Vous êtes sur la page 1sur 37

Bachelor Thesis

The behaviour of graphene under


one-dimensional compression

T.V. de Ruijsscher

Supervisor: External advisor:


Prof. Dr. A. Fasolino Dr. P. Christianen

Radboud University Nijmegen

Institute for Molecules and Materials,

Theory of Condensed Matter

Summer 2012
2
Contents

Preface 5

1 Graphene 9

2 Monte Carlo simulation 13


2.1 How to evaluate physical properties at finite temperature . . . . . . . . . . . 13
2.2 Monte Carlo procedure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
2.3 Wavemoves . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
2.4 Discrete Fourier Transform . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17

3 Results: Deformation under compression 21


3.1 Periodicity dependence on compression and temperature . . . . . . . . . . . . 21
3.2 Periodicity dependence on the length . . . . . . . . . . . . . . . . . . . . . . . 25
3.3 The effect of periodic boundary conditions . . . . . . . . . . . . . . . . . . . . 31

4 Conclusions 33

5 Outlook 35

Bibliography 37

3
4 CONTENTS
Preface

Motivation
The interest for the behaviour of graphene under pressure was first born when reading the
article Multiple-length-scale elastic instability mimics parametric resonance of non-linear os-
cillators by F. Brau et al. from the university of Mons [1]. This article describes the nature
of rigid membranes under pressure. What Brau showed, is that ‘compressing a membrane
resting on a soft foundation creates a regular pattern of sinusoidal wrinkles with a broad
distribution of energy’, where, for stronger compression, period-doubling bifurcations occur,
as is shown in fig. 1 and 2. Since at the department of Theory of Condensed Matter at
the Radboud University, where this research project took place, special interest goes to the
study of graphene, the discoveries in this article made us curious whether graphene would
behave in such a way as well. According to Brau, the bifurcations are ‘induced by the intrin-
sic non-linearity of the elasticity equations’. For that reason, I worked closely together with
Marco Stevens, who accounted for the more mathematical part of this problem. He showed
among other things that even if the graphene would not rest on any substrate, there should
be non-linearity and therefore bifurcations [2].
But why would you even bother to research a subject like this? As is often the case in
fundamental physical research, no applications are available at this moment. However, most
of modern facilities would not be possible if years beforehand some fundamental research
had not been done. In this specific case, many scientists all over the world are unravelling
properties of graphene. And why not? It seems reasonable to know as much as possible of a
new material that has enough opportunities for application in, for instance, electrical devices
as solar cells and touch screens [3]. One could imagine that by creating a regular, static,
wrinkle pattern one could modify the electronic transport or create a template for growth of
molecular crystals.

5
6 CONTENTS

Figure 1: From reference [1]. This picture shows the pattern formed by a thin stiff PDMS
membrane resting on a thick soft PDMS substrate. (A) With a small compression ratio,
a regular sinusoidal waveform is established. (B) By increasing the compression ratio, the
wavelength decreases, whereas the amplitude increases. (C) At a certain compression ratio,
the waveform, instead of being described by one sinusoidal function with one wavelength,
has become a combination of two such functions. This event of period-doubling is called
bifurcation.

Figure 2: From reference [1]. This picture shows the amplitudes (A1 and A2 ) as a function of
the compression ratio δ. System 1, dealt with in fig. 1, is showed by the open triangles, whereas
system 2 of a PDMS foundation cured with UV/ozone is reported by the filled symbols. A
bifurcation is visible with increasing compression: the system splits in two branches.
CONTENTS 7

Build-up
This thesis should give a final view of my bachelor internship at the department of Theory
of Condensed Matter under supervision of Prof. Dr. Annalisa Fasolino. Most of all, the used
methods and the results are emphasized. In order to make this picture as complete as possi-
ble, I will first discuss some basic properties of graphene in chapter 1. In chapter 2, the Monte
Carlo method is highlighted, together with the used procedure, finishing with an explanation
of the method of the Discrete Fourier Transform. After the theoretical background, we face
the results in chapter 3, consisting of the wrinkle periodicity dependence on compression,
temperature and length. Furthermore, some remarks on the use of periodic boundary con-
ditions are made. Based on the results described, some conclusions are drawn in chapter 4.
The last chapter gives an outlook with suggestions for future research.
8 CONTENTS
Chapter 1

Graphene

Since this whole thesis is about graphene, we start by introducing the main structural prop-
erties. Graphene is at first sight a simple construction, being no more than a sheet consisting
of one layer of carbon atoms arranged in a honeycomb crystal lattice. Or as Andre Geim
and Konstantin Novoselov described it [4]: it is ‘a flat monolayer of carbon atoms tightly
packed into a two-dimensional (2D) honeycomb lattice. [. . . ] It can be wrapped up into 0D
fullerenes, rolled into 1D nanotubes or stacked into 3D graphite.’ These last properties make
graphene a sort of ‘mother of carbon structures’, as shown in fig. 1.1.
Since the listing of carbon as a chemical element by Antoine Lavoisier in 1789, a lot has
been discovered in the scientific world of carbon. In the 20th century, almost all variants of
this element had been discovered, including 0D, 1D and 3D varieties. Only the 2D allotrope1
was missing, although named ‘graphene’ already in 1962 by Hanns-Peter Boehm [5]. It was
not until 2004 that the long-expected final jigsaw piece of the carbon family was found, by the
earlier mentioned physicists Geim and Novoselov. For this discovery they earned the Nobel
Prize in Physics in 2010 [6].

After this very brief history, let us now look a bit closer to graphene. When one takes a
sample of, say, 840 atoms in a rectangular cell of 170.9 Å by 11.4 Å, one can see in fig. 1.2a
and fig. 1.2b that graphene is completely flat at the temperature T = 0. However, when
temperature rises a bit, it starts to show so-called ‘ripples’: out of plane fluctuations. These
ripples are an intrinsic property of graphene, crucial to ensure thermal stability [7] and can
be seen in fig. 1.2c and fig. 1.2d.
To be able to simulate the graphene numerically, more knowledge about the bonding
between the carbon atoms is required. The atoms interact with each other by a strong,
covalent sp2 bond, 3-fold and directional, as shown in fig. 1. The low coordination rules
out the possibility of describing it by means of a radial potential that always favours high
coordination. To achieve a proper description, the so-called LCBOP is used, which stands
for Long-range Carbon Bond-Order Potential [8]. This is a rather complex potential, which
depends not only on the position of the atoms, but also on the angles between them and
on their coordination number (i.e. the number of nearest neighbours). The difference in
coordination number between the carbon allotropes diamond (4) and graphene (3), can be
1
Allotropy is the phenomenon that a particular element can occur in nature in different forms, because of
a difference in their crystal structure. The allotropes of carbon are diamond (3D), graphite (3D), graphene
(2D), nanotubes (1D) and fullerenes (0D).

9
10 CHAPTER 1. GRAPHENE

Figure 1.1: From reference [4]. Different allotropes of graphene. On the top, the 2D allotrope
graphene as the ’mother of carbon structures’. Furthermore, from the left to the right the
0D, 1D en 3D allotropes (fullerene, nanotube and graphite respectively).

(a) Top view of the sample at T = 0.

(b) Side view of the sample at T = 0.

(c) Top view of the sample at T = 300 K.

(d) Side view of the sample at T = 300 K.

Figure 1.2: A graphene sample of N = 840 atoms with dimensions Lx = 11.4 Å and Ly =
170.9 Å at two different temperatures, showing the ‘ripples’ at non-zero temperature.
11

(a) Graphene has a coordination number of 3 and (b) Diamond has a coordination number of 4 and
bonds forming angles of 120◦ . bonds forming angles of 109.5◦ .

Figure 1.3: The allotropes graphene and diamond have a different coordination number.

seen in fig. 1.3.


Since this thesis focusses on stretching and bending properties of graphene, it is useful
to describe its free energy. According to Marco Stevens in his Bachelors’ thesis [2] the free
energy can be split in three parts as

F = Fbend + Fstretch + Fline , (1.1)


where we can neglect the last term only depending on the boundary of the surface for
very large systems, leaving

F = Fbend + Fstretch . (1.2)


The bending and stretching terms can be written as


κ ( )2
Fbend = d2 x ∇2
2
∫ (1.3)
1 ( )
Fstretch = d2 x λu2αα + 2µu2αβ ,
2

according to M.I. Katsnelson in his book Graphene, carbon in two dimensions [9], where
κ is the bending rigidity, λ and µ are the so-called Lamé coefficients and uαβ is the strain
tensor. This results in the total free energy
∫ { ( )
1 2 ( )}
F = d2 x κ ∇2 + λu2αα + 2µu2αβ . (1.4)
2
According to K.V. Zakharchenko et al. [10] κ increases with temperature as shown in
fig. 1.4, and is therefore lowering when cooling down. This means that the bending component
of the free energy F becomes less important with decreasing temperature.
12 CHAPTER 1. GRAPHENE

Figure 1.4: From reference [10]. Temperature dependence of the bending rigidity κ of
graphene, where the solid line with circles corresponds with single layer graphene.
Chapter 2

Monte Carlo simulation

2.1 How to evaluate physical properties at finite temperature


When simulating the behaviour of graphene computationally, we have to somehow reach
states with the lowest Helmholtz free energy, as required for thermal equilibrium [11]. This
free energy is given by
F = −kB T ln(Z) , (2.1)
which leaves us with the task to calculate the partition function Z. However, for a complex
system like graphene, this partition function, defined as

Z= e−Ei /kB T (2.2)
i

over all configurations of the atoms, is impossible to solve. Therefore we have to look for a
method that bypasses this quantity. One way of achieving this, is by calculating averages in
the canonical ensemble. The average of e.g. the quantity A can be calculated by

A(r) exp (−βE(r))dr
hAi = ∫ , (2.3)
exp (−βE(r))dr
where N is the number of particles, r the 6N -dimensional coordinates of the phase space and
1
β=
kB T
the inverse temperature with kB the Boltzmann constant. Now the integral in (2.3) can be
approximated by replacing it by a sum over all configurations M , where we have to correct
by dividing through this number:

1
∑i Ai exp (−βEi )
hAi = . (2.4)
M i exp (−βEi )
Of course not all the configurations are of equal probability, and some are negligible
compared to others. In order to be able to count only the contributions to the average that we
cannot neglect, we assign a weight factor pi to each contribution. When evaluating the average
with this factor and only counting the configurations that give a noticeable contribution to
hAi, say M , it remains
∑M −1
1 i=1 pi pi Ai exp (−βEi )
hAi = ∑M . (2.5)
M i=1 exp (−βEi )

13
14 CHAPTER 2. MONTE CARLO SIMULATION

In the case of the canonical ensemble, the Boltzmann distribution turns out to be a good
choice, i.e.
exp (−βEi )
pi = , (2.6)
Z
which leaves us with

1 ∑
M
hAi = pi Ai
M
i=1
(2.7)
1 ∑M
= A∗i ,
M
i=1

where A∗i are the sampled values when using the Boltzmann distribution as weight factor.
In the method described above, it can be seen that the Boltzmann distribution and the
partition function are used in the sampling, using (2.6). However, we want this sampling
without using the Boltzmann distribution explicitly. To achieve this, a method called ‘Metro-
polis sampling’ or the ‘M(RT)2 algorithm’ (after the last names of the five authors) comes in
sight. In this thesis, the background of this non-trivial sampling technique will not be dealt
with. However, in the next section the procedure will be described step by step.
After the sampling, we want to minimize the Helmholtz free energy, given by both (2.1)
and
F = E − TS , (2.8)
with S the entropy of the system. As can be seen immediately, the internal energy E → F
as T → 0. Both the sampling according to Metropolis and the minimization of the internal
energy give a considerably simplification of the calculations, where the results are still a good
description of physical reality.

2.2 Monte Carlo procedure


To start simulating the behaviour of graphene under 1-dimensional compression, we have to
start with a particular graphene sample. In this case, we start with an elongated sample,
using periodic boundary conditions in both x- and y-direction. By doing this, we can limit
ourselves to no more than five honeycomb cells in the y-direction, which saves us a remarkable
amount of calculation time. We used among other things the sample shown in fig. 2.1, with
dimensions 687.4 Å by 11.4 Å.
According to the method of Metropolis sampling, the Monte Carlo procedure is done in
different steps, which will be dealt with in the following. One note has to be made beforehand:
the steps to produce the algorithm are really non-trivial and a full derivation goes far beyond
the goal of this thesis. Therefore, only the useful equations will be given.

Move 1 atom
The first step is to move one atom. However, one cannot simply move one atom without
restrictions. According to the method of Metropolis sampling [12], we can take a uniform
trial probability (i.e. the probability to have a proposed move from y to x) of
{
1/∆ |x − y| < ∆/2
P (x|y) = , (2.9)
0 otherwise
2.2. MONTE CARLO PROCEDURE 15

(a) Top view of the sample.

(b) Zoom of fig. 2.1a to show the small size


of Ly .

(c) Side view of the sample.

Figure 2.1: Two views of the used sample (LS ) of N = 3360 atoms with Lx = 11.4 Å and
Ly = 687.4 Å.


where ∆ ∝ T.

Determine the energy difference


To be able to decide whether a move has to be accepted to contribute to the sum in (2.3)
to evaluate averages in the canonical ensemble, we have to determine the difference U 0 − U
between the internal energy after and before the trial move.

Criteria for acceptance or rejection


Once done a trial move and determined the appropriate energy difference, one has to decide
whether this move should be accepted. Therefore, the method of Metropolis sampling offers us
a tool based on (2.9). According to this theory, the acceptance probability (i.e the probability
that a proposed move from y to x is accepted) is given by
[ ]
f (x)
A(x|y) = min 1, , (2.10)
f (y)

where f (x) is the probability for the system to be in state x. Using (2.10) and the fact that
f (r) ∝ exp (−βU (r)) for the specific case of the canonical ensemble, one can conclude that
the acceptance probability becomes
[ 0
]
A(r0 |r) = min 1, e−β[U (r )−U (r)] . (2.11)

Now it becomes clear why the energy difference is a useful thing to know, since this is the
only quantity the acceptance probability depends on.
The algorithm now makes use of a random number ξ ∈ [0, 1], generated by the computer,
which helps us determining whether to accept the trial move or to reject it.

ˆ If A(r0 |r) = 1 the move is accepted.


16 CHAPTER 2. MONTE CARLO SIMULATION

Figure 2.2: The same sample as in fig. 2.1, but with a built-in sinusoidal pattern.

ˆ If 1 > A(r0 |r) > ξ the move is accepted.

ˆ If A(r0 |r) < ξ the move is rejected and r0 = r (the new configuration equals the old
one).

The total simulation


In the total simulation, all previous steps are repeated until the system stabilizes. Of course,
all the steps have to be done by the computer, making use of the program LCHBOPMC by
Jan Los. In this program, we can use the earlier mentioned LCBOP potential [8] to describe
the graphene sample. Furthermore, the program can do millions of Monte Carlo steps (1
Monte Carlo step is defined as N moves, where N is the number of atoms) in order to make
the sample converge to a stable free energy minimum.
Since the radius ∆ rises as the temperature rises, and therefore the energy difference too,
it is wise to simulate at relatively high temperatures, to make sure that the fluctuations are
high enough. At the end, the whole sample can be cooled down and stabilized again, to be
sure that the minimum in internal energy corresponds to the minimum in Helmholtz free
energy, according to (2.8). As told already in the introduction, Brau wrote about ‘a regular
pattern of sinusoidal wrinkles with a broad distribution of energy’ [1]. For this reason, it
could be worth trying to start with a sinusoidal sample instead of a flat one, like in fig. 2.2.
However, it is no good to be groping in the dark and just put some periods in it. One can
use periods visible in the sample after a simulation starting from flat, or make use of the
information that the Fourier transform gives us. I will come back on this last subject at the
end of this chapter.

2.3 Wavemoves
Although the Monte Carlo method described above gives us great opportunities to simulate
the behaviour of graphene under 1-dimensional compression, there is one improvement that
could make the convergence occur earlier, which could save a lot of time: the implementation
of so-called ‘wavemoves’.
The main purpose of this method is to make it easier to create a sinusoidal pattern of long
wavelengths on the sample. With our original Monte Carlo model, it is extremely improbable
to happen. However, this addition to our model makes it possible to move all atoms with a
wave-like displacement in the z-direction. Since the short wavelengths are already dealt with
by our original model, we limit this new implementation to use values corresponding to the
largest wavelengths. Furthermore, we only want wavelengths that make a number of whole
sines fit nicely on the sample as to have zero displacement at the edges.
To achieve this, one can define wavevectors q as
( )
2π 2π
q = (qx , qy , 0) = mx , my , 0 , (2.12)
Lx Ly
2.4. DISCRETE FOURIER TRANSFORM 17

where mx and my are integers and Lx and Ly the dimensions of the sample in the x- and y-
direction, respectively. Fortunately, Jan Los’s program has an extension that makes it possible
to set these so-called ‘wavemoves’ on and (because there is only compression in the x-direction)
set the parameter qx,max which defines the minimum wavelength of the wavemoves.
Unfortunately, the method of wavemoves failed to give a good result, not because of the
program, but because of a fault in the initial conditions. Together with a lack of time, the
results with the method of wavemoves will not be included in this thesis.

2.4 Discrete Fourier Transform


When one has to do with sinusoids of different frequency, it is always a useful thing to use the
method of Fourier transforms (FT). Basically, it translates a spatial spectrum to a frequency
spectrum as ∫ ∞
1
g(ω) ≡ F[f (x)] = √ f (x)e−2πiωx dx . (2.13)
2π −∞
However, since we are dealing with a discrete signal with a limited spatial reach having the
spatial coordinates of the N atoms as our measuring points, we need to use the so-called
discrete Fourier transform. We therefore define a step in the frequency by

∆ω = , (2.14)
L
where L is the considered interval. With this information, the discrete FT is defined as
(according to A first course in Computational Physics [13], where a detailed deviation is
given)


N −1
f (j∆x)e−i2πj N
m
g(m∆ω) =
j=0
(2.15)

N −1 [ ( m) ( m )]
= f (j∆x) cos 2πj − i sin 2πj ,
N N
j=0

with m = 0, . . . , N − 1.
For we are only interested in the amplitude of the discrete FT, we write the real and
imaginary parts


N −1 ( m)
<(g(m∆ω)) = f (j∆x) cos 2πj ,
N
j=0
(2.16)

N −1 ( m)
=(g(m∆ω)) = − f (j∆x) sin 2πj ,
N
j=0

using as final transform

|FT| ≡ |g(ω)| = <(g(m∆ω))2 + =(g(m∆ω))2 . (2.17)

with m = 0, . . . , N − 1 and f (rj,x ) = rj,z . rj,ξ being the ξ-coordinate of the j th atom.
18 CHAPTER 2. MONTE CARLO SIMULATION

Since we will only present discrete FT in the following we will use FT as an abbreviation for
discrete Fourier transform.

In this thesis, we are dealing with a graphene sheet of initial length Lx . We are interested
in the out-of-plane positions of the atoms. Therefore we define f (m∆ω) = rj,z and rj,ξ 0 the

in-plane ξ-coordinate of the undistorted sample. Using this information and the definition of
the wavevectors q in (2.12), we can write the FT as

rj,z e−i(rj,x qx +rj,y qy )
0 0
g(qx ) =
all atoms j

rj,z e−irj,x qx
0
= (2.18)
all atoms j
∑ [ ( 0 ) ( 0 )]
= rj,z cos rj,x qx − i sin rj,x qx ,
all atoms j

making use of the simplification that we only consider qy = 0.


The only thing left is control the implementation of this procedure. Therefore, to test
the program performing the FT, we have constructed a sample modulated by three added
sines of different period and a sample with an added step function and calculated the FT
with the results shown in fig. 2.3 and fig. 2.4 respectively. As can be seen, they both give the
expected result. For the periodic pattern in fig. 2.3b, there are peaks at qmx = 0.0730 Å−1 ,
qmx = 0.1095 Å−1 and qmx = 0.2555 Å−1 , which correspond to mx = 2, mx = 3 and mx = 7
respectively according to (2.12). For the step function, fig. 2.4b shows multiple peeks that
fit nicely on a sinc function1 being the continuous Fourier transform of the step function.
Although the model gives zero at qx = 0 instead of a large peak, this model is still very
useful, because these limit cases are not important in this research project.
This whole exercise gives the opportunity to define which frequencies are most present in
the final configuration of the graphene sample, namely the peaks in the FT. This information
can be used not only to justify whether or not a period doubling bifurcation occurs, but
also to put these frequencies as sinusoidal patterns as starting sample in the Monte Carlo
simulations, as mentioned earlier. This procedure could lead to a gain of time in stabilizing
and probably to results that are more stable.
Unfortunately, this last method does not seem to work that well, because after equilibration
the peaks in the Fourier spectrum remain at those qx -values that are put in the initial sample.

1 sin(x)
The sinc function is defined as sinc(x) = x
.
2.4. DISCRETE FOURIER TRANSFORM 19

(a) the sample

1.8E+03

1.6E+03

1.4E+03
Fourier transform per particle

1.2E+03

1.0E+03

8.0E+02

6.0E+02

4.0E+02

2.0E+02

0.0E+00
0.0000 0.0730 0.1460 0.2190 0.2920 0.3649
qx (Å−1)

(b) the DFT

Figure 2.3: (a) The sample of 840 atoms and (b) the FT of a sum of a 7-period, a 3-period
and a 2-period sine.
20 CHAPTER 2. MONTE CARLO SIMULATION

(a) the sample

2.5E+05

2.0E+05
Fourier transform per particle

1.5E+05

1.0E+05

5.0E+04

0.0E+00
0.0000 0.0730 0.1460 0.2190 0.2920 0.3649
−1
qx (Å )

(b) the DFT

Figure 2.4: (a) The sample of 840 atoms with a step and (b) the FT of that step function,
with dashed the sinc function, which is the continuous Fourier transform of a step function.
Chapter 3

Results: Deformation under


compression

In this chapter, the results of the different simulations are presented. To give a total picture
of the results, they are divided into different sections dealing with different dependencies.
The simulations are done for samples of different lengths (called LS , LM and LL in increasing
order of length, with specific values shown in table 3.1), and different compressions (3%, 5%
and 8%). During the simulations, the temperature decreased from T = 700 K to T = 0.1 K.

name N Lx (Å) Ly (Å)


LS 3360 687.4 11.4
LM 5880 1203.9 11.4
LL 8400 1720.4 11.4

Table 3.1: The number of atoms N and the length in both x- and y-direction (Lx and Ly
respectively) for the three used samples LS , LM and LL .

3.1 Periodicity dependence on compression and temperature


In order to study the dependence of the periodicity of the wrinkles on compression and tem-
perature, we took a fixed length (LS ). After that, the temperature was kept constant at
varying compression and vice versa. In fig. 3.1 we show the configurations at different tem-
peratures during the cooling down process. The results of variation in both temperature and
compression are shown in fig. 3.2, showing the periodicity in the sample and the corresponding
FT at different temperatures and under different compressions.
As we can see in fig. 3.1, at high temperature (T = 700 K), the different configurations
during the simulation differ quite a lot. This can be seen by comparing the first plot (1
configuration) with the second one (20 configurations plotted one over the other, showing a
broad band). However, when cooling down the shape becomes more stable, as can be seen by
the band narrowing with temperature.
As we can see in fig. 3.2, for both 8% and 3% compression the highest peak in the FT is
at qx = 0.05474 Å−1 , corresponding to a 114.8 Å period1 . This is roughly a sixth of the total
1
In this thesis, the periods are in measures of the (initial) sample before compression.

21
22 CHAPTER 3. RESULTS: DEFORMATION UNDER COMPRESSION

length of the sample and thus explains the global six-period pattern.
At T = 300 K visual inspection suggests period doubling for a compression of 8%, which
is clearly less present at the lower compression of 3%. This fact is confirmed by the FT
on the right-hand side of fig. 3.2. For 8% there are clearly two dominant peaks, the main
one at qx = 0.05474 Å−1 (corresponding to 114.8 Å) and a second one at qx = 0.04562 Å−1
(corresponding to 137.7 Å). For 3% instead there is no clear second periodicity, but only many
other smaller components. At T = 0.1 K the height of the main peak is more pronounced
with respect to the second. The second periodicity is also less evident in the shape of the
wrinkle. One reason for this behaviour could be that the bending contribution to the energy
is temperature dependent, as already shown in fig. 1.4 on page 12. This could cause the
period-doubling effect to vanish at temperatures low enough.
Moreover, for the 8% compression, the FT are negligible for qx > 0.11861 Å−1 (periods <
53.0 Å), whereas for 3%, they are negligible for qx > 0.09124 Å−1 (periods < 68.8 Å). This
means that for a higher compression, smaller wavelengths become more important. This
result makes the description of the wrinkle formation in terms of a bifurcation process less
straightforward. However, this may point at a point too far in the bifurcation scheme. That
is, each branch in fig. 2 splits again in two branches and so on. This would cause a very
complex situation for high compression.
15
10 1 configuration, T = 700 K
5
0

z (Å)
−5
−10
−15

15
10 20 configurations, T = 700 K
5
0

z (Å)
−5
−10
−15

15
10 21 configurations, T = 500 K till T = 300 K
5
0

z (Å)
−5
−10
−15

15
10 20 configurations, T = 0.1 K
5
0

z (Å)
−5
−10
−15
0 100 200 300 400 500 600
x (Å)

Figure 3.1: Representative configurations of the sample LS during the same simulation (with a compression of 8%). Starting at the
3.1. PERIODICITY DEPENDENCE ON COMPRESSION AND TEMPERATURE

top, one can see one configuration at T = 700 K, 21 configurations at T = 700 K, 20 configurations during the cooling down from
T = 500 K till T = 300 K and 21 configurations at T = 0.1 K.
23
24
15 2.5E+05
3%, T = 500 K till T = 300 K 8%, T = 500 K till T = 300 K T = 500 K till T = 300 K
10
2.0E+05 8%
3%
5
1.5E+05
0

z (Å)
1.0E+05
−5

5.0E+04

Fourier transform per particle


−10

−15 0.0E+00
15 2.5E+05
3%, T = 0.1 K 8%, T = 0.1 K T = 0.1 K
10 2.0E+05 8%
3%
5
1.5E+05
0

z (Å)
1.0E+05
−5

5.0E+04
Fourier transform per particle

−10

−15 0.0E+00
0 100 200 300 400 500 600 0 100 200 300 400 500 600 0 0.04 0.08 0.12
x (Å) x (Å) qx (Å−1)
CHAPTER 3. RESULTS: DEFORMATION UNDER COMPRESSION

Figure 3.2: Representative configurations of the smallest sample (LS ) at different temperatures (20 configurations from T = 500 K
till T = 300 K up and 20 configurations at T = 0.1 K down) and different compressions (3% left and 8% in the middle). Notice that
the amplitude of the wrinkles increases with compression and that for 8% two periods seem to be present. On the right-hand side the
corresponding Fourier transforms per particle are drawn, with a plus (+) for 8% and a cross (×) for 3%.
3.2. PERIODICITY DEPENDENCE ON THE LENGTH 25

LL LM LS
qx (Å) P (Å) qx (Å) P (Å) qx (Å) P (Å)
0.04379 143.5 0.06256 100.4
T = 500 K till T = 300 K 0.06934 90.6 0.06778 92.7 0.05474 114.8
0.08029 78.3 0.08863 70.9
0.04379 143.5
T = 0.1 K 0.06934 90.6 0.06256 100.4 0.05474 114.8
0.08029 78.3 0.06778 92.7

Table 3.2: The x-component of the wave vectors where the Fourier transform peaks for the
different samples LL , LM and LS at different temperatures at 8% compression. For some
lengths and temperatures more than one peak is not negligible. In that case, the values for
the highest peaks are given, where the most important value is underlined. Moreover, this
table shows the corresponding periods P for the listed qx -values.

3.2 Periodicity dependence on the length


From fig. 3.3 – 3.6 we can learn some things about how the periodicity depends on the length
of the sample. Therefore, we use a constant temperature range (i.e. T = 500 K till T = 300 K
and T = 0.1 K) and compression (i.e. 8% and 3%) while varying the length (LL , LM and LS
in decreasing order, with specifications again in table 3.1 on page 21).
First of all, we see that at low temperatures (T = 0.1 K) the shape of the sample is pretty
stable, something we had already noted in fig. 3.1. Moreover, the spreading of the wrinkles is
comparable for different lengths at the same compression and temperature, so it seems that
the time needed to approach equilibrium under compression is independent of the length of
the sample.
When comparing the three samples at both high (T = 500 K till T = 300 K) and low
(T = 0.1 K) temperature, one can see that they have all approximately the same period.
However, there is no clear repeated pattern visible. To get more information, we have a look
at the FT. The FT of the samples of different lengths shows only for LS a clear double peak as
discussed in section 3.1. The largest peaks in the FT of the three samples at 8% compression
are listed in table 3.2, from which it is clear that for LM and LL there are several peaks.
Again this result makes the description of the wrinkle formation in terms of a bifurcation
process less straightforward, dependent on the length of the sample.
What is very clear though, is the period doubling effect at 8%, that stays at low temper-
ature only for a longer sample. There are different reasons why such a phenomenon would
occur. The first one is that the system is not close enough to equilibrium and as a consequence
the double period is still disappearing. This is a plausible opportunity, because the energy is
still lowering (as can be seen in fig. 3.7) and the double period effect is clearly much less than
at T = 300 K.
The second option is that a long sample at a compression high enough still shows some period
doubling. This can be underpinned by the very small variation of z as a function of Monte
Carlo time during this last part of the simulation. To decide whether the former or the latter
is true, further investigation is required.
26 CHAPTER 3. RESULTS: DEFORMATION UNDER COMPRESSION

15
10 LL
5
z (Å)

0
−5
−10
−15

15
10 LM
5
z (Å)

0
−5
−10
−15

15
10 LS
5
z (Å)

0
−5
−10
−15
0 200 400 600 800 1000 1200 1400
x (Å)
(a) The shape of 20 configurations for different lengths.

120000
L
M
S

100000
Fourier transform per particle

80000

60000

40000

20000

0
0 0.02 0.04 0.06 0.08 0.1 0.12 0.14
qx (Å−1)

(b) The Fourier transforms of the samples in fig. 3.3a

Figure 3.3: (a) The shape of the configurations and (b) the Fourier transform of samples of
different lengths at temperatures of T = 500 K till T = 300 K under a compression of 8%.
3.2. PERIODICITY DEPENDENCE ON THE LENGTH 27

15
10 LL
5
z (Å)

0
−5
−10
−15

15
10 LM
5
z (Å)

0
−5
−10
−15

15
10 LS
5
z (Å)

0
−5
−10
−15
0 200 400 600 800 1000 1200 1400
x (Å)
(a) The shape of 21 configurations for different lengths.

250000
L
M
S

200000
Fourier transform per particle

150000

100000

50000

0
0 0.02 0.04 0.06 0.08 0.1 0.12 0.14
qx (Å−1)

(b) The Fourier transforms of the samples in fig. 3.4a

Figure 3.4: (a) The shape of the configurations and (b) the Fourier transform of samples of
different lengths at a temperature of T = 0.1 K under a compression of 8%.
28 CHAPTER 3. RESULTS: DEFORMATION UNDER COMPRESSION

15
10 LL
5
z (Å)

0
−5
−10
−15

15
10 LM
5
z (Å)

0
−5
−10
−15

15
10 LS
5
z (Å)

0
−5
−10
−15
0 200 400 600 800 1000 1200 1400 1600
x (Å)
(a) The shape of 20 configurations for different lengths.

45000
L
M
40000 S

35000
Fourier transform per particle

30000

25000

20000

15000

10000

5000

0
0 0.02 0.04 0.06 0.08 0.1 0.12 0.14
qx (Å−1)

(b) The Fourier transforms of the samples in fig. 3.5a

Figure 3.5: (a) The shape of the configurations and (b) the Fourier transform of samples of
different lengths at temperatures of T = 500 K till T = 300 K under a compression of 3%.
3.2. PERIODICITY DEPENDENCE ON THE LENGTH 29

15
10 LL
5
z (Å)

0
−5
−10
−15

15
10 LM
5
z (Å)

0
−5
−10
−15

15
10 LS
5
z (Å)

0
−5
−10
−15
0 200 400 600 800 1000 1200 1400 1600
x (Å)
(a) The shape of 21 configurations for different lengths.

60000
L
M
S

50000
Fourier transform per particle

40000

30000

20000

10000

0
0 0.02 0.04 0.06 0.08 0.1 0.12 0.14
qx (Å−1)

(b) The Fourier transforms of the samples in fig. 3.6a

Figure 3.6: (a) The shape of the configurations and (b) the Fourier transform of samples of
different lengths at a temperature of T = 0.1 K under a compression of 3%.
30 CHAPTER 3. RESULTS: DEFORMATION UNDER COMPRESSION

-7.34839

-7.3484

-7.34841

-7.34842
E per particle (eV)

-7.34843

-7.34844

-7.34845

-7.34846

-7.34847

-7.34848

-7.34849
4.6e+06 4.65e+06 4.7e+06 4.75e+06 4.8e+06 4.85e+06 4.9e+06 4.95e+06 5e+06
tMC

Figure 3.7: The energy development per particle at low temperature (T = 0.1 K) for the long
sample LL . The steps are due to the limited accuracy up to which the program calculates
the energy.
3.3. THE EFFECT OF PERIODIC BOUNDARY CONDITIONS 31

3.3 The effect of periodic boundary conditions


In all the previous simulations, periodic boundary conditions were imposed. However, it is
not clear whether this effects the simulation and if, in which way. Therefore, also a simulation
without periodic boundary conditions in the x-direction has been done. By mistake, in this
case only a simulation with a compression of 14% has been done, which is not included in the
previous research, because this gave no clear results and smaller compressions seemed to be
better.
Despite these disadvantages, some important things can be seen in fig. 3.8, which shows
configurations of the sample LS under a compression of 8% at different temperatures. The
most important thing is that at the boundaries, the maxima of the out-of-plane displacements
stays the same as in the middle, whereas for the situations with periodic boundary conditions
(for instance in fig. 3.1 on page 23) these are smaller. Moreover, the outer left and right part
of the sample have a bigger out-of-plane displacement.
The last remarkable thing is that the superposition of different periods seems less evident.
That, again, is a reason for further investigation.
32
15
10 1 configuration, T = 700 K
5
0

z (Å)
−5
−10
−15

15
10 20 configurations, T = 700 K
5
0

z (Å)
−5
−10
−15

15
10 21 configurations, T = 500 K till T = 300 K
5
0

z (Å)
−5
−10
−15

15
10 20 configurations, T = 0.1 K
5
0

z (Å)
−5
−10
−15
0 100 200 300 400 500
x (Å)

Figure 3.8: Configurations of sample LS , under 14% compression without periodic boundary conditions. Starting at the top, one can
see 1 configuration at T = 700 K, 21 configurations at T = 700 K, 20 configurations cooling down from T = 500 K till T = 300 K and
21 configurations at T = 0.1 K.
CHAPTER 3. RESULTS: DEFORMATION UNDER COMPRESSION
Chapter 4

Conclusions

We have looked at the formation of wrinkles under compression, for which we saw stable
shapes appear. We have analysed in terms of the FT and found the existence of more than
one period, qualitatively supporting the bifurcation picture of Brau et al. [1] in fig. 1 and
fig. 2 on page 6. This is particularly for the small sample at high compression (8%). All other
situations seem more complex. Also for larger samples (LM and LL ) it seems to be more
complex than just two periods that are present at higher compression.
By analysing all results I have realised that possibly the compressions I have used were
already too heigh to see the first period doubling. In that case we are looking at higher
bifurcations, which creates a more complex situation. Also the effect of periodic boundary
conditions seems to be important. Nevertheless, the present results do show some interesting
behaviour of graphene under compression and support the steps to follow for a more complete
understanding of this phenomenon.

My internship at the department of Theory of Condensed Matter was a very interesting


period for me, because I have learnt a lot of things concerning research. First of all, I learned
to do calculations and predictions on a real system, which is certainly different than in the
courses I had before. Moreover, this internship gave me a nice inside in the scientific world
of research. I learned for instance what kind of effort and time it takes to finally publish a
paper and how people work together to achieve this. Finally, the internship made me more
confident, because I initially thought I would not be able to do a research like this.

33
34 CHAPTER 4. CONCLUSIONS
Chapter 5

Outlook

Although I am finishing this research with the writing of this thesis, the research on this
specific topic is not over at all. This thesis is just a beginning and there are lots of things
that has to be deepened. Therefore, I am very glad that Prof. Dr. A. Fasolino is intending
to do some further research on this subject. Some projects that can still be done with this
thesis in mind are listed below.

Method
There are some project I suggest that can be done involving the method of stabilizing the
final sample as quick as possible.

ˆ The first project is about the method of wavemoves that failed in an earlier state. I still
believe that it must be possible to reach final stable configurations at a shorter time
using this method. However, it is important to choose the right maximum qx -value (as
to deal with only the largest wavelengths) and to have a box in the z-direction that is
large enough. When all this is correctly prepared, it could be interesting to look at the
dependence of the final configurations (so at T = 0.1 K) on the input variables.

ˆ It seems that the final shape depends on the initial periods if we take a superposition
of some sines (of different period) as initial sample. Although we do not want such a
behaviour for this project, it can be interesting for other research goals to know what
this dependence looks like.

ˆ As showed shortly in section 3.3, the behaviour of graphene under compression seems
to be different when we impose periodic boundary conditions as if we do not. In this
research project, all have been done with periodic boundary conditions. However, it
might be better to run all simulations without periodic boundary conditions.

ˆ It seems that in my research, the smallest compression was already too high, resulting
in a complex superposition of many periods. However, to see the first bifurcation,
smaller compressions should be imposed. Finding the right compression to make the
first bifurcation visible is a very important subject in research following this thesis.

35
36 CHAPTER 5. OUTLOOK

Dependence on temperature and compression


The first dependences we have been looking at, are the dependences on temperature and
compression. In this part of the subject, still a lot can be done to deepen and clarify things.

ˆ With my data, it was not possible to get a nice picture of how the periods depend on
the compression. However, it would be very interesting to see whether the appearance
and disappearance of certain peaks is visible.

ˆ Because there are clearly two periods visible for the sample LS under a compression of
8% at T = 300 K (both in the shape of the configurations and in the FT), where only
one is really clear at T = 0.1 K, it is interesting to find the reason for this phenomenon.
In this thesis, it is already suggested that this could be caused by the temperature
dependence of the bending rigidity κ of graphene (fig. 1.4). However, there is a lot
more work to be done in order to prove or reject this hypothesis. In addition to this,
it is useful to show whether the double period pattern completely disappears at low
temperature or only lowers, because for longer samples this is not the case.

Dependence on length
As for the length of the sample, only for the smallest sample a bifurcation pattern was
observed. However, the longer samples LM and LL show many peaks in the FT that are
more or less equal, which points at a more complex situation. Further research on this topic
is needed, and it may be a good idea to start with lowering the compression and compare the
different lengths again.
Bibliography

[1] F. Brau, H. Vandeparre, A. Sabbah, C. Poulard, A. Boudaoud, and P. Damman.


Multiple-length-scale elastic instability mimics parametric resonance of nonlinear oscil-
lators. Nature Phys., 7:56–60, 2011.

[2] Marco Stevens. Wrinkled patterns in graphene. Bachelor’s thesis, Radboud University
Nijmegen, 2011.

[3] Yong P. Chen and Qingkai Yu. Nanomaterials: Graphene rolls off the press. Nature
Nanotech., 5:559–560, 2010.

[4] A.K. Geim and K.S. Novoselov. The rise of graphene. Nature Mater., 6:183–191, 2007.

[5] H. P. Boehm, A. Clauss, G. O. Fischer, and U. Hofmann. Das adsorptionsverhalten sehr


dünner kohlenstoff-folien. Zeitschrift für anorganische und allgemeine Chemie, 316:119–
127, 1962.

[6] Nobelprize.org. The nobel prize in physics. http://www.nobelprize.org/nobel_


prizes/physics/, September 2012.

[7] A. Fasolino, J.H. Los, and M.I. Katsnelson. Intrinsic ripples in graphene. Nature Mater.,
6:858–861, 2007.

[8] Jan H. Los, Luca M. Ghiringhelli, Evert Jan Meijer, and A. Fasolino. Improved long-
range reactive bond-order potential for carbon. i. construction. Phys. Rev. B, 72:214102,
2005.

[9] M.I. Katsnelson. Graphene, carbon in two dimensions. Cambridge University Press,
2012.

[10] K.V. Zakharchenko, J.H. Los, M.I. Katsnelson, and A. Fasolino. Atomistic simulations of
structural and thermodynamic properties of bilayer graphene. Phys. Rev. B, 81:235439,
2010.

[11] R. Bowley and M. Sanchez. Introductory Statistical Mechanics. Oxford University Press,
2002.

[12] M.E. Tuckerman. Statistical Mechanics: Theory and Molecular Simulation. Oxford
University Press, 2010.

[13] P.L. de Vries. A first course in Computational Physics. John Wiley & Sons, 1993.

37

Vous aimerez peut-être aussi