off-magic number clusters
(red colors).

We confirm the existence of accumulated defects in experiment by means of 360°

electron tomography (ET) in scanning transmission electron microscopy (STEM) mode.
Several colloidal clusters were deposited on a Lacey carbon grid and selected a cluster
with highly-ordered surface pattern. Transferring it onto a tip with a plateau on a
micropillar matching the respective particle size was performed by T. Przybilla from the
Spiecker group.63 The freestanding particle-on-tip geometry enables to perform 360° ET,
which allows for high-precision 3D analyses without “missing wedge” artefacts116. Details
about the transfer procedure can be found in reference 63. The surface rendering of the
tomographic reconstruction of the colloidal cluster (Figure 23a) shows clear local five-
fold patterns and anti-Mackay features with rectangular patterns of width three particles.
The colloidal cluster is viewed along a two-fold axis and contains about 5000 particles, as
determined manually in Visual Reality (VR) visualization software (credit of S. Englisch
from Spiecker group).

The interior structure of the cluster is revealed by slices at different positions through the
three-dimensional reconstruction. The slice at the upper hemisphere cuts through eight
domains of fcc grains (Figure 23b). Most particles are in crystalline domains including
two grains showing {111} crystal planes. The revealed plane exhibit overall two-fold
symmetry. Two small disordered regions at opposing sites are marked by red lines in
Fig. 6b. As the slicing plane moves towards the lower hemisphere, the disordered region
extends towards the middle. Once the slice reaches the lower hemisphere (Figure 23c),
the defect regions (marked in red) join through the center of the image that separates the
crystalline domains, corroborating with the wedge-like distribution of defects predicted
from the model.

A simulated cluster of 5088 particles (credit of C. Mbah), shown in Figure 23d, has
identical structural features as the experimentally investigated cluster underlining that our
two-step simulation scheme accurately replicates the experiment. Cross-sections of this
simulated cluster taken at identical heights as the tomographic data reveal a similar

58
distribution of defect (Figure 23e,f). The upper hemisphere of the simulated cluster shows
a high degree of crystallinity (Figure 23e). Most particles are in crystalline domains
(marked dark green). In the lower hemisphere, the defect wedge appears across the
cluster (Figure 23f). Overall, we conclude that the position of the defect wedge agrees
well between experiment, simulation and geometric model

Figure 23, Electron tomographic reconstruction of a colloidal cluster confirms the accumulation of defects in
a wedge. a, Surface rendering of electron tomographic reconstruction of colloidal cluster with a primary particle size of
about 190 nm. The typical surface tiling with rectangular patterns (width three particles) characteristic for highly ordered
colloidal clusters is observed. b, Virtual slice through the reconstructed cluster at its upper hemisphere perpendicular
to a two-fold axis reveals the interior structure with eight ordered grains. Two small disordered regions (marked by red
lines) are visible at opposing sides. c, Slice at the lower hemisphere. Only six ordered grains are seen, covering a
reduced area. The defect region (marked by red lines) extends from two opposing sides (present in (b)) towards the
center separating the ordered regions into two parts. The defect region lies along the two-fold axis (oriented in viewing
direction). d, A simulated off-magic number cluster (5088 spheres) is shown for comparison. Particles are colored by
the number of nearest neighbors. Particles in dark green are crystalline with 12 neighbors. Particles in white and light
green are in defect regions with lower numbers of neighbors. e,f, Slices through the upper (e) and lower (f) hemisphere
reveal a wedge-like defect region, agreeing with the observations in the reconstructed experimental colloidal cluster
(b,c).

We systematically analyze all simulated clusters containing between 1000 and 5000
particles. For off-magic number clusters defects accumulate in a wedge as described
above whereas in magic number clusters much fewer defects are present. We plot the
fraction of defect particles as a function of system size and compare it to the free energy

59
landscape in Figure 24. The fraction of defects fluctuates as cluster size increases and
is minimal in the magic number regions where free energy has a minimum. As the system
size transitions out of a magic number region, the fraction of defect particles drastically
increases. The correlation between the fraction of defects in the hard-sphere simulations
and the free energy landscape underlines the entropy-driven magic number effect in
colloidal clusters.

Figure 24, The fraction of defect particles in colloidal clusters is correlated to the free energy landscape. The
calculated free energy of colloidal clusters is shown in green, the fraction of defect particles is shown in green. Only
particles in the cluster interior (75% of the radius) are used for the analysis to exclude effects of surface tiling.
Interpolated lines are guides to the eye. In magic number regions, the fraction of defect particles is low. In off-magic
number regions, corresponding to higher free energy, the clusters contain a higher fraction of defect particles.

In Chapter 4, we use the framework of magic numbers to understand the structure of

colloidal clusters at magic numbers. In this chapter, we focus on the scenarios between
the exact magic numbers. Just as 14 spheres cannot pack into an icosahedron, geometric
rules only allow certain numbers of spheres to assembly perfectly icosahedral concentric
shell structures. We analyze the free energy landscape of colloidal clusters as a function
of system size and identified regions with low free energy that correspond to magic
number clusters with closed shell structure. In contrast to atomic magic number clusters
that show discrete free energy minima, the free energy minima in colloidal clusters are
broadened. Three mechanisms are found that contribute to the broadening of minima in
the free energy landscape of colloidal clusters. Whenever possible, our confined self-
assembling system attempts to maintain global icosahedral symmetry with closed,
concentric shells but can efficiently accommodate fluctuations up to several hundred

60
particles via adjustments within anti-Mackay layers, mixing anti-Mackay layers, and
fluctuations in order near the vertices of the cluster. Outside the magic number regions,
icosahedral symmetry is clearly broken. In these regions, the system accumulates defects
in a wedge separating grains instead of evenly distributing stress throughout all 20 grains.
A defected icosahedron model is proposed to account for the distribution of defects in off-
magic number clusters. The tomographic reconstruction of an experimentally obtained
cluster and defect analysis of simulated clusters confirm this defect accumulation. The
fluctuation of free energy is shown to be associated with the fraction of defect particles in
the cluster.

61
 Chapter 6
The Breakdown of Magic Numbers in Colloidal Clusters
A manuscript based on the results described in this chapter will be submitted as:
Junwei Wang, Chrameh Fru Mbah, Praveen Bommineni, Michael Engel, and Nicolas
Vogel
The Breakdown of Magic Numbers in Colloidal Clusters (in preparation)

I performed the colloidal cluster experiments, proposed the geometric model and
performed the theoretical calculation. C.F. Mbah performed the simulation and free
energy calculation. P. Bommineni contributed to the data analysis. M. Engel and N. Vogel
supervised the work.

In Chapter 4 we demonstrate that magic numbers exist in colloidal clusters, even though
interparticle potential is hard sphere-like. In Chapter 5 we find that when the numbers of
particles do not allow the formation of magic number structures, in other words, for off-
magic numbers, excess defect particles become inevitable in the cluster, destroying
crystalline order. The framework of magic numbers already helps to advance the
understanding of structure formation in confined colloidal self-assembly. As magic
number is inherently a finite feature in small systems, it must disappear in large systems.
However, the boundary between small and large system is unclear. For some physical
systems, magic numbers have been observed in clusters up to 20,000 building
blocks,109,117 although Mackay initially proposed that an increasing geometric strain
renders large magic number structures unfavorable.51 To date, the upper limit of system
sizes forming magic number structures is unknown. It is also not clear when and how the
magic number effect breaks down. Second, as the number of building blocks increases,
characteristic properties of finite systems eventually transition into bulk properties, which
poses the question on whether the breakdown of magic numbers is directly associated
with this transition. In this chapter, we discuss the breakdown of magic numbers.

62
Colloidal clusters with non-close shells

To investigate the evolution of magic number configurations in colloidal clusters, we start

by comparing the structure of such clusters with different system sizes. Figure 25a shows
a small icosahedral Mackay colloidal cluster with N~2000. As discussed in Figure 12,
such clusters exhibit a characteristic surface tiling of 20 close-packed hexagonal {111}
planes. Five hexagons share their sides and surround the vertex regions at the five-fold
symmetry axes of the icosahedron. The sphere packing model accurately describes the
observed clusters and highlights its close shell structure at the surface (Figure 25b). The
interior close shell structure is illustrated by the constituent tetrahedral grains of the
cluster (Figure 25b, left), where layers of close packed spheres are stacked in an
ABCABC sequence from the center to the surface.

With increasing numbers of particles, medium size colloidal clusters exhibit a more
complex surface tiling (Figure 25c). Characteristic rectangular facets of {100} planes
(yellow) share sides with hexagonal facets of {111} planes (green). The rectangular
regions are characteristic for the presence of anti-Mackay shells near the surface.52,118,119
These shells are produced by a twinning plane over the interior Mackay shells. 56,57 The
inclusion of anti-Mackay shells in the cluster improves the packing efficiency in the
spherical confinement.56,57,120 The additional facets are a compromise to efficient packing
within the spherical curvature while maintaining close shell structures, as highlighted in
the model (Figure 25d). The number of anti-Mackay shells determines the area of
rectangular facets at the surface. In this example, four anti-Mackay shells (light blue)
twinned over Mackay shells (blue) give rise to surface rectangle of width of five spheres
(Figure 25c).

Intuitively, to maintain both icosahedral symmetry and close shell structure, the inclusion
of anti-Mackay shells could continue for even larger clusters, resulting in increasingly
larger {100} facets at the surface. However, large colloidal clusters show no such surface
rectangles. Instead, they consist of discrete terraces of {111} facets separated by {110}
facets (Figure 25e, N~50.000). The shell closure is clearly broken at the cluster surface.
With no anti-Mackay shells present, such clusters are modelled as truncated Mackay
icosahedra (Figure 25f). The failing of shell closure at the few outermost layers is

63
highlighted by the presence of distinct steps of the surface terraces in the tetrahedral
grains of the cluster (Figure 25f, left). As the magic number effect is manifested by the
formation of close shells, this change in cluster structure hints at the breakdown of magic
numbers with increasing system sizes. Noteworthily, bulk properties are not completely
recovered at this system sizes in spherical confinement: the typical fcc symmetry known
from bulk colloidal crystals is reached at much larger system sizes (N>100.000). 48,83,121

Figure 25, three types of Icosahedral colloidal clusters in spherical confinement. a, small clusters show
hexagonal surface tiles around local five-fold axes. b, the sphere packing model and one of its constituent tetrahedral
grains highlights the prevalence of connected {111} surface facets and the concentric close shell structure. c, medium
clusters maintain close shell structures with hexagonal ({111}) and rectangular ({100}) surface tiles. d, the sphere
packing model and one tetrahedral grain shows the surface anti-Mackay shells (light blue) produced by twinning over
the interior Mackay shells. These anti-MacKay shells allow the formation of a close shell cluster by increasing its
sphericity. e, large clusters do not show a close shell surface structure but separated terraces of {111} planes and
parallel strips of particles at {110} facets. f, a sphere packing model of a truncated Mackay icosahedron reproduces
these surface features, without inclusion of any surface twinning shells. Scale bars: 2 μm.

Geometric requirement for shell closure

To understand the disappearance of close shells, we examine the sphere packing into an
icosahedron cluster in detail. Figure 26a shows model substructures of colloidal clusters
after applying spherical truncation to the anti-Mackay icosahedron model (more detail
description in Figure 12). For a certain truncation radius, the substructure a(i) shows a

64
close shell surface tiled with hexagons sharing sides with rectangles. The presence of
anti-Mackay shells results in the extra 30 rectangular {100} facets, compared to the 20
hexagonal {111} facets in small clusters without anti-Mackay shells (Figure 25a). These
anti-Mackay shells reduce the volume of the gap between cluster surface and spherical
confinement by reducing the area of each flat facet. As a result, such anti-Mackay clusters
pack more efficiently and closer to the curved interface of the spherical confinement and
are the dominant structures observed for colloidal clusters. As shown in Figure 26a(ii),
slightly increasing the truncation radius includes one extra adatom-like sphere (green) at
the surface. Continuous increase of the truncation radius provides space between cluster
surface and curved confinement to add additional spheres to the facets, forming islands
of seven spheres on the faces (green) and two spheres on the edges (yellow, Figure
26a(iii)). Such adatoms and terraces, which disrupt local order and cause entropy
penalties, are avoided by colloidal clusters and never observed in experiments and
simulation at this system size.48 Further increase of the truncation radius result in even
larger islands at the surface (Figure 26a(iv)), until these islands merge to form a new
complete close shell (Figure 26a(v)). The surface rectangle in the new close shell
increases its width (from three to four) and decreases its length (from six to five, Figure
26a(i) to Figure 26a(v)).

Figure 26, Geometric requirements to maintain close shell structures. a, spherical truncation of the sphere packing
model with increasing radii creates substructures that numerate all possible cluster configurations with icosahedral
symmetry. The images show magnified areas of the cluster model, focusing on the evolution of {111} and {100} facets.

65
i, a certain truncation radius produces a close shell structure with {111} and {100} facets. ii – iv, slightly increasing the
truncation radius enlarges the gap between the cluster surface and the curved confinement, which now includes
additional, isolated particles. v, further increase of the truncation radius forms a new, close shell magic number structure
by connecting the previously isolated regions of additional particles. vi, further increasing the truncation radius again
includes extra spheres at the cluster surface.

Evidently, all possible icosahedral substructures are determined only by the size of
Mackay icosahedron and the truncation radius (Chapter 4, Figure 12). Mathematically,
these two parameters can be infinitely large, but, as we show in the following, geometric
consideration shows that shell closure becomes impossible at a critical size. To simplify
the calculations, we used regular tetrahedra to represent the non-crystallographic grains
in the icosahedral cluster in Figure 27 (such as ABCO and ABCO’).51 Spherical truncation
produces a triangle DEF (green) and a rectangle EDGH (yellow), which represent the
hexagonal {111} and rectangular {100} facets at the cluster surface. The truncation radius
relates to the number of surface anti-Mackay shells (compare Figure 26a, i and v). The
distance between the cluster surface and the curved confinement is represented by line
segments JK and MN- these segments describe the distance to the faces of the
icosahedron, DFace and the distance to the edges of icosahedron, DEdge, respectively).

As illustrated in in Figure 27, for shell closure, this gap must not be large enough to fit
extra spheres, which translates to DFace and DEdge being simultaneously smaller than their
critical values. These critical values correspond to the distance between {111} planes and
{100} planes in a fcc lattice, respectively. Shell closure is violated when the distances
become large enough to fit an additional sphere over the icosahedral faces (Figure 26a(ii);
DFace too large), over the edges (Figure 26a(vi); DEdge too large) or over both (Figure
26a(iii) and Figure 26a(iv)).

66
Figure 27, Geometric calculation of the gap between cluster facets and confinement. The slightly deformed
tetrahedral grain of identical sphere packing in the icosahedron model is represented with regular tetrahedra ABCO
and twinned with other tetrahedra. The rectangle DEHG represents the rectangular facet in the cluster, the triangle
DEF represents the hexagonal facet in the cluster. The gap between the facets and the curved confinement is
represented by line segment MN (Dedge) and JK (Dface). Since any additional spheres on the facets must be avoided,
close shell structure is attainable when this length is shorter than the diameter of the constituent particle.

Now consider in Figure 27 that the regular tetrahedron ABCO has an edge OC of length
𝑠, twinned with tetrahedron ABCO’. The spherical truncation centers at point O and
intersect the tetrahedron ABCO’ at point F, the radius OF has a length of 𝑅, line segment
FC has length of 𝑥. For triangle FCO, according to laws of cosines:

𝑅 2 = 𝑥 2 + 𝑠 2 − 2𝑥𝑠 cos ∠𝐹𝐶𝑂

67
Where the angle ∠𝐹𝐶𝑂 equals ∠𝑂′𝐶𝑂 and equals two times the tetrahedron face-vertex-
1
edge angle α = cos −1 . hence:
√3

𝑑(𝐹𝐶) = 𝑥 = 𝑠 cos 2𝛼 ± √(𝑠 cos 2𝛼)2 + 𝑅 2 − 𝑠 2

As the truncation radius OF must be larger than OC and smaller than OO’ to have an
intersection:

𝑥 = 𝑠 cos 2𝛼 + √(𝑠 cos 2𝛼)2 + 𝑅 2 − 𝑠 2

where 𝑠 < 𝑅 < 2𝑠 sin 𝑎.

The line segment 𝑂𝐽 is normal to the triangle 𝐴𝐵𝐶 and 𝐷𝐸𝐹, and has the length of the
truncation sphere radius. Hence the distance between truncation sphere and cluster
surface is:

𝑑(𝐽𝐾) = 𝑑(𝑂𝐽) − 𝑑(𝑂𝑃) − 𝑑(𝑃𝐾)

where 𝑑(𝑂𝐽) = 𝑅, 𝑑(𝑂𝑃) = 𝑠 sin 𝛼, which is the height of the tetrahedron ABCO, 𝑑(𝑃𝐾)
is the distance between triangle ABC and DEF, which equals to the distance from point F
to triangle ABC.

Therefore, the distance between the truncation sphere and cluster surface can be
expressed by the tetrahedron and the truncation radius:

𝑑(𝐽𝐾) = 𝑅 − 𝑠 sin 𝛼 − 𝑥 sin 𝛼

𝑑(𝐽𝐾) = 𝑅 − 𝑠 sin 𝛼 − (𝑠 cos 2𝛼 + √(𝑠 cos 2𝛼)2 + 𝑅 2 − 𝑠 2 ) sin 𝛼

where 𝑠 < 𝑅 < 2𝑠 sin 𝑎.

Similarly, the distance between truncation sphere to cluster surface at the rectangle
DEHG can be expressed as:

𝑑(𝑀𝑁) = 𝑑(𝑂𝑀) − 𝑑(𝑂𝑄) − 𝑑(𝑄𝑁)

where the length of OM equals the truncation radius 𝑅, the length of 𝑂𝑄 is the distance
√3
between tetrahedron vertex O to midpoint Q in the opposing edge 𝑠.
2

68
To simplify the calculation, we consider the tetrahedron ABO’O’’ to be a regular
tetrahedron, neglecting that the dihedral angle at the edge AB is in fact expanded by
about 7.4°. The length of line segment QN is the distance between rectangle DEHG and
midpoint Q in AB. Due to triangle similarity, it follows:

𝑑(𝑄𝑁) 𝑑(𝐴𝐷)
=
𝑑(𝑄𝑇) 𝑑(𝐴𝑂′ )
1
where the length of QT is the edge to edge distance in tetrahedron 𝑠, the length of AO’
√2

is the edge of tetrahedron 𝑠, 𝑑(𝐴𝐷) = 𝑑(𝐶𝐹) = 𝑥, hence:

√3 𝑥
𝑑(𝑀𝑁) = 𝑅 − 𝑠−
2 √2

√3 𝑠 cos 2𝛼 + √(𝑠 cos 2𝛼)2 + 𝑅 2 − 𝑠 2

𝑑(𝑀𝑁) = 𝑅 − 𝑠−
2 √2

where 𝑠 < 𝑅 < 2𝑠 sin 𝑎.

For sphere packing (diameter of each individual sphere = 1) in face center cubic crystal,
√6 √2
the distance between {111} plane is , the distance between {100} plane is , the
3 2

boundary condition to ensure no additional sphere can fit into the gap between cluster
surface and truncation sphere is:

√6 √2
𝑑(𝐽𝐾) ≤ 𝑎𝑛𝑑 𝑑(𝑀𝑁) ≤
3 2

Discretizing the length of tetrahedra 𝑠 to be integer from 3 to infinite (Mackay icosahedron

of size 3 to infinitely large), the radius of truncation sphere is in the range 𝑠 < 𝑅 < 2𝑠 sin 𝑎.

As shown above, the distance between cluster surface and confinement (over the cluster
face, DFace of the value 𝑑(𝑀𝑁) and over the cluster edge, DEdge of the value 𝑑(𝐽𝐾)) can
be expressed analytically using the size of the tetrahedron 𝑠 and the truncation radius 𝑅
as two independent parameters.

69
Figure 28, numerical calculation of the gap between the cluster surface and the confinement as a
function of the truncation radius.

We perform the numerical calculation in a Python code. As show in Figure 28a, over the
faces of the cluster, the grey color represents truncation radii where JK or Dface is greater
than its critical distance, forming non-close shell structures where extra sphere can fit in
the gap. Green represent truncation radii where JK or Dface is shorter than the critical
length, forming close shell structures. The critical length is determined by the distance
between {111} crystal planes in a fcc lattice. The increase of truncation radius 𝑅 is gradual,
but the number of shells is discrete integer. Hence the gap distance D face is not a
continuous function of truncation radius. The shown calculation is based on an extended
Mackay icosahedron (m = 10, Figure 12), where the spherical truncation includes at most
10 surface shells (a = 10) in the substructures, hence the 10 individual segments in the
diagrams. Figure 28b shows the gap over the edge of the cluster (MN or D edge) as a
function of the truncation radius. Note that here the critical length is the distance between
{100} crystal plane in a fcc lattice. Considering both gaps over the faces and edges of the
cluster (Figure 28a,b), close shell structures can only exist when both JK and MN are
shorter than their critical length, only some truncation radius satisfy this condition, which
are marked by orange line segments at the X axis in Figure 28c. Note that the 7 possible
close shell structures are calculated based on the icosahedral model for the case of
m = 10 (Figure 12), when we perform the same calculation for m = 1, 2, 3, …, all possible
icosahedral close shell structures under spherical confinement can be enumerated.

70
Breakdown of magic numbers

We plot the theoretical prediction of all possible icosahedral close shell structure as a
function of cluster radius in Figure 29a (orange dots). The prediction of the model
quantitatively agrees with experimental observation of approximately 140 different
colloidal clusters (blue dots). The interior structure of experimental clusters can be
accurately determined from their surface patterns, as established by tomography studies
described in Chapter 4 and Chapter 5 (Figure 13, Figure 14).56,57 For small clusters
(radius < 5), the gap between the cluster surface and the curved confinement is too small
to fit in any extra particles. The icosahedral shell closure is achieved without any surface
anti-Mackay shells, as for example seen in the small cluster shown in Figure 25a. As the
cluster radius increases, the gaps between surface facets and droplet confinement (both
DFace and DEdge) also increases. To avoid the energy penalty associated with isolated
particles on the facets, the cluster includes more anti-Mackay shells, resulting in the
characteristic tiling of triangular {111}- and rectangular {100} facets, seen e.g. in the
medium-sized cluster of Figure 25c. Geometrically, DFace decreases monotonically as a
function of the spherical truncation radius, which means more anti-Mackay shells are
always favorable, as they reduce the gap over the faces of the icosahedral cluster.
However, DEdge has a minimum and increases at large truncation radii, which limits the
number of anti-Mackay shells, as they may increase the gap over the edges of the
icosahedral cluster. These opposing trends result in a critical cluster radius, above which
DFace and DEdge cannot simultaneously remain below their critical values. This failure of
shell closure is predicted and observed for a cluster radius at about 20. For larger clusters
exceeding this critical radius, the inclusion of anti-Mackay shells creates more interior
twinning but fails to form close shells. Therefore, it is no longer entropically favorable.
Instead, the clusters adopt truncated Mackay icosahedron structures and the number of
anti-Mackay shells abruptly drops to zero. Importantly, and in contrast to very small
clusters, these Mackay shells now do not form a close shell but separated terraces
(Figure 25e).

71
We compute the free energy of hard sphere colloidal clusters in hard spherical
confinement up to large system sizes up to 20,000 (Figure 29b, credit of C. Mbah) at a
packing fraction of 0.52. The fluctuations of the free energy per particle in the colloidal
clusters reveals the magic number effect as discussed in Chapter 4. When the number
of constituent spheres allows close shell structures, the clusters exist in local minima of
free energy. The periodic occurring energy minima in the free energy landscape with the
increasing cluster radius correspond to the inclusion of increasing numbers of close shells
(Figure 26a-f). The strength of the magic number effect is indicated by the magnitude of
free energy minima. The energy gain for each shell closure slowly attenuates to zero with
increasing radius. At a cluster radius around 14, no fluctuations in free energy are
observed, marking the breakdown of the magic number effect. This system size at this
breakdown in simulations coincides with the first occurrence of truncated icosahedron
without close shells in experiments (marked by small-dash lines). Interestingly, above a
radius of 15, close-shell clusters with large numbers of anti-Mackay and truncated
icosahedral clusters without close shells are both observed in experiments (between
small- and large-dash line in Figure 29a). This coexistence agrees with the free energy
calculations, which do not show an energy gain for the formation of closed shells. For
larger clusters with radii exceeding 20, close shell clusters vanish in experiments (large-
dash line in Figure 29a), in agreement with the geometrically predicted inability to form
such close shells.

Combining free energy calculations, experiments and theory, the failure of shell closure
marks the breakdown of magic numbers in colloidal clusters, rooted in their icosahedral
geometry. For a wide size range, close shell structures are attainable and favorable as
surface twinning improves local packing near the confinement. At larger system sizes, the
formation of close shell structures becomes geometrically impossible and the colloidal
clusters adopt by forming truncated Mackay icosahedra with low-coordinated surfaces
and separated terraces. The failure of shell closure is directly encoded in the nature of
icosahedral symmetry as the gap sizes between the cluster facets and the spherical
confinement evolve differently over the faces and edges. Coincidently, for colloidal cluster
exceeding the critical radius, we occasionally observe clusters with fcc symmetry in
experiments,48 but most clusters maintain icosahedral symmetry. This suggests that the

72
breakdown of magic numbers is not directly associated with the transition between finite
and bulk properties. This observation indicates that kinetic effects rather than
thermodynamics play a major role for the transition from finite size- to bulk structures in
spherical self-assembling systems.122 We will discuss such kinetic effect in Chapter 8.

Figure 29, evolution of colloidal cluster geometry as a function of system size. a, the number of close shells at
the cluster surface increases with cluster radius, until an abrupt breakdown, where shell closure becomes geometrically
impossible. This behavior is predicted by theory (orange) and observed in experiments (blue). b, the free energy of
colloidal clusters fluctuates as a function of cluster radius. Alternating energy minima correspond to magic number
regions that allows formation of close shell structures. The magic number effect, indicated by the magnitude of the free
energy well, attenuates to zero towards larger cluster size, coinciding with the breakdown of close shell structures
observed in experiments (dashed line).

73
 Chapter 7
Structural Color as a Tool to Investigate Structures and
Dynamics
The results described in this chapter were published in:
Junwei Wang, Umair Sultan, Eric S.A. Görlitzer, Chrameh F. Mbah, Michael Engel,
Nicolas Vogel
Structural color of colloidal clusters as a tool to investigate structure and dynamics.
Advanced Functional Materials 30, 1907730 (2020)

I performed the colloidal cluster experiments, optical characterization and generate the
sphere packing model. U. Sultan and E.S.A. Goerlitzer contributed to the optical
measurements. C.F. Mbah performed the simulation. M. Engel and N. Vogel supervised
the work.

The previous chapters discuss the structure of colloidal clusters in detail. One key
component throughout our investigation is the accurate identification of cluster structure
and geometry. Such detailed structural information can be obtained from electron
tomography, as shown, e.g. in Figure 13 and Figure 23. This method is time consuming,
technically demanding and limited to only a small number of samples. We also heavily
rely on SEM to identify cluster structure by examining their surface. Although the surface
contains ample information, even to the extent to indicate the interior Mackay and anti-
Mackay shell combinations (Figure 12 and Figure 15), they can sometimes be
misleading. Especially, as shown in Figure 22, even defected colloidal clusters with
system sizes that do not allow shell closure can appear well-ordered in most surface
regions. An alternative to investigate cluster structure is to not look at individual particles,
but their ensemble. As the size of the constituent colloidal particles forming the cluster is
in the range of wavelengths of visible light, their periodic arrangement diffracts light
constructively to give rise to structural color (Chapter 1). In this chapter, we show that the
structural color emerging in the colloidal clusters can be used to identify symmetries and
structures. More importantly, it can be used to monitor the dynamics of such clusters,

74
showing, for example the rotation of consolidated clusters in liquid, or provide a
convenient way to follow the crystallization pathways of colloidal particles in droplets.

Colloidal clusters with three types of symmetries

Typically colloidal clusters obtained from fast droplet drying are spherical clusters with a
hexagonally close-packed monolayer of colloidal particles containing grain boundaries 42
at the cluster surface (Chapter 3).49,123,124 Commonly, a few concentric monolayers are
stacked near the surface, forming onion-ring-like shells.49 Upon illumination, the
periodicity of these stacked monolayers generates structural color when the size of the
colloidal particles matches the constructive interference conditions with visible light. As a
result of the spherical symmetry, such clusters show an isotropic color motif consisting of
a colored circle at the cluster center when observed in a microscope.49,123 With decreasing
drying speed, the system can equilibrate into its minimum energy configuration (Chapter
3). We have introduced that upon reaching a critical packing fraction, the colloidal
particles undergo a phase transition from a fluid into a crystalline cluster,48 where
symmetry is dictated by the system size. Icosahedral clusters are dominantly observed
when the number of constituent colloidal particles is realtively small (Chapter 4).
Decahedral clusters occur less frequently and are irregularly distributed over a broad
range of sizes117 (details will be discussed in Chapter 8). Larger numbers of particles
within the confinement favor fcc clusters,48 which will be discussed in this chapter. Here,
before describing the observed, anisotropic structural color motifs resulting from these
defined, crystalline colloidal clusters, we first introduce and compare all three sphere
packing models describing the three-dimensional structures of the colloidal clusters of
three different symmetries (Figure 30).

The sphere packing model of a large icosahedral cluster (Figure 30a), as discussed
extensively in previous chapters, is based on a Mackay icosahedron51 (upper inset),
which consists of 20 slightly deformed tetrahedral fcc grains. Spherical truncation results
in a quasi-spherical cluster shape that exhibits the characteristic features of twenty
hexagonally close-packed {111} patches surround twelve five-fold symmetric axes at the
cluster surface. An SEM image of a complete icosahedral cluster (Figure 30b) shows five
{111} patches surrounding a five-fold axis at the surface. In previous chapters, we

75
characterized the three-dimensional structure of smaller icosahedral colloidal clusters by
electron tomography (Figure 13 and Figure 23). Clusters as shown in Figure 30 are too
large to be penetrated by electron beams. We therefore resort to a simpler method to
reveal their interior structure. We ultrasonicate the dispersion containing the clusters and
examine their fractured remains. A broken piece of an icosahedral cluster contains
remains of six tetrahedral grains (Figure 30c). The model opened at a similar fracture
point (inset) matches the experimental fracture surface.

The sphere packing model of a decahedral cluster (Figure 30d) is based on a pentagonal
bipyramid, which consists of five slightly deformed tetrahedral fcc grains (upper inset),
each twinned with two of its neighbors.125,126 The thermodynamics and formation
pathways of decahedral clusters will be discussed in detail later (Chapter 8). The grains
in decahedral clusters are also slightly deformed in comparison to the ideal fcc lattice.
While {111} planes form an angle of cos(1/3) ≈ 70.5° in the fcc lattice, the dihedral angle
of grains in the decahedral cluster is 360°/5 = 72°. The surface of the truncated structure
(lower inset) exhibits ten {111} patches in direction of the face normals of the pentagonal
bipyramid, five each in the upper and lower hemisphere, forming two rings surrounding
the sole five-fold axis. Perpendicular to this axis, five square-packed {100} patches are
found. The model is rotated to facilitate comparison with an experimental cluster in Figure
30e. In the SEM image, only one region with five-fold symmetry identified by the presence
of five {111} patches surrounding a five-fold vertex is visible at the cluster surface. The
remaining structure has alternating {100} and {110} patches perpendicular to the five-fold
axis. The absence of other regions with five-fold symmetry is the main difference to the
icosahedral cluster. The interior of a fractured decahedral cluster reveals two {111}
planes joined at an angle (Figure 30f). Its structure agrees well with a constructed
decahedral model from three tetrahedral grains with 144° dihedral angle (inset).

The sphere packing model of a fcc cluster (Figure 30g) is based on a single-crystalline
octahedron-shaped grain without deformation, twinning, or stacking fault (upper inset).
The octahedron has eight faces, which, after spherical truncation (lower inset) are
converted to eight {111} patches. The six vertices of the octahedron are truncated to
reveal six {100} patches. Truncating the twelve edges reveals {110} patches. The

76
surface features of a fcc cluster in SEM are alternating {111} and {100} patches (Figure
30h). A fractured fcc cluster (Figure 30i) has highly crystalline interior consisting of a
single grain in agreement with the model.

Figure 30, Surface and interior structure of crystalline colloidal clusters. Rows show the three main crystalline
cluster types: a-c, icosahedral cluster; d-f, decahedral cluster; g-i, fcc cluster. Columns show structure models with two
steps of the construction process in insets (a,d,g), SEM images of an intact cluster (b,e,h), and SEM images of fractured
cluster fragments obtained by ultrasonic fracturing (c,f,i). PS particle size is 244 nm. Scale bars: 2 μm.

We now analyze the structural color motifs of the three crystalline clusters with an optical
microscope in reflection mode. Structural color is predominantly observed whenever a
{111} plane is oriented perpendicular to incident light so that light reflected from the
colloidal cluster is recollected by the objective lens (Chapter 1).49,83,127 Only light waves
that match the Bragg condition interfere constructively, thus producing structural color
with wavelength determined by the size and refractive index of the colloidal particles. We

77
now analyze colloidal clusters to identify crystal planes matching this condition and
subsequently correlate the model structure to the observed color motifs.

Structural color of fcc clusters

The model of an ideal fcc cluster without defects is a single fcc grain (Figure 31a). A few
hexagonally close-packed {111} layers at the top (green) are lifted for illustration. All
{111} layers throughout the cluster follow the ABCABC packing sequence (Figure 31b).
However, the fcc clusters we find are rarely perfect single grains. This is seen in molecular
dynamics simulations that assemble clusters that closely match the experimentally
prepared clusters48,56 and therefore allow a detailed investigation of their interior structure
(credit of C. Mbah). A large fcc cluster assembled in simulation exhibits a sequence of
twinned fcc domains (Figure 31c). The first fcc domain consist of three stacked layers in
ABC sequence (green). The next two layers are stacked in BA sequence (grey) and thus
are twinned to the first domain, and so on. Importantly, all twinning planes are parallel
throughout the fcc cluster, creating a laminated texture. We now focus on three exemplary
illumination scenarios of the fcc cluster as indicated in the figure.

In the first scenario, incident light impinges on the cluster from the top, perpendicular to
the {111} plane (Figure 31c, incident light 1). All stacked layers are positioned
periodically with identical distance, which is not affected by the twinning between
laminated fcc domains. Figure 31d schematically illustrates this scenario by showing the
cluster from the direction of the incident light. Particles that match the interference
condition discussed above are marked in green. In this orientation, all particles within the
cluster contribute to the structural color.83,128 Figure 31e characterizes an fcc clusters by
SEM and optical microscopy. The orientation perpendicular to the {111} plane can be
seen from the close-packed layer right at the top center of the cluster. It agrees well with
the structure of the simulated cluster. The entire cluster shows structural color in optical
microscope images underlining that the fcc structure (with laminated twinning planes)
persists throughout the cluster. The wavelength of structural color can be tuned over the
visible range by changing the size of the particles (Figure 31e, inset). Structural color is
observed throughout the spherical outline of the three clusters demonstrating a high
degree of crystallinity for all colloid sizes.

78
In the second scenario, the direction of incident light is rotated by 60° and impinges on
the cluster from the side (Figure 31c, incident light 2). Due to the symmetry of the fcc
lattice, incident light from this angle meets the first domain perpendicular to another {111}
plane (marked in green). However, because twinning changes the orientation of adjacent
fcc domains, only every other domain is oriented in a way to create structural color. The
simulated cluster depicted from the direction of the incident light (Figure 31f) then
contains stripes alternating between domains that reflect light (marked in green) and
domains that do not reflect light (marked in grey). Close inspection of experimental fcc
clusters with an orientation matching the orientation of the simulated cluster reveals
multiple twinning planes as inferred from the surface structure (Figure 31g). Optical
microscopy produces the expected stripe-like motifs (Figure 31g) corroborating the
structure model. As before, the wavelength of structural color can be tuned via the size
of the colloidal particles (Figure 31g, inset). Note that all twinning planes in fcc clusters
are parallel, which is in contrast to earlier observations of multi-grain textures with
uncorrelated orientations caused by independent nucleation events. 78,81

In the third scenario, the incident light is not aligned perpendicular to any {111} crystal
plane of the cluster (Figure 31c, incident light 3). Although higher order crystal planes
may be oriented to fulfill a Bragg condition, their larger separation distance does not
produce structural color in the visible range. The analysis of the simulated cluster predicts
no structural color (Figure 31h), which is also confirmed by optical microscopy for all
particle sizes (Figure 31i). Fcc clusters suspended in a liquid phase are mobile. They
continuously change their color appearance by rotating through the different illumination
scenarios.

79
Figure 31, Structural color of fcc colloidal clusters. a, Structure model highlighting stacked hexagonally closed-
packed {111} layers. b, ABC stacking sequence of the layers. c, Simulated fcc cluster from 90,000 hard spheres in
spherical confinement after removing surface defects and shrinking the spheres for clarity. The side view image reveals
a sequence of fcc domains with alternating orientation (green and grey) separated by stacking faults. Three illumination
scenarios are indicated. d,f,h, Views of the simulated cluster along the three incident light directions. Particles in {111}
planes perpendicular to the direction of incident light are marked in green. Only these particles contribute to the
structural color of the cluster. e,g,i, Experimental characterization of colloidal clusters oriented in the three incident light
directions using SEM (left images) and optical microscopy in reflection mode for colloids of three sizes (small insets).
PS particle size: 230 nm unless noted differently (“col”). Scale bars: 10 μm.

Structural color of decahedral clusters

In the model of a decahedral cluster one grain is marked in green and lifted for illustration
(Figure 32a). As before, structural color is observed only if the incident light impinges
perpendicular to a {111} plane. Two types of cluster orientations afford such conditions
and therefore produce structural color. In Figure 32b, a fan-shaped color motif covering
the center of the cluster is observed in optical microscopy when one of the five grains is
oriented with its {111} plane perpendicular to the incident light. The side view image of
the model shows how the layers of the grain penetrate through the cluster, while the top

80
view image illustrates the expected structural color motif (green particles). Model and
optical microscopy (OM) match. Figure 3c depicts a different orientation affording
structural color. The decahedral cluster is oriented with one of its five twinning planes
towards the incident light, the vertex region points to the side. In this orientation, two
neighboring grains are oriented with their {111} plane perpendicular to the incident light
(Figure 32c, side view). The resulting color motif is a semicircle (Figure 32c, top view),
which matches experimental observation.

Figure 32, Structural color of decahedral colloidal clusters. a, Structure model consisting of five twinned fcc grains
reminiscent of the segments of an orange. b,c, Appearance of structural color under two illumination scenarios. Model
clusters are depicted perpendicular to (side view, particles shrunken for clarity) and along (top view) the incident light.
Green spheres in top view illustrate the geometry of the expected color motif. Experimental color motifs in reflection-
mode optical microscopy (OM) agree with the models. PS particle size: 230 nm. Scalebar: 5 μm.

Structural color of icosahedral clusters

We have shown in Chapter 4 that icosahedral clusters at intermediate system sizes (from
a few hundred to about 10,000 particles) are accurately described by a Pentakis
dodecahedron model (Figure 12). In Figure 33a, one of the 20 tetrahedral grains is lifted
to show the different regions of such clusters. Three anti-Mackay layers (dark green) are
twinned with a tetrahedral grain of the Mackay core over the icosahedral face. A few

81
additional particles in the anti-Mackay region (yellow) locate over icosahedral edges. The
symmetry of the icosahedral cluster produces a number of complex structural color motif.

In Figure 33b, the icosahedral cluster is oriented with a three-fold axis parallel to the
incident light, as most clearly seen at the top of the side view image of the model. In this
scenario, two of the 20 tetrahedral grains have a {111} plane oriented perpendicular to
the incident light and thus contribute to the structural color (marked in green). The
expected color motif is a triangle (top view) in agreement with optical microscopy (OM).

In Figure 33c, the incident light impinges on a two-fold axis of the icosahedral cluster. In
this scenario, two tetrahedral grains located in the cluster interior (marked in green) are
oriented with {111} planes perpendicular to the incident light (side view). In addition, a
part of the surface anti-Mackay layers (marked in yellow) aligns with a {110} plane
perpendicular to the incident light. The crystal planes have a slightly larger lattice distance
compared to the one of the Mackay core resulting in a redshift of the structural color to
yellow. The relative position of the two tetrahedral grains and the anti-Mackay layers
generates a bowtie motif, in which a narrow rectangle connects two triangles (top view,
OM).

The signature five-fold symmetry of icosahedral clusters can also be captured in a

structural color motif (Figure 33d). When observed along a five-fold axis, no tetrahedral
grain fully aligns with the incident light. Nevertheless, the surface anti-Mackay layers in
the lower hemisphere of the cluster have {111} planes that are nearly perpendicular to
the incident light (side view) resulting in star-like motif with five dots at the rim of the
cluster (top view, OM). This is confirmed by observing the same cluster in SEM and under
microscope. Five-fold motifs are difficult to observe in the optical microscope because
they are highly sensitive to cluster orientation and are strongly affected by structural
defects, particularly when clusters outside magic number regions with broken icosahedral
symmetry are expected to show distorted five-fold structural color motifs (Figure 22,
Figure 23 and Figure 24).

82
Figure 33, Structural color of icosahedral colloidal clusters. a, Structure model consisting of 20 tetrahedral fcc
grains in the Mackay core and anti-Mackay layers. b-d, Appearance of structural color under three illumination
scenarios. Model clusters are depicted perpendicular to (side view, particles shrunken for clarity) and along (top view)
the incident light. Green spheres in top view illustrate the geometry of the expected color motif. Experimental color
motifs in reflection-mode optical microscopy (OM) agree with the models. PS particle size: 230 nm. Scalebar: 5 μm.

Dynamics of colloidal cluster monitored by structural color

Now we show that structural color is a useful tool to monitor the dynamics of colloidal
clusters suspended in liquids. To study such dynamic effects, we suspend a group of
colloidal clusters in oil and deposit the suspension on a glass substrate. Oil evaporation
induces convective flow that sets selected clusters into motion. Figure 34a, showing time-
lapse images recorded with a high-speed camera shows details of the cluster motion. In
about 0.04 seconds, an icosahedral cluster with diameter of about 6 μm rotated about 2
μm away from original position. Its bowtie-shape motif disappeared and reappeared with
around 60 ° in-plane rotation. As the change of color motifs is dictated by cluster symmetry
and rotation axis, we infer that the cluster rotated around an axis perpendicular to the

83
incident light (viewing angle). Around this axis, every 36 ° rotation produces the
reoccurrence of bowtie motifs. When the translation of the cluster is caused solely by
rotation on substrate, it equals to cluster circumference 6  19 μm, multiplied by the
amount of rotation 36°/360°, which gives 1.9 μm, in close agreement with experiment.
We therefore deduce that the cluster is rolling on the substrate. Further following this
icosahedral cluster (Figure 34b), after 0.3 seconds, its bowtie motif rotates in-plane
continuously. In this case, the rotation axis is parallel to viewing angle, otherwise triangle
or star-like motifs would have appeared. As in-plane rotation does not contribute to
translation, the displacement of the cluster is caused by cluster sliding, likely due to the
loss of direct contact to the substrate or low friction due to the suspending oil. In a second
experiment, a thin oil film containing two icosahedral clusters is deposited on a glass
substrate (Figure 34c). Three-fold symmetry motifs are visible. During drying, the
receding meniscus forms a circular shape and induces capillary attraction between the
two clusters (Figure 34d). The meniscus has normal component to the substrate which
cause stronger friction between clusters and substrate. At the end of drying, capillary
forces rotate the clusters while pulling them together as evidenced by the change in color
motif (Figure 34e). As these examples demonstrate, anisotropic color motifs of colloidal
clusters are a simple indicator for rotation. Potential applications are the design and
testing of microfluidic devices, investigation of rotational diffusion at liquid interfaces, 129
in colloid suspension,130,131 and comparison of sliding and friction on solid
substrates.132,133

84
Figure 34, Structural color for monitoring the dynamics of colloidal clusters. Optical microscopy in reflection
mode of colloidal clusters suspended in oil on a glass substrate. a,b, Snapshots of an icosahedral colloidal cluster
showing out-of-plane and in-plane rotations, separated by 0.4 seconds. Clusters are set in motion by convective flow.
Supplementary Video S5 contains the full evolution in 1 second. c-e), Time series of two icosahedral colloidal clusters
under capillary attraction during oil evaporation. PS particle size: 244 nm. Scale bars: 10 μm.

The evolution of structural color, paired with the ability to identify the cluster type from its
structural color motif, can be utilized to observe and analyze confined colloidal
crystallization in real time. Figure 35 shows a time series of optical microscope images.
The droplet contains approximately 500,000 colloidal particles. The system size is large
enough to approximate bulk behavior, which favors crystallization into the fcc structure.48
At the beginning of the evaporation process, the droplet has a whitish overall appearance
(Figure 35a). The colloidal particles are in the liquid state and cause random scattering

85
of incident light. As the droplet shrinks and the packing fraction of colloids increases, a
homogenous low-intensity red hue emerges (Figure 35b). This hue originates from short-
range correlations developing among the colloidal particles. 134,135 Upon further increase
in packing fraction, a central red spot appears, accompanied by five regions of grating-
like color49,82,136 near the rim (Figure 35c). This color motif is characteristic of an onion-
ring-like concentric layer structure near the droplet surface, usually observed in spherical
photonic balls fabricated by fast droplet drying (Chapter 3).49,123,124 It is surprising that
long-range correlation is present already at this early stage when the majority of the
colloids still remains in the liquid state. The five-fold symmetry of the motif even suggests
icosahedral ordering within these surface layers.

Figure 35d captures the onset of colloid crystallization within the droplet. A red quasi-
circular region extends from the center to the circumference. The color intensity of the
quasi-circular region keeps increasing (Figure 35e). This suggest that nucleation and
growth occur predominantly from one region, in contrast to the nucleation of icosahedral
clusters.48 Because optical images are two-dimensional projections, it is difficult to decide
whether nucleates starts at the center or at the droplet surface. However, the central red
dot, present already in Figure 35c, remains unaffected by the changes in color around it,
demonstrating that the onion-ring-like surface layers persist through the initial
crystallization process. Figure 35f finally shows uniform color throughout the droplet
outline. At this stage, crystallization into a fcc colloidal crystal has been completed. We
note that some experiments see the onion-ring-like surface layer structure remaining until
the end of the crystallization process even though the interior exhibits fcc ordering. This
suggests that in these cases all or parts of the surface layers remain intact around an fcc
crystalline core. Returning to the time series, after the end of the crystallization process
the colloidal cluster consolidates further. The interparticle distances continuously
decreases as the remaining water evaporates. The result is a blueshift of structural color
from red to orange to yellow to green (Figure 35g-h). With the help of structural color, we
can identify the structure of colloidal clusters easily with an optical microscope within
seconds, which allows more reliable statistics, i.e. concerning the percentage of a certain
symmetry at some system sizes. Structural color also enables the investigation of kinetics
in the crystallization process, which is otherwise inaccessible for our experimental

86
systems. Apart from the particle size being too small, at the late stage of droplet drying,
the interparticle distance decreases eventually to zero, too small for conventional confocal
microscope to image and track. In addition, there are up to 100,000 particles in the
droplets, it is challenging to obtain 3-dimensional position for such a large number of
particles to extract information about their nucleation and crystallization kinetics. Using
structural color, we are able to monitor the growth of a decahedral colloidal cluster in-situ
in Chapter 8.

Figure 35, Structural color shows the crystallization process of colloidal clusters in real time. Time series of an
aqueous droplet of a colloidal dispersion suspended in oil and observed under optical microscope in reflection mode.
Over the course of 30 minutes, the droplet shrinks and undergoes a series of structural transformations as revealed by
the analysis of its structural color. Blueshift of structural color is an indicator of the reduction of interparticle distance.
PS particle size: 230 nm. Scale bars: 10 μm.

87
 Chapter 8
Competing Magic Numbers: Icosahedron vs Decahedron
A manuscript based on the results described in this chapter will be submitted as:
Chrameh Fru Mbah*, Junwei Wang*, Silvan Englisch, Erdmann Spiecker, Nicolas Vogel
and Michael Engel
The Path Less Taken – Decahedral or Icosahedal Symmetry in Colloidal Clusters (in
preparation)

I performed the colloidal cluster experiments, proposed the probability model. C.F. Mbah
performed the simulation, free energy calculation and grain development detection. S.
Englisch perform the X-ray tomography. E. Spiecker, N. Vogel and M. Engel supervised
the work.

Up to now, we have focused on icosahedral (Ih) colloidal clusters extensively. However,

another type of clusters with five-fold symmetry is observed in experiments but much less
frequently. Structural color helps to identify these clusters to have decahedral symmetry.
As a decahedron has only one five-fold axis, it has a lower symmetry than an icosahedron.
Recall that the icosahedral sphere packing model consists of concentrically stacked close
shells, corresponding to discrete magic numbers (Chapter 4). The distribution of these
magic numbers is associated with the icosahedral (Ih) symmetry. As shown in Figure 30,
the decahedral sphere packing model also consists of concentric close shells, which gives
rise to another, and distinctive set of magic numbers associated with decahedral (Dh)
symmetry. For a certain number of colloidal particles in droplets, the particles need to
choose one symmetry over the other. Do icosahedral magic numbers offer more low
energy configurations than decahedral magic numbers, hence the more frequent
occurrence of icosahedral clusters? Or are there aspects other than thermodynamics that
controls the structure formation for colloidal clusters in spherical confinement?

88
Decahedral colloidal clusters

At close inspection, we observe decahedral (Dh) clusters for a wide range of system sizes
up from a few thousands to a hundred thousand (Figure 36a-c), although their occurrence
is generally low. The spherical droplet confinement enforces the highly-crystalline cluster
to produce characteristic surface features observed under the scanning electron
microscope (SEM). In Figure 36b, the Dh cluster exhibits five patches of hexagonally
close-packed regions surrounding a five-fold symmetric axis and patches of square-
packed regions on the surface, highlighted with yellow hexagons and squares. Two edges
of the square align parallel to the edges of the two hexagons near the poles on the cluster
surface with parallel lines of particles in between (Figure 36c). As already discussed in
Chapter 7 Figure 30, The structure of colloidal Dh cluster can be explained by a
pentagonal bipyramid model of equal spheres. The model consists of five grains of slightly
deformed face-cubic-center (fcc) packing, joined at the five-fold axis. Similar to
icosahedral clusters (Chapter 4), spherical truncation is applied to generate quasi-
spherical model structure that accurately reproduce the surface features of observed
clusters (Figure 36f,g). Interestingly, even intricate surface twinning, also observed in
decahedral bimetallic nanoparticles137, similar to the anti-Mackay layers in icosahedral
clusters48,56,118,138, can be captured using extended decahedron model. The interior
structure of colloidal Dh clusters is confirmed and resolved by X-Ray tomography
performed in a ZEISS Xradia 810 Ultra.62 Since Dh clusters are rare, a transfer and search
technique was developed by Silvan Englisch from the Spiecker group, 62 who also
performed the tomography. Figure 36d shows a selected cluster viewed along its five-
fold axis displaying evidently columns of particles arranged pentagonally around the axis,
unique for decahedral symmetry. By morphology and grayscale threshold segmentation
algorithm and manual counting in virtual reality in the software Arivis, the cluster is
determined to have 2286 particles (credit of S. English from the Spiecker Group). Its
reconstructed surface, viewed along five-fold axis, is shown in Figure 36h.

89
Figure 36, Decahedral colloidal clusters. a-c, scanning electron microscope images of decahedral colloidal clusters
crystallized from slow drying of polystyrene particle aqueous emulsion droplets, whose system size ranges from 5,000
to 100,000 particles. The surface is tiled characteristically by ten {111} and five {100} crystal planes, marked by yellow
hexagons and squares. Only one five-fold symmetric axis is visible in b due to random viewing angle. e-g,
corresponding models of decahedral clusters. d, X-ray image reveals multi-twinned crystalline grains sharing one five-
fold axis in the decahedral cluster. h, X-ray tomography acquires individual particle positions, showing five {111} planes
at the surface. Colloidal particle diameter, 230 nm in a-c, 500 nm in d, scale bar, 2 µm.

We first examine mechanism of strain relaxation in colloidal Dh clusters. As the dihedral

angle of grains in the cluster is 360°/5 = 72°, slightly larger than the angle of cos(1/3) ≈
70.5° between {111} planes in the fcc lattice, Dh clusters are known to exhibit internal
geometric strains139,140. We extract particle positions from the experimentally obtained
clusters by X-Ray tomography (Figure 36, credit of S. Englisch). While other grains share
a twinning plane, a stacking fault plane with hcp environment is clearly seen between two
grains (marked in dark green, Figure 37a, a few disordered surface particles are
removed). We focus on the core of the cluster for better illustration. The edge of the core
decahedron consists of four particles; the plane between neighboring grains of six close-
packed particles. The stacking fault plane (marked by dashed line) locates closer to the
Dh center in comparison to an ideal decahedron model (Figure 37b), which leads to slight
lift of the six particles from the side view (Figure 37c). Importantly, the entire column of
particles at the five-fold center is missing, which allows the formation of the plane defect

90
throughout the multi-twinned cluster to relax strain. Absenting central particles in the five-
fold center can increase overall packing fraction51 and is indeed observed in all simulated
clusters (Figure 37d, left, grains are colored randomly). Occasionally the five-fold axis
split to allow two more twinned fcc grains in the cluster (Figure 37d, right), which has
been proposed and observed in decahedral nanoparticles141,142.

Figure 37, Defects in decahedral clusters. a, plane defect (dark green) in a decahedral colloidal cluster, viewed alone
five-fold axis. Particle positions are extracted from X-ray tomography reconstruction. b, the core decahedron in the
reconstructed cluster (light green) shows absence of a column of particles in the five-fold center, allowing the defect
plane to locate closer to the center, compared to ideal decahedron (yellow). c, the defect plane is lifted slightly to ensure
close packing. d, absence of central column of particles is observed in all simulated decahedral clusters (left, grains
are colored randomly). Occasionally, the five-fold axis is split by additional grains twinned with two neighboring grains
which further relax the geometric strain (right).

Magic numbers in icosahedral and decahedral clusters

In Chapter 1 and Chapter 4, we established that Ih symmetry is entropically favored for

hard sphere-like particles in spherical confinement until the system size approaches bulk
values, where fcc dominates48. However, we observe Dh clusters throughout system size
ranging from a few hundred up to a hundred thousand, although it occurs rarely compared
to Ih clusters. To understand the competition between Ih and Dh symmetries, free energy
of Mackay, anti-Mackay Ih clusters (blue) and Dh clusters (yellow) from ideal geometric
construction was calculated (credit of C. Mbah in the Engel Group) in the range of 2600
to 4700 spheres (Figure 38a). Both curves show magic number effects - the free energy
fluctuates as system size increases. This effect arises in finite confined system because

91
geometry dictates that only certain numbers of spheres can form closed-shell structures
or their variations, which adapts more efficiently to the curved confinement interface.
Higher frequency and lower magnitude in the fluctuation in the free energy landscape of
Dh cluster result from its lower symmetry. Importantly, its minima locate at different
positions than those of Ih clusters. Interestingly, the free energy of simulated clusters
(red), obtained by spontaneous crystallization in the confinement, has regions of higher
energy than the Dh clusters, but not the Ih clusters. This suggests that the simulated
clusters often do not equilibrate into Dh clusters, even though they are lower energy
structures. We then examine the occurrence of Ih and Dh clusters at all system sizes,
since each data point in the red curve is averaged over 100 simulations. Figure 38b
shows that Ih clusters appears almost 100% around the minima of Ih curve (blue) in the
magic number region, and gradually decreases to zero when moving away from the
minima into the off-magic number region, as the Ih symmetry is broken by the defect
wedges, as described in Chapter 5. The occurrence of Dh clusters remains lower than
3% and only rise up to about 10% near the minima of Dh curve (yellow).

Figure 38, Competition of magic numbers associated with decahedral and icosahedral symmetry. a, free energy
of icosahedral (blue, Mackay and anti-Mackay model) and decahedral clusters (yellow) from constructed models
fluctuates as a function of cluster size, the latter with smaller magnitude and higher frequency. The free energy

92
landscape of clusters from simulations (red, each point averaged over 120 simulations) is smoother but does not follow
the lower boundary of ideal icosahedral and decahedral curves. b, occurrence of icosahedral clusters from simulations
reaches almost 100% within the minima and decreases to almost 0% between its minima in the icosahedral free energy
curve. c, decahedral cluster occurs at most around 10% (at confinement radius 9.2 and 10.1), although its free energy
reaches minima, lower than icosahedral and averaged simulated clusters. The low occurrence of decahedral cluster
even when free energy favors it suggests strong kinetic effect (credit of C. Mbah from the Engel Group).

Formation pathways

The low occurrence of Dh clusters even when entropy favors it indicates dominant
dynamic effect over thermodynamics in the crystallization process. To investigate kinetics
of cluster formation, an algorithm was developed by C. Mbah from the Engel group to
detect nucleating events in simulation. In experiment, as we established in Chapter 7,
structural color can be used to monitor the crystallization in droplets, as the sizes of
colloidal particles are chosen to match the wavelengths of visible light to give Bragg
diffraction. Our simulation observes that layered fluid near the spherical confinement
already shows Ih ordering, while the interior remains as amorphous fluid, long before
reaching critical packing fraction of crystallization. Similar to the Ih ordering of hard
spheres on two dimensional spherical surface44 (Chapter 1) and the layering structure of
atoms and colloidal particles during quenching49,121,143,144.

The crystallization starts with random nucleating and melting events near the
interface48,143,144, a few layers into the ordered fluid. Afterwards, some persisting grains
become stable and form five compact grains near the interface (Figure 39a, top view,
particles in fluid and unstable nucleates are made transparent for clarity, grains are
colored randomly). These quasi-tetrahedral shaped grains twin with its neighbors and
point towards the confinement center (Figure 39a, side view, rotated 90 degrees from the
top view). This arrangement is reflected by a sharp spot in the center, surrounded by 10
equally-spaced blurred spots at the rim in the bond order diagram (BOD) (Figure 39a,
inset in top view). The diffuse background in the BOD indicates that most particles are
still in fluid phase. To better visualize the grain arrangement, we project the particles with
crystalline environment (small nucleating grains as well as growing grains) to a spherical
surface (Figure 39a, projection map, credit of C. Mbah). The five large triangular domains
in the center result from the compact five grains, separated by five lines, originating from

93
particles in the twinning planes. Clusters of dots can be seen on the projection map,
originating from small unstable nucleus. As the crystallization proceeds, the compact five
grains remain in the same position, but grow past the center towards the other pole of the
confinement, with intensified BOD pattern (Figure 39b, top, side view). The compact five
grains continue to grow, not only towards the other pole but also laterally, until consuming
the majority of particles in the confinement, resulting in a BOD with clearly resolved points
(Figure 39c, top, side view). As a Dh cluster only has one five-fold axis, only two local
five-fold patterns – one in the center and one split at the left and right edge of the stereo
map can be observed. Interestingly, although the two or three outermost layers near the
interface nucleate first and initialize cluster formation, they quickly melt and remain in
liquid state even after complete crystallization (Figure 39c, side view). In addition, some
small grains in these layers, although undergoing nucleating and melting frequently, are
observed to align to the compact five grains on the other side without contact. This effect
is observed more in smaller system sizes, which suggests large fluctuation and long-
range correlation in strongly confined fluid.

We observe the formation of Dh colloidal clusters by structural color in real time under
optical microscope. While the droplet appears overall whitish due to random scattering
of colloidal particles of diameters in 230 nm in liquid state, a triangular region extending
from the rim to the center appears yellow (Figure 39d, top). This region, indicating
crystallized grains, continues to grow past the center towards the other pole, eventually
filling half the circle (Figure 39d middle, bottom) - the characteristic structural color
pattern of colloidal Dh clusters (Figure 32). The droplets are immobilized in a sealed
PDMS chamber for slow drying. Starting from packing fraction of 1%, the complete drying
takes approximately 3 days, but the crystallization process completes in about 30 minutes.

In contrast, an Ih cluster has different formation pathway. The crystallization starts

similarly with a few twinned quasi-tetrahedral grains near the interface, but often not in
compact configuration, as in the case of Dh clusters (Figure 39e, projection map,
triangular domains are masked in white for clarity). The absence of compact five grains
results in less than 10 blurred spots at the rim of the BOD (Figure 39e, inset in top view).
As crystallization continues, the initially formed grains maintain the same position and
size, while more tetrahedral grains nucleate near them (Figure 39f, top, side view). The

94
projection map in Figure 39f shows 8 grains surrounding two local five-fold patterns. The
nucleation of new grains continues by the faces of existing grains and eventually forms a
complete Ih cluster, whose projection map shows 12 local five-fold patterns resulting from
20 grains (Figure 39g). Interestingly, clusters with off-magic numbers follows the same
formation pathway but leaves a gap of defects in the end, which explains the accumulation
of defect into a wedge domain, shown in Chapter 5.

We also observe Ih cluster formation in experiment. The Ih cluster, viewed along its two-
fold axis, shows a unique bow-tie like pattern, where each triangle in the bow-tie originates
from one tetrahedral grain (Chapter 7 Figure 33). Other grains are not aligned with
incident light to generate structural color. Figure 39h (top to bottom) shows the triangle
on the left forms from its left edge, followed by the triangle on the right forming from its
lower edge. This agrees with observations in simulation that new grains nucleate
sequentially by the faces of existing grains. However, due to Ih geometry, no more than
two grains can be unambiguously identified from the same viewing angle by structural
color.

95
Figure 39, Crystallization pathways of decahedral and icosahedral clusters. a-c, growth pathway of decahedral
clusters. At the beginning, five stable grains nucleate near the confinement interface, forming a compact local five-fold
twinned structure (top view in a, particles in liquid state and unstable grains are made transparent for clarity), which
give rise to a bright central dot and 10 equally spaced dots in the rim in the bond order diagram (inset). These quasi-
tetrahedral shaped grains point towards the center (side view in a). To visualize relative positions of grain during
formation, bonds between nearest neighbor in all crystalline domains are projected to the spherical surface (projection
map in a). As crystallization proceeds, the compact five grains remain in their position (top view in b) but grow towards
the other side past the center (side view in b). Since no new grains form, the pattern in projection map remains similar.
At last, the five grains reach to the other side and consumes most particles in the confinement into their grains (top,
side views in c). There are two local five-fold symmetric pattern in the projection map in c (one in the center and one
split onto both sides) as decahedral cluster has only one five-fold axis. d, a droplet containing 230 nm in diameter
colloidal particles gives overall whitish appearance under optical microscope due to random scattering of particles in
liquid state, a yellow triangular domain indicates the crystalline grains (top image). This domain grows and eventually
reaches the other side, forming characteristic half-circle color pattern of decahedral clusters (middle, bottom image). e-
g, growth pathway of icosahedral clusters. Five stable grains nucleate near the confinement interface but does not form
compact structure (top, side views in e), crystalline regions are highlighted in white triangles in the projection map (e).
As crystallization continues, more tetrahedral grains subsequently nucleate by the side of existing grains (top, side
views in f), evidently shown in the projection map (f). The crystallization completes after nucleation of twenty grains
(top, side views in g). As icosahedral cluster has six five-fold axes, twelve local five-fold patterns are found in the
projection map (g). h, icosahedral cluster shows typical bow-tie color pattern where each triangle indicates a tetrahedral

96
grain (bottom image). Real-time observation confirms that the two triangles nucleate subsequently instead of
simultaneously (top, middle image).

The contrastingly different crystallization pathways prompt the following questions – when
do the particles decide to choose the Dh or the Ih path? To answer this, we simulate a
cluster with 4399 particles at 0.52 packing fraction (credit of C. Mbah), a system size
where entropy does not favor Dh or Ih symmetry. While the pressure slowly decreases
with time (Figure 40a), particles rearrange and spontaneously crystallize into a Dh cluster.
The first pressure-drop region relates to the formation of layered fluid near the
confinement, the second resulting from first-order phase transition. As EDMD simulation
is retractable, for each time frame of this particular cluster, we use its configurations as
input for 192 new simulations and count the occurrence of Ih clusters as its probability.
We find that the fate of this cluster is determined in a very short time interval at the very
beginning of the crystallization (marked by grey bar in Figure 40a). When small quasi-
tetrahedral grains start forming near the confinement interface, the cluster is most likely
to become icosahedral (Figure 40b, unstable grains and particles in fluid phase are made
semi-transparent). When more grains form around a five-fold tip (Figure 40c), the
probability to become Ih cluster drops to 50%. It drops near zero when these five grains
reach towards the center and broaden laterally, their shape becoming more cylindrical
than tetrahedral (Figure 40d). Once the five grains grow past the center (Figure 40e),
the evolution towards Dh cluster is unstoppable.

To rationalize the low occurrence of Dh clusters even when entropy favors its formation,
we hypothesize a simple model, considering the probability to form five-compact triangles
on an Ih template in two-dimension (Figure 40f). The Ih template may be related to the
layered fluid near the confinement interface, which reduce the energy barrier of nucleation.
Given that new triangles only form sequentially by the side of existing triangles, similar to
observations grain nucleation in three-dimension, the first three formed triangles have
only one possible configuration (Figure 40f, in yellow). The fourth triangle can form at five
possible positions by the side of the existing triangles (Figure 40g, blue). After the fourth
triangle selects one position (Figure 40h, marked by a star), six positions become
possible for the fifth triangle. Only when the fifth triangle selects position one or six, will
the first five triangles form compact structures (Figure 40i), which relates strongly to Dh

97
cluster formation. This configuration only forms when position one or five in Figure 40g
2×2
and position one or six in Figure 40h are selected, with the probability 5×6 ≈ 13%.

Figure 40, The unusual pathway to decahehedral clusters in a stochaistic process a, the pressure during the
formation of a selected decahedral cluster is plotted against simulation time. The first pressure-drop region relates to
layered fluid formation near the spherical confinement, the second first-order phase transition of hard spheres from
liquid to solid. Configurations in each simulation time is taken as input for a hundred new simulations. The occurrence
of icosahedral clusters after complete crystallization is counted to represent its formation probability. The inset in a
shows that the probability to form icosahedral cluster drop from nearly 100% to 0% within a very narrow time window
at the beginning the crystallization process (marked by grey bar in the red curve). b, this probability is very high when
the first one or two quasi-tetrahedral shape grains nucleates near the confinement interface (unstable nucleus and
liquids are made transparent for clarity). c, the probability drops to about 50% when five stable grains form compact
structures. d, the probability drops to nearly 0% when these grains reaches the confinement center and become more
cylindrical shape. e, the probability becomes 0% after the five grains grow past the confinement center. f, consider a
simple model where triangles appear sequentially by the side of existing triangles on an icosahedral template (marked
in grey). The first three triangles (yellow) has only one possible configuration. g, there are five sides where the fourth
triangle can appear (marked in blue). h, when the fourth triangle (marked by a star) takes position 1 in g, six sides are
possible for the fifth triangle to appear. i, the first five triangles form a compact structure only when position 1 or 6 in h
is taken, the probability being 13%.

98
In this chapter, we show that the unusual low occurrence of decahedral cluster is not
dictated by thermodynamics – in fact, such clusters often exhibit a lower free energy than
icosahedral clusters. Instead, it is caused by kinetics, as probability biases against a
compact five-grain seed towards decahedral symmetry. We identify the critical moment
in the formation pathway for colloidal clusters, a defining bifurcation after which the cluster
can only evolve towards either icosahedral or decahedral symmetry.

99
 Chapter 9
Mechanics of Colloidal Supraparticles under Compression
A manuscript based on the results described in this chapter was submitted as:

Junwei Wang, Jan Schwenger, Andreas Ströbel, Patrick Feldner, Patrick Herre,
Stephan Romeis, Wolfgang Peukert, Benoit Merle and Nicolas Vogel
Mechanics of Colloidal Supraparticles under Compression (under revision)

I synthesized all colloidal clusters used in the expeirments. I also processed, analyzed
and interpreted all measurement data and developed the scaling analysis presented
below. A. Ströbel performed the nanoindentation measurement in ambient condition
under my supervision, J. Schwenger and P. Herre performed the in-situ compression in
SEM, P. Feldner, P. Herre, S. Romeis contributed to the indentation measurement and
discussion. W. Peukert, B. Merle and N. Vogel supervised the work.

In the previous chapters we focused on the thermodynamics and kinetics of colloidal

clusters in spherical droplet confinement, which provided insights into structure formation
processes of such finite systems. We term the assembled structure colloidal clusters due
to their crystalline nature as a result of sufficient equilibration time. As we show in Chapter
3, the sufficiently-slow droplet drying often takes more than a few days. Most spherical
assembly of colloidal particles, despite the various methods, do not result in clusters of
such high order. Here, we termed the spherically arranged assembly of colloidal particles
as supraparticles (Chapter 1). The vast diversity of colloidal particles, with different
material, surface chemistry and interaction, together with the possibility to control their
structure in a finite space, opens doors for applications of colloidal supraparticles.
However, one prerequisite for any meaningful application involving colloidal
supraparticles is that they must be mechanically robust. As their properties arise from
well-defined internal structure, colloidal supraparticles must maintain structural integrity,
throughout the fabrication, handling and in the application environment. The material may
fail, catastrophically or gradually, for example by impact shock from the solid substrates,

100
shearing from the suspended liquid, stretching of the surrounding matrix, or by
compression from external forces during application.

However, the mechanical response of supraparticles is extremely complex as it involves

hierarchical levels of forces. At the individual primary particle level, the mechanical
properties are determined by the material, potentially modified by size effects, which can
be probed by nanoindentation on single colloidal particles.145 At this single particle level,
theoretical models are needed to interpret data and extract relevant moduli, since most
mechanical properties, such as Young’s modulus, are defined using specifically shaped
specimen – and not spherical objects. As an example, the Hertz model has been used to
obtain the elastic modulus of a single silica sphere of a few hundred nanometer from the
load and displacement data during compression.146,147 Attractive forces act between the
individual primary particles in contact, especially van der Waals forces at particle contact,
capillary forces from liquid bridges at the neck of contact caused e.g. by water
condensation,148–151 and potential solid bridges forming from impurities in the drying
dispersion152. Static and dynamic friction132,153 between the particles may further change
the resulting mechanical properties. At the assembly level, the force network of the
interacting primary particles determines the propagation of external pressure, and the
arrangement of particles may thus influence crack propagation or dislocation
movement.154,155 All of these individual contributions are coupled dynamically to the
structure of the supraparticle during its mechanical response to external forces.

There have been continuous theoretical and experimental efforts to address the
mechanics of particle assemblies.156,157 Most relevant model for supraparticles is
proposed by Rumpf,158,159 which reasons that the fracture of agglomerates under load is
caused by tensile stress within the structure and predicts the critical fracture stress by
summing all adhesive forces between interparticle bonds in a fracture plane. This model
provides important rule of thumb predictions on agglomerate stability as it can involve
important parameters such as van der Waals and capillary forces, solid bridges, particle
interlocking, binders, as well as geometric shape and packing fraction of the agglomerate.
However, it also assumes a clear fracture plane that disintegrates the whole agglomerate
into two pieces, which is usually not the case in ductile deformation, or when fracture
occurs gradually within the structure.156,157 In addition, the size of the agglomerate does

101
not play a role in the model. From the experimental side, the nanomechanical behavior
of granular materials in bulk or in specifically shaped specimens has received growing
attention.160–166 Pioneering micromechanical studies focused on failure and fracture of
spherical granulates or agglomerates under compression,167–172 incorporating contact
models using discrete element method.173–175 However, a common obstacle is the lack of
control in the geometric features in such assemblies, both in terms of primary particle size
and agglomerate size, which hinders interpretation of data. A detailed understanding of
the mechanical response, especially tailored for colloidal supraparticles, is largely missing
to date.

In this chapter, we discuss the mechanics of colloidal supraparticles under compression,

which is the most relevant mechanical properties in terms of applications. We fabricate
colloidal supraparticles with uniform, well-defined geometric features using droplet-based
microfluidics as described in Chapter 3. This method creates uniform droplets of a
colloidal dispersion within a continuous oil phase. The diameters of these droplets can be
accurately controlled,65 which enables precise, independent control of primary particle
size, material and the subsequent supraparticle sizes. We use this ideal model systems
to investigate the mechanical stability of supraparticles under compression using
nanoindentation and observe the deformation both in-situ and post-measurement in
scanning electron microscope.

The fracture of colloidal supraparticles under compression

Figure 41b shows an optical microscopy image of PS SPs underlying the high uniformity
(PP = 230nm). All SPs exhibit the characteristic structural color signature of a
consolidated, spherical symmetry with defined, onion-like {111} layers and an amorphous
core,49,50 indicating that the internal structure of the particles is similar and uniform. The
inset shows a scanning electron microscope (SEM) image revealing the ordered surface
structure of the SPs.

We deposit SPs on flat silicon substrates and use an advancing flat intender tip to
investigate the mechanical stability of SPs under compression (Figure 41c). The large tip
diameter (90 μm) ensures that entire SP is always compressed between two parallel walls.

102
The mechanical response of the SPs, especially the deformation resistance and fracture,
is reflected by the force and displacement recorded at the indentation tip.

Figure 41, Uniform supraparticles prepared from droplet-based microfluidics used as model systems. a,
Emulsion droplets of an aqueous colloid dispersion in oil fabricated by microfluidics. b, Optical microscope image of
colloidal supraparticles exhibiting structural color. The scanning electron microscope image in the inset reveals the
ordered surface of the supraparticles and their constituent primary particles. c, Side view of a colloidal supraparticle on
a substrate under the flat indentation tip. The force and displacement of the tip is recorded during compression to study
the mechanical properties at a single particle level.

A typical force and displacement curve during the compression of colloidal SPs is shown
in Figure 42a (SP = 10 μm, PP = 244 nm). The nanoindentation is performed in ambient
condition at 40% humidity in speed control mode at 50 nm/s (credit of A. Ströbel, together
with me). Shortly after the tip contacts the SP (displacement less than about 100 nm), the
SP deforms elastically, which can be fitted by the Hertzian model (green dashed curve). 147
This agreement is already interesting as the Hertzian contact model assumes a
continuum solid sphere, while the SP is a porous assembly of individual spherical PPs.
After the yield point, the SP enters the elastic plastic deformation regime. The slope of
the linear region indicates the magnitude of resistance against deformation of the SP
under compression. The force applied to the indentation tip is stopped after the tip

103
advances 600 nm into the SP, but the SP continues to deform slightly, indicating a typical
creeping behavior of viscoelastic polymeric material.176 When the indentation tip retracts
from the SP, the recorded resistance force of the SP also decreases, as shown in the
unloading curve. The elastic plastic loading index (EPL)177 can be calculated by the area
underneath the unloading and loading curve, which represents, respectively, the energy
stored in the reversible elastic deformation and the total energy including the additional
irreversible plastic deformation. A completely elastic material has as EPL of 0, a
completely plastic material of 1. The colloidal SP measured here has an EPL of 0.72,
suggesting a strong plastic deformation. We note that the elastic deformation regime of
SP take place within 100 nm displacement, which is half the PP diameter. This length
scale is very small compared to the SP diameter, and therefore only has limited relevance
for the mechanical properties in typical applications of SPs. We therefore focus on the
mechanics of SPs at larger deformation.

Figure 42b shows four SPs (SP = 10 μm, PP = 244 nm) under compression at large
deformation. We develop a cleaning protocol to ensure no pieces of SP or PP residue
adhered on the indentation tip to affect the compression process (Figure S1). To discuss
the mechanics of SPs, stress and strain needs to be converted from force and
displacement measurement. However, the contact area under the indenter tip is dynamic
and unknown, as the SP undergoes shape change during compression. Instead of actual
stress and strain. Although real stress and strain cannot be obtained, we obtain nominal
stress by dividing the force by the initial cross-sectional area of the SP, and the nominal
strain by dividing displacement by the initial diameter of the SP. The nominal stress and
strain curves of the four SPs are highly reproducible and almost overlap (Figure 42b).
The initial elastic deformation is not visible as the SP yields at about 1% (Figure 42b,
arrow c). We examine the SPs after compression in the SEM and found no visible
deformation in the individual PP or SP level in the initial, elastic regime (Figure 42c). As
compression continues, the deformation is characterized by a linear region up to around
15% nominal strain at 20 MPa nominal stress, until the curve flattens. We define the slope
of the stress-strain curve in this linear regime as the deformation resistance of SPs. In
this regime, the periphery of the SP remains spherical, but the individual PPs are visibly
deformed at the contact area (Figure 42d). Further compression results in multiple cracks

104
visible extending from the periphery to the center, but the deformed SP remained as one
piece (Figure 42e). Multiple plateaus are present in the nominal stress-strain curve,
presumably caused by the formation of consecutive cracks. We define the fracture stress
and strain from the first turning point of the stress-strain curve. The slope after each
plateau becomes steeper, indicating that the SP becomes more resistant to external
forces as it deforms, probably due to the enlarged contact area which is also observed in
consecutive loading-unloading compression cycles. After compression, the individual
PPs are heavily deformed without positional rearrangement, evident in the flattened SP
surface (Figure 42e). The almost vertical unloading curve indicates little stored elastic
energy and heavy plastic deformation of the SP. Intriguingly, both the development of the
stress-strain curve and multiple fracture pattern of the PS SP bears striking resemblance
to individual micron-sized amorphous silica or titania particles, 146,178 which may suggest
a universality of plasticity in particulate matter and atomic material. 164 We note that the
mechanical response of SPs is addible, which allows estimation of mechanical stability of
assemblies of SPs such as monolayer (Figure 41b) or even superlattices at higher
hierarchies.

105
Figure 42, Mechanical properties of PS colloidal supraparticles under compression in ambient humidity. a,
Load-displacement curve of a typical colloidal supraparticle consisting of 244 nm primary PS particles during loading
and unloading. After initial elastic deformation, the supraparticle undergoes a linear region of elastic-plastic deformation,
the slope of which represents the resistance against deformation under compression. The area under loading and
unloading curve indicates toughness and the stored elastic energy during deformation. b, Nominal stress-strain curves
for four individual supraparticles, highlighting the reproducible compression behavior. c-e, SEM images at different
stages of the compression experiment. The supraparticle (c) first deforms plastically at the beginning of compression
(d), then undergoes multiple fracturing indicated by plateaus as the material fails (e). The supraparticle is ductile and
individual primary particles are heavily deformed.

The scaling between fracture and supraparticle geometry

We investigate the effect of SP size on the deformation resistance against compression.

Using colloidal PP with a size of 244 nm, we fabricate SPs with four different diameters
(Figure 43a). We first performed indentation tests on 21 SPs (SP = 10 μm, light blue).
The solid line depicts the averaged nominal stress strain value, the color band represents
the standard deviation. The reproducible mechanical response of SPs reflects the high

106
uniformity of the fabricated samples and allows accurate representation with smaller
numbers of measurements. The slope of the nominal stress-strain curves for all SPs (SP
= 7, 10, 15 and 20 μm) overlaps completely, independent of the SP size (Figure 43a).
This indicates that the deformation resistance is an inherent material property of SPs.
However, the fracture event depends on SP sizes. The load at which fracture occurs
scales with SP diameter with a power law exponent of 2.5 (Figure 43b, deviation shown
in error bar, for SP = 7 and 10 μm, the error bar is smaller than the circle). The
displacement at which fracture occurs scales with SP diameter with a power law exponent
of 1.5. This suggest a simple linear relationship between load and displacement when SP
fracture which allows prediction of SP mechanics by measuring only a few different
diameters. In addition, we found that the elastic-plastic loading (EPL) index EPL
decreases with increasing SP sizes, implying that at 5% strain, compressive energy
dissipates more into elastic deformation for larger SPs.

Having established that the deformation resistance against compression is independent

of SP diameters, we can conveniently compare the effect of different PP sizes, without
the dilemma to fix either the SP size or the number of PPs in the SP (Figure 43c). We
find that decreasing PP sizes (PP = 1000, 345, and 244 nm, grey to green to blue in
Figure 43c) increases SP deformation resistance, represented by the slope of the linear
region in the nominal stress strain curve before fracture. Following this trend, a solid PS
sphere can be considered as a SP consisting of infinitely small PPs. Indeed, we measured
that large, micrometer-sized solid PS spheres (diameter of 10, 20 and 30 μm, yellow, pink
and red curves) have higher deformation resistances than SPs. Interestingly, similar to
SPs, the nominal stress strain curves of these solid PS microspheres also collapse into
a master curve, but their fracture points are not affected by the different diameter. In
addition, multiple plateaus are also observed for solid PS spheres during compression,
suggesting multiple cracking events during the deformation. These similarities suggest
that a PS colloidal SP may be approximated as a large porous PS sphere and its
deformation resistance is an inherent material property, which depends only on the pore
size (given by the PP size). Indeed, our data suggests that the deformation resistance of
SP scales inversely with PP size (Figure 43d).

107
Continuing the analysis, we find that the fracture stress scales with SP diameter with a
power law exponent of 0.5 and with PP diameter with a power law exponent of -1 (Figure
43f). In addition, we find that the stress-strain curves of all measurements (Figure 43a,
Figure 43c) can be rescaled reasonably well (Figure 43e), considering the influence of
SP and PP diameter on the fracture stress and strain (for visual clarity, only averaged
value is shown, standard deviation is not shown). This generality suggests that the SP’s
deformation resistance and fracture can be approximated by a simple mechanism.

We apply the Griffith fracture mechanics176,179 to the fracture of SPs under compression.
Both the pop-in events in the quantitative load and displacement data and the in-situ
observations of deformation point out that the failure of the SPs is controlled by the
formation and propagation of a crack. According to Griffith criterion of fracture, the
initiation of the crack propagation is predicted by:

2𝐸𝛾
𝜎𝑐 = √ ∙𝑓
𝜋𝑎

where 𝜎𝑐 is the critical stress at fracture, 𝐸 is the Young’s modulus of the sample, 𝑓 the
geometry factor, 𝛾 the surface energy density, 𝑎 the size of the initial defect.

The energy required to create two free surfaces per area is represented by 𝛾, here it is
the energy required to break the bonds between neighboring PPs, overcoming their
cohesion forces. Therefore, it scales with the surface energy density of such interparticle
bonds:

𝛾 ∝ (𝐹𝑐𝑎𝑝𝑖𝑙𝑙𝑎𝑟𝑦 + 𝐹𝑣𝑑𝑤 ) ∙ 𝑑𝑏𝑜𝑛𝑑𝑠

The bond density 𝑑𝑏𝑜𝑛𝑑𝑠 is the number density of bonds per surface area, therefore it
scales with the number of bonds between PPs at the surface of the SP, divided by the
SP surface area:

𝑁𝑃𝑃 𝐴𝑆𝑃 1 𝐷𝑆𝑃 2 1 𝑁𝑐

𝑑𝑏𝑜𝑛𝑑𝑠 = 𝑁𝑐 = 𝑁𝑐 ∙ = 𝑁𝑐 2 ∙ 2 =
𝐴𝑟𝑒𝑎 𝐴𝑃𝑃 𝐴𝑆𝑃 𝐷𝑃𝑃 𝐷𝑆𝑃 𝐷𝑃𝑃 2

108
where 𝑁𝑐 is a constant representing the coordination number of particles on the SP
surface, 𝐷𝑆𝑃 and 𝐷𝑃𝑃 are the diameters of the supraparticles and primary particles,
respectively.

The cohesion forces between PPs has two contributions: the capillary force from the
meniscus of condensed water in ambient humidity between two PPs, 𝐹𝑐𝑎𝑝𝑖𝑙𝑙𝑎𝑟𝑦 , and the
van der Waals attraction between two PPs, 𝐹𝑣𝑑𝑤 . For spherical particles, it was shown
that both the van der Waals and capillary force scale with their radius. 21,180 In our case,
the interparticle cohesion forces scale with PP diameters:

𝐹𝑐𝑎𝑝𝑖𝑙𝑙𝑎𝑟𝑦 + 𝐹𝑣𝑑𝑤 ∝ 𝐷𝑃𝑃

It follows that:

1
𝛾 ∝ (𝐹𝑐𝑎𝑝𝑖𝑙𝑙𝑎𝑟𝑦 + 𝐹𝑣𝑑𝑤 ) ∙ 𝑑𝑏𝑜𝑛𝑑𝑠 ∝
𝐷𝑃𝑃

The parameter 𝑎 in the Griffith equation describes the defect size from which a crack is
initiated. Recall that the geometric feature of SP consists of identical PPs arranged in
concentric spherical shell, and in each shell the PPs arrange in hexagonal close packing,
confirmed by SEM image of SP (Figure 1b), it’s cross-section49 and the emerging
structural color.121 During compression, the SP experience compressive stress along the
direction of indenter tip, and tensile stress along the periphery – the closer to the SP
surface, the larger the tensile stress. The tensile stress, responsible for the fracture, is
distributed via the contacts between neighboring PPs on the surface of the SP. We indeed
observe that the crack initiates from the surface of the SP (supplementary video 1),
breaking the bonds between neighboring PPs and separating them without causing
deformation of the PPs. Therefore, the initial defect size in SP is determined by small
variations at the contact, 𝐷𝑐𝑜𝑛𝑡𝑎𝑐𝑡 . Theoretically, two rigid spheres in direct contact have
zero contact area. The real contact area is determined by particle surface roughness,
which is at molecular length scale of a few nanometers,21 regardless of the PP diameter
( 𝐷𝑃𝑃 ). Therefore, the fracture of the hierarchical SP is insensitive to the constant
𝐷𝑐𝑜𝑛𝑡𝑎𝑐𝑡 .181 However, during compression, tensile stress are concentrated at the
circumference of contact circle at the SP surface and distributed among all PP contacts

109
in the circumference, whose number scales linearly with SP diameter (𝐷𝑆𝑃 ). This requires
higher stress to break a PP contact in larger SP. Thus, the effective initial defect size is:

𝐷𝑐𝑜𝑛𝑡𝑎𝑐𝑡 1
𝑎∝ ∝
𝐷𝑆𝑃 𝐷𝑆𝑃

This agrees with our measurements (Figure 3a, 3b). With the same PP (hence the same
the Young’s modulus 𝐸 and surface energy density 𝛾), the fracture stress scales with SP
diameter with a power law exponent of 0.5.

For a particulate material such as our SPs, the Young’s modulus 𝐸 is unknown. Other
material properties that can be used to deduce the Young’s modulus, i.e. the Poisson
ratio, are also unknown. However, our measurements indicate that the slope of the SP
nominal stress-strain curve (up to 10% strain, Figure 3c) follows the trend of the initial
elastic region (less than 1% strain before yielding, Figure 2a), which can approximate SP
Young’s modulus. This relation suggests that, the modulus is inversely proportional to
𝐷𝑃𝑃 (Figure 3d):

1
𝐸∝
𝐷𝑃𝑃

Altogether, applying Griffith mechanics, the critical fracture stress depends upon the SP
geometry:

2𝐸𝛾 𝐷𝑆𝑃 0.5

𝜎𝑐 = √ ∙𝑓 ∝
𝜋𝑎 𝐷𝑃𝑃

We replot all our measurement data and find this scaling relationship accurately predicts
the fracture stress of all measurements with different SP and PP diameters (Figure 3e).
Although Griffith’s theory is originally proposed to account for brittle fracture in elastic
continuum material,179 the surface energy in the model can be replaced by strain energy
release rate developed by Irwin,182 where nonlinear inelastic effects observed in our SPs
are also considered. Our scaling analysis based on these theories shows that the
mechanical properties of SP can be tailored and designed by choosing PP and SP
diameters. For example, large solid PS spheres fracture at about 10% strain (Figure 3c)

110
regardless of its diameter, but the fracture of PS SPs can be varied from 10% to 20%
strain, depending on the SP diameter.

Figure 43, Influence of supraparticle and primary particle diameter on the mechanical response of
supraparticles under compression. a, Deformation resistance is independent of supraparticle diameter (primary
particle diameter of 244 nm), indicated by the linear region of nominal stress-strain curves. b, The fracture load of
supraparticle scales with its diameter via a power law with an exponent of 2.5. c, Deformation resistance of
supraparticles increases as the primary particle diameter decreases (grey to green to blue), approaching the value of
single large PS particles (yellow, pink, red). d, The deformation resistance of supraparticles scales inversely with the
primary particle diameter. e, Normalization and rescaling collapse the load and displacement curves of all
measurements into a master curve. f, Scaling relationship between fracture stress and the diameters of supra- and
primary particles.

Influence of cohesive forces between primary particles

We estimate the individual contribution of the van der Waals and capillary forces to the
adhesive forces in SPs. The contribution of attractive van der Waal forces 21,180 between
neighboring PPs can be expressed by:

𝐴𝐷𝑃𝑃
𝐹𝑣𝑑𝑤 =
24𝑑2

111
where 𝐴 is the Hamaker constant and 𝑑 is the separation between the surface of
constituent particles. The attractive capillary forces between neighboring particles21,180 is
given by:

𝐹𝑐𝑎𝑝𝑖𝑙𝑙𝑎𝑟𝑦 = 2𝜋𝛾𝐿 𝐷𝑃𝑃 𝑐𝑜𝑠𝜃

where 𝛾𝐿 is the interfacial tension of the liquid, 𝑐𝑜𝑠𝜃 is the contact angle between liquid
and the particle. According to Rumpf’s model, the critical stress for fracture in granular
material is:

(1 − 𝜀) 𝐹
𝜎𝑅𝑢𝑚𝑝𝑓 = 1.1
𝜀 𝐷𝑃𝑃 2

where 𝜀 is the void fraction in the granular material. Assuming literature values for the
Hamaker constant 𝐴 of PS to be 7 × 10−20 𝐽 ,180 the separation between 244 𝑛𝑚
particles 𝑑 to be 0.15 𝑛𝑚,180 the void fraction 𝜀 to be 0.26 due to well-ordered interior of
SP,49 the van der Waals forces between two particles is about 30 𝑛𝑁, which contribute to
a fracture stress of approximately 2 𝑀𝑃𝑎. Assuming 𝛾𝐿 of water to be 72 𝑚𝑁/𝑚, the
contact angle to be 30°,183 the capillary force between two particles is about 90 𝑛𝑁,
contributing to a fracture stress of approximately 6 𝑀𝑃𝑎 . Combining both forces, a
fracture stress of approximately 8 𝑀𝑃𝑎 is expected.

This value is significantly lower than the measured value of about 15 to 30 𝑀𝑃𝑎 (SP = 7
to 20 μm, PP = 244 nm). The discrepancy may originate from additional effects occurring
at the interparticle bonds that are not captured in the models. On the PP level, both van
der Waals and capillary forces are known to deform nanometer-sized PS particles and
may enhance the adhesion due to enlarged contact area. 184–186 During compression,
contact area increases significantly due to deformation of PS particles which further
increases adhesive forces. On the SP level, the Rumpf model assumes brittle fracture
along a fracture plane that disintegrates the granular material. Our experiment clearly
shows a ductile deformation, and the SP remains in one piece after fracture, which allows
significantly more energy dissipation by deformation of primary particles, and hence
higher fracture stress than predicted from Rumpf’s theory. Furthermore, there may be a
size effect related to the small dimension of SP, in contrast to almost infinite number of
particles in the Rumpf model, which predicts no size effect of SP. 187,188 Additionally,

112
surfactants, which may deposit in the interstitial sites of the supraparticles upon drying,
may additionally strengthen the interparticle bonds.

The state of the capillary meniscus between PPs depends on the amount of water content
in the SPs, hence the humidity around SPs.157,189 A reduced humidity should therefore
reduce the mechanical stability of SPs under compression. Although the humidity inside
the nanoindentation setup cannot be precisely controlled, we measured the same batch
of SP samples in summer and winter where the humidity drops from around 50% to 20%.
Indeed, in less humid environment, the deformation resistance of SPs reduces by more
than half, and the fracture pressure drops from around 20 MPa to 8 MPa. In addition, the
fracture mode changed from ductile to brittle. However, our previous findings still hold
true (Figure 43f) at reduced humidity – the deformation resistance is independent of SP
size and approaches the value of solid PS microspheres with decreasing PP size.

Indentations on SPs in the ultrahigh vacuum inside an SEM chamber was performed,64
where capillary forces can be excluded, to further investigate their contributions to the
stability of SPs (credit of J. Schwenger from the Peukert group). We equilibrate the
sample under vacuum inside the SEM chamber for over 24 hours. We verify that both
devices (in ambient condition and inside SEM) give quantitatively comparable results by
compressing large solid PS microspheres (20 μm) as reference. Importantly, this means
that the mechanical properties of individual PS microspheres are not influenced by
humidity - any changes in the SPs measured in vacuum must be related to their internal
structure, or, the forces acting between the PPs.

A typical nominal stress-strain curve of SP (SP = 10 μm, PP = 244 nm) compressed and
recorded by the customized indenter in the high vacuum of the SEM is shown in Figure
4a (dark blue curve). The inset shows the same sample under ambient conditions for
comparison (light blue curve, taken from Figure 42c). From our previous estimation, the
van der Waals forces contribute approximately 25% to the total adhesive forces between
PPs, the capillary forces 75%. Complete removal of water content should therefore cause
a four-fold reduction in the critical fracture stress. Indeed, the stress at fracture drops from
15 MPa in humid condition to about 6 MPa in vacuum. Note that the value does not
decrease linearly with decreasing humidity (from 20, 8, to 6 MPa at 50%, 20% and 0%

113
humidity in Figure 43a, Figure 44a), a known phenomenon in wet granular matter
because the distribution of water and the state of capillary bridges are complex function
of the water content.21,155,190 Our customized setup enables in-situ observation of the
deformation process. We ensured that the imaging in SEM does not affect the mechanical
properties by comparing measurements with and without electron beam. Figure 44b-e
shows SEM images taken during different stages of compression (indicated in Figure
44a). In the initial states of compression, where the nominal stress – nominal strain curve
enters the linear region, the contact area increased, and the SP bulged outwards at its
equator (Figure 44c). At fracture, a crack initiated from the SP periphery (Figure 44d),
which caused a drop in the force measured at the indentation tip (Figure 44a, arrow d).
While significant particle rearrangement occurred in the contact area with the indenter
and particularly close to the crack (Figure 44d), no PP rearrangement in other regions
was observed, supporting our scaling analysis. The compression continued until 40%
nominal strain. The unloading revealed little elastic energy stored in the deformation
process (Figure 44a, arrow e) as the deformed SP recovers only slightly after indenter
tip retreat (Figure 44e).

For small PPs (PP = 244 nm), the deformation resistance is independence of SP diameter
(Figure 44f), similar to ambient conditions (Figure 43a), although the fracture stress is
reduced. We increase the PP diameter from 244 nm to 500 nm and observe a reduced
deformation resistance and lowered stress and strain at fracture. Noteworthily, the
recorded data scattered significantly, and the deformation resistance was no longer
uniform for the different samples. This scattering suggests a brittle fracture behavior with
catastrophic failure, suggesting that the interparticle adhesion weakens significantly for
large PPs in vacuum, leading to changes in the fracture mode.

114
Figure 44, Mechanical properties of SPs in vacuum. a, Nominal stress-nominal strain curve of PS supraparticles
(PP diameter = 244 nm) measured in high vacuum inside an SEM chamber. The absence of humidity significantly
reduces the SP deformation resistance (humid conditions shown in inset). c-f) In situ SEM characterization of the SP
upon deformation. The SP (c) deforms at the contact with the nanoindenter tip and the substrate (d), followed by
fracture from the surface (e), and barely resumes shape after unloading due to mostly plastic deformation (f). b, the
deformation resistance is independent of SP diameter in vacuum. Curves of individual measurements are shown.

Influence of primary particle material

Intuitively, it may be anticipated that the mechanical stability of SPs can be improved by
using PPs of stronger material. We use silica colloidal particles, whose toughness, elastic
and compressive modulus are magnitudes higher than those of PS particles. 176
Surprisingly, as shown in Figure 45 in humid ambient conditions, silica SPs (SP = 7 μm,
PP = 230 nm) showed a significantly lower deformation resistance against compression
compared to PS SPs (SP = 7 μm, PP = 244 nm, identical as in Figure 2c). The nominal
fracture stress dropped from about 15 MPa (PS SPs) to about 4 MPa (Silica SPs) and the
nominal fracture strain from about 12% to 5%. In contrast to PS particles, the mechanical
properties of silica SPs remained largely unaffected by humidity. Measurements in

115
ambient humidity and in vacuum showed similar deformation resistance and fracture
stress. We find that the fracture stress for both silica and PS SPs are similar in vacuum
(Figure 45b). This similarity indicates that in vacuum the only adhesive forces in the SP
results van der Waals attraction, as the Hamaker constant for silica180 is similar to PS at
about 7 × 10−20 .

Importantly, we notice that silica SP shows brittle fracture behavior both in ambient
condition (Figure 45c) and in vacuum (Figure 45d), contrasting with the ductile
deformation of PS SPs, especially in ambient conditions. Corroborating this difference in
fracture mode, the individual silica PPs do not show any deformation after the SP is
fractured in ambient condition (Figure 45c) and in vacuum (Figure 45d). In contrast, PS
PPs deform heavily after compression in ambient condition, but much less in vacuum.

a 20 b 8
normalized load / MPa

normalized load / MPa

15 6

10 4

5 2

0 0
0 0.04 0.08 0.12 0.16 0 0.04 0.08 0.12 0.16
normalized displacement / - normalized displacement / -

c d e

2 μm

Figure 45, Comparison of PS and silica supraparticles, a, Silica supraparticles fracture at much lower strain and
stress than PS supraparticles in humid condition. Blue and green line show the averaged stress-strain data, the colored
band indicates the standard deviation. The larger standard deviation for silica particles indicates a brittle fracture

116
behavior. b, Vacuum does not influence the deformation resistance of silica supraparticles but significantly reduces the
deformation resistance for PS supraparticles. In vacuum, silica and PS supraparticles fracture at similar stress. c-e)
Scanning Electron Microscopy image of the different supraparticles after compression. Silica supraparticles fractured
brittlely into two pieces in both humid condition(c) and in vacuum (d), their primary silica particles are not deformed at
individually particle level. In contrast to the large deformation in humid condition, PS particle show little deformation in
vacuum (e).

We now aim to provide a general picture of the mechanical properties of SPs under
compression. The external compressive energy must be dissipated within the SP
structure. In extreme cases, the energy can dissipate either completely via deformation
of individual PPs, in which case no fracture with any positional rearrangement occurs, or
completely via breakage of interparticle bonds, without any PP deformation. In reality,
both mechanisms are at play. However, the weaker component, either the individual PPs
or the interparticle bond, dominates the deformation process and determines the fracture.
If the PPs are sufficiently soft, they yield and deform before an interparticle bond is broken;
if the PPs are sufficiently hard, the interparticle bond will break first. In humid condition,
capillary bridges between PPs significantly strengthen the interparticle bonds compared
to vacuum, where only van-der-Waals forces are assumed to act between the particles.
Silica particles has a compressive strength 30 times higher than PS particles, 146 hence
much more difficult to be deformed at individual particle level. Therefore, for silica SP, the
compressive energy is predominantly dissipated by breaking interparticle bonds, causing
particle rearrangement and hence early catastrophic brittle fracture without much PP
deformation. For PS SP, the energy is primarily dissipated by deformation of the PPs –
the SP fracture ductile and remain as one piece. Since adhesion scales inversely with PP
diameter, the larger the PPs, the more compressive energy dissipates by breaking the
interparticle bonds rather than deforming individual PPs. As a result, SPs with larger PPs
fracture at smaller stress and strain and more brittle. Reducing humidity has a similar
effect. In vacuum, the interparticle adhesion is reduced four-fold. In this case, PS SPs
fracture at lower stress and in a brittle mode without observable deformation of the PPs.
We note that the surfactant stabilizing the water in oil droplet during SP fabrication may
play role in the mechanical properties, which is not included in our discussion. The
surfactant used in our system is an amphiphilic poly(perfluoropropylene glycol)-
poly(ethylene glycol)-block-poly(propylene glycol)-block-poly(ethylene glycol)-

117
poly(perfluoropropylene glycol) triblock copolymer (Chapter 3).54 It has a block that is an
industrial fluorinated lubricant, and a block that is hygroscopic, which may facilitate water
condensation. In addition, the polypropylene block is known to shows a preferable
adsorption to hydrophobic surfaces such as PS instead of hydrophilic surfaces such
silica.75 This may be related to the unexpected persistence of the mechanical properties
of silica SP stability in the absence of humidity. The effect of the surfactant, its interaction
with the particles, and its distribution through the SP, merits a detailed, separate analysis.
To simplify the discussion, we resorted to the use of the minimal amount of surfactant
required to stabilize the emulsion and neglected its contribution to the mechanical
properties.

In this chapter, we establish an important basis for applications of supraparticles, that

their mechanical stability is surprising high – at ambient conditions, a supraparticle
consisting of PS primary particles mechanically resembles a single large porous PS
sphere. investigate the mechanical properties of well-defined supraparticles that are held
together by physical interparticle forces. Our analysis shows that the mechanical
properties under compression result from an intricate interplay between interparticle
adhesion and the material of the primary particles. We find that the resistance against
compression is inversely proportional to primary particle size, while fracture stress and
strain scales with supraparticle size. Using Griffith’s theory, we provide a scaling analysis
that successfully relates the mechanical properties to the sizes of supraparticle and
primary particles. We further examine the contribution of adhesive forces in the
supraparticles. Reducing humidity lowers the mechanical stability of PS SP, both on the
deformation resistance and the fracture stress and strain. Counterintuitively, we
demonstrate that primary particles with higher mechanical stability, such as silica, make
less stable supraparticles, as such primary particles cannot deform to dissipate energy.

118
Summary
In this thesis, we aimed to understand the number – structure – property relationship for
colloidal clusters self-assembled in spherical confinement. The confinement limits the
number of interacting particles and its spherical geometry regulates the explorable space
that alters particles equilibrium structures from their bulk counterpart. The number
determines the structure, which, in turn, determines their properties, such as minimum
energy structure and structural color patterns. In Chapter 3, we established the basic
experimental setting where colloidal particles are encapsulated in aqueous emulsion
droplets suspended in continuous oil phase. There are two requirements for colloidal
clusters to reach equilibrium in a droplet – that particles do not adsorb to the droplet
interface and that the drying of droplet is sufficiently slow. We developed a simple method
to produce water in oil in water double emulsions. Osmotic pressure controls the direction
of water diffusion in and out of the droplet, which not only guarantees the homogenous
and complete crystallization of colloidal clusters, but also allows studying their melting. In
Chapter 4, we examined icosahedral colloidal clusters, the most dominant type for small
and intermediate numbers. A sphere packing model was developed based on Mackay
icosahedron that precisely describes the cluster’s concentric icosahedral shell structure.
We found that when the number of particles in the confinement allows such close shell
structures, free energy of the cluster reaches local minima, compared to other numbers.
We termed these structures magic numbers. Since magic numbers are associated with
close shell sphere packing, they appear discretely and sparsely in the integer spectrum
– the number of any colloidal clusters is unlikely to be an exact magic number. We
focused on these clusters in Chapter 5. We found three mechanisms to expand a single
magic number into a magic number region, in which colloidal clusters maintains low free
energy and icosahedral symmetry by tolerating excess or deficient particles. We found
off-magic number regions where defects accumulate in a wedge-domain in the cluster
with broken icosahedral symmetry and high free energy. As cluster size increases in the
free energy landscape, their particle number resides in magic or off-magic number
regions periodically with low or high energy, forming increasing number of complete or
incomplete close shell structures. As magic number configurations are a finite size effect,

119
they must disappear when the system size approaches bulk dimensions. In Chapter 6
we showed that the energy gain of forming close shells attenuates to zero with increasing
cluster size, where non-close shell structures prevail. We demonstrated that this failure
of shell closure is inevitable due to the icosahedral sphere packing under spherical
confinement, marking a critical cluster size that provide an upper limit for the occurrence
of finite size effects. In Chapter 7 we identified colloidal clusters of three types of
symmetry – icosahedral, decahedral and octahedral, based on their structural color
pattern arising from different grain arrangement. We demonstrated how to use structural
color to monitor the dynamics of clusters, such as rotation, as well as the formation
pathway during the crystallization in-situ. This helped us to investigate the role of kinetics
in colloidal clusters formation in Chapter 8. We showed that the dominance of icosahedral
over decahedral symmetry in colloidal clusters is not caused by favorable
thermodynamics but kinetic bias. We found that the cluster formation bifurcates into two
paths at very early stage of crystallization, after which no cross-over is observed. We
rationalize the higher probability of selecting the icosahedral pathway as a result of the
stochastic nucleation event under the spherical confinement. In Chapter 9, we focused
on mechanical stability of kinetically arrested spherical colloidal clusters, which are
commonly known as colloidal supraparticles. We demonstrated their high stability under
compression and provide a predictive scaling that relates the deformation resistance,
fracture stress and strain to their geometric features. We also discussed the interplay
between the mechanical properties and the interparticle adhesive forces and the particle
material.

What is the behavior of some small spheres inside a large sphere? This is perhaps one
of the simplest questions one can ask and the answers are complex and fascinating. In
hindsight, almost all behaviors can be predicted by doodling some packing of spheres
and some imagination. This demonstrates the deep connection between physics and
geometry; as our system is finite and the spheres discrete, between numbers and
properties. Despite our thorough understanding of these colloidal clusters, it still seems a
miracle that nature allows icosahedral cluster to exist in spherical confinement where the
competing globality and locality reaches harmony. Globally, if the cluster consists of only
one grains, it is too anisotropic for a sphere, where does nucleation starts and how does

120
it break the spherical symmetry? In this sense, the existence of large fcc cluster is even
more mysterious and deserves further investigations. If the cluster consists of multiple
grains that orient towards the confinement center, the grain must have a pointy vertex, as
curvature increases to infinity in the center. These grains must also share boundary at a
certain lattice plane; hence the associated extra energy must not be high. These grains
also share a boundary with the curved confinement, now at different faces and hence
different lattice planes. Locally, spheres pack densest into face center cubic (or hexagonal
close packed), which happens to shape into a tetrahedron – it’s pointy, twinning energy
is low, and all faces are equal. It all matches! This explains the robustness of icosahedral
packing of spheres in spherical confinement. One can now imagine and design other
shapes, lattices or confinement geometries – when these requirements can be fulfilled,
the structure formation is robust. It also explains how these hard spheres, which only
experience collision from nearest neighbors, coordinate perfectly in forming globally
symmetric structures, particularly for magic numbers. How does one particle know the
total number of particles in the confinement, to know what it should do to minimize energy
for the entire cluster? It probably does not need to. It only needs to optimize its position
around its local surrounding, forming face center cubic grain that complies to the local
curvature. Such grains will nucleate one by one at its side until consuming all particles in
the confinement – at icosahedral magic numbers, the grains fit perfectly together, giving
rise to low energy icosahedral close shells; at off-magic numbers, some particles will be
left in a wedge domain, not able to crystallize. However, the particles pay a price when
local optimization does not lead to global optimum. When decahedral symmetry give rise
to even lower energy structure for some particle numbers (which covers a large range in
the number spectrum), forming locally-favorable tetrahedral grains impedes the pathway
towards decahedral clusters, due to the inability for global consideration of individual
particles. The price is even higher at large system size when a single grain face center
cubic cluster has lower energy than icosahedral clusters. Preliminary experiments
indicate that at low packing fraction, particles already form a layered fluid with global
icosahedral order near the spherical confinement. With every incremental increase of
packing fraction, the particles travel further in the pathway towards icosahedral cluster –
the crossover to global face center cubic cluster is perhaps unfavorable at this moment,

121
until the icosahedral cluster forms and the energy barrier to transition into face center
cubic cluster is too high. This thesis emphasized on the thermodynamics of colloidal
clusters. Aspects of kinetics, as well as the influence of particle potential and flexibility of
the confinement, will require further analysis.

Sophisticated and intriguing physical phenomena often arise when two competing
aspects are at play at the same time. Colloidal clusters in spherical confinement
demonstrate the interplay between crystallinity and curvature, locality and globality, as
well as kinetics and thermodynamics. We hope the knowledge acquired in this thesis
provides some general insights for structure and its formation in confinement.

122
Reference
1. Thomson, J. J. XXIV. On the structure of the atom: an investigation of the stability and
periods of oscillation of a number of corpuscles arranged at equal intervals around the
circumference of a circle; with application of the results to the theory of atomic structure.
London, Edinburgh, Dublin Philos. Mag. J. Sci. 7, 237–265 (1904).
2. Bowick, M., Cacciuto, A., Nelson, D. R. & Travesset, A. Crystalline Order on a Sphere and
the Generalized Thomson Problem. Phys. Rev. Lett. 89, 185502 (2002).
3. Miller, W. L. & Cacciuto, A. Two-dimensional packing of soft particles and the soft
generalized Thomson problem. Soft Matter 7, 7552 (2011).
4. Baletto, F. & Ferrando, R. Structural properties of nanoclusters: Energetic, thermodynamic,
and kinetic effects. Rev. Mod. Phys. 77, 371–423 (2005).
5. Rickard, D. T. The origin of framboids. Lithos 3, 269–293 (1970).
6. Ohfuji, H. & Rickard, D. Experimental syntheses of framboids - A review. Earth-Science Rev.
71, 147–170 (2005).
7. Ohfuji, H. & Akai, J. Icosahedral domain structure of framboidal pyrite. Am. Mineral. 87,
176–180 (2002).
8. Nozawa, J. et al. Magnetite 3D colloidal crystals formed in the early solar system 4.6 billion
years ago. J. Am. Chem. Soc. 133, 8782–8785 (2011).
9. Kimura, Y. et al. Vortex magnetic structure in framboidal magnetite reveals existence of
water droplets in an ancient asteroid. Nat. Commun. 4, 2–4 (2013).
10. Giammona, J. & Campàs, O. Physical constraints on early blastomere packings. PLOS
Comput. Biol. (2021). doi:10.1371/journal.pcbi.1007994
11. Yao, H. et al. Molecular Architecture of the SARS-CoV-2 Virus. Cell 183, 730-738.e13 (2020).
12. Wood, W. W. & Jacobson, J. D. Preliminary Results from a Recalculation of the Monte Carlo
Equation of State of Hard Spheres. J. Chem. Phys. 27, 1207–1208 (1957).
13. Alder, B. J. & Wainwright, T. E. Phase Transition for a Hard Sphere System. J. Chem. Phys.
27, 1208 (1957).
14. Frenkel, D. Order through entropy. Nat. Mater. 14, 9–12 (2014).
15. Stöber, W., Fink, A. & Bohn, E. Controlled growth of monodisperse silica spheres in the
micron size range. J. Colloid Interface Sci. 26, 62–69 (1968).
16. Ugelstad, J. Emulsion polymerization: initiation of polymerization in monomer droplets. J.
Polym. 11, 503–513 (1973).
17. Van Blaaderen, A. & Vrij, A. Synthesis and characterization of colloidal dispersions of
fluorescent, monodisperse silica spheres. Langmuir 8, 2921–2931 (1992).
18. Yoshida, H., Ito, K. & Ise, N. Localized ordered structure in polymer latex suspensions as

123
 studied by a confocal laser scanning microscope. Phys. Rev. B 44, 435–438 (1991).
19. Bosma, G. et al. Preparation of Monodisperse, Fluorescent PMMA–Latex Colloids by
Dispersion Polymerization. J. Colloid Interface Sci. 245, 292–300 (2002).
20. Van Blaaderen, A. & Wiltzius, P. Real-space structure of colloidal hard-sphere glasses.
Science. 270, 1177–1179 (1995).
21. Butt, H.-J. & Kappl, M. Surface and Interfacial Forces. Surface and Interfacial Forces (Wiley-
VCH Verlag GmbH & Co. KGaA, 2010). doi:10.1002/9783527629411
22. Verwey, E. J. W. Theory of the Stability of Lyophobic Colloids. J. Phys. Colloid Chem. 51,
631–636 (1947).
23. Pusey, P. N. & van Megen, W. Phase behaviour of concentrated suspensions of nearly hard
colloidal spheres. Nature 320, 340–342 (1986).
24. Vrij, A. Polymers at Interfaces and the Interactions in Colloidal Dispersions. Pure Appl.
Chem. 48, 471–483 (1976).
25. Gasser, U., Weeks, E. R., Schofield, A., Pusey, P. N. & Weitz, D. a. Real-space imaging of
nucleation and growth in colloidal crystallization. Science 292, 258–262 (2001).
26. Peng, Y. et al. Two-step nucleation mechanism in solid-solid phase transitions. Nat. Mater.
14, 101–108 (2015).
27. Lu (陸述義), P. J. & Weitz, D. A. Colloidal Particles: Crystals, Glasses, and Gels. Annu. Rev.
Condens. Matter Phys. 4, 217–233 (2013).
28. Cates, M. E. & Manoharan, V. N. Celebrating Soft Matter ’s 10th anniversary: Testing the
foundations of classical entropy: colloid experiments. Soft Matter 11, 6538–6546 (2015).
29. Li, B., Zhou, D. & Han, Y. Assembly and phase transitions of colloidal crystals. Nat. Rev.
Mater. 1, 15011 (2016).
30. Goerlitzer, E. S. A., Klupp Taylor, R. N. & Vogel, N. Bioinspired Photonic Pigments from
Colloidal Self-Assembly. Adv. Mater. 30, 1–15 (2018).
31. Vukusic, P. & Sambles, J. R. Photonic structures in biology. Nature (2003).
doi:10.1038/nature01941
32. SANDERS, J. V. Colour of Precious Opal. Nature 204, 1151–1153 (1964).
33. Vukusic, P., Sambles, J. R., Lawrence, C. R. & Wootton, R. J. Quantified interference and
diffraction in single Morpho butterfly scales. Proc. R. Soc. London. Ser. B Biol. Sci. 266,
1403–1411 (1999).
34. Aguirre, C. I., Reguera, E. & Stein, A. Tunable colors in opals and inverse opal photonic
crystals. Adv. Funct. Mater. 20, 2565–2578 (2010).
35. Zhao, Y., Xie, Z., Gu, H., Zhu, C. & Gu, Z. Bio-inspired variable structural color materials.
Chem. Soc. Rev. 41, 3297–3317 (2012).
36. Forster, J. D. et al. Biomimetic Isotropic Nanostructures for Structural Coloration. Adv.

124
 Mater. 22, 2939–2944 (2010).
37. Jiang, P. & McFarland, M. J. Large-scale fabrication of wafer-size colloidal crystals,
macroporous polymers and nanocomposites by spin-coating. J. Am. Chem. Soc. 126,
13778–13786 (2004).
38. Vogel, N., Retsch, M., Fustin, C.-A., del Campo, A. & Jonas, U. Advances in Colloidal
Assembly: The Design of Structure and Hierarchy in Two and Three Dimensions. Chem. Rev.
115, 6265–6311 (2015).
39. Phillips, K. R. et al. A colloidoscope of colloid-based porous materials and their uses. Chem.
Soc. Rev. 45, 281–322 (2016).
40. Vogel, N., De Viguerie, L., Jonas, U., Weiss, C. K. & Landfester, K. Wafer-scale fabrication of
ordered binary colloidal monolayers with adjustable stoichiometries. Adv. Funct. Mater.
21, 3064–3073 (2011).
41. Mino, Y., Watanabe, S. & Miyahara, M. T. Fabrication of Colloidal Grid Network by Two-
Step Convective Self-Assembly. Langmuir 27, 5290–5295 (2011).
42. Bausch, A. R. et al. Grain boundary scars and spherical crystallography. Science. 299, 1716–
8 (2003).
43. Einert, T., Lipowsky, P., Schilling, J., Bowick, M. J. & Bausch, A. R. Grain boundary scars on
spherical crystals. Langmuir (2005). doi:10.1021/la0517383
44. Guerra, R. E., Kelleher, C. P., Hollingsworth, A. D. & Chaikin, P. M. Freezing on a sphere.
Nature 554, 346–350 (2018).
45. Manoharan, V. N., Elsesser, M. T. & Pine, D. J. Dense Packing and Symmetry in Small
Clusters of Microspheres. Science. 301, 483–487 (2003).
46. Lauga, E. & Brenner, M. P. Evaporation-driven assembly of colloidal particles. Phys. Rev.
Lett. 93, 1–4 (2004).
47. Lacava, J., Born, P. & Kraus, T. Nanoparticle clusters with Lennard-Jones geometries. Nano
Lett. 12, 3279–3282 (2012).
48. de Nijs, B. et al. Entropy-driven formation of large icosahedral colloidal clusters by
spherical confinement. Nat. Mater. 14, 56–60 (2014).
49. Vogel, N. et al. Color from hierarchy: Diverse optical properties of micron-sized spherical
colloidal assemblies. Proc. Natl. Acad. Sci. 112, 10845–10850 (2015).
50. Zhao, Y., Shang, L., Cheng, Y. & Gu, Z. Spherical colloidal photonic crystals. Acc. Chem. Res.
47, 3632–3642 (2014).
51. Mackay, A. L. A dense non-crystallographic packing of equal spheres. Acta Crystallogr. 15,
916–918 (1962).
52. Kuo, K. H. Mackay, Anti-Mackay, Double-Mackay, Pseudo-Mackay, and Related
Icosahedral Shell Clusters. Struct. Chem. 13, 221–230 (2002).
53. Manoharan, V. N. Colloidal matter: Packing, geometry, and entropy. Science. 349, 1253751

125
 (2015).
54. Holtze, C. et al. Biocompatible surfactants for water-in-fluorocarbon emulsions. Lab Chip
8, 1632–1639 (2008).
55. Etienne, G., Kessler, M. & Amstad, E. Influence of Fluorinated Surfactant Composition on
the Stability of Emulsion Drops. Macromol. Chem. Phys. 218, (2017).
56. Wang, J. et al. Magic number colloidal clusters as minimum free energy structures. Nat.
Commun. 9, 5259 (2018).
57. Wang, J. et al. Free Energy Landscape of Colloidal Clusters in Spherical Confinement. ACS
Nano 13, 9005–9015 (2019).
58. Bitzek, E., Koskinen, P., Gähler, F., Moseler, M. & Gumbsch, P. Structural relaxation made
simple. Phys. Rev. Lett. 97, 170201 (2006).
59. Glaser, J. et al. Strong scaling of general-purpose molecular dynamics simulations on GPUs.
Comput. Phys. Commun. 192, 97–107 (2015).
60. Anderson, J. A., Lorenz, C. D. & Travesset, A. General purpose molecular dynamics
simulations fully implemented on graphics processing units. J. Comput. Phys. 227, 5342–
5359 (2008).
61. Frenkel, D. & Ladd, A. J. C. New Monte Carlo method to compute the free energy of
arbitrary solids. Application to the fcc and hcp phases of hard spheres. J. Chem. Phys. 81,
3188–3193 (1984).
62. Englisch, S. et al. Scale-Bridging 3D-Analysis of Colloidal Clusters Using 360° Electron
Tomography and X-Ray Nano-CT. Microsc. Microanal. 25, 392–393 (2019).
63. Przybilla, T. et al. Transfer of Individual Micro- and Nanoparticles for High-Precision 3D
Analysis Using 360° Electron Tomography. Small Methods 2, 1700276 (2018).
64. Romeis, S., Paul, J., Ziener, M. & Peukert, W. A novel apparatus for in situ compression of
submicron structures and particles in a high resolution SEM. Rev. Sci. Instrum. 83, (2012).
65. Shang, L., Cheng, Y. & Zhao, Y. Emerging Droplet Microfluidics. Chem. Rev. 117, 7964–8040
(2017).
66. Utada, A. S. et al. Dripping, Jetting, Drops, and Wetting: The Magic of Microfluidics. MRS
Bull. 32, 702–708 (2007).
67. Dangla, R., Gallaire, F. & Baroud, C. N. Microchannel deformations due to solvent-induced
PDMS swelling. Lab Chip 10, 2972 (2010).
68. Salmon, A. R. et al. Microcapsule Buckling Triggered by Compression-Induced Interfacial
Phase Change. Langmuir 32, 10987–10994 (2016).
69. Toor, A. et al. Mechanical Properties of Solidifying Assemblies of Nanoparticle Surfactants
at the Oil-Water Interface. Langmuir 35, 13340–13350 (2019).
70. Datta, S. S., Shum, H. C. & Weitz, D. A. Controlled buckling and crumpling of nanoparticle-
coated droplets. Langmuir 26, 18612–18616 (2010).

126
71. Rey, M., Yu, T., Guenther, R., Bley, K. & Vogel, N. A Dirty Story: Improving Colloidal
Monolayer Formation by Understanding the Effect of Impurities at the Air/Water Interface.
Langmuir 35, 95–103 (2019).
72. McGorty, R., Fung, J., Kaz, D. & Manoharan, V. N. Colloidal self-assembly at an interface.
Mater. Today 13, 34–42 (2010).
73. Park, B. J. et al. Direct measurements of the effects of salt and surfactant on interaction
forces between colloidal particles at water-oil interfaces. Langmuir 24, 1686–1694 (2008).
74. Etienne, G., Vian, A., Biočanin, M., Deplancke, B. & Amstad, E. Cross-talk between emulsion
drops: How are hydrophilic reagents transported across oil phases? Lab Chip 18, 3903–
3912 (2018).
75. Hueckel, T., Hocky, G. M., Palacci, J. & Sacanna, S. Ionic solids from common colloids.
Nature 580, 487–490 (2020).
76. Stolnik, S. et al. Adsorption behaviour and conformation of selected poly(ethylene oxide)
copolymers on the surface of a model colloidal drug carrier. Colloids Surfaces A
Physicochem. Eng. Asp. 122, 151–159 (1997).
77. Pitto-Barry, A. & Barry, N. P. E. Pluronic® block-copolymers in medicine: From chemical
and biological versatility to rationalisation and clinical advances. Polym. Chem. 5, 3291–
3297 (2014).
78. Kim, S. H., Park, J. G., Choi, T. M., Manoharan, V. N. & Weitz, D. a. Osmotic-pressure-
controlled concentration of colloidal particles in thin-shelled capsules. Nat Commun 5,
3068 (2014).
79. Doufène, K., Tourné-Péteilh, C., Etienne, P. & Aubert-Pouëssel, A. Microfluidic Systems for
Droplet Generation in Aqueous Continuous Phases: A Focus Review. Langmuir 35, 12597–
12612 (2019).
80. Lee, S. S., Kim, S. K., Won, J. C., Kim, Y. H. & Kim, S. H. Reconfigurable Photonic Capsules
Containing Cholesteric Liquid Crystals with Planar Alignment. Angew. Chemie - Int. Ed. 54,
15266–15270 (2015).
81. Shirk, K., Steiner, C., Kim, J. W., Marquez, M. & Martinez, C. J. Assembly of colloidal silica
crystals inside double emulsion drops. Langmuir 29, 11849–11857 (2013).
82. Kim, S. H., Jeon, S. J. & Yang, S. M. Optofluidic encapsulation of crystalline colloidal arrays
into spherical membrane. J. Am. Chem. Soc. 130, 6040–6046 (2008).
83. Choi, T. M. et al. Photonic Microcapsules Containing Single-Crystal Colloidal Arrays with
Optical Anisotropy. Adv. Mater. 31, 1–8 (2019).
84. Vian, A., Reuse, B. & Amstad, E. Scalable production of double emulsion drops with thin
shells. Lab Chip 18, 1936–1942 (2018).
85. Arriaga, L. R., Amstad, E. & Weitz, D. A. Scalable single-step microfluidic production of
single-core double emulsions with ultra-thin shells. Lab Chip 15, 3335–3340 (2015).
86. Echt, O., Sattler, K. & Recknagel, E. Magic numbers for sphere packings: Experimental

127
 verification in free xenon clusters. Phys. Rev. Lett. 47, 1121–1124 (1981).
87. Martin, T. P. Shells of atoms. Phys. Rep. 273, 199–241 (1996).
88. Mackay, A. L. A dense non-crystallographic packing of equal spheres. Acta Crystallogr. 15,
916–918 (1962).
89. Farges, J., de Feraudy, M. F., Raoult, B. & Torchet, G. Noncrystalline structure of argon
clusters. II. Multilayer icosahedral structure of Ar N clusters 50 < N < 750. J. Chem. Phys.
84, 3491–3501 (1986).
90. Northby, J. A. Structure and binding of Lennard-Jones clusters: 13 ≤ N ≤ 147. J. Chem. Phys.
87, 6166–6177 (1987).
91. Doye, J. P. K., Wales, D. J. & Berry, R. S. The effect of the range of the potential on the
structures of clusters. J. Chem. Phys. 103, 4234–4249 (1995).
92. Hendy, S. C. & Doye, J. P. K. Surface-reconstructed icosahedral structures for lead clusters.
Phys. Rev. B 66, 235402 (2002).
93. Jensen, K. M. Ø. et al. Polymorphism in magic-sized Au144(SR)60 clusters. Nat. Commun. 7,
11859 (2016).
94. Laasonen, K., Panizon, E., Bochicchio, D. & Ferrando, R. Competition between icosahedral
motifs in AgCu, AgNi, and AgCo nanoalloys: A combined atomistic–DFT study. J. Phys. Chem.
C 117, 26405–26413 (2013).
95. Mayoral, A., Llamosa, D. & Huttel, Y. A novel Co@Au structure formed in bimetallic
core@shell nanoparticles. Chem. Commun. 51, 8442–8445 (2015).
96. Ferrando, R., Jellinek, J. & Johnston, R. L. Nanoalloys: From theory to applications of alloy
clusters and nanoparticles. Chem. Rev. 108, 845–910 (2008).
97. de Nijs, B. et al. Entropy-driven formation of large icosahedral colloidal clusters by
spherical confinement. Nat. Mater. 14, 56–60 (2015).
98. Radu, M. & Schilling, T. Solvent hydrodynamics speed up crystal nucleation in suspensions
of hard spheres. Europhys. Lett. 105, 26001 (2014).
99. Roehm, D., Kesselheim, S. & Arnold, A. Hydrodynamic interactions slow down
crystallization of soft colloids. Soft Matter 10, 5503–5509 (2014).
100. Furukawa, A. & Tanaka, H. Key role of hydrodynamic interactions in colloidal gelation. Phys.
Rev. Lett. 104, 1–4 (2010).
101. de Graaf, J., Poon, W. C. K., Haughey, M. J. & Hermes, M. Hydrodynamics strongly affect
the dynamics of colloidal gelation but not gel structure. arXiv:1808.01722 (2018).
102. Lauga, E. & Brenner, M. P. Evaporation-driven assembly of colloidal particles. Phys. Rev.
Lett. 93, 238301 (2004).
103. Nam, H.-S., Hwang, N. M., Yu, B. D. & Yoon, J.-K. Formation of an icosahedral structure
during the freezing of gold nanoclusters: surface-induced mechanism. Phys. Rev. Lett. 89,
275502 (2002).

128
104. Guerra, R. E., Kelleher, C. P., Hollingsworth, A. D. & Chaikin, P. M. Freezing on a sphere.
Nature 554, 346–350 (2018).
105. Schilling, T. & Schmid, F. Computing absolute free energies of disordered structures by
molecular simulation. J. Chem. Phys. 131, 231102 (2009).
106. van Anders, G., Klotsa, D., Ahmed, N. K., Engel, M. & Glotzer, S. C. Understanding shape
entropy through local dense packing. Proc. Natl. Acad. Sci. 111, E4812–E4821 (2014).
107. Mayer, M. G. On Closed Shells in Nuclei. II. Phys. Rev. 75, 1969–1970 (1949).
108. Mayer, M. G. On Closed Shells in Nuclei. Phys. Rev. 74, 235–239 (1948).
109. Martin, T. P., Bergmann, T., Göhlich, H. & Lange, T. Observation of electronic shells and
shells of atoms in large Na clusters. Chem. Phys. Lett. 172, 209–213 (1990).
110. Haxel, O., Jensen, J. H. D. & Suess, H. E. On the ‘Magic Numbers’ in Nuclear Structure. Phys.
Rev. 75, 1766–1766 (1949).
111. Fujii, S. et al. Platonic micelles: Monodisperse micelles with discrete aggregation numbers
corresponding to regular polyhedra. Sci. Rep. 7, 1–8 (2017).
112. Martin, T. P., Näher, U., Schaber, H. & Zimmermann, U. Clusters of fullerene molecules.
Phys. Rev. Lett. 70, 3079–3082 (1993).
113. Hsia, Y. et al. Design of a hyperstable 60-subunit protein icosahedron. Nature 535, 136–
139 (2016).
114. Gozzo, F. C. et al. Gaseous supramolecules of imidazolium ionic liquids: ‘Magic’ numbers
and intrinsic strengths of hydrogen bonds. Chem. - A Eur. J. 10, 6187–6193 (2004).
115. Xu, F. et al. Correlating the magic numbers of inorganic nanomolecular assemblies with a
molecular-ring Rosetta Stone. Proc. Natl. Acad. Sci. 109, 11609–11612 (2012).
116. Midgley, P. A. & Weyland, M. 3D electron microscopy in the physical sciences: The
development of Z-contrast and EFTEM tomography. Ultramicroscopy 96, 413–431 (2003).
117. Rahm, J. M. & Erhart, P. Beyond Magic Numbers: Atomic Scale Equilibrium Nanoparticle
Shapes for Any Size. Nano Lett. 17, 5775–5781 (2017).
118. Hendy, S. C. & Doye, J. P. K. Surface-reconstructed icosahedral structures for lead clusters.
Phys. Rev. B Condens. matter 66, 235402 (2002).
119. Noya, E. G. & Doye, J. P. K. Structural transitions in the 309-atom magic number Lennard-
Jones cluster. J. Chem. Phys. 124, 104503 (2006).
120. van Anders, G., Klotsa, D., Ahmed, N. K., Engel, M. & Glotzer, S. C. Understanding shape
entropy through local dense packing. Proc. Natl. Acad. Sci. 111, E4812–E4821 (2014).
121. Wang, J. et al. Structural Color of Colloidal Clusters as a Tool to Investigate Structure and
Dynamics. Adv. Funct. Mater. 30, 1907730 (2020).
122. Chen, Y. et al. Morphology selection kinetics of crystallization in a sphere. Nat. Phys. 17,
121–127 (2021).

129
123. Kim, S. H., Lee, S. Y., Yi, G. R., Pine, D. J. & Yang, S. M. Microwave-assisted self-organization
of colloidal particles in confining aqueous droplets. J. Am. Chem. Soc. 128, 10897–10904
(2006).
124. Zhao, Y., Shang, L., Cheng, Y. & Gu, Z. Spherical colloidal photonic crystals. Acc. Chem. Res.
47, 3632–3642 (2014).
125. Marks, L. D. Modified Wulff constructions for twinned particles. J. Cryst. Growth 61, 556–
566 (1983).
126. Ino, S. Stability of Multiply-Twinned Particles. J. Phys. Soc. Japan 27, 941–953 (1969).
127. Ohnuki, R., Isoda, S., Sakai, M., Takeoka, Y. & Yoshioka, S. Grating Diffraction or Bragg
Diffraction? Coloration Mechanisms of the Photonic Ball. Adv. Opt. Mater. 7, 1–7 (2019).
128. Ge, J. et al. Magnetochromatic microspheres: Rotating photonic crystals. J. Am. Chem. Soc.
131, 15687–15694 (2009).
129. Wang, A., Rogers, W. B. & Manoharan, V. N. Effects of Contact-Line Pinning on the
Adsorption of Nonspherical Colloids at Liquid Interfaces. Phys. Rev. Lett. 119, 1–5 (2017).
130. Edmond, K. V., Elsesser, M. T., Hunter, G. L., Pine, D. J. & Weeks, E. R. Decoupling of
rotational and translational diffusion in supercooled colloidal fluids. Proc. Natl. Acad. Sci.
U. S. A. 109, 17891–17896 (2012).
131. Martin, S., Reichert, M., Stark, H. & Gisler, T. Direct observation of hydrodynamic rotation-
translation coupling between two colloidal spheres. Phys. Rev. Lett. 97, 1–4 (2006).
132. Heim, L.-O., Blum, J., Preuss, M. & Butt, H.-J. Adhesion and Friction Forces between
Spherical Micrometer-Sized Particles. Phys. Rev. Lett. 83, 3328–3331 (1999).
133. Bohlein, T., Mikhael, J. & Bechinger, C. Observation of kinks and antikinks in colloidal
monolayers driven across ordered surfaces. Nat. Mater. 11, 126–130 (2012).
134. Magkiriadou, S., Park, J.-G., Kim, Y.-S. & Manoharan, V. N. Disordered packings of core-
shell particles with angle-independent structural colors. Opt. Mater. Express 2, 1343–1352
(2012).
135. Rojas-Ochoa, L. F., Mendez-Alcaraz, J. M., Sáenz, J. J., Schurtenberger, P. & Scheffold, F.
Photonic properties of strongly correlated colloidal liquids. Phys. Rev. Lett. 93, 1–4 (2004).
136. Rastogi, V. et al. Synthesis of light-diffracting assemblies from microspheres and
nanoparticles in droplets on a superhydrophobic surface. Adv. Mater. 20, 4263–4268
(2008).
137. Rodríguez-López, J. L. et al. Surface reconstruction and decahedral structure of bimetallic
nanoparticles. Phys. Rev. Lett. 92, 196102–1 (2004).
138. Mayoral, A., Llamosa, D. & Huttel, Y. A novel Co@Au structure formed in bimetallic
core@shell nanoparticles. Chem. Commun. (Camb). 51, 8442–5 (2015).
139. Johnson, C. L. et al. Effects of elastic anisotropy on strain distributions in decahedral gold
nanoparticles. Nat. Mater. 7, 120–124 (2008).

130
140. Hofmeister, H. Fivefold Twinned Nanoparticles. 3, 431–452 (2004).
141. Goris, B. et al. Measuring Lattice Strain in Three Dimensions through Electron Microscopy.
Nano Lett. 15, 6996–7001 (2015).
142. Chen, C. C. et al. Three-dimensional imaging of dislocations in a nanoparticle at atomic
resolution. Nature 496, 74–77 (2013).
143. Nam, H.-S., Hwang, N., Yu, B. & Yoon, J.-K. Formation of an Icosahedral Structure during
the Freezing of Gold Nanoclusters: Surface-Induced Mechanism. Phys. Rev. Lett. 89, 1–4
(2002).
144. Opletal, G., Feigl, C. A., Grochola, G., Snook, I. K. & Russo, S. P. Elucidation of surface driven
crystallization of icosahedral clusters. Chem. Phys. Lett. 482, 281–286 (2009).
145. Paul, J., Romeis, S., Tomas, J. & Peukert, W. A review of models for single particle
compression and their application to silica microspheres. Adv. Powder Technol. 25, 136–
153 (2014).
146. Paul, J. et al. In situ cracking of silica beads in the SEM and TEM - Effect of particle size on
structure-property correlations. Powder Technol. 270, 337–347 (2015).
147. Hertz, H. Über die berührung fester elastischer Körper (On the contact of rigid elastic
solids). J. reine und Angew. Math. 92 156–171 (1896).
148. Rabinovich, Y. I., Esayanur, M. S. & Moudgil, B. M. Capillary forces between two spheres
with a fixed volume liquid bridge: Theory and experiment. Langmuir 21, 10992–10997
(2005).
149. Willett, C. D., Adams, M. J., Johnson, S. A. & Seville, J. P. K. Capillary bridges between two
spherical bodies. Langmuir 16, 9396–9405 (2000).
150. Gallego-Gómez, F., Morales-Flórez, V., Morales, M., Blanco, A. & López, C. Colloidal crystals
and water: Perspectives on liquid–solid nanoscale phenomena in wet particulate media.
Adv. Colloid Interface Sci. 234, 142–160 (2016).
151. Kralchevsky, P. A. & Nagayama, K. Capillary Forces between Colloidal Particles. Langmuir
10, 23–36 (1994).
152. Abi Ghanem, M. et al. Longitudinal eigenvibration of multilayer colloidal crystals and the
effect of nanoscale contact bridges. Nanoscale 11, 5655–5665 (2019).
153. Ling, X., Butt, H. J. & Kappl, M. Quantitative measurement of friction between single
microspheres by friction force microscopy. Langmuir 23, 8392–8399 (2007).
154. Goldenberg, C. & Goldhirsch, I. Force Chains, Microelasticity, and Macroelasticity. Phys.
Rev. Lett. 89, 1–4 (2002).
155. Kudrolli, A. Sticky sand. Nat. Mater. 7, 174–175 (2008).
156. Reynolds, G. K., Fu, J. S., Cheong, Y. S., Hounslow, M. J. & Salman, A. D. Breakage in
granulation: A review. Chem. Eng. Sci. 60, 3969–3992 (2005).
157. Bika, D. G., Gentzler, M. & Michaels, J. N. Mechanical properties of agglomerates. Powder

131
 Technol. 117, 98–112 (2001).
158. Rumpf, H. C. H. The theory of tensile strength of agglomerates during power transmission
at contact points. Chemie Ing. Tech. 42, 538–540 (1970).
159. Davies, D. K. Surface charge and the contact of elastic solids. J. Phys. D. Appl. Phys. 6, 1017–
1024 (1973).
160. Roth, M., Schilde, C., Lellig, P., Kwade, A. & Auernhammer, G. K. Colloidal aggregates tested
via nanoindentation and quasi-simultaneous 3D imaging. Eur. Phys. J. E 35, 1–12 (2012).
161. Yin, J., Retsch, M., Thomas, E. L. & Boyce, M. C. Collective mechanical behavior of
multilayer colloidal arrays of hollow nanoparticles. Langmuir 28, 5580–5588 (2012).
162. Huang, P., Zhang, L., Yan, Q., Guo, D. & Xie, G. Size Dependent Mechanical Properties of
Monolayer Densely Arranged Polystyrene Nanospheres. Langmuir 32, 13187–13192
(2016).
163. Strickland, D. J., Huang, Y. R., Lee, D. & Gianola, D. S. Robust scaling of strength and elastic
constants and universal cooperativity in disordered colloidal micropillars. Proc. Natl. Acad.
Sci. U. S. A. 111, 18167–18172 (2014).
164. Cubuk, E. D. et al. Structure-property relationships from universal signatures of plasticity
in disordered solids. Science. 358, 1033–1037 (2017).
165. Giuntini, D. et al. Defects and plasticity in ultrastrong supercrystalline nanocomposites. Sci.
Adv. 7, 1–11 (2021).
166. Dreyer, A. et al. Organically linked iron oxide nanoparticle supercrystals with exceptional
isotropic mechanical properties. Nat. Mater. 15, 522–528 (2016).
167. Schilde, C., Westphal, B. & Kwade, A. Effect of the primary particle morphology on the
micromechanical properties of nanostructured alumina agglomerates. J. Nanoparticle Res.
14, 745 (2012).
168. Zellmer, S., Lindenau, M., Michel, S., Garnweitner, G. & Schilde, C. Influence of surface
modification on structure formation and micromechanical properties of spray-dried silica
aggregates. J. Colloid Interface Sci. 464, 183–190 (2016).
169. Sekido, T. et al. Controlling the Structure of Supraballs by pH-Responsive Particle Assembly.
Langmuir 33, 1995–2002 (2017).
170. Berggren, J., Frenning, G. & Alderborn, G. Compression behaviour and tablet-forming
ability of spray-dried amorphous composite particles. Eur. J. Pharm. Sci. 22, 191–200
(2004).
171. Cheong, Y. S., Adams, M. J., Hounslow, M. J. & Salman, A. D. Microscopic interpretation of
granule strength in liquid media. Powder Technol. 189, 365–375 (2009).
172. Raichman, Y., Kazakevich, M., Rabkin, E. & Tsur, Y. Inter-nanoparticle bonds in
agglomerates studied by nanoindentation. Adv. Mater. 18, 2028–2030 (2006).
173. Kozhar, S., Dosta, M., Antonyuk, S., Heinrich, S. & Bröckel, U. DEM simulations of

132
 amorphous irregular shaped micrometer-sized titania agglomerates at compression. Adv.
Powder Technol. 26, 767–777 (2015).
174. Russell, A., Müller, P. & Tomas, J. Quasi-static diametrical compression of characteristic
elastic-plastic granules: Energetic aspects at contact. Chem. Eng. Sci. 114, 70–84 (2014).
175. Antonyuk, S., Tomas, J., Heinrich, S. & Mörl, L. Breakage behaviour of spherical granulates
by compression. Chem. Eng. Sci. 60, 4031–4044 (2005).
176. Callister, W. D. Materials science and engineering: An introduction (2nd edition). Mater.
Des. 12, 59 (1991).
177. Etsion, I., Kligerman, Y. & Kadin, Y. Unloading of an elastic-plastic loaded spherical contact.
Int. J. Solids Struct. 42, 3716–3729 (2005).
178. Herre, P. et al. Deformation behavior of nanocrystalline titania particles accessed by
complementary in situ electron microscopy techniques. J. Am. Ceram. Soc. 100, 5709–
5722 (2017).
179. Griffith;, A. A. & Eng, M. VI. The phenomena of rupture and flow in solids. Philos. Trans. R.
Soc. London. Ser. A, Contain. Pap. a Math. or Phys. Character 221, 163–198 (1921).
180. Israelachvili, J. Intermolecular and Surface Forces. Intermolecular and Surface Forces
(Elsevier, 2011). doi:10.1016/C2009-0-21560-1
181. Gao, H., Ji, B., Jäger, I. L., Arzt, E. & Fratzl, P. Materials become insensitive to flaws at
nanoscale: Lessons from nature. Proc. Natl. Acad. Sci. U. S. A. 100, 5597–5600 (2003).
182. Irwin, G. R. Analysis of Stresses and Strains Near the End of a Crack Traversing a Plate. J.
Appl. Mech. 24, 361–364 (1957).
183. Vogel, N. et al. Direct visualization of the interfacial position of colloidal particles and their
assemblies. Nanoscale 6, 6879–6885 (2014).
184. Butler, M., Box, F., Robert, T. & Vella, D. Elasto-capillary adhesion: Effect of deformability
on adhesion strength and detachment. Phys. Rev. Fluids 4, 1–27 (2019).
185. Wang, X. D., Chen, B., Wang, H. F. & Wang, Z. S. Adhesion between submicrometer
polystyrene spheres. Powder Technol. 214, 447–450 (2011).
186. Rumpf, H., Sommer, K. & Steier, K. Mechanismen der Haftkraftverstärkung bei der
Partikelhaftung durch plastisches Verformen, Sintern und viskoelastisches Fließen. Chemie
Ing. Tech. 48, 300–307 (1976).
187. Arzt, E. Size effects in materials due to microstructural and dimensional constraints: a
comparative review. Acta Mater. 46, 5611–5626 (1998).
188. Merle, B., Schweitzer, E. W. & Göken, M. Thickness and grain size dependence of the
strength of copper thin films as investigated with bulge tests and nanoindentations. Philos.
Mag. 92, 3172–3187 (2012).
189. Pietsch, W. Size Enlargement by Agglomeration. in Handbook of Powder Science &
Technology 202–377 (Springer US, 1997). doi:10.1007/978-1-4615-6373-0_6

133
190. Scheel, M. et al. Morphological clues to wet granular pilestability. Nat. Mater. 7, 189–193
(2008).

134
List of Publications
 Wang, J., Mbah, C.F., Bommineni, P., Engel, M., and Vogel, N.,
The Breakdown of Magic Numbers in Colloidal Clusters (in preparation)
 Wang, J., Schwenger, J., Ströbel, A., Feldner, P., Herre, P., Romeis, S., Peukert,
W., Merle, B. and Vogel, N.
Mechanics of Colloidal Supraparticles under Compression (under revision)
 Wang, J., Amstad, E., and Vogel, N.,
Thin Shell Double Emulsion by Centrifugal Atomization (in preparation)
 Mbah, C.F.*, Wang, J.*, Englisch, S., Spiecker, E., Vogel, N. and Engel, M.,
The Path Less Taken – Decahedral or Icosahedal Symmetry in Colloidal Clusters
(in prepration)
 Wang, J., Sultan, U., Goerlitzer, E.S., Mbah, C.F., Engel, M. and Vogel, N.,
Structural color of colloidal clusters as a tool to investigate structure and
dynamics. Advanced Functional Materials, 30(26), p.1907730 (2020).
Highlighted in: Special Issue: Assembly of Materials Building Blocks into
Integrated Complex Functional Systems
 Englisch, S., Wirth, J., Przybilla, T., Zubiri, B.A., Wang, J., Vogel, N. and
Spiecker, E.,
Scale-Bridging 3D-Analysis of Colloidal Clusters Using 360° Electron
Tomography and X-Ray Nano-CT.
Microscopy and Microanalysis, 25(S2), pp.392-393 (2019).
 Wang, J., Mbah, C.F., Przybilla, T., Englisch, S., Spiecker, E., Engel, M. and
Vogel, N.,
Free energy landscape of colloidal clusters in spherical confinement.
ACS Nano, 13(8), pp.9005-9015 (2019).
Highlighted in: ACS Nano Cover
 Wang, J., Mbah, C.F., Przybilla, T., Zubiri, B.A., Spiecker, E., Engel, M. and
Vogel, N.,
Magic number colloidal clusters as minimum free energy structures.
Nature Communications, 9(1), pp.1-10 (2018).
Highlighted in: FAU Research, Science et Vie, phys.org, EurekAlert! and idw
 Egly, S., Fröhlich, C., Vogel, S., Gruenewald, A., Wang, J., Detsch, R.,
Boccaccini, A.R. and Vogel, N.,
Bottom-Up Assembly of Silica and Bioactive Glass Supraparticles with Tunable
Hierarchical Porosity.
Langmuir, 34(5), pp.2063-2072 (2018).

135
 I. Supervised students:
 S. H. (master)
Colloidal Self-Assembly in Double Emulsions Produced by Droplet-Based
Microfluidics
 W. T. L. (master)
Magnetic Colloidal Photonic Crystals with Tunable Optical Properties
 U. S. (master)
Drying Kinetics of Spherical Colloidal Photonic Crystals
 A. S. (master)
Mechanical Properties of Spherical Colloidal Crystals
 L. A. (bachelor)
Interactions Between Colloidal Particles, Surfactants and a Water-Oil Interface in
Emulsion Droplets
 T. G. (internship)
Magic Number Across the Scales
 B. T. (miniproject)
Particle Adsorption at the Oil-Water Interface under the Influence of Surfactant
 E. G. (miniproject)
Shrinkage Rate of W/O Emulsion Droplets and Crystallization of PS Photonic
Balls
II. Teaching:
 Tutor for Polymer Science and Engineering exercise class (students from
Chemical and Bio-Engineering Department (CBI) and international program of
Advanced Materials and Process (MAP))
III. Conference contribution:
Oral presentation:

 DPG Spring Meeting (Dresden, Germany, 2017)

 ProcessNet (Halle, Germany, 2017)
 Zsigmondy Colloquium (Saarbrücken, Germany, 2017)
 Soft Matter Interface (Lugano, Switzerland, 2017)
 IACIS (Rotterdam, Netherlands, 2018)
 ECIS (Ljubljana, Slovenia, 2018)
 MRS Fall Meeting (online, 2020)

136
 Poster presentation:

 EAM Symposium (Erlangen, Germany, 2016, Best Poster Prize),

 DPG Summer School (Bad Honnef, Germany, 2016),
 ICEAM (Erlangen, Germany, 2017),
 Particle-Based Material Symposium (Erlangen, Germany, 2018, Best Poster Prize)
 Particle-Based Material Symposium (Ulm, Germany, 2019),
 GRC Soft Matter Physics (New England, USA, 2019),
 Zsigmondy Colloquium (Düsseldorf, Germany, 2020)

Invited Seminar talk:

 Satoshi Watanabe’s Group (Kyoto University, 2018)

 Syuji Fujii’s Group (Osaka Institute of Technology, 2018)
 Esther Amstad (EPFL, 2018)
 Matthias Kolle’s Group (MIT, 2019)
 Special Kavli Seminar (Harvard, 2019)
 Vinothan Manoharan’s Group (Harvard, 2021)

Organization:

 Initiator and organizer for one-day symposium Young Researchers Day (Erlangen,
Germany, 2018) in the framework of German Clusters of Excellence Engineering
of Advanced Materials (EAM)
 Assistant for Particle-Based Materials Symposium (Erlangen, Germany, 2018)

137
Acknowledgement
I remember sitting in a rather empty office with a scent of newness, a youngish looking
man to my left, enthusiastically doodling some sketches on a blank paper, along with a
brisk pace of speech. “Pick your favorite”, he leaned back to his chair. I stared at the
papers scattered randomly on the small round table - the handwriting was almost
indiscernible, but I knew what it was about. Two topics were offered to me for my master
thesis, one was to structure glass surface by colloidal particles for transparent liquid
repellent surface, the other to mix small carbon nanoparticles with larger polymer particles
in a water droplet. Both sounded interesting, I had no preference. Recalling from his
exhilarated monologue that the latter was “unclear what would happen”, I chose it. This
was where the journey began.

It is a long and wonderful journey. Starting from my earliest memory, the scene has
changed throughout the years– more books stacked in the shelf, more SEM pictures were
put on the office wall, more collaborators were sitting around the small round table. Niki
managed to slow down his pace of speech and I understood his handwriting without much
effort. My countless memory of various scenery in the office overlaps and condensates
into that little yellow round table – the constant that witnesses all the events and changes
around it, like an unspecial cobblestone that witness our evolution and history. It was not
only my own endeavor, but a journal together with Niki, a fellowship of collaborators and
friends. From not knowing what to expect, we are now close to understand spherical
particles in a droplet. Ironically, until this moment, I have never mixed any carbon
nanoparticles with polymer particles in a water droplet – the first experiment planned for
my master thesis project that extends into a PhD.

I’m immensely grateful for Nicolas Vogel, my PhD advisor. Words fail to express my
gratitude, it usually goes unspoken. One’s research always carries a personal trait. To
some extend it’s like a novel, the author has to decide what part of the world to explore
and how to portrait it. Niki has given me utmost freedom to explore my interest and gather
all resources to support me. He makes himself always available for my problems in
research and in life. As year goes by, I am increasingly amazed by how he can have deep
insights in all broad range of researches in our group, to capture, critically analyze and

138
generate new ideas. I’m also amazed by how he can effortlessly sort through complexity,
spot missing or weak link in the logic, and establish a clearer logic, how he can convey
ideas to different audience. For my many years of knowing Niki, he always knows the
right thing to do, in different situations such as conferences, meetings and exams, with
different people such as collaborators, students and reviewers, as different roles such as
a scientist, a husband and a father. Spending time with him is always inspiring and joyful.
Most important to me, he is a good person with warmness and kindness. He treats people
equally and always try to help those who needs help, working towards creating an
environment that is equal, fair and inclusive for everyone. Niki is not only a great advisor
to me, but also a friend in life. I am extremely happy to be a part of our group and am
eager to continue working together to create this environment.

I’m also extremely grateful for Michael Engel, my coadvisor and also a friend in life. He is
always available when I seek help in research, career and life, also when I just impulsively
wanted to talk. At some point during my master thesis, I showed Niki Michael’s work on
icosahedral quasicrystal and was excitedly shocked at hearing “he will come here soon”.
What are the odds! I am extremely lucky to work closely with such great experimentalist
like Niki and simulator/theorist like Michael. In almost all our discussions, Michael either
knew the answer, or had an accurate intuition that turned out to be correct. He eventually
talked me out of reproducing his simulation with real colloids, saying the potential is
difficult to be realized in experiment. I hope one day I can prove him wrong. Michael
always asks “why should we do this”, which forces me to put my work into a broader and
higher perspective. For countless times I complain about reviewers, about journals and
academia, about people and opinions that I don’t agree and comprehend, he is always
willing and trying to understand and rationalize them. To me this demonstrates how
wisdom is obtained via benevolence – to understand things as they are without
involvement and selfness.

During my PhD, I received tremendous help from my students. I was fortunate to have
known my first master student Andi. I often joked that he is the average of this German
generation, to which he only grinned embarrassingly and could not find response. I meant
it as the highest compliment. We talked a lot. He always has his opinions, but is never
strongly opiniated, which I compared to the Hobbits, the only kind that can resist the

139
temptation of the ring. Umair is great fun, he always finds the most proper jokes and witted
comebacks in our group’s daily bantering, sparkling, humorous but not offensive. This
balance can only be kept by an agile mind and a sincere respect for others. Wai Theng
pays great attention to details, considers as many as possible aspects of a problem and
is always planning ahead. She took on challenges with courage and proceeded to
success through hard work, intelligence and resilience. Simon is quick to understand
everything and adsorbs knowledge like a sponge. He is always eager to initiate, to change
and to perfect things with energy and activeness. I also had some short-term students.
Bogdan can single out subtle variables and execute sensitive experiments without
mistake; Theo can grasp ideas from wide-ranging fields and is always open for new ideas;
Lisa can find a clear line of logic, sorting through unknown experiment parameters. In
their master, bachelor or short-term projects, all of my students exhibited capacity and
personality that I admire and wish to have. From our close interaction, they all have left
me great legacies to pick up. I was only their supervisor for thesis, they were my teachers
for work and life.

My memories in our group somehow breaks into several pieces. The first was when our
group started to form. It felt like a small family, with me, Stefen, Houda, Yara and Steffi
like a big sister. I remember Steffi introducing me the wonderful world of microfluidics in
an empty lab where I lay everything out on the working bench. It felt like spring, I
remembered a dim office with them where green crept up outside. The small family grew
a little with Karina as the big sister, single handedly taking over tasks to ensure the smooth
running of a lab, which after she left had to be shared by all employed members. With
Marcel R. of strong determination to success and excellent management skills. With
visiting Grant who lighted up the room and Sunny who was pleasant and comprehensively
presentable. The family grew larger with Nemanja, Denise, Roman and Melli. Nemanja
was an old friend around whom I was very comfortable and relaxed, who always mocked
how I said “busy” and whose name I could never pronounce properly. Denise was always
nice, Roman was always creative. Melli was sincere and free. I remember boldly
inspecting every corner in her parents’ house in her birthday parties and out gathering in
the Berg after she left the group. Then the family grew even larger, my life became more
dramatic: Andi with his warm and silly grin, Josing, Taotao and Chrissy mysteriously

140
gossiping at the corner, Marcel D. pulling his boy stunt, Lutz babbling like a radio, Jo and
Flo effectively encouraging each other to do childish things, Umair fighting to take back
control of his playlist, Yara recording proof of Umair’s fake promises, Reza shining laser
through his gin, Tobi whipping nitrogen-cooled ice cream, Teresa energetically calling for
another round of drinks, Salva transitioning through stages of intoxication from overly
excited jokes to incomprehensible loud voices. Eric waking up with a half full glass of gin
on his belly without spilling, Niki enthusiastically wanting to talk just about anything with
anyone at 4am, which the experienced skillfully dodged, tittering at the inexperienced who
took the impact. It felt like in a cliché teen comedy, it felt like summer. It was the pool
outside the lab, it was the long benches on the lawn, it was the Berg. My memories were
vague mosaic of Queen music from a far-away loudspeaker, out-of-tone shouting, racing
heartbeats, drunken zombies filling the street, a dark narrow lab corridor barely fitting the
long bench, from tidy dishes to scattered cups, from carefully balancing on the table to
recklessly stumbling, from warm sunshine in blue sky and green grass, to cool air in
twilight, to chill and shivering but loud darkness, and of course, the headache and regret
in the morning. I advocated a few times to instore a camera in the floor’s eventful kitchen
and have our own reality show to bring some extra funding to our research activities, it
was never considered seriously. It would have captured more memories that were
beginning to wear out in my head - only a small fraction was stored in people’s phone or
mind, each in its own perspective. It was always fun and enlightening to talk about past
parties, cross examining different witnesses’ testimony, putting pieces of evidence, being
shocked at what actually happened, and finally realized “oh that’s why”.

The good companionship nurtured a good working environment. Eric has great execution
and actively offer his help. Salva almost feels obliged to help others. Yara prioritize her
responsibilities for others even before her own experiments. Reza managed the group at
Niki’s absence and show unparalleled courage in life. Teresa put the interest of others
above hers when office space is limited. Guddi speak up and defend others. Julius acts
immediately to help although joking oppositely. Herbert maintains diligence and
management throughout the years. Johannes minds no one else business yet ready to
help when requested. Giulia takes initiative and always ready to share her experience
and expertise with those who need them. Vaibhav shows deep sympathy for the less

141
fortunate and works towards change. The list would go on for too long if I were to include
every member that have taught me something and have shared a fond memory with me.
Lukas, Melissa, Meichen, Moritz, Tamara, Irene, Katharina, Sophie, Carina, Laura,
Benedikt, Suky, Saskia, Robert, Maike, Betty, Fredi, Maria-Nefeli, Florian, Maret, Jacob
and Arvind, my interaction were limited to some of these members, either due to the
nature of our projects, the duration of their stay, or the restriction from the Corona
lockdown. But they all brought their unique elements, and together shaped the group into
a diverse, inclusive, and loving family that tolerated, educated and nurtured me.

I am also grateful for my collaboration partners. Chrameh is my closest collaborator since

I started the PhD, my work depends on his. I lost count of the time we sat together,
examining his simulation data together, figuring out what my colloid experiment cannot
reveal. He is always kind and patient, responding to the request and ideas that came to
my mind out of nowhere (and often led to nowhere). Praveen joined later. Praveen makes
difficult task seem effortless and became immediately excited when we brainstormed new
ideas. Many resulted in wonderful discovery, which I still owed him to realize them in
experiments. I was a frequent visitor to the Engel group, Marco often joined our discussion
with his insight and sharp intuition. Engel group is the place to find answers, Spiecker
group is the place to find proof. I remembered the first time I met Thomas, asking his help
for tomography. Next thing I knew I was sitting next to him, mesmerized by his masterfully
operation at the FIB. I remember sitting in front of a TEM with him, Benjamin and Ming
before Christmas, our first trial with the first cluster sample gave us breathtakingly
beautiful images – luck has rewarded their expertise. It remained the one of the most
beautiful things I’ve seen. Then Silvan came to help with his magic to identify and relocate
the cluster in NanoCT. He is calm, reliable and always has a solution. I spent a lot of time
in the Spiecker group and became acquainted with more of their members. Once I met
some of them in a bar, they sat next to our group members and asked for our names and
roles in the Vogel group. Eventually they asked Niki – “so you are also working in the
Vogel group, huh?” It was very amusing when Niki said, “yes, I am Vogel.” I’m also grateful
for Prof. Spiecker to grant me access to his facilities. He speaks softly, the slow pace of
speech always comforting and reassuring, giving me confidence that things will work.
With the Engel and Spiecker group, working is always full of excitement and fun, they are

142
all welcoming and capable, sincere and friendly. I knew they have my back. They are like
the Aladdin lamp, any problem I threw at them will be solved, any wishes will be granted.

I was also fortunately to have some academic exchange. My four weeks in Japan left me
one of the greatest memories in life. The beautiful Kyoto city, metropolitan Tokyo area,
lovely houses and skyscrapers, cleanest street and disordered poles, temples and
pensais, delicious raman next to the Katsura campus, for someone who grew up watching
Japanese animation and manga, I was living in a childhood dream. Japan was so different
yet similar. Satoshi and Atsushi welcomed me with great hospitality. Satoshi and I talked
about everything, Atsushi has the most diligent work ethnics and became a dear friend.
Satoshi’s group were also extremely inclusive. Later in that year, I spent another month
in Switzerland in Esther’s lab. She is so sharp and full of energy. Gianluca had been so
generous in helping me with surfactant and microfluidic chips. I remembered the hike with
her group, and a weekend getaway with Taotao by the beautiful lake of Lausanne. It is
the most beautiful city I would like to live in. I also want to thank Joanna for arranging a
special Kavli seminar and Mathias in his group seminar present my work when I visited
Boston, Anna and Soeren for showing me around Joanna’s Lab, Anthony in Mathias’ lab,
Yinan and Zizhao in Dave’s lab and giving me confocal training and microfluidic spacers,
Ming in Vinny’s lab and later invited to a talk and Chrisy, who showed me the good food
in Boston. I met Peter Lu who is impressively knowledgeable and fiercely energetic. I
admired his overwhelming originality and freedom in his work. They were all incredibly
intelligent and kind-hearted people who were inspiring for me, even for such a short time
– I only wish I could spend more time with them. I had a bit taste of my childhood American
dream. I remain in close contact with some of them. Ming Xiao is always helping me with
career advice, Chrisy Du became a close friend with whom I talked on a weekly basis. I
would share with her my enthusiasm to write some fan fictions of a manga, and in the
next minute joined her in a meeting with Niki and Julia to organize an academic
symposium. Similarly, I met Da Wang in a conference, and remain close contact. He is
not only an expert in confinement and electron tomography but knows just about every
paper out there. It is simply amazing.

I was fortunate to receive many more collaborations. Benoit had spent tremendous
amount of time to help me examine the mechanics data measured by Andi with the help

143
of his student Patrick Feldner. He is a careful scientist, always helpful and insightful. Prof.
Peukert also went to great length to proofread our manuscript, providing expert advice to
bring out the hidden science in the mass amount of data obtained by Andi, Jan, Patrick
Herre and Stefan. I had received some amazing particles, the hematite cubes from Laura
Rossi, the core-shell particles from Matthias Karg, the iron oxide nanoparticles from Alex
and Gerold Schneider, and polyhedral MOF particles from Yang, Inhar Imaz and Daniel
Maspoch. All of them were doing wonders in droplets, and we were beginning to
understand their behavior and utilize their properties. I am grateful that they have trust in
us to work together to explore the possibilities of supraparticles.

I could not have carried out my work without the ardent help from Michael Auer and Paula,
who took charge of safety, waste disposable, SEM training, contracts, purchases, and
many other bureaucratic burdens that I have avoided only due to their help. I was always
greeted with smile even when I was always bringing them work. The same goes to the
workshop, Julia, Tarek, Margita, Saman, and Cornelia. I am grateful for their hard work
that make mine easier. Vanessa, Frederik, Flo, Julia Seifert, Holger, Monica, Uwe, Tobias,
Nabi … There were more colleagues in the institute that I would like to thank for their
friendly conversation and company in the Berg. It was a great time. Particularly Andi V.,
who was helping me with the ordering, both for consumables in the lab and for Japanese
snacks in online stores when I need someone to join for free delivery. He is incredibly
kind.

I also owe a great deal to my friends in life, apart from those in our lab. Taotao, Mo and
Xiaokun are by my side in my ups and downs throughout the years. I could not imagine
surviving my PhD without them. Since I came to Germany, I have more regular video or
wechat calls with my parents, we talked about everything like friends. They are my heroes
and the greatest people I know. I am beyond grateful.

I had many travel plans after my PhD, I wanted to visit Da Wang’s family in Utrecht, to
visit Chrameh’s family in Stockholm, to visit Atsushi in Kyoto, to visit Chrisy in Boston
(now in Philadelphia), none is possible now due to the pandemic. The pandemic has
completely changed the world, but even before the pandemic, the world was tumbling
and sliding towards a dangerous direction. I was fortunate enough to be surrounded by

144
loving people, but when I looked outside my small circle of life, there were growing
conflicts between races, nations, religions and classes. Even in my surroundings, the
anxiety of the pandemic is slowly corrupting people’s goodwill. The colloids in droplets
taught me some lessons. It demonstrates the interplay between local and global optimum
– spheres are able to pack optimally into local fcc grains, which also happen to pack
tightly globally in spherical confinement, at a small sacrifice at the boundaries. Other
shapes may have a difficulty to reach local and global optimum at the same time, only
leaving a mess. If we all work only to maximize our own interest, without caring for others,
we may end up far away from a global optimum, which eventually hurts our local
environment, as we are all living in this small world. The colloids also show that always
taking path of immediate and maximal energy decrease in the short term may deviate the
system from ending up in fcc packing, the optimal in the long term. We must not be blinded
by short term interests and choose less appealing, sometimes difficult path that would
eventually lead us to our desire destination.

Wisdom is our solution to problems and conflicts. To obtain wisdom one must understand
the whole instead of the part. To do so one must acknowledge that others exist, then one
must understand that we are all the same human being, be it the colleagues around us,
or the people across the ocean. I then realized communication is the solution. The image
of anyone can vary contrastingly from one month to another, it could be glorified or
demonized. We all carry many tags. When focused on these tags, abstraction easily leads
to dehumanization. Exposure and communication reveal that we are all the same people
who just want to be happy, to love and be loved. But communication can also deepen
differences, evidence can reinforce bigotry, if we decide to close our eyes, we will never
see. The only way is to want to understand that we don’t understand, be it people or
things. Achieving wisdom is not an ability, is a choice, a choice to show kindness to others.
Love your enemies, my younger shelf would laugh at this concept, but our ancestors have
laid out the solution, so repeated it became cliché. Kindness to others is redemption to
ourselves, not a redemption from sin, error or evil, but redemption from pain and suffering.
Kindness is our only way out.

145