Upload
others
View
1
Download
0
Embed Size (px)
Citation preview
Physics of a Limit-Periodic Structureby
Zongjin Qian
Department of PhysicsDuke University
Date:Approved:
Joshua Socolar, Supervisor
Henry Greenside
Stephen Teitsworth
Thesis submitted in fulfillment of the requirements for Graduation with Distinctionin the degree of Bachelor of Science in the Department of Physics
in the Undergraduate School of Duke University2013
Copyright © 2013 by Zongjin QianAll rights reserved except the rights granted by the
Creative Commons Attribution-Noncommercial Licence
Abstract
This study concerns the properties of physical systems that can spontaneously form
non-periodic structures. It focuses on the Socolar-Taylor tiling model in which a
single space-filling prototile forces a limit-periodic pattern–a state made up of the
union of infinite levels of periodic structures with ever-increasing sizes–through local
interaction rules. A two-dimensional lattice model, which possesses the Socolar-
Taylor tiling as its ground state is constructed. It is known that during a slow
quench from an initial high-temperature, disordered phase, the ground state of the
model emerges through an infinite sequence of phase transitions. As temperature is
decreased, sublattices with periodic structures of increasing lattice constants become
ordered. In this study, we construct a theory based on one-dimensional Ising model to
explain the time scales required for equilibrium to be reached at a given temperature
by sublattices of increasing lattice constants. We observe a discrepancy in the scaling
behavior predicted by our theory and obtained from simulation of the tiling model,
which is likely due to finite size effect of the tiling model. We find that during a rapid
quench, the energy barriers created by competing domain walls cause the system to
fall out of equilibrium. Two types of domain wall with different physical structures
and energy costs are found in the system. The associated energetics of each type
of domain wall is discussed, and a particular type of domain wall is identified as
responsible for slowing down the ordering of the tiling system.
iii
Contents
Abstract iii
List of Tables vi
List of Figures vii
Acknowledgements ix
1 Introduction 1
1.1 Quasicrystals: Systems With Ordered Non-Periodic Structures . . . . 3
1.2 Tiling Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.3 Behavior in a Rapid Quench . . . . . . . . . . . . . . . . . . . . . . . 8
1.4 Questions of Study . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
1.5 Outline . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
2 Previous Findings 11
2.1 Order Parameter . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
2.2 Slow Quench . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
2.3 Scaling Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
3 Relaxation Time 18
3.1 Scaling Relation for Relaxation Time . . . . . . . . . . . . . . . . . . 20
3.2 tn{tn´1 by MC Simulation of Tiling Model . . . . . . . . . . . . . . . 21
3.3 Theory Prediction of tI;n{tI;n´1 . . . . . . . . . . . . . . . . . . . . . 25
3.3.1 Theory Calculation for τ2{τ1 . . . . . . . . . . . . . . . . . . . 28
iv
3.3.2 τn{τn´1 Ratios by MC Simulation of 1D Ising Systems . . . . . 35
4 Domain Dynamics 38
4.1 Domain Walls . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
4.2 Energy Costs of Infinite Domain Walls . . . . . . . . . . . . . . . . . 45
4.3 Domain Wall Energetics . . . . . . . . . . . . . . . . . . . . . . . . . 51
5 Conclusion 58
Bibliography 60
v
List of Tables
3.1 The ratios between the relaxation constants of different levels of struc-ture of the tiling model compared with theoretical predictions. . . . . 35
vi
List of Figures
1.1 A section of a perfect rhombic Penrose tiling. . . . . . . . . . . . . . 2
1.2 The prototile and its mirror image with color matching rules. . . . . . 7
2.1 The rules for assigning staggered tetrahedral spin vector. . . . . . . . 12
2.2 Order vs. temperature during a slow quench. . . . . . . . . . . . . . . 15
2.3 Scaling collapse of Fig. 2.2 data. . . . . . . . . . . . . . . . . . . . . . 17
3.1 The initial configuration for MC simulation of level 3. . . . . . . . . . 23
3.2 Φn vs. T . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
3.3 Φn vs. scaled T for n “ 1, 2, 3. . . . . . . . . . . . . . . . . . . . . . . 25
3.4 A diagram illustrating Ising systems at level 3. . . . . . . . . . . . . . 27
3.5 The configurations for a level 1 Ising System. . . . . . . . . . . . . . . 28
3.6 The configurations for a level 2 Ising System. . . . . . . . . . . . . . . 30
4.1 Domain walls present after a rapid quench. . . . . . . . . . . . . . . . 40
4.2 Configurations of sitting at T “ 0.6 for 69 ˆ 104 mcs. . . . . . . . . . 42
4.3 The first type of domain wall. . . . . . . . . . . . . . . . . . . . . . . 43
4.4 The second type of domain wall. . . . . . . . . . . . . . . . . . . . . . 44
4.5 Number of black stripe mismatches vs y for horizontal domain walls. 46
4.6 Number of color stripe mismatches vs y for horizontal domain walls. . 46
4.7 A horizontal domain wall of type II spanning a 64 ˆ 64 lattice withminimum mismatches. . . . . . . . . . . . . . . . . . . . . . . . . . . 47
4.8 Portion of a horizontal domain wall of type II with minimummismatches. 48
vii
4.9 Number of black stripe mismatches vs x for slanted domain walls. . . 49
4.10 Number of color stripe mismatches vs x for slanted domain walls. . . 50
4.11 A slanted domain wall of type I spanning a 64 ˆ 64 lattice with lownumber of mismatches. . . . . . . . . . . . . . . . . . . . . . . . . . . 51
4.12 Portion of a slanted domain wall of type I with low number of mis-matches. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52
4.13 A configuration that gets stuck in ordering due the presence of aslanted domain wall of type I. . . . . . . . . . . . . . . . . . . . . . . 53
4.14 Portion of a slanted domain wall of type I from the configuration inFig. 4.13 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
4.15 Portion of a slanted domain wall of type I with no mismatches in blackstripe. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
4.16 Configurations of a sudden quench to T “ 0.6 and sitting for 5 ˆ 105
mcs followed by sitting at T “ 1.0 for 5 ˆ 104 mcs. . . . . . . . . . . 56
viii
Acknowledgements
I would like to acknowledge and extend my heartfelt gratitude to the following per-
sons who have made the completion of this thesis possible. My deepest gratitude goes
to Dr. Joshua Socolar, for giving me the opportunity to do research with him, and for
being abundantly helpful and offering invaluable assistance, support and guidance
along the way. Special thanks to Dr. Henry Greenside, for being a wonderful DUS,
and for his vital encouragement and support throughout my undergraduate study.
I would like to convey thanks to the department of physics for providing excellent
computing facilities, and to all faculty members and staff who extended help and
inspiration in one way or another. I wish to express my love and gratitude to my
beloved families and friends, for their understanding and endless love, through the
duration of my studies.
This work has been funded by a summer research fellowship from Triangle Materials
Research Science and Engineering Center (MRSEC).
ix
1
Introduction
This thesis addresses the properties of materials that can spontaneously form or-
dered non-periodic structures. Such structures are realized in several real physical
systems, most notably, in quasicrystals. Quasicrystals, short for quasiperiodic crys-
tals, are very different from the usual crystals as their structures can exhibit certain
symmetries such as the five-fold symmetry which are strictly forbidden in normal
crystalline structures. These unusual symmetries explain many of quasicrystals’ elec-
tronic, elastic and frictional properties. Since the date of their discovery, hundreds of
quasicrystalline samples have been made in laboratories, and technologically interest-
ing applications such as quasicrystalline coating on non-sticking, corrosion-resistant
cookware have been made. Today the structures of quasicrystals, their physical prop-
erties, and their spontaneous formation remain subjects of intense study.
In this thesis, I study the spontaneous formation of ordered non-periodic structures
by focusing on a recently discovered tiling model [1]. The idea of thinking ordered
non-periodic structures such as those displayed in quasicrystals as tiling models is
not novel. In fact, when the theoretical concept of quasicrystals was first conceived,
1
the initial inspiration was the two-dimensional Penrose tiling model [2] shown in
Fig. 1.1. Penrose had identified a pair of tiles that can fill the two-dimensional space
non-periodically, forming a self-similar patter full of five-fold symmetries. Several
theorists independently proposed the idea that the structures exhibited in Penrose
tiling may have analogies in solids [3–5].
Figure 1.1: A section of a perfect rhombic Penrose tiling. [6]
In this thesis, we focus on the Socolar-Taylor tiling which presents a single space-
filling prototile that forces a limit-periodic pattern (a state made up of the union of
infinite levels of periodic structures with ever-increasing sizes [7]) through local in-
teraction rules. This model is the first known example that a single two-dimensional
prototile can fill the entire two-dimensional Euclidean space with no overlap to form
a non-periodic pattern. Because of its simplicity, it has the significance of providing
a template for materials that can self-assemble into non-periodic structures using
just a single building block. It has been suggested that it may be possible for solid
state materials, colloidal systems, and collections of macroscopic units to achieve
structures displayed in the tiling model [8]. The allowed global tiling in the model
2
can be mapped to the states of a statistical mechanics model. Studying the ordering
dynamics of such a model in forming the limit-periodic phase would allow us to gain
insights into the unusual and remarkable properties of the tiling structures if they
were realized in real physical systems.
1.1 Quasicrystals: Systems With Ordered Non-Periodic Structures
For centuries, rigorous mathematical theorems dictated that crystals are ordered
solids with periodic structures that can be characterized by long-range translational
and rotational symmetries. It was commonly thought that solids come in two forms,
ordered and disordered. Ordered implies periodicity, while disordered solids describe
amorphous materials such as glass. It was well established by the classic theorems
of Schoenflies and Fyodorov proven in the 19th century that five-, seven- and any
higher-fold symmetry in two dimensions and icosahedral symmetry in three dimen-
sions are incompatible with periodicity and thus forbidden in crystals. Thus it is
not surprising that when Dan Shechtman et al. [9] announced the discovery of an
aluminum–manganese alloy with the forbidden icosahedral symmetry–the most fa-
mous forbidden symmetry in crystals–in 1984, the overall reaction of the scientific
community was strong skepticism. In a remarkable coincidence, the theoretical con-
cept of an ordered phase of matter that escapes periodicity and possesses long-range
icosahedral symmetry was proposed around the same time by Dov Levine and Paul
Steinhardt [5]. As an example, a three-dimensional quasicrystal can be constructed
using polyhedral units analogous to the Penrose tiles. The new hypothetical ordered
phase of matter was named quasicrystal, short for quasiperiodic crystal. The newly
proposed concept of quasicrystal accounts for the diffraction pattern of the type that
Shechtman observed. Furthermore, quasiperiodicity opens up an infinite possibility
of ordered solids with forbidden symmetries to crystals.
3
However the idea of a quasicrystal as a real physical phase of matter was not read-
ily accepted initially. First, X-ray studies [10] revealed that the diffraction peaks of
Shechtman’s icosahedral aluminum–manganese phase are not truly point-like, as pre-
dicted by the quasicrystal theory. Instead the diffraction peaks have experimentally
resolved finite widths. In addition, Shechtman’s icosahedral aluminum–manganese
phase was very unstable. Annealing could not sharpen the diffraction peaks and leads
to crystallization. Most critics also thought that the quasicrystal phase is physically
impossible. It was argued that in a quasiperiodic structure, no two atoms or clusters
could occupy identical positions, so they cannot self-organize into a perfect quasicrys-
talline structure. However in 1987, the synthesis of the first bona fide quasicrystal
sample in laboratory by Tsai et al [11] firmly established quasicrystal as a real phys-
ical phase of matter. Tsai et al. discovered a quasicrystal phase (Al63Cu24Fe13)
that exhibits sharp diffraction peaks and long-range icosahedral symmetry. Later
on, hundreds of high-quality quasicrystal samples with different forbidden symme-
tries have been identified.
Until very recently, all known samples of quasicrystals have been synthetic. It has
long remained a question of debate that whether quasicrystals could be metastable
states of matter and many contend that quasicrystals are too complicated to be sta-
ble and that all ground states of matter are crystalline. However, Luca Bindi and
Paul Steinhardt [12] recently found a natural quasicrystal in a meteorite whose com-
position is measured to be Al63Cu24Fe13, the same composition as the first bona fide
quasicrystal sample discovered by Tsai et al. The discovery of a natural quasicrys-
tal gives strong support for the argument that quasicrystals can be as energetically
stable as crystals, and can be formed under natural conditions, as predicted by the
original quasicrystal theory.
4
1.2 Tiling Model
The theory of tiling concerns using non-overlapping copies of a set of tiles to cover
the entire space of the Euclidean plane or some other geometric settings. It is not
hard to find a set of tiles that admits a periodic pattern characterized by transla-
tional and rotational symmetry. It is also not hard to find a set of tiles that admits
a nonperiodic pattern, but often the tiles employed could also be used to construct
periodic patterns. However, it is much more difficult to find a set of tile-types or
“prototiles” that can only be used to construct a nonperiodic pattern. Such sets are
called “aperiodic”. There are few known examples of aperiodic tilings, perhaps the
most famous one is the Penrose tiling [2]. In 2010, Joshua Socolar and Joan Tay-
lor [1] discovered a single prototile that covers the entire Euclidean plane only in a
nonperiodic way with a set of matching conditions. This is the first known aperiodic
tiling set with matching conditions made of just a single prototile in two-dimensional
Euclidean space. Historically, the discovery of metallic alloys which shares the essen-
tial structure of the Penrose tiling [9,11] proved that ordered nonperiodic structures
could form spontaneously. In material physics, the tiles may represent large building
blocks or clusters of atoms, and the matching rules which determine how tiles fit
together may represent the energetics of a physical system [7]. Hence a key moti-
vation for Socolar and Taylor’s search for a single prototile is the hope that their
tiling model could help to predict the properties of materials that can self-assemble
into ordered nonperiodic structures using just the same cluster of atoms or a uniform
building block.
The Socolar-Taylor tiling model presents a single space-filling prototile which is a
regular hexagon decorated with markings. A version of the prototile and its mirror
5
image is shown in Fig. 1.2(a). Each tile has a set of black stripe and colored stripe
decorations, and can be placed in one of its twelve possible orientations obtainable
by rotations of π{3 and reflection. Notice that the set of colored stripes are the set
of black stripes rotated by π{2 and scaled by a factor of?3. The model specifies a
set of rules to determine how nearest and next-nearest tiles should be oriented based
on the markings on the tiles. The rules for placing the tiles are as follows:
1. Nearest tiles must form continuous black stripes;
2. Next-nearest tiles that sit at the opposite ends of a tile edge must form con-
tinuous colored stripes.
The rules are illustrated in Fig. 1.2(b). A Hamiltonian can be defined for the tiling
model based on the“matching rules” for placing the adjacent tiles and next-nearest
neighbor tiles on a close-packed lattice. For the nearest neighbor pairs, energy zero
is assigned if the black stripes are continuous across the boundary of the tiles and
energy �1 otherwise. For the next-nearest neighbor pairs, energy zero is assigned if
the colored stripes at the closest corners are connected and energy �2 otherwise. It
has been shown that the ground state is a zero-energy structure displaying an infinite
hierarchy of triangular lattices with ever increasing lattice constant [8]. A portion of
the infinite ground state is shown in Fig 1.2(c). Each level of periodic sublattice is
scaled by a factor of 2 compared to its previous level. The ground state is however
non-periodic as there is no single sublattice with the largest lattice constant. The
non-periodicity is a direct consequence of the local “matching rules”. The rules en-
forces a simple periodic lattice at the smallest scale using a subset of the markings
of the tiles. The rules then effectively operate in the same manner at larger scales
to generate the same periodic lattices with arbitrarily large lattice constants [8].
A closer look at the infinite ground state of the model reveals that the black triangular
6
(a)
(b)
(c)
Figure 1.2: The prototile and its mirror image with color matching rules. (a) Thetwo tiles are reflected about a vertical axis. (b) For zero energy, the nearest neighbortiles must form continuous black stripes and the next-nearest neighbor tiles mustform continuous color stripes. The arrow indicates where the color stripes from twonext-nearest neighbor tiles are joined to form a continuous stripe. (c) A portion ofthe infinite zero-energy ground state.
sublattices of increasing lattice constants play a key role in explaining the behavior
of the tiling model. The sublattice made of the smallest triangles shall be referred as
the level 1 structure from now on, the sublattice made of the next smallest triangles
as the level 2 structure, etc. The edge of a black triangle at level n crosses kn ´ 1
tiles, where kn ” 2n´1. At each level, the black triangles form a periodic pattern in
which the lines connecting the centers of all the triangles form a honeycomb lattice.
Further, each sublattice is an exact replica of the previous one, scaled by a factor by 2.
7
1.3 Behavior in a Rapid Quench
A quenching procedure is a process in which one starts a system in an initial disor-
dered, high-temperature phase, and lower the temperature to monitor how the system
evolves toward the final state. For the tiling model we study, a quenching procedure
is very useful in investigating how the ordered limit-periodic ground state is formed
dynamically. Some preliminary evidence has been found that a rapid quench of the
tiling system leads to disordered states [8]. It has been shown that when the tiling
models orders during a quench, there is a hierarchy of ordering for sublattices of
increasing lattice constants, i.e. as the temperature is lowered, the level 1 structure
orders before the level 2 structure, which orders before the level 3 structure, etc.
However, if the quenching rate is too high, the level 1 could not become reasonably
well formed before the level 2 structure starts to order. As a result neither the level
1 nor the level 2 structure would become well ordered. Ref. [8] suggests that a rapid
quench would lead the system to some disordered final states due to the competition
between two or more levels of structures. The kinetic barriers created by a level of
structure when it fails to order would cause simultaneous frustration of the ordering
of the subsequent levels of structure. However, the reason that the system falls out of
equilibrium and ends up in some disordered phase is far from clear from any previous
findings.
1.4 Questions of Study
Previous work [8] has focused on the study of the equilibrium properties of the tiling
model, where the goal is to characterize the ordered phases that arise below the
critical temperatures in the tiling system. The usefulness of studying the tiling model
in the equilibrium framework is limited in understanding the physical properties
of the limit-periodic structures exhibited in the tiling model, especially if one is
8
interested in finding in nature or synthesizing materials that can spontaneously form
limit-periodic structures. A natural question to ask is how slowly one has to lower the
temperature of the system during a quench in a order to have all levels of structure
fully ordered so that the limit-periodic ground state could be reached? This is the
first question addressed in this thesis. The answer to this question tells the minimum
time scale required to achieve a certain number of perfectly ordered levels of structure
in the tiling model. We also know that a rapid quench results in disordered phases.
However, the origin of such disordered states are poorly understood. A second set
of questions addressed by this thesis are: what causes the system to fall out of
equilibrium if the temperature is lowered too fast? Once the system gets into a
disordered phase, how high is the energy barrier to get out and go to an equilibrium
state. The answer to these questions give insights on the nonequilibrium dynamics
of the tiling model.
1.5 Outline
An outline of the remainder of this chapter is as follows. Chapter 2 discusses in
detail some important previous findings on the properties of the tiling model. The
definition of an order parameter to quantify the ordering of the system is presented.
The behavior of the system in a slow quench is discussed and a scaling law that
maps the behavior of the tiling system from one level of sublattice to another is
introduced. In Chapter 3, we construct a theory based on one-dimensional Ising
model to find a scaling relationship on the time scales required for equilibrium to be
reached at a given temperature by sublattices of increasing lattice constants. The
scaling relationship is found through numerical simulation of the tiling model and
predicted using a theoretical model. Results suggest that there is a discrepancy be-
tween our theoretical prediction and the actual time scales. Chapter 4 discusses the
nonequilibrium dynamics of the tiling model. The formation of competing domains is
9
identified as the cause for the system’s ordering getting frozen during a rapid quench.
The two types of domain walls present in the system are examined. This is followed
by analyzing the domain wall energetics associated with each type of domain wall.
Finally, we identify a special type of domain wall as responsible for slowing down the
system’s recovery from a disordered state to an equilibrium state.
10
2
Previous Findings
2.1 Order Parameter
Defining a proper order parameter is an important first step towards understanding
the intricate equilibrium and dynamical properties of the tiling model. As we lower
the temperature of the tiling system slowly, the level 1 triangles form following a
second-order phase transition. As the temperature of the tiling system is further
lowered, subsequent levels of structure form following a hierarchy of phase transi-
tions. A family of vector order parameters Φn which characterizes the ordering of
the tiling system has been found [8]. There is one order parameter associated with
each level of structure. The level n order parameter saturates to 1 when the level
n triangles are perfectly formed. The definition of an order parameter for the tiling
system is described here.
For the level 1 structure, each tile can be assigned to one of four different sublattices,
denoted as A, B, C, D as shown in Fig. 2.1. Notice that a quarter of the tiles did not
contribute to the formation of black triangles in the level 1 structure. Each of the
11
four different sublattices could constitute the non-contributing sublattice at level 1.
Fig. 2.1 shows an example in which the tiles sitting on sublattice A do not contribute
to the the level 1 black triangles. At level 2, if the level 1 structure is perfectly or-
dered, the corners of the level 2 black triangles must come from the non-contributing
sublattice of level 1. The system size for level 2 is thus effectively 1{4 of that of level
1. Note that the level 1 non-contributing sublattice is an exact copy of the original
lattice only that the distance between tiles are scaled by a factor of 2. At level 2,
the four sublattices can again be assigned for the level 2 tiles. A non-contributing
sublattice can also be defined and it gives the tiles to form the level 3 structure. Such
construction can be repeated ad infinitum so that the tiles of the level n structure
come from the non-contributing sublattice of level n ´ 1.
Figure 2.1: The pattern of the level 1 triangles formed when the non-contributingtiles are from those of the A sublattice and the rules for assigning spin vectors fortiles of different sublattices. For explanation of the dashed lines and grey bar, seetext. Adapted with permission from “Hierarchical Freezing in a Lattice Model” byT. W. Byington and J. E. S. Socolar, Physical Review Letters, vol. 108 (2012), pp.045701.
It has been shown the an order parameter can be explicitly defined for each level of
structure [8] by associating a “staggered tetrahedral spin” vector with each tile. The
rules for assigning the spin vector for each tile is illustrated by Fig. 2.1. See also
12
Ref. [8] for details. Each tile j at level n is assigned a spin vector σjn “ �eX , where
�eX is a unit vector pointing in one of the four corners of a reference tetrahedron
labeled “X”. The spin for each tile is determined by both its sublattice and the
orientation of the line connecting the two black triangle corners. Specifying the spin
vector for a tile does not completely determined the orientation of the tile. There
are four compatible orientations associated with a spin vector as shown in Fig. 2.1.
For a tile that has spin vector �eA, there are two possible locations of its long black
stripe as indicated by the two dashed grey lines, and there are two possible locations
of its long colored stripe as indicated by the grey diagonal arrow. When a perfectly
ordered level 1 structure has sublattice A as its non-contributing sublattice, all the
tiles on sublattice B, C, and D have spin vector �eA while the tiles on sublattice A
does not contribute to the ordering so they could have any of the other three spin
vectors. The situation is exactly the same for any subsequent level of structure. The
average spin for level n is simply �σ�n “ 1Nn
řj σ
jn, where Nn is the number of tiles
at level n and is determined by the relation Nn “ Nn´1{4 as discussed before. The
order parameter, Φn for level n can now be defined very easily. For each level of
structure, It is given by the maximum of the projection of the average spin along the
four tetrahedral spin vectors: Φn ” maxX
r�eX ¨ �σ�ns. The order parameter Φn goes to
its maximum value 1 when level n triangles become perfectly ordered. The tiles on
the non-contributing sublattice X could be in any orientations.
2.2 Slow Quench
The study of how the tiling system evolves from an initial random configuration at
finite temperatures reveals some remarkable properties of the model. Long range or-
der emerges in the tiling model during a slow quench from high to low temperatures
13
as a consequence of the local interactions of the tiles. This is an example of the phe-
nomenon known as “Emergence”, the hallmark of many self-organizing systems [13].
In many complex systems, there are multiple agents interacting dynamically in many
ways, following local updating rules and paying little attention to any higher level
instructions. In our tiling model, the nearest pairs and next-nearest pairs of tiles
play the role of the dynamically interacting agents. The systems become “emergent”
when the local interactions results in discernible macroscopic behaviors. As Murray
Gell-Mann once said: surface complexity arising out of deep simplicity.
Ref. [8] shows that during a slow quench from a high temperature, disordered phase,
the limit-periodic ground state emerges through an infinite sequence of second-order
phase transitions. Further, there is an hierarchy of phase transitions. As temperature
is decreased, sublattices with periodic structures of increasing lattice constant become
ordered. These results are reproduced here in this thesis as a confirmation and are
used to illustrate the physical properties of the tiling model. In our study, we look at
the special case �1 “ �2 “ � “ 1. Fig. 2.2 shows the behavior of Φn for n “ 1, 2, 3, 4 for
a slow quench. The simulations are done on a 64 ˆ 64 rhombus lattice with periodic
boundary conditions using the standard Metropolis algorithm. In each Monte Carlo
Step (mcs), each tile in the lattice has an equal probability of being chosen and N
random tile selections are made, where N “ 64 ˆ 64 is the total number of tiles
in the rhombus domain. For each tile selection, one of the twelve possible states is
randomly chosen as a possible move. The tile is then changed into the proposed state
according to the distribution:
ωiÑj “ min�1, e´∆EiÑj{T(
(2.1)
If the chosen move lowers the energy of the system, it is accepted with certainty,
14
otherwise, the move is accepted with a probability of e´∆EiÑj{T , where ∆EiÑj is the
change in energy of the proposed flip. The temperature T is lowered by δT “ 0.01
after every r “ 1.2 ˆ 106 mcs, where r controls the quenching rate. Fig. 2.2 shows a
clear sequence of phase transitions during a slow quench. Each subsequent level of
structure becomes well ordered only if its previous level is perfectly ordered.
�������������������������������������������������
�
�
�������������������
�����������������������������������������������������������������������������������������������������������������������������������
��������������������������������������������������������������������������������������������������������������������������
�
�������������������
�����������������������������������������������������������
�����������������������������������������������������������������������������������������������������������������������������������������������������
�
�����������
����������������������������������������
���������������������������������������������������������������������������������������������������������������������������������������������������������������
��
�
��
�������������������������������������
� n�1� n�2� n�3� n�4
0.0 0.5 1.0 1.5 2.00.0
0.2
0.4
0.6
0.8
1.0
Temperature T�Ε�kB�
�n
Figure 2.2: Behavior of the order parameter Φn vs. the temperature T during aslow quench. Simulations were performed on a 64ˆ64 rhombic domain for �1 “ �2 “ 1and r “ 1.2 ˆ 106.
2.3 Scaling Theory
The tiling model follows a hierarchy of phase transitions to become fully ordered
during a slow quench. For the level n structure to become fully ordered, it requires
the level n ´ 1 structure to be almost perfectly ordered before level n transition
occurs. A consequence is that assuming level n ´ 1 is perfectly ordered, the phase
15
transition of the level n is identical to that of the level n ´ 1 up to rescaling of the
temperature. Ref. [8] finds a relation between Tn and T1 such that the behavior of
the system at level n at temperature Tn is identical to that at level 1 at temperature
T1. It is given as:
tanh
ˆ�
2T1
˙“
„tanh
ˆ�
2Tn
˙kn
(2.2)
where kn ” 2n´1, as defined before. Or, equivalently, for all n,
tanh
ˆ�
2Tn
˙“
„tanh
ˆ�
2Tn`1
˙2
(2.3)
One immediate usefulness of this formula is that one can deduce the level n phase
transition temperature Tc;n for any n, given the level 1 transition temperature Tc;1.
As shown by Fig. 2.3, an excellent data collapse for n “ 1, 2, 3, 4 is obtained using
Eq.(2.2) by plotting ΦnpTnq as a function of T1pTnq. Note the system size for the
level 4 structure is only 8 ˆ 8, and the deviation of the level 4 data points for high
temperatures is due to the finite size effect. The deviated points represent projec-
tions of onto different tetrahedral spin vectors.
16
�������������������������������������������������
�
�
����������������������
�������������������������������������������������������������������������������������������������������������������������������
����������������
�
�������
����������������������������������������������������������������������
������
�
�
����
����������������������������������������������
��
�
��
������������������������������������ � n�1
� n�2� n�3� n�4
0.0 0.5 1.0 1.5 2.00.0
0.2
0.4
0.6
0.8
1.0
Temperature T�Ε�kB�
�n
Figure 2.3: Behavior of the order parameter Φn as a function of the rescaledtemperature T1pTnq during a slow quench. The deviation of the level 4 data pointsis due to the finite size of the system.
17
3
Relaxation Time
This chapter addresses the first question raised at the end of Sec. 1.4. The question
asks how slowly the temperature should be lowered during a slow quench in order for
the system to reach the limit-periodic ground state? It is a fact that the system goes
through a sequence of phase transitions before it reaches the ground state. Further,
the order with which the phase transitions happen follow a strict hierarchy. The level
n ´ 1 structures must be sufficiently ordered before the level n structures starts to
order in order for both levels to reach equilibrium. Based on these facts, a quenching
procedure that guarantees the well ordering of all levels of structure so that the
ground state emerges as the final state is as follows:
1. Start the system at a high temperature T ą Tc;1, where the tiles are completely
disordered; here Tc;1 is the transition temperature for the level 1 structure.
2. Instantaneously cool the system to a lower temperature T1 satisfying Tc;2 ăT1 ă Tc;1; the level 1 structure starts to equilibrate but the level 2 structure
remains disordered as T1 is above the level 2 transition temperature, Tc;2.
3. Wait enough time until the level 1 structure becomes fully ordered; At this
18
point, the level 2 structure remains completely disorientated.
4. Instantaneously cool the system to a lower temperature T “ T2pT1q, whichis the rescaled temperature of T1 given by the Rescaling Theory as stated
in Eq.(2.2). T2 satisfies Tc;3 ă T2 ă Tc;2 as a consequence of the Rescaling
Theory; the level 2 structure starts to equilibrate but the level 3 structure
remains disordered.
5. Repeat step 2 and 3 ad infinitum, each time instantaneously cool the system
to the temperature T “ TnpT1q, where Tn satisfies Tc;n`1 ă Tn ă Tc;n. wait for
the level n structure to reach equilibrium before lowering the temperature to
T “ Tn`1pT1).
Note Tn satisfying Tc;n`1 ă Tn ă Tc;n is a direct consequence of the Rescaling Theory
which says the physics of the level n structure at the T1 rescaled temperature, Tn is
exactly the same as that of the level 1 structure at T1; Since T1 satisfies Tc;2 ă T1 ăTc;1, Tn too should be in between the transition temperature of its corresponding
level of structure and the level above. The above procedure would produce a final
state of the limit-periodic phase. Given an infinite system, the waiting time required
to generate the desired final state would be infinite since there is no largest sublattice
in the tiling model which means the procedure has to be carried on forever. However
it is still meaningful to ask how long one has to wait to have a certain number,
say k levels of structure fully ordered. The minimum waiting time, tmin is given by
tmin “ řkn“1 tn, where tn is the minimum time required for level n structure to reach
equilibrium at temperature Tn, given that all previous n ´ 1 levels have been fully
ordered following our quenching procedure. Now the question we originally asked
could be sharpened as the following: What is the minimum time scale required
for levels of structure of increasing lattice constants to reach equilibrium at the
19
temperatures tTnukn“1? Answering this question effectively breaks the total dynamics
of the tiling model into just the individual dynamics of the different levels of structure.
3.1 Scaling Relation for Relaxation Time
To answer the above question, the key is the sublattice displayed in each level of
structure. Each level is made of the black and color triangles. The black triangles
in the level n structure has an edge that crosses kn ´ 1 tiles, where kn ” 2n´1 as
discussed in Sec. 1.2. The set of color triangles associated with the level n struc-
ture is just the set of black triangles rotated by π{2 and scaled by a factor of?3.
Referring to the configuration shown in Fig. 1.2, the level n structure is bound to
become fully ordered once the corners of the level n black triangles are locked in. At
that stage, the middle tiles on an edge of the black triangle can be any of the four
possible orientations corresponding to two locations of the long black stripe and two
locations of the long colored stripe. Their role is to mediate the interactions until
one corner of the triangle develops a correlation with the other corner. Once this
correlation is developed, the level n structure will always come to equilibrium in a
short time scale compared to the time required to develop correlation between the
corners of the level n triangles (The system does not exhibit hysteresis [8]).
If every level m ă n is perfectly ordered, it is expected that the level n dynamics is
identical to the level 1 dynamics in every aspect except the time required for correla-
tion to develop between the corners of the black triangles. The correlation time for a
certain level of structure is then determined by the edge length of its black triangles.
At level n, each edge of the triangle crosses kn ´1 tiles, with black stripes joining the
nearest neighbor tiles. There are altogether kn possible mismatches in black stripes
in each edge, hence kn nearest neighbor bonds between two corner tiles. Each edge
can effectively be treated as a 1D Ising system of length kn ` 1. Given that every
20
level m ă n is perfectly locked in, the minimum time tn required for equilibrium to
be reached by the level n structure at Tn is essentially determined by the time it
takes for two end spins in the Ising system to correlate to each other. As the length
of the Ising system scales with the number of level of structure, a scaling relation
is expected to exist for the correlation time of end spins. Determining the scaling
relation for the correlation time of Ising systems of increasing length would then
provide us with a scaling relation for the sequence of minimum time, ttnukn“1.
More precisely, if the size of each level of structure, i.e. the number of black triangles,
is controlled to be the same, then the only difference between the different levels of
structure is really just the edge length of the black triangles. We make the following
hypothesis: given that the system sizes for both levels are the same and all lower levels
of structure are perfectly ordered, the ratio between the minimum time required for
level n and n´1 to reach equilibrium, tn{tn´1, is determined by the ratio between the
time it takes to reach equilibrium by the 1D Ising systems corresponding to the two
levels of structure, tI;n{tI;n´1. To investigate the truth of this hypothesis, we carry
out theory calculations and numerical simulations for the Ising systems to obtain the
ratio, tI;n{tI;n´1, and compare it with the ratio, tn{tn´1 obtained from the simulation
using the tiling model.
3.2 tn{tn´1 by MC Simulation of Tiling Model
To test our hypothesis about the correspondence between the scaling behavior of the
different levels of tiling structure and that of 1D Ising systems, we first carry out
Monte Carlo simulation of the tiling model using the quenching procedure discussed
at the beginning of Chapter 3. The procedure guarantees the well ordering of all
levels of structure of the tiling model so that the limit-periodic ground state emerges
in the end. It converts the dynamics of the tiling model at each temperature in tTnu
21
to just the dynamics of the level n structure as all previous n ´ 1 are already locked
in when the level n structure starts to order. Monte Carlo Simulation of the tiling
model following the quenching procedure is used to find the ratio, tn{tn´1, between
the minimum time required for level n and n ´ 1 to reach equilibrium at the tem-
perature Tn and Tn´1 respectively. We define the number of mcs run as T . For
sufficiently large T , the order parameter, Φn for the level n structure approaches its
equilibrium value exponentially. The relaxation constant τn could be extracted from
the curve of Φn as a function of T . As the time scale of the transient behavior of the
curve for ΦnpT q is very small compared to the time scaled required for Φn to reach
its equilibrium value exponentially, we could neglect the initial part of the curve for
ΦnpT q where the transient behavior occurs. Hence taking the ratio τn{τn´1 gives a
satisfactory estimate of the ratio tn{tn´1.
According to the quenching procedure, the method of simulation of the tiling model
is as follows: Let T1 “ 1.0. For the level n structure, level 1 through level n ´ 1
structures are set to be perfectly ordered, while the level n structure is set to be
completely random; The tiling system is then equilibrated at the temperature Tn,
the rescaled temperature of T1 given by the Rescaling Law in Eq. (2.2).
A snapshot of a typical starting configuration for the level 3 structure is showcased in
Fig. 3.1. Note that the level 1 and 2 structures are perfectly in place. It is intended
that the level 3 structure is initially random, however in order to lock in the level
1 and 2 structures, the long black stripes of the middle tiles of level 3 are initially
always parallel to the potential edges of the level 3 black triangles. During the MC
simulation of the level 3’s relaxation to equilibrium, the level 3 middle tiles would go
to orientations such that the long black stripes are nonparallel to the potential edges
of the level 3 black triangles with negligible probability due to high energy cost of
22
those moves. The only allowed positions of the long black stripes are as illustrated
in Fig. 2.1 by the dashed grey lines, corresponding to the states |Ó� and |Ò� in our
analogy of 1D Ising systems.
Figure 3.1: A portion of the initial configuration used for the MC simulation ofthe level 3 structure. The level 1 and 2 structures are perfectly locked in, while thelevel 3 structure is random.
MC simulations of relaxation to equilibrium are done for the level 1, 2 and 3 struc-
tures. Curves of Φn as a function of mcs, T for n “ 1, 2, 3 are shown in Fig. 3.2. The
tiling system sizes used for the relaxation of level 1, 2 and 3 are 32ˆ32, 64ˆ64, and
128 ˆ 128 respectively such that the number of tiles that need to be ordered dur-
ing the relaxation of each level is the same. The level 1 structure is equilibrated at
T1 “ 1.0; The level 2 and 3 structures are equilibrated at the rescaled temperatures
of T1 at T2 “ 0.603335 and T3 “ 0.427097 respectively. Together these conditions
ensure the difference in the relaxation time required for each level is expected to
23
be only due to the different time required for the corners of the black triangles at
different levels to develop correlations with each other.
�
�
�
���������������������������������������
�������������������������������������������������������������������������������������������������������������
�
����������������������������������������������
������������������������������
������������
������������������
������������������������
�����������������������
� n�1� n�2� n�3
0 5000 10000 15000 20000 250000.0
0.2
0.4
0.6
0.8
1.0
�
�n
Figure 3.2: The plot of Φn vs. T for n “ 1, 2, 3. The tiling system sizes used forthe relaxation of level 1, 2, and 3 are 32 ˆ 32, 64 ˆ 64, and 128 ˆ 128 respectively.The temperatures of equilibration for level 1, 2, and 3 are T1 “ 1.0, T2 “ 0.603335and T3 “ 0.427097 respectively.
We neglect the transient behavior of ΦnpT q and compare the curves for ΦnpT q for
Φn ě 0.8. The relaxation constant τn is extracted for each curve and their ratios are
listed in the first row of Tab. 3.1. In Fig. 3.3, the curve for Φ1pT q has been scaled in
T by the factor of τ2{τ1 ˆ τ3{τ2 “ 13.8 ˆ 10.7. The curve for Φ2pT q has been scaled in
T by the factor of τ3{τ2 “ 10.7. The scaled curves of Φ1pT q, Φ2pT q and Φ3pT q showvery good agreement with each other, indicating a good estimate of the ratio τ2{τ1and τ3{τ2. As discussed earlier, the ratio τ2{τ1 and τ3{τ2 approximates the ratio t2{t1and t3{t2 respectively.
24
�
�
������������������������������������������
����������������������
����������������������������
�
�������������������������������������������������������
������������������������������������
�������������������������������������
�������������������������������������������������������������������
���������������������������������������
����������������������������������������������������������������
� n�1� n�2� n�3
0 400000 800000 12000000.80
0.85
0.90
0.95
1.00
�
�n
Figure 3.3: The plot of Φn vs. scaled T for n “ 1, 2, 3. The ratio between therelaxation constants for level 1 and 2 gives τ2{τ1 “ 13.8, while the ratio betweenthe relaxation constants for level 2 and 3 gives τ3{τ2 “ 10.7. The three curves showgood agreement with each other after the mcs T for level 1 is scaled by the factorτ2{τ1 ˆ τ3{τ2, and the mcs T for level 2 is scaled by the factor τ3{τ2.
3.3 Theory Prediction of tI;n{tI;n´1
The next step of testing our hypothesis is to find tI;n{tI;n´1, the ratio between the
time required for 1D Ising systems that correspond to two subsequent levels of struc-
ture in the tiling model to reach equilibrium. The ratio tI;2{tI;1 is found through a
direct theoretical calculation, while the ratio between higher levels of Ising systems
are obtained through MC simulation of the 1D Ising systems using the Metropolis
Algorithm, the same algorithm we used in the simulation of the tiling model. The
settings of the problem we need to solve in order to obtain the desired ratios are as
follows: In each level of structure, let the total number of 1D Ising systems, s to be
the same since we want to control the system size for each level to be the same. At
level n, the length of the Ising systems is kn ` 1. For each Ising system, the first
25
spin is fixed in the up orientation with the state |Ò�, and the rest are initialized to
random orientations. Spin updating rules follow the standard Metropolis Algorithm
except that only two out of the three spins (the first spin is fixed) are allowed to
be flipped. One Monte Carlo Step (mcs) is defined as potentially updating s ˆ kn
spins-the number of movable spins-among s 1D Ising systems asynchronously. For
each pair of spin bond, energy is 0 if the pair of spins is aligned, and 1 if the pair
is anti-aligned. An Ising system is ordered if the first and last spins are aligned, i.e.
both in the |Ò� state. When level n structure reaches equilibrium, an equilibrium
fraction of the Ising systems will be in the ordered configurations. The exact value
of the fraction is determined from Boltzmann statistics by taking the ratio between
the Boltzmann factor for the ordered configuration and the sum of the Boltzmann
factors for all the possible configurations of the Ising system. For level n, If we
plot the fraction of Ising systems that are ordered as a function of T , it is expected
that for sufficiently large T , the Ising systems approaches the equilibrium fraction
exponentially. The the relaxation time constant, τn can then be extracted from the
exponential curve. τn determines the rate of approach to equilibrium for the level
n Ising systems, hence we can approximate the ratio tI;n{tI;n´1 by taking the ratio
τn{τn´1 as discussed before.
The analogy between the settings of the problem for the Ising systems and for the
tiling model is apparent. The various conditions in the Ising systems are intended to
reproduce those in the tiling system. Fig. 3.4 gives an illustration of the Ising systems
at level 3. The bonds between pairs of spins in the Ising system represent the bonds
created by the black stripes joining the nearest neighbor tiles along the edges of the
black triangles. An aligned pair of spins represents a pair of nearest neighbor tiles
with matched black stripes joining them. Besides, the energy assignments for the
Ising spin bonds are in accordance with those stated in the local “matching rules”
26
for the tiles. Further, the definition for one mcs for the Ising systems matches that
for the tiling model. In every mcs, on average there is one attempt to update every
spin, whereas in the tiling model, on average there is also one attempt to update
every tile in one mcs.
Mismatched corners
Matched corners
Level 3
Figure 3.4: A diagram illustrating Ising systems at level 3. The Ising system at thetop has anti-aligned end spins, corresponding to an unordered edge of black triangle,while the Ising system at the bottom has aligned end spins, corresponding to anordered edge of black triangle. The arrows indicate the spin states represented bythe end and middle tiles in their respective orientations. Each anti-aligned pair ofspin bond represents a mismatch in black stripes of neighboring tiles. The fractionα is greater than β as the end tiles have more energetically favorable positions to goto.
To complete the analogy, one more parameter needs to be introduced to the 1D Ising
systems: Suppose a spin was in the state |Ò� originally. In the standard Metropolis
Algorithm for the 1D Ising model, when the spin is chosen to be updated during
a mcs, it is tested whether it should be flipped to the opposite state |Ó�. Suppose
that there is only a probability of ρ that state |Ó� is generated to be tested for the
spin chosen. With probability 1´ ρ, the spin would just stay in its original state |Ò�without going through the Metropolis updating procedure. This extra parameter ρ is
introduced because when a tile on an edge of a black triangle is chosen to be updated,
as discussed in Sec. 3.2, among the 12 possible orientations it could be tested with,
27
only a fraction of the orientations correspond to a flip of the allowed positions of the
long black stripe which are parallel to the edges of the black triangles. Furthermore,
this fraction is different for the middle and end tiles on an edge of black triangle
because the end tiles have more allowed positions of the black stripe than the middle
tiles. The middle tiles participate in the formation of black triangles of the previous
levels, hence only have two energetically favorable positions of the long black stripe,
corresponding to spin states |Ò� and |Ó�. We let ρ “ α for the end spins in the 1D
Ising systems, and let ρ “ β for the middle spins, where α and β are constants.
3.3.1 Theory Calculation for τ2{τ1
For the simple case of level 1 and level 2 Ising systems , the relaxation time con-
stant, τ can be solved exactly. For higher level structures, MC simulations using the
settings stated above for the problem of Ising systems are used to find τ .
At level 1, each 1D Ising system has 2 spins, corresponding to two tiles on each edge
of the level 1 black triangles, with the leftmost spin fixed in the state |Ò�. Fig. 3.5
shows each configuration of the Ising system and how each configuration could be
updated after one spin has been flipped. The numbers in the parentheses are the
corresponding energies of the configurations.
(1)(0) (0)(1)Figure 3.5: On the left side are the possible configurations of a level 1 Ising system,and on the right side are the possible configurations after one spin has been flipped. Inthe case of level 1, only the second spin could be flipped since the first is always fixed.The numbers in the parenthesis are the corresponding energies of the configurations.
Define at time step t (after 1 ˆ t spins have been updated), the number of Ising
systems with the ordered configuration |ÒÒ� as nt, its fraction among a total of s
28
Ising systems as qt. Then nt`1 is related to nt by:
nt`1 “ αp pnt ´ 1q qt ` αp1 ´ pq nt qt ` p1 ´ αq nt qt
` p1 ´ αqnt p1 ´ qtq ` α pnt ` 1q p1 ´ qtq
“ nt ´ α rpp ` 1qqt ´ 1s
(3.1)
Where p “ e´1{T1 is the Boltzmann factor which indicates the probability that a
move is accepted if it increases the energy of the Ising system by 1 at temperature
T1. Divide both sides by s, the total number of Ising systems, then:
qt`1 “ qt ´ α
srpp ` 1qqt ´ 1s (3.2)
At equilibrium, let q˚ be the equilibrium value of qt, then:
q˚ “ q˚ ´ α
srpp ` 1qq˚ ´ 1s (3.3)
Solve for q˚ to get:
q˚ “ 1
p ` 1(3.4)
Let qt “ q˚ ` �t, qt`1 “ q˚ ` �t`1, where �t and �t`1 are the deviations from the
equilibrium fraction at time t and t ` 1 respectively. It follows from Eq.(3.2) and
Eq.(3.4) that:
�t`1 “ �t”1 ´ α
spp ` 1q
ı(3.5)
One mcs is defined as potentially updating all the movable spins in the Ising systems.
At level 1, the total number of movable spins is s. It follows that:
�T `1 “ �T”1 ´ α
spp ` 1q
ıs(3.6)
29
Take the limit where s Ñ 8 as the size of level 1 becomes infinitely large, then:
limsÑ8
�T `1 “ �T e´αpp`1q (3.7)
Applying the above relation recursively, one arrives at:
�T “ �0e´T αpp`1q (3.8)
Where �0 is the deviation from the equilibrium fraction q˚ at time 0. The equation
for �T has the form of an exponential function, so the relaxation constant for level 1
can now be extracted:
τ1 “ 1
αpp ` 1q (3.9)
Now for level 2, each 1D Ising system has 3 spins, corresponding to three tiles on
each edge of the level 2 black triangles, with the leftmost spin fixed in the state
|Ò�. Fig. 3.6 shows all possible configurations of the Ising system and how each
configuration could be updated after one spin has been flipped.
(1)(0) (0)(1)(2) (1)
(1)(2) (2)(1)(0) (1)
Figure 3.6: On the left side are the possible configurations of a level 2 Ising system,and on the right side are the possible configurations after one spin has been flipped.The first spin is fixed in the up orientation as always. The numbers in the parenthesisare the corresponding energies of the configurations.
30
Define at time step t (after 2 ˆ t spins have been updated), the number of Ising
systems with configurations |ÒÒÒ�, |ÒÓÒ�, |ÒÒÓ� as n1,t, n2,t, n3,t respectively. Their
fractions among a total of s Ising systems as q1,t, q2,t, q3,t respectively. n1,t`1, n2,t`1,
n3,t`1 can be related to n1,t, n2,t, n3,t respectively:
n1,t`1 “ 1
2
`2n1,t ´ q1,tpαp1 ` βp2q ` βq2,t ` αq3,t
˘` 1
2αp1 pn1,t ´ 1q q1,t
` 1
2αp1 ´ p1qn1,tq1,t ` 1
2βp2 pn1,t ´ 1q q1,t ` 1
2βp1 ´ p2qn1,tq1,t ` 1
2p1 ´ αqn1,tq1,t
` 1
2p1 ´ βqn1,tq1,t ` 1
2p1 ´ βqn1,tq2,t ` 1
2β pn1,t ` 1q q2,t ` n1,tq2,t
2` 1
2α pn1,t ` 1q q3,t
` n1,t`1 ´ q1,t ´ q2,t ´ q3,t
˘` 1
2p1 ´ αqn1,tq3,t ` n1,tq3,t
2
“ 1
2
“2n1,t ´ q1,tpαp1 ` βp2q ` βq2,t ` αq3,t
‰
(3.10)
n2,t`1 “ 1
2βp1 ´ p2qn2,tq1,t ` 1
2βp2 pn2,t ` 1q q1,t ` 1
2p1 ´ βqn2,tq1,t ` n2,tq1,t
2
` 1
2α pn2,t ´ 1q q2,t ` 1
2β pn2,t ´ 1q q2,t ` 1
2p1 ´ βqn2,tq2,t
` 1
2αp1 pn2,t ` 1q
`1 ´ q1,t ´ q2,t ´ q3,t
˘` 1
2p1 ´ αqn2,tq2,t ` n2,tq3,t
` 1
2αp1 ´ p1qn2,t
`1 ´ q1,t ´ q2,t ´ q3,t
˘` 1
2p1 ´ αqn2,t
`1 ´ q1,t ´ q2,t ´ q3,t
˘
` 1
2n2,t
`1 ´ q1,t ´ q2,t ´ q3,t
˘
“ 1
2
“2n2,t ` αp1 ` q1,tpβp2 ´ αp1q ´ αp1q2,t ´ αp1q3,t ´ αq2,t ´ βq2,t
‰
(3.11)
31
n3,t`1 “ 1
2αp1 ´ p1qn3,tq1,t ` 1
2αp1 pn3,t ` 1q q1,t ` 1
2p1 ´ αqn3,tq1,t ` n3,tq1,t
2
` n3,tq2,t ` 1
2α pn3,t ´ 1q q3,t ` 1
2p1 ´ αqn3,tq3,t ` 1
2β pn3,t ´ 1q q3,t
` 1
2β pn3,t ` 1q
`1 ´ q1,t ´ q2,t ´ q3,t
˘` 1
2n3,t
`1 ´ q1,t ´ q2,t ´ q3,t
˘
` 1
2p1 ´ βqn3,tq3,t ` 1
2p1 ´ βqn3,t
`1 ´ q1,t ´ q2,t ´ q3,t
˘
“ 1
2
“2n3,t ` β ` q1,tpαp1 ´ βq ´ βq2,t ´ αq3,t ´ 2βq3,t
‰
(3.12)
Where p1 “ e´1{T2 and p2 “ e´2{T2 are the Boltzmann factors which indicate the
probabilities that a move is accepted if it increases the energy of the Ising system by
1 and 2 respectively at temperature T2. Divide both sides of Eq.(3.10)–(3.12) by s,
the total number of Ising systems, then:
q1,t`1 “ q1,t ´ 1
2s
“q1,tpαp1 ` βp2q ` βq2,t ` αq3,t
‰(3.13)
q2,t`1 “ q2,t ` 1
2s
“αp1 ` q1,tpβp2 ´ αp1q ´ αp1q2,t ´ αp1q3,t ´ αq2,t ´ βq2,t
‰(3.14)
q3,t`1 “ q3,t ` 1
2s
“β ` q1,tpαp1 ´ βq ´ βq2,t ´ αq3,t ´ 2βq3,t
‰(3.15)
By following the same procedure as in the calculation for level 1, we standardize
q1,t, q2,t and q3,t around their equilibrium values q˚1 , q
˚2 and q˚
3 respectively. Let �1,t,
�2,t and �3,t be the deviations from the equilibrium fractions at time t respectively,
then three recursive relations can be obtained for the deviations from the equilibrium
fractions. Written in matrix form:
»
–�1,t`1
�2,t`1
�3,t`1
fi
fl “
»
——–
2s´p1α´p2β2s
β2s
α2s
p2β´p1α2s
2s´p1α´α´β2s ´p1α
2s
p1α´β2s ´ β
2s2s´α´2β
2s
fi
ffiffifl
»
–�1,t�2,t�3,t
fi
fl (3.16)
Or:
32
�t`1 “ A�t (3.17)
WhereA is the 3 by 3 matrix in Eq.(3.16). One mcs is defined as potentially updating
all the movable spins in the Ising systems. At level 2, the total number of movable
spins is s ˆ k2 “ 2s. It follows that:
�T `1 “ A2s�T (3.18)
Let �0 be the deviation from the equilibrium fraction vector q˚ at time 0. Applying
Eq.(3.18) recursively to arrive at:
�T “ pA2sqτ�0 (3.19)
Suppose the eigenvalues and eigenvectors of A are tλiu3i“1 and tviu3i“1 respectively.
�0 can be written as a linear combination of the eigenvectors of A,
�0 “3ÿ
i“1
civi (3.20)
Where tciu3i“1 are real constants. Eq.(3.19) becomes:
�T “3ÿ
i“1
cipλ2si qτvi (3.21)
For illustrative purpose, let us set α “ 1 and β “ 1 and proceed with the calculation.
It follows:
A “
»
——–
2s´p1´p22s
12s
12s
p2´p12s
2s´p1´22s ´p1
2s
p1´12s ´ 1
2s2s´32s
fi
ffiffifl (3.22)
The eigenvalues of A can be solved:
33
λ1 “ 1 ´ 1
s
λ2,3 “ 1 ´ 1
4s
ˆ2p1 ` p2 ` 3 ˘
bp22 ´ 2p2 ` 5
˙ (3.23)
Raise the eigenvalues to the power of 2s and take the limit where s Ñ 8:
limsÑ8
λ2s1 “ e´2
limsÑ8
λ2s2,3 “ e
´ 12
´2p1`p2`3˘
?p22´2p2`5
¯ (3.24)
For sufficiently large number of mcs, or T , the term with λ2s3 dominates in Eq. (3.21):
�T » c3pλ2s3 qT v3
“ c3e´ 1
2
´2p1`p2`3´
?p22´2p2`5
¯T v3
(3.25)
An Ising system is ordered if the first spin and last spin are aligned, i.e. in one of
the configurations of |ÒÒÒ�, |ÒÒÓ�. We denote the total fraction deviation from the
equilibrium fraction of ordered systems, q˚ after T mcs as �tot;T . Then for sufficiently
large T :
�tot;T “ c3pv31 ` v32qe´ 12
´2p1`p2`3´
?p22´2p2`5
¯T (3.26)
From Eq.(3.26), the relaxation time constant τ2 for level 2 in the case that both
α “ 1 and β “ 1 can be extracted:
τ2 “ 2
2p1 ` p2 ` 3 ´ap22 ´ 2p2 ` 5
(3.27)
Now the value of τ2{τ1, which approximates tI;2{tI;1, can be calculated using the above
calculations for τ1 and τ2. Set T1 “ 1.0 and T2 “ 0.603335, which is the rescaled
temperature of T1 for the level 2 Ising systems. For each edge of the black triangles,
the number of tile orientations that correspond to a flip of the allowed positions of
34
the long black stripe is in fact 2 out of 12 for the middle tiles, and 12 out of 12
for the end tiles. This gives the fraction α “ 1 and β “ 16 . However, the situation
is more complicated for the middle tiles. The long color stripe of each middle tile,
which is perpendicular to the long black stripe, could have mismatches with the the
long color stripes directly above and below it. Depending on the positions of the
long color stripes above and below each middle tile, 1 of the 2 allowed orientations of
the middle tile could be energetically unfavorable, and thus forbidden. This means
the fraction β could be as low as 112 and as high as 1
6 . Setting β “ 112 , we obtain a
theoretical ratio of τ2{τ1 “ 14.464; Setting β “ 19 , τ2{τ1 “ 11.033; Setting β “ 1
6 ,
τ2{τ1 “ 7.618.
3.3.2 τn{τn´1 Ratios by MC Simulation of 1D Ising Systems
The τn{τn´1 ratios for the higher levels are obtained from MC simulations of the 1D
Ising systems using the settings as stated at the beginning of Sec. 3.3. We set the
temperatures of the Ising systems for level 1, 2, and 3 as T1 “ 1.0, T2 “ 0.603335,
and T3 “ 0.427097 to match the temperatures used for equilibration of the corre-
sponding levels of structure of the tiling model. The results for α “ 1 and β “ 112
are summarized in Tab. 3.1 for level 1, 2, and 3.
τ2{τ1 τ3{τ2Simulation of Tiling Model 13.8 10.7Simulation of Ising Model 14.45 4.10Direct Calculation (Ising) 14.46 -
Table 3.1: The ratios between the relaxation constants for the level 1, 2, and 3structure of the tiling model obtained from MC simulation of the tiling model arecompared with the ratios predicted by simulation of 1D Ising systems as well as fromdirect calculations.
The ratios between the relaxation constants from the MC simulations of the tiling
model are also listed in Tab. 3.1 for comparison. For τ2{τ1, the predicted value from
35
theory calculation and simulation of 1D Ising systems are in good agreement with
each other. Further, they agree reasonably well with the value from simulation of the
tiling model. However, there is an obvious discrepancy between the predicted value
of τ3{τ2 and that from simulation of the tiling model. We have tried using different
values of α and β unsuccessfully. In fact, setting β to any value higher than 1{12 de-
creases the predicted value of τ3{τ2, thus widens the discrepancy. This disagreement
is still not well understood at the current stage of research. To resolve the discrep-
ancy, a better understanding is needed for how the long color stripes of the middle
tiles on each edge of black triangles interact with their next-nearest neighbors above
and below them. We have also hypothesized that different values of β may have
to be set for different middle tiles along the edge of a black triangle. However, an
analysis which examines how often each middle tile is flipped during MC simulations
of the tiling model does not support this hypothesis. An ongoing effort to resolve
the discrepancy involves removing the color stripes of the tiling model altogether,
i.e. setting �2 “ 0, so that the color stripes no longer play a role in the dynamics of
the tiling model. Preliminary evidence suggests that without the color stripes, the
tiling model would still be able to relax to its limit-periodic ground state. Finding
the τn{τn´1 ratios in the case �2 “ 0 would test the validity of using 1D Ising systems
to model the dynamics of the tiling model.
The most likely explanation of the observed discrepancy in the scaling behavior is
due to the finite-size effect in MC simulations of the tiling model. Our theory based
on 1D Ising model assumes the collective physics effects of ordering aside from the
the development of correlation between the corners of the triangles are the same
for different levels of structure. This assumption is valid if the system size is large
enough. The collective physics effects of each level’s ordering include the domain
dynamics. Due to the limited sizes of the system we used for MC simulations of
the tiling model, the domain dynamics may not be the same for level 1, 2 and 3,
thus the scaling behavior of the time required for equilibration by different levels
of structure is affected by the finite-size effect. We shall return to this discussion
36
in the concluding paragraph after we study the domain dynamics in the next chapter.
37
4
Domain Dynamics
Some preliminary evidence has been found that a rapid quench of the tiling system
leads to disordered states [8]. If the rate of dropping the temperature is too high so
that the temperature drops below the level n transition temperature Tc;n before Φn´1
can reach a sufficiently high value, then neither level n nor level n ´ 1 could become
well ordered. The result is that the ordering the system gets frozen and the system
ends up in some disordered configuration. Intuitively, this could be understood if we
look at the non-contributing sublattice for each level of structure. When the level 1
structure fails to order during a rapid quench, the value of its order parameter, Φ1
would be far from its maximum value 1. When the temperature drops to T2;c so that
the level 2 structure starts to order, level 2 would be ignorant about which is the
non-contributing sublattice for level 1 and hence on which sublattice to establish the
level 2 structure. As a result, level 2 would also fail to become fully ordered. Any
subsequent level would face the same difficulty of choosing a sublattice to establish
its order, and hence the system as a whole would fail in getting ordered. The behav-
ior of the system during a rapid quench is reminiscent of glass formation, where any
nonzero quench rate would eventually lead to the system getting trapped in some
38
nonequilibrium states [14].
Several questions remain poorly understood from any previous findings. First, why
does the system falls out of equilibrium during a rapid quench? More precisely, what
are the kinetic barriers created by a certain level of structure when it fails to or-
der that cause the simultaneous frustration of the ordering of the subsequent levels?
Second, when the system goes into a disordered phase during a rapid quench, does
the system get stuck there? Could the system return to equilibrium if we relax the
system at some finite temperature for a reasonable amount of time? If the answer is
yes, then what is the energy cost of removing the kinetic barriers created by that level
of structure when it fails to order. These are the questions addressed by this chapter.
4.1 Domain Walls
A closer look at the configuration of the tiling system resulted from a rapid quench
reveals that domain walls form when the ordering of a certain level of structure gets
frustrated. Fig. 4.1 shows that several domains form after a size 32 ˆ 32 system
goes through a rapid quench in which the initial temperature, T “ 2.0 is lowered by
δT “ 0.01 after every r “ 10 mcs.
The level 1 tiles colored in red have spin vector �eA, and they form a domain for which
the non-contributing sublattice for level 1 is sublattice A. The tiles in sublattice A
in the red domain could have any of the other three spin vectors, however they are
colored in dark red to emphasize that they are part of the red domain. Similarly, the
purple, green and yellow domains have sublattice B, C, D as their non-contributing
sublattices respectively. The color stripes are omitted in Fig. 4.1 to reduce visual
distraction, but they do participate in forming the domain walls. The four domains
39
Figure 4.1: Four domain walls are present after a rapid quench from T “ 2.0.Temperature is lowered by δT “ 0.01 after every r “ 10 mcs. The red, purple,green and yellow domains have sublattice A, B, C and D as their non-contributingsublattice respectively.
compete with each other as each wants its non-contributing sublattice to be the
non-contributing sublattice for the entire level 1 structure, and there is no single
dominating domain during the rapid quench. Clearly the energy barriers created by
the domain walls prevented the system from getting further ordered. Even though
the domains are only colored for level 1 here, they could form in any level of struc-
ture. Level 2 domains could develop within each of the four level 1 domains shown
in Fig. 4.1, and level 3 domains could develop within the domains of level 2, etc.
The domain walls created when a level of structure fails to order prevent the system
from getting further ordered during a rapid quench, but would the system go back
40
to an equilibrium state if we relax the system at a finite non-zero temperature for
a reasonable amount of time such that a single domain emerges by conquering the
others (We require the sitting time to be reasonable, as if one waits for an astronom-
ical amount of time, there is always a small chance that the system equilibrates)?
Fig. 4.2 illustrates such an example. However, as we will see later, this is not al-
ways the case. Whether the system could equilibrate by sitting at a finite non-zero
temperature for a reasonable amount of time depends on the temperature and the
type of domain walls present in the system. An example in which the system gets
stuck in a configuration, and the ordering of the system freezes is discussed in Sec. 4.3.
Fig. 4.2(a) shows a configuration of a size 64ˆ64 system after a sudden quench from
an initial high temperature to T “ 0.6, followed by sitting at that temperature for
5 ˆ 105 mcs. While Fig 4.2(b–f) show the subsequent configurations of the system
after further sitting at T “ 0.6. As the sitting temperature T “ 0.6 is below the
level 2 transition temperature, T2;c and above the level 3 transition temperature,
T3;c, the highest level of structure that can be formed within each domain is level 2.
An examination of these configurations reveals that there are two types of domain
walls present in the tiling model. A type I domain wall is shown in Fig. 4.3, which
is a zoomed in view of the domain wall between the yellow and green domains in
the lower left part of the configuration in Fig. 4.2(a). A type I domain wall is made
of tiles of alternating colors arranged along a straight line. The black triangles from
two different domains are separated by the long black stripes of the tiles along the
domain wall. Long correlation in the long black stripes often develops such that it
is possible to have no mismatches in the black stripes along the domain wall.
A type II domain wall is shown in Fig.4.4, which is a zoomed in view of the domain
41
II
II
II
II
IIII
(a) T “ 50 ˆ 104 (b) T “ 54 ˆ 104
(c) T “ 66 ˆ 104 (d) T “ 67 ˆ 104
(e) T “ 68 ˆ 104 (f) T “ 69 ˆ 104
Figure 4.2: The configurations of a size 64 ˆ 64 system after a rapid quench toT “ 0.6 and sitting at that temperature for 69ˆ104 mcs. The black and color stripesare omitted here. The roman numerals indicate the types of domain wall. (a) Theconfiguration after being suddenly quenched to T “ 0.6 and sitting for 50ˆ 104 mcs.(b–f) Subsequent configurations after furthering sitting at T “ 0.6.
42
Figure 4.3: The type I domain wall shown from a zoomed in view of the domainwall between the yellow and green domains in the lower left part of the configurationin Fig. 4.2(a).
walls between the yellow and purple domains in the lower right part of the config-
uration in Fig. 4.2(a). Along the type II domain wall, the black triangles from two
different domains get intercepted by each other. In many cases, the incomplete level
1 black triangles from one domain are connected to the incomplete black triangles of
higher levels from the other domain. In this case, the highest level of black triangles
formed is level 2, so an entanglement between the level 1 and 2 incomplete black
triangles across two different domains is observed.
Observing how the domain walls in the configurations in Fig. 4.2 have evolved over
a period of 69ˆ104 mcs, it is found that the type I domain walls are much harder to
move around than the type II domain walls. The middle yellow domain is bounded
by two domain walls of type I as indicated by the labels “ I” in Fig 4.2(a). The rest
43
Figure 4.4: The second type of domain wall shown from a zoomed in view of thedomain walls between the yellow and purple domains in the lower right part of theconfiguration in Fig. 4.2(b).
of the domain walls are of type II as indicated by labels “II”. The yellow domain has
expanded through moving its type II domain walls while its type I domain walls have
changed in length but maintained their shapes and positions. Apparently, the type
I domain wall is intrinsically harder to move than the type II domain wall. The part
of the green domain pointed by an arrow in Fig. 4.2(a) is bounded by one domain
wall of type I and one of type II. It has shrunk over time despite the fact that its
type I domain wall remains stable. However the blue domain on the other side of
its type II domain wall gets taken by the yellow domain, which then pushed the
type II domain wall of the green domain inward. The invasion of the yellow domain
subsequently consumes the type I domain wall of the green domain which served as
a defense. Sitting at T “ 0.6 for more time, it is observed that the yellow domain
eventually conquers the other domains through the expansion of its type I domain
44
walls.
4.2 Energy Costs of Infinite Domain Walls
The energetics associated with the two types of domain wall are different, and this
accounts for the difference in the difficulty of move the two types of domain wall.
Here we construct flat domain walls that span the entire system in different direc-
tions, and examine the energy costs of moving the domain walls around. The type
of domain wall arisen in each direction and the corresponding number of mismatches
along the domain walls give insights into the energy barriers created by the domain
walls and their associated dynamics.
First, two flat domain walls are constructed in the horizontal direction to span a size
64 ˆ 64 system in a rhombus domain. The system is made of two domains, a red
and a yellow one. Each domain is in the limit-periodic ground state, i.e. all levels of
structure are well ordered. Note that under the periodic boundary conditions, there
are at least two mismatches present in the system [8]. In fact one mismatch in black
stripe is each present in the red and yellow domain. Define a coordinate system for
the rhombus domain such that the x-axis is parallel to the horizontal direction, and
the y-axis is parallel to the slanted direction. The origin is set at the bottom right
corner of the domain (the direction of axes and origin are marked in Fig. 4.7). The
lower domain wall is fixed at y “ 15 and the upper wall is initially set at y “ 40.
The upper domain is subsequently moved upward to the position y “ 60 in steps of
∆y “ 1. The number of mismatches in both the black and color stripes on the upper
domain wall are plotted as a function of the y position respectively in Fig. 4.5 and
Fig. 4.6.
45
�
�
�
��
�
�
�
�
�
�
�
�
�
�
�
�
�
�
� �
40 45 50 55 6040
60
80
100
120
y
Num
.ofblack
stripemismatches
Figure 4.5: Number of black stripe mismatches vs y for horizontal domain wallsseparating perfectly ordered red and yellow domain.
�
� � �
�
� � �
�
� � �
�
� � �
�
� � �
�
40 45 50 55 60
130
140
150
160
170
180
190
y
Num
.ofcolorstripemismatches
Figure 4.6: Number of color stripe mismatches vs y for horizontal domain wallsseparating perfectly ordered red and yellow domain.
46
The minimum mismatches in black tripe on the upper horizontal domain wall is 62
and, is obtained when y “ 50 and 54; While the minimum number of mismatches in
color stripe on the upper horizontal domain wall is 127 when y “ 54 and 58. There
are other y values at which the number of mismatches in black and color stripes are
just off by 1 or 2 from the minimum numbers. It is clear that periodic behavior
is present in the number of mismatches in both the black and color stripes when
the y value is changed. For the black stripe, the number of mismatches remains
relatively high for three steps of ∆y “ 1 increment and reaches around the minimum
on the fourth step of increment; While for the color stripe, the number of mismatches
remains stable around the minimum for three steps of ∆y “ 1 increment and shoots
to a high value on the fourth step of increment. The resultant horizontal domain
walls are found to be of type II.
y
x
Figure 4.7: A configuration showing two domains separated by a horizontal upperdomain wall of type II at y “ 54 with minimum mismatches. The black and colorstripes are omitted here.
A configuration that corresponds to the upper domain wall having the minimum
47
Figure 4.8: Zoomed in view of a portion of the upper horizontal domain wall oftype II in Fig. 4.7. Following the blue path, a mismatch in black stripe is connectedto another mismatch in black stripe along the domain wall. Following the greenpath, a mismatch in color stripe is connected to another mismatch in color stripealong the domain wall.
number of mismatches in both the black and color stripe decorations at y “ 54 is
shown in Fig. 4.7. A closer view of the upper domain wall is shown in Fig. 4.8. Ex-
amining the mismatches in the black stripe decoration reveals that each mismatch in
black stripe along the domain wall is always connected to another mismatch in black
stripe along the domain wall. One such example is highlighted by the blue path in
Fig. 4.8. The total number of mismatches in black stripe along the upper domain is
64, that is 1 mismatch in black stripe per tile. For the mismatches in color stripe, it
is also the case that each mismatch must have a companion along the domain wall,
connected by a path of incomplete colored triangle. One example is highlighted by
48
the green path in Fig. 4.8. There is a total of 128 mismatches in color stripes along
the upper domain wall, that is 2 mismatches in color stripe every tile.
Next, two flat domain walls are constructed in the slanted direction to separate a
red and a yellow domain. Each domain is in the limit-periodic ground state. The
left slanted domain wall starts at x “ 42 and rises at an angle of 60˝ with respect to
the x-axis to span the entire 64ˆ64 rhombus lattice. The right domain wall initially
starts at x “ 6, and the starting position is subsequently moved to x “ 36 in steps of
∆x “ 1. The number of mismatches in both the black and color stripes on the right
slanted domain wall are plotted as a function of the starting x position respectively
in Fig. 4.9 and Fig. 4.10. The resultant slanted domain walls are of type I.
�
�
��
�
�
��
�
�
�
��
�
���
��
�
�
�
� �
�
�
�
���
�
5 10 15 20 25 30 3520
40
60
80
100
x
Num
.ofblack
stripemismatches
Figure 4.9: Number of black stripe mismatches vs x for slanted domain wallsseparating perfectly ordered red and yellow domain.
The minimum mismatches in black stripe on the right slanted domain wall is 38 for
49
�
�
��
�
�
���
�
�
�
�
�
��
�
�
�
�
�
�� ��
�
�
��
�
�
5 10 15 20 25 30 35100
120
140
160
180
200
x
Num
.ofcolorstripemismatches
Figure 4.10: Number of color stripe mismatches vs x for slanted domain wallsseparating perfectly ordered red and yellow domain.
starting position x “ 31 and 34. The minimum number of mismatches is much lower
than that in the case of horizontal domain walls. This is because a long straight
line made of the long black stripes belonging to tiles of alternate colors forms part
of the domain wall. There could be no mismatches in black stripe along the straight
line separating the two domains as long correlation in black stripe develops. This
is illustrated by a zoomed in view of the domain wall in Fig. 4.12. Later we will
see that the minimum mismatches in black stripe could be even lower if the red
and yellow domains do not have all their levels of structure perfectly ordered. The
minimum number of mismatches in color stripe is 130 at starting position x “ 32.
Here, it is also the case that a mismatch in black stripe or in color stripe must have
a companion mismatch along the domain wall, as illustrated by the blue and green
paths in Fig. 4.12 respectively. The periodic behavior in the number of mismatches
as the starting x position changes is less obvious in the case of the slanted domain
walls. However, it is still clear that the number of mismatches in both the black
50
stripe and the color stripe reaches a relatively low value every few steps of increment
in the starting x position.
y
x�
Figure 4.11: A configuration showing two domains separated by a slanted rightdomain wall of type I at starting position x “ 31 with low number of mismatches.The black and color stripes are omitted here.
4.3 Domain Wall Energetics
Now we are ready to answer why the system’s ordering gets frozen during a rapid
quench. Domain walls often form during a rapid quench of the tiling system. We
have seen a case in Fig. 4.2 that the system goes into a configuration made of several
domains separated by domain walls of different types, and one domain eventually
conquers the others after long enough time. However, there are also cases in which the
system would just get stuck in a particular configuration, and it takes an astronomical
amount of time to move the domain walls around. Fig. 4.13 presents such an example.
The configuration is formed by a sudden quench of a 64 ˆ 64 system from an initial
high temperature with a random configuration to the temperature T “ 0.6. The
51
Figure 4.12: Zoomed in view of a portion of the right slanted domain wall of typeI in Fig. 4.11. Following the blue path, a mismatch in black stripe is connected toanother mismatch in black stripe along the domain wall. Following the green path, amismatch in color stripe is connected to another mismatch in color stripe along thedomain wall.
system is then equilibrated at this temperature for 5 ˆ 105 mcs. At T “ 0.6, the
level 2 structure is well ordered as is evident in Fig. 4.14. The resultant configuration
has two long slanted domain walls of type I separating a red and a yellow domain.
It is found that the domain walls can hardly be moved by keeping sitting at the
temperature T “ 0.6. As a result, the system gets stuck in this configuration and
could not reach equilibrium at T “ 0.6 within a reasonable amount of time.
52
Figure 4.13: A configuration that gets stuck in ordering due the presence of aslanted domain wall of type I that spans the entire 64ˆ64 system. The configurationis resulted from a sudden quench from an initial high temperature to T “ 0.6 andsitting for 5 ˆ 105 mcs. The black and color stripes are omitted here.
Figure 4.14: Zoomed in view of a domain wall of type I from the configuration inFig. 4.13.
53
Siting at temperatures even lower than T “ 0.6 makes it even harder to move the
stable domain walls of type I. This is because the chance of a tile going into an
orientation that does not lower the system’s energy and thus upsets the energetically
favorable domain walls becomes smaller and smaller at lower temperatures. It is
noteworthy that sitting at a temperature lower than T “ 0.6 allows higher levels
of structure to develop within each domain until the size of the triangular lattice
becomes too large to be accommodated by the domain sizes. As shown by Fig. 4.15,
there could be no mismatches in black stripes at all along the domain wall. All the
mismatches are in color stripe decorations. The domain wall of type I obtained in
this case is more energetically favorable than the one we constructed in Fig. 4.12
due to zero mismatches in black stripes. However, an examination of the entire
configuration that is partly shown in Fig. 4.15 reveals that even though all levels of
structure that can be fitted into each domain are ordered, not every black triangle
is perfectly formed. In other words, there are mismatches in black stripes within
each domain. While the configuration we constructed in Fig. 4.12 has all levels of
structure perfectly ordered within each domain, and mismatches in black stripes are
only along the domain wall.
The only way to disrupt the stable domain wall of type I is to increase the tempera-
ture of the system. A simulation of equilibrating the configuration shown in Fig. 4.13
at T “ 1.0 demonstrates the red domain could eventually conquer the yellow domain
in 5 ˆ 104 mcs. The subsequent configurations of the system are shown in Fig. 4.16.
The relative difficulty of moving the domain walls around and thus of one domain to
conquer the others is clearly determined by the energetics of the domain walls present
in the system. Due to the different nature of the two types of domain wall, the en-
ergetics associated with the two types of domain wall are also different. Clearly the
54
Figure 4.15: This is the same portion of the slanted domain wall of type I as inFig. 4.14 after dropping the temperature of the configuration in Fig. 4.13 to T “ 0.05and sitting for another 5 ˆ 104 mcs. There is no mismatches in black stripe alongthe domain wall.
domain walls of type I are energetically more favorable and represent local minima
on the free energy map of the tiling system. This is not surprising as a domain wall of
type I could minimize the number of mismatches by having no mismatches in black
stripes along the domain wall. To disrupt a domain wall of type I, the black triangles
on the two sides of the domain wall that are right next to the long black straight
line have to be broken. This would turn a long straight domain wall of type I into a
wiggly shape, setting the system free from the energy minima imposed by the domain
wall of type I. Part of a domain wall of type I could turn into a domain wall of type
II, which is much easier to move around and break through. However, the process
of breaking all the black triangles next to the domain wall could cost a significant
amount of energy. Thus a domain wall of type I could present an insurmountable
energy barrier. In the case discussed above, sitting at T “ 1.0 allows only the level 1
55
(a) T “ 0 (b) T “ 1 ˆ 104
(c) T “ 2 ˆ 104 (d) T “ 3 ˆ 104
(e) T “ 4 ˆ 104 (f) T “ 5 ˆ 104
Figure 4.16: (a) A configuration of a size 64ˆ 64 system after a sudden quench toT “ 0.6 and sitting at that temperature for 5 ˆ 105 mcs. (b–f) Configurations aftera sudden increase of the temperature to T “ 1.0 and sitting for every 1 ˆ 104 mcs.The black and color stripes are omitted here.
56
structure to develop. The energy of cost of breaking the level 1 triangles right next
to the domain wall of type I is not so high, plus the fact that the temperature is
high, the probability of having tiles along or right next to the domain wall go to ori-
entations that would disrupt the domain wall is high, determined by the Boltzmann
factor e´∆E{T , where ∆E is the change in energy of the move. Thus it is relatively
easy to break a type I domain wall. However as the temperature is decreased, more
and more levels of structure develope within each domain, and the energy cost of
breaking the higher levels of triangle right next to the domain wall becomes very high.
Combined with the effect of the low temperature, the Boltzmann factor e´∆E{T is too
low to allow a significant amount of tiles to go to orientations that would disrupt a
domain wall of type I. Hence, the type I domain walls become practically immovable.
Often a configuration resulted from a rapid quench would have a mixture of domain
walls of both types. As seen in examples shown in Fig. 4.2, one domain conquers an-
other most easily by breaking the domain walls of type II, whereas the type I domain
walls remain intact and stable unless they get consumed by domains expanding in
a direction that is parallel to the type I domain wall. It is found that for each pair
of colored domains, there exists a special direction in which the type I domain walls
could arise. In a large size system, many domains could be formed during a rapid
quench. The system tends to relax to a configuration in which the type I domain
walls establish themselves wherever possible in the special directions, while the type
II domain walls form in the other directions. At a given temperature, with more
type I domain walls present, it is more difficult to restore the system to equilibrium
as the type I domain walls slow down the ordering of the tiling system.
57
5
Conclusion
The Socolar-Taylor tiling model uses a single space-filling prototile to force a limit-
periodic pattern through local matching rules. A two-dimensional lattice model
which possesses the this tiling as its ground state is constructed. During a slow
quench from an initial high-temperature, disordered phase, the ground state of the
model emerges through an infinite sequence of phase transitions following a strict
hierarchy. As temperature is decreased, sublattices with periodic structures of in-
creasing lattice constants become ordered. A theory based on one-dimensional Ising
model is constructed to explain the time scales required for equilibration at a given
temperature by sublattices of increasing lattice constants. Compared with the re-
sults from the simulation of the tiling mode, the theory successfully predicts the
ratio of the time scales required by the level 1 and 2 structures to reach equilib-
rium. However, discrepancies exist for higher levels of structure in predicting the
scaling behavior of the time scales to reach equilibrium. We see that the domain
dynamics determines the extreme slowness of the system’s relaxation to equilibrium
during thermal quenches. In building the theory based on Ising model to predict
the scaling behavior, we have assumed the domain dynamics is equivalent for all
58
levels of structure. Even at level 2 and 3, it requires systems of very large sizes for
this assumption to be true, so that the finite size effect is negligible. However, the
sizes of the tiling model used for the level 2 and 3 simulation are only 64 ˆ 64 and
128ˆ128 respectively, which are not large enough to see the full domain dynamics of
the model. This is the most likely cause of the observed discrepancies in the scaling
behavior between the theory prediction and the results of simulation.
During a rapid quench, the energy barriers created by competing domain walls pre-
vent the system from getting further ordered and cause the system to fall out of
equilibrium. Two types of domain wall with different physical structures and energy
costs are found in the system. During a domain expansion, the associated energetics
makes the type I domain walls much harder to be broken through or moved around
than the type II domain walls. In a large size system, domain walls of type I are es-
tablished wherever possible in special directions during a rapid quench. The relative
difficulty of moving the type I domain walls slows down the relaxation of the system
to equilibrium.
59
Bibliography
[1] J. E. Socolar and J. M. Taylor, An aperiodic hexagonal tile, Journal ofCombinatorial Theory, Series A 118 (2011) no. 8, 2207 – 2231.
[2] R. Penrose, The role of aesthetics in pure and applied mathematical research,The Institute of Mathematics and its Applications Bulletin 10 (1974) no. 7/8,266–271.
[3] A. L. Mackay, Crystallography and the penrose pattern, Physica A: StatisticalMechanics and its Applications 114 (1982) 609 – 613.
[4] P. Kramer and R. Neri, On periodic and non-periodic space fillings of Em
obtained by projection, Acta Crystallographica Section A 40 (1984) no. 5,580–587.
[5] D. Levine and P. J. Steinhardt, Quasicrystals: A New Class of OrderedStructures , Phys. Rev. Lett. 53 (1984) 2477–2480.
[6] Penrose Quilt Challenge, . http://domesticat.net/quilts/penrose.
[7] J. Socolar and J. Taylor, Forcing Nonperiodicity with a Single Tile, TheMathematical Intelligencer 34 (2012) 18–28.
[8] T. Byington and J. Socolar, Hierarchical Freezing in a Lattice Model , Phys.Rev. Lett. 108 (2012) 045701.
[9] D. Shechtman, I. Blech, D. Gratias, and J. W. Cahn, Metallic Phase withLong-Range Orientational Order and No Translational Symmetry , Phys. Rev.Lett. 53 (1984) 1951–1953.
[10] P. A. Bancel, P. A. Heiney, P. W. Stephens, A. I. Goldman, and P. M. Horn,Structure of Rapidly Quenched Al-Mn, Phys. Rev. Lett. 54 (1985) 2422–2425.
60
[11] A.-P. Tsai, A. Inoue, and T. Masumoto, A Stable Quasicrystal in Al-Cu-FeSystem, Japanese Journal of Applied Physics 26 (1987) no. Part 2, No. 9,L1505–L1507.
[12] P. Steinhardt and L. Bindi, In search of natural quasicrystals , Reports onProgress in Physics 75 (2012) no. 9, 092601.
[13] S. Johnson, Emergence: the connected lives of ants, brains, cities, andsoftware. Scribner, 2001.
[14] P. Debenedetti, F. Stillinger, et al., Supercooled liquids and the glasstransition, Nature 410 (2001) no. 6825, 259–267.
61