Master of Science Thesis
Magnetic Monopoles in Spin Ice
Axel Nordstr¨om
Supervisor: Patrik Henelius
Department of Theoretical Physics, School of Engineering Sciences Royal Institute of Technology, SE-106 91 Stockholm, Sweden
Stockholm, Sweden 2014 Typeset in LATEX
Examensarbete inom ¨amnet teoretisk fysik f¨or avl¨aggande av civilingenj¨orsexamen inom utbildningstprogrammet Teknisk fysik. Graduation thesis on the subject Theoretical Physics for the degree of Master of Science in Engineering from the School of Engineering Sciences.
TRITA-FYS 2014:26 ISSN 0280-316X ISRN KTH/FYS/–14:26–SE
© Axel Nordstr¨om,May 2014 Printed in Sweden by Universitetsservice US AB, Stockholm May 2014 Abstract
In this thesis, we investigate the behaviour of magnetic monopoles in spin ice when an external magnetic field is applied. We find that steady state direct currents of magnetic monopoles cannot be maintained for long and consider the possibility of alternating magnetic currents by investigating the alternating current susceptibility using both analytical and Monte Carlo techniques. Moreover, we look at the transition that occurs when a magnetic field is ap- plied in a 111 direction. We show that the transition is a continuous crossover rather thanh a phasei transition in the nearest neighbour model and we study the behaviour of the system during the crossover, especially at the critical field where a temperature independent state appears. Using Monte Carlo methods and analyti- cal methods based on the Bethe approximation, we find that the mean monopole density is 0.4 monopoles per tetrahedron in the temperature independent state at the critical field.
Keywords: spin ice, magnetic monopoles, phase transitions.
iii iv Preface
This thesis is the result of my degree project at the Department of Theoretical Physics at the Royal Institute of Technology (KTH) during the spring semester of 2014. The work concerns magnetic monopoles in frustrated pyrochlore magnets – spin ice – and their statics and dynamics in an applied magnetic field.
Overview
The thesis is divided into six chapters and three appendices. Chapter 1 briefly introduces the subject of magnetism and spin ice, whereas we in Chapter 2 go into some more detail on the generalities of spin ice, introducing the pyrochlore lattice and the model used. Chapter 3 concerns Monte Carlo simulations and we introduce the Metropolis algorithm used in all simulations throughout this thesis. In Chapters 4 and 5, the main results of the work are presented. In Chapter 4, we investigate the dynamics of magnetic monopoles in an applied magnetic field and in Chapter 5 we address the potential phase transition when a magnetic field is applied to spin ice in a 111 direction. Finally, the results are summarised in Chapter 6. Theh firsti appendix contains detailed information regarding the simulations per- formed throughout the thesis. The other two appendices contain brief reviews of rare earth magnetism and complexity classes in computer science, respectively. Knowledge of these topics is not required in order to understand the results pre- sented in the thesis, but may help putting models and problems into context.
Conventions
It is conventional in the field of frustrated magnetism to use the same unit for energy and temperature since it makes comparison of energy scales convenient.Thus we put kB = 1 throughout the thesis and measure energy in kelvins. Otherwise we use standard SI units.
v vi Acknowledgements
First and foremost, I would like to thank my supervisor Assoc. Prof. Patrik Henelius for giving me the opportunity to work with this thesis, for valuable guid- ance and discussions as well as for helping me with practical matters surrounding the thesis. I would also like to thank Mikael Twengstr¨omwith whom I have shared office this past semester and with whom I have had many fruitful discussions on coding, simulations and physics. Many thanks to my colleagues at the Department of Theoretical Physics at KTH for welcoming me among them. Finally I want to thank my friends and family for their encouragement and support.
vii viii Contents
Abstract ...... iii
Preface v
Acknowledgements vii
Contents ix
1. Introduction 1
2. Background theory 3 2.1. Magnetism ...... 3 2.1.1. Para-, ferro- and antiferromagnetism ...... 3 2.1.2. Frustration ...... 4 2.2. Spin ice ...... 5 2.2.1. The pyrochlore lattice and the ice rules ...... 5 2.2.2. Residual entropy ...... 8 2.2.3. Magnetic monopoles ...... 8 2.3. Modelling spin ice ...... 9 2.3.1. The nearest neighbour model ...... 9 2.3.2. The dipole model ...... 10 2.3.3. Effect of magnetic fields ...... 10
3. Monte Carlo simulations 13 3.1. Pseudo-random numbers ...... 13 3.2. Monte Carlo algorithms ...... 13 3.2.1. Requirements on Monte Carlo algorithms ...... 13 3.2.2. Single spin flips ...... 14 3.2.3. The Metropolis algorithm ...... 14 3.2.4. Limitations of Metropolis dynamics ...... 15
ix x Contents
4. Magnetricity 17 4.1. Magnetic monopole current ...... 17 4.2. Expression for the monopole current ...... 18 4.2.1. Magnetisation ...... 21 4.2.2. AC susceptibility ...... 22 4.2.3. Analogy to an RL-circuit ...... 22 4.3. Spin-lattice relaxation ...... 23 4.3.1. Monopole current ...... 25 4.3.2. Interpretation of χS ...... 26 4.4. Numerical determination of the susceptibility ...... 26 4.5. Predicting the monopole current ...... 28
5. Phase transitions in a 111 magnetic field 31 5.1. Introduction to phaseh transitionsi ...... 31 5.2. Kagom´eice ...... 32 5.2.1. Value of the critical field ...... 33 5.2.2. Magnetisation plateau ...... 36 5.2.3. Phase diagram of the nearest neighbour model ...... 37 5.2.4. Low temperature simulations ...... 40 5.2.5. Phase transition or continuous crossover? ...... 41 5.3. Temperature independent state at Hc ...... 42 5.3.1. Monopole density for larger systems ...... 44 5.3.2. Bethe approximation ...... 46 5.3.3. Onset of the temperature dependence ...... 49 5.4. Phase transitions in the dipole model ...... 50
6. Conclusions 53 6.1. Magnetricity ...... 53 6.2. Phase transitions ...... 54
A. Simulations 55 A.1. Details ...... 55 A.2. Simulation parameters ...... 55
B. Rare-earth magnetism 59 B.1. Generalities ...... 59 B.2. Hund’s rules ...... 59
C. Complexity classes 61 C.1. P, NP and #P problems ...... 61 C.2. NP- and #P-complete problems ...... 61 C.3. The P versus NP problem ...... 62
Bibliography 63 Chapter 1
Introduction
The phenomenon of magnetism has been known since about 500 BC, possibly even longer. Since the 12th century, ferromagnets have been used as compasses for navigation and much modern technology, such as hard disks and techniques for medical imaging, relies heavily on magnetism. Although magnetism has been used in technology for nearly a millennium, it was not until the 20th century that the microscopic mechanisms of magnetism were properly understood and the discoveries of other types of magnetism than the usual ferromagnetism were made. To this day, magnetism remains a much studied phenomenon and the research into magnetic systems is vast to say the least. One specific kind of systems that have acquired must attention over the past few decades are so-called frustrated magnetic systems. In these systems, it is impossible to minimise the energy of all interactions simultaneously and the ground state is massively degenerate (see Section 2.1.2). When Harris et al. [1] performed susceptibility measurements and neutron scat- tering on the rare earth pyrochlore compound Ho2Ti2O7 in 1997, it turned out that this material was not only frustrated, but also had a net ferromagnetic near- est neighbour interaction. No such system had previously been observed. Due to an analogy to the frustrated structure of water ice, the family of frustrated pyrochlore magnets to which Ho2Ti2O7 belongs was dubbed ”spin ice”. Since ideal frustrated systems retain their degeneracy even when their temper- ature reach absolute zero, they seem to violate the third law of thermodynamics – which states that the entropy of a perfect crystal, at absolute zero kelvin, is exactly equal to zero. It is, however, widely believed that real frustrated systems have a phase transition to an ordered state at low temperatures, due to small perturba- tions in the system starting to play an important role. In many cases, though, the phase transition would occur at a temperature below what is experimentally attainable today. Electricity and magnetism are intimately related via Maxwell’s equations. How- ever, whereas there exist free electric charges which are sources of the electric field,
1 2 CHAPTER 1. INTRODUCTION magnetic fields seem only to be generated by magnetic multipoles. The question thus arises: do magnetic charges, or monopoles, exist? The existence of magnetic monopoles was proposed by Paul Dirac in 1931 [2], but no experiment has been able to detect any magnetic monopoles to this day. However, in 2008 it was suggested that elementary excitations in spin ice materi- als behave much like classical analogues of Dirac’s magnetic monopoles [3]. This discovery raises the questions of whether or not it is possible to create currents of magnetic monopoles, as is possible with electric charges, and if it even would be possible to construct magnetic equivalents of electronic circuits in condensed matter systems despite the apparent rareness or lack of natural magnetic monopoles. Chapter 2
Background theory
Here we introduce different kinds of magnetism and how these can give rise to frustration. We also introduce the spin ice compounds, the pyrochlore lattice and present the models used to describe spin ice.
2.1 Magnetism
2.1.1 Para-, ferro- and antiferromagnetism
Magnetism occurs when the intrinsic magnetic moments of elementary particles, spins, interact – either with a magnetic field or with each other. If the spins do not interact among themselves, but only with external mag- netic fields, we are dealing with paramagnetism. The spins tend to align with the magnetic field according to M = χH, (2.1) where M is the sample magnetisation, H is the external magnetic field and χ is the susceptibility. In purely paramagnetic materials, the susceptibility obeys Curie’s law [4] 1 χ , (2.2) ∝ T where T is the temperature. If we consider a simple exchange interaction between nearest neighbour spins, we get two different cases: ferromagnetism and antiferromagnetism. In a ferro- magnet, the spins tend to align parallel to each other below the Curie temperature
3 4 CHAPTER 2. BACKGROUND THEORY
TC . Above this temperature, the behaviour is essentially paramagnetic. However, Equation 2.2 must be substituted for the Curie-Weiss law [4] 1 χ . (2.3) ∝ T TC − In antiferromagnets, the spins instead align anti-parallel to each other below the N´eeltemperature TN . As in the ferromagnet, the antiferromagnet becomes paramagnetic above TN , but obeys 1 χ , (2.4) ∝ T + θ where θ is a material dependent parameter in the order of TN [4]. Equations 2.2, 2.3 and 2.4 can thus be written as ∼
Θ = 0 paramagnet 1 CW χ where ΘCW > 0 ferromagnet (2.5) ∼ T ΘCW − ΘCW < 0 antiferromagnet, and thus we can determine what kind of interactions dominate at low temperature by measuring the susceptibility as a function of temperature in the paramagnetic phase and determining the Curie-Weiss temperature, ΘCW [5].
2.1.2 Frustration When a system cannot minimise the energy of all its interactions simultaneously, we say that the system is frustrated. This implies that the ground state of frustrated systems will be degenerate, since there are several ways to minimise the total energy of the system. The link between frustration and degeneracy is so fundamental that the appearance of degeneracy sometimes is taken as a definition of frustration [6].
Random frustration One way to achieve frustration in a magnetic system is by introducing random- ness in the system. If spins interact ferromagnetically with some neighbours, but antiferromagnetically with others, it may not be possible to satisfy every bond on the lattice, possibly giving rise to frustration. Random frustration is for example present in spin glasses [5].
Geometrical frustration Geometrical frustration arises when all bonds between spins cannot be satisfied because of the lattice geometry. A simple example is the Ising antiferromagnet on a triangular lattice shown to the left in Figure 2.1. In this case, each spin wants both its neighbours to anti-align, which is not possible for all three spins on a single triangle. 2.2. SPIN ICE 5 ?
Figure 2.1: Examples of geometrical frustration. To the left is a triangular unit cell with antiferromagnetic interactions making one spin point up and one down. The last spin can be either up or down, being unable to satisfy both of its bonds. To the right is a primitive unit cell of the pyrochlore lattice. Here the spins interact ferromagnetically resulting in the frustrated ground state with two spins pointing into the tetrahedron and two spins pointing out of it.
In spin ice, the frustration arises from the geometry of the pyrochlore lattice (see Section 2.2.1). Essentially, four spins located on the vertices of a tetrahedron are forced to point either into or out of the tetrahedron. The spins interact ferro- magnetically meaning that if a spin points into the tetrahedron, it wants all other spins to point out of the tetrahedron, or vice versa. This is obviously not possible to satisfy for all four spins in the tetrahedron, and the compromise leading to the lowest energy of the system is two spins pointing into and two spins pointing out of the tetrahedron, as in the right panel of Figure 2.1.
2.2 Spin ice
2.2.1 The pyrochlore lattice and the ice rules Spin ice is the name given to a family of magnetically frustrated pyrochlore oxides 4+ 4+ such as Ho2Ti2O7, Dy2Ti2O7, Ho2Sn2O7 and Dy2Sn2O7. The Ti and Sn ions are non-magnetic, but the Ho3+ and Dy3+ ions are both rare earth elements with large magnetic moments in the order of 10µB [7]. Here, µB is the Bohr magneton, e defined as µB = ~ , where e is the elementary charge, is the reduced Planck’s 2me ~ constant and me is the electron mass. It is the large magnetic moment of the ions, the large crystal field splitting of the compounds and the fact that the magnetic ions reside on a pyrochlore lattice that give rise to the interesting properties of spin ice. The pyrochlore lattice is a face-centered cubic (fcc) Bravais lattice with a four site basis. The atoms are oriented such that each primitive unit cell forms a tetra- hedron by itself, and when several unit cells are combined we get a lattice of vertex- 6 CHAPTER 2. BACKGROUND THEORY
[001]
[010]
[100]
Figure 2.2: The pyrochlore lattice. The magnetic ions reside on the vertices of vertex-sharing tetrahedra. sharing tetrahedra as in Figure 2.2. Each tetrahedron’s center occupies a site on a diamond lattice [3]. The diamond lattice dual to the pyrochlore lattice is drawn in blue in Figure 2.3. Due to holmium and dysprosium being rare earth elements, the unpaired elec- trons contributing to their magnetic moment are of 4f type. The wavefunctions of these electrons do not stretch as far from the nucleus as for example 5s and 5p electrons [8] which are also present in Ho3+ and Dy3+. The relatively confined elec- tronic wavefunctions make the exchange interaction relatively small, in the order of a few kelvins, whereas the crystal field splitting is around 200 to 300 K [7]. The large crystal field completely eclipses the exchange interaction and forces the spins to point along their local 111 axes, that is, into or out of a tetrahedron. h i However, the exchange interaction still has a role to play. Whereas the crystal 2.2. SPIN ICE 7
Figure 2.3: The tetrahedra of the pyrochlore lattice form a diamond lattice. The diamond lattice sites are marked with blue spheres.
field pins the spins to the local 111 axes, it is the spin-spin interactions (mainly nearest neighbour interactions, suchh i as exchange) that determines whether a spin should point into or out of a tetrahedron. With the net nearest neighbour interac- tion being ferromagnetic [9], the energy of one spin pointing into a tetrahedron is minimised if the other spins of the tetrahedron point outwards. Thus it is impos- sible to minimise the energy of all nearest neighbour interactions simultaneously and the system is frustrated. The ground state is when two spins point into and two spins point out of each tetrahedron – the so-called ice rules or two-in-two-out rule [1] – and is massively degenerate. The net nearest neighbour interaction is known to be ferromagnetic since mea- surements of the Curie-Weiss temperature gave a positive ΘCW +1.9 K which implies ferromagnetism as noted in Section 2.1.1. Spin ice was the≈ first ferromag- netic system to show frustration – all previously known frustrated magnetic systems had been antiferromagnetic, such as the triangular Ising antiferromagnet mentioned previously [1]. 8 CHAPTER 2. BACKGROUND THEORY
The name spin ice and the ice rules come from an analogy to water ice. In water ice, the oxygen atoms reside on a diamond lattice with four hydrogen atoms tetrahedrally coordinated [10]. The hydrogen atoms are thus situated on the py- rochlore lattice, dual to the diamond lattice. In the ground state, two of these hydrogen atoms are covalently bound to the oxygen atom and two form hydrogen bonds – analogous to the two-in-two-out rules of spin ice. Thus arises geometrical frustration which gives rise to a large degeneracy of the water ice ground state.
2.2.2 Residual entropy Due to the degeneracy of the spin ice ground state, spin ice compounds – just like water ice – possess a zero-point entropy. This entropy was estimated for water ice by Pauling in 1935 [10] by estimating the number of possible ground states, Ω. For each tetrahedron there are four spins that can point either in or out, giving rise to 24 = 16 unique configurations. The ones obeying the ice rules, i.e. two-in- 4 two-out, can be calculated as 2 = 6. If we now take N tetrahedra, we have 2N spins and thus 22N possible configurations, but since only a fraction 6/16 of the states per tetrahedron obey the ice rules, we obtain
6 N 3N Ω = 22N = , (2.6) 16 2 which gives the entropy 3 S = ln Ω = N ln . (2.7) 2 Experiments have confirmed that spin ice compounds possess a residual entropy close to Pauling’s estimate [11, 12].
2.2.3 Magnetic monopoles As stated above, the ground state of spin ice compounds is a state obeying the ice rules two-in-two-out. The elementary excitation occurs when a single spin is flipped, creating one tetrahedron with three-in-one-out and one tetrahedron with one-in-three-out. However, this excited state is also degenerate, and we can con- tinue flipping certain spins without increasing the energy of the system. In this way we can separate the three-in-one-out and the one-in-three-out tetrahedra by a macroscopic distance as in Figure 2.4. In 2008, Castelnovo, Moessner and Sondhi [3] argued that these elementary excitation can be viewed as magnetic monopoles, carrying a magnetic charge and being under the influence of a magnetic Coulomb potential. Since only a finite amount of energy is required to separate the monopoles infinitely, the monopoles are deconfined, unlike similar excitations in ordered ferromagnets [3]. Whereas these excitations effectively can be seen as magnetic monopoles, they are not equivalent to the magnetic monopoles suggested by Dirac in 1931 [2]. The 2.3. MODELLING SPIN ICE 9
Figure 2.4: Magnetic monopoles in spin ice. A chain of flipped spins (in white) gives rise to two defects of three-in-one-out and one-in-three-out respectively. These monopoles are deconfined and can be separated indefinitely at a finite energy cost. so-called Dirac string connecting two magnetic monopoles must be unobservable for Dirac monopoles, but in the case of spin ice, the Dirac string is a chain of flipped spins (in white in Figure 2.4) connecting the magnetically charged tetrahedra and thus it is actually observable [3]. Furthermore, the monopole charge is not quantised in spin ice. The charge is defined as qm = 2µ/ad, where µ is the rare earth ion ± magnetic moment and ad is the diamond lattice parameter. Thus, it is possible to change the monopole charge by applying pressure to the sample – i.e. changing the lattice parameter. Dirac monopoles, on the other hand, must be quantised in units of h/µ0e where h is Planck’s constant, µ0 is the vacuum permeability and e is the elementary charge [2, 3]. Furthermore, magnetic monopoles in spin ice always appear in pairs of one positive and one negative monopole.
2.3 Modelling spin ice
2.3.1 The nearest neighbour model
The simplest way to describe the magnetic properties of spin ice materials is to employ the nearest neighbour model. This is essentially the Ising model [13] on a 10 CHAPTER 2. BACKGROUND THEORY pyrochlore lattice. The Hamiltonian of the Ising model is given by X H = J Si Sj, (2.8) − · hi,ji where i, j indicates summation over nearest neighbour spins Si and Sj. The spins areh vectorsi of unit length directed in the 111 directions. J is the interac- tion constant. If J > 0, the interaction is ferromagnetic,h i whereas if J < 0 it is antiferromagnetic. For the two most common spin ice compounds, Ho2Ti2O7 and Dy2Ti2O7, J equals 5.4 K and 3.3 K [7], respectively. The nearest neighbour model is used in all simulations throughout this thesis and is – despite its simplicity – surprisingly successful in reproducing the qualitative behaviour of real spin ice compounds.
2.3.2 The dipole model As one might suspect, it turns out that the nearest neighbour model is a simplifi- cation of reality. Due to the large magnetic moments of the Ho3+ and Dy3+ ions, long range dipole interactions are bound to play an important role in the behaviour of spin ice [9]. When accounting for dipole-dipole interactions, the Hamiltonian is given by [9]
X 3 X Si Sj 3 (Si rij)(Sj rij) H = J Si Sj + Drnn · 3 · 5 · , (2.9) − · rij − rij hi,ji j>i | | | | where D is an interaction constant for the dipole-dipole interaction, rnn is the nearest neighbour distance and rij is the vector pointing from site i to site j. Using the above model, it actually turns out that J < 0 – i.e. the exchange interaction is antiferromagnetic. D is, however, positive and large enough to make the net nearest neighbour interaction ferromagnetic [9]. −3 For simulations, the long range interactions due to the rij dependence of the dipole term are problematic. Attempts have been made to| truncate| the interaction after the 5th or 12th nearest neighbour [14, 15] but in these cases even the nearest neighbour model yields better results due to screening of medium- to long-range interactions in the real system. In order to properly account for the long range interactions, one often turns to Ewald summation techniques [9].
2.3.3 Effect of magnetic fields Applying a magnetic field to the sample is equivalent to adding a Zeeman term to the Hamiltonians of the respective models in Equations 2.8 and 2.9 given by [8] X HZ = µ0H µSi, (2.10) − · i 2.3. MODELLING SPIN ICE 11
where µ0 is the vacuum permeability, µ is the magnetic moment of the ions and H is the applied magnetic field. The spins will thus seek to align with the magnetic field in order to minimise their free energy. 12 Chapter 3
Monte Carlo simulations
This chapter is concerned with Monte Carlo algorithms. We introduce some general requirements on Monte Carlo algorithms and present the single spin flip Metropolis algorithm which is used in the simulations throughout the thesis.
3.1 Pseudo-random numbers
Monte Carlo simulations are based on random numbers and are therefore highly dependent on our ability to generate large amounts of random numbers on a com- puter. Unfortunately, computers are by construction deterministic in their nature, i.e. they are unable to generate pure random numbers. However, by making use of clever algorithms we can generate simulated random numbers – pseudo-random numbers [16]. Hence, when referring to random numbers throughout this thesis, we actually mean pseudo-random numbers.
3.2 Monte Carlo algorithms 3.2.1 Requirements on Monte Carlo algorithms In order to construct a Monte Carlo algorithm that properly describes reality we generally impose two requirements on the algorithm: it must be ergodic and it must obey detailed balance. Ergodicity is fulfilled if the system – being in any allowed state – can reach all other states in a finite time [16]. That is, all states of our system must be connected in phase space – we do not allow separate ”islands” of states. Ergodicity ensures that we sample a fraction of phase space that is representative of the entire phase space. The condition of detailed balance comes from the interpretation of Monte Carlo sampling as Markov chains. A Markov chain is a chain of stochastic events that is
13 14 CHAPTER 3. MONTE CARLO SIMULATIONS memoryless. This means that the transition probability from state n to state m at time t, Wn→m only depends on the current state of the system, n(t), and not on any previous states, n(t0 < t) [16]. The probability of being in state n at time t, Pn(t), obeys the master equation
dPn(t) X = [P (t)W P (t)W ] . (3.1) dt m m→n n n→m m −
The first term in the brackets of Equation 3.1 accounts for the change in Pn(t) due to transitions from a state m to the state n, whereas the last term represents the change due to transitions from n to m, hence the minus sign. Finally, we sum over all possible states in order to account for all possible transitions in the system1. dPn(t) In the steady state – equilibrium – the derivative, dt , must be zero. The simplest way to satisfy this is to say that all terms in the sum must equal zero, which leads to the condition for detailed balance [16]
Pn(t)Wn→m = Pm(t)Wm→n. (3.2)
Imposing detailed balance when choosing transition rates for a Monte Carlo al- gorithm ensures that we are sampling the equilibrium properties of the system. Detailed balance is thus a sufficient condition to fulfil the steady state master equa- tion. However, it is not a necessary condition – it is possible to fulfil the steady state master equation by mixing different terms in the sum of Equation 3.1, although this is considerably more complicated and generally, detailed balance is preferred.
3.2.2 Single spin flips Throughout our simulations we have used single spin flip methods. This means that the accepted moves are the flipping of single spins with a probability 0 < W 1. Since this probability is strictly non-zero, single spin flip algorithms are trivially≤ ergodic.
3.2.3 The Metropolis algorithm The algorithm of choice for our simulations of nearest neighbour interactions in spin ice is the Metropolis algorithm [17]. The essential part of the algorithm is to choose the transition rate according to
e−β∆E if ∆E > 0 W = (3.3) n→m 1 otherwise,
−1 where ∆E = Em En is the change in energy due to the spin flip and β = T is the inverse thermal− energy. While we for standard non-importance sampling Monte
1Note that for n = m, the two terms cancel so we have no problems with transitions to the same state. 3.2. MONTE CARLO ALGORITHMS 15
Carlo would weight the contribution from each configuration by its Boltzmann fac- tor, choosing the transition rate according to Equation 3.3 introduces an importance sampling and we instead weight all states equally, but sample the state with larger Boltzmann factor more often. Equation 3.3 also guarantees that detailed balance is obeyed. Since the probability of being in a state n is given by 1 P (t) = e−βEn , (3.4) n Z where Z is the partition function [16], Equation 3.2 reduces to 1 1 e−βEn e−β(Em−En) = e−βEm , (3.5) Z Z if Em En, or ≥ 1 1 e−βEn = e−βEm e−β(En−Em), (3.6) Z Z if Em En, both of which are trivially true, and thus detailed balance is obeyed. Schematically,≤ we can describe a Monte Carlo simulation of an Ising-like model utilising the Metropolis algorithm as follows:
1. Set up the lattice and assign each spin a random direction. 2. Perform a number of Monte Carlo steps to equilibrate the system.
3. Perform a number of Monte Carlo steps during which the quantities of interest are calculated.
A Monte Carlo step means 1. Select a random spin from your lattice.
2. Calculate the transition rate, Wn→m, according to Equation 3.3. 3. Generate a uniformly distributed random number 0 < r < 1.
4. If Wn→m > r, the spin flip is accepted. Else it is rejected and nothing happens. 5. For a system of N spins, repeat steps 1 to 4 N times.
3.2.4 Limitations of Metropolis dynamics A rule of thumb for optimising Monte Carlo algorithms is to have an acceptance rate of about 50% for the moves [18] – in our case the single spin flips. However, in the 16 CHAPTER 3. MONTE CARLO SIMULATIONS low temperature limit, β , the Metropolis transition rate goes to zero, meaning that barely any moves will→ ∞ be accepted making the algorithm very inefficient in sampling a representative part of the system states. In the limit of high temperatures, β 0, the transition rate rather goes to 1. This means that every spin flip will be→ allowed, which risks giving rise to larger correlations in time than should be the case. If one is to simulate high temperatures one may instead choose the Glauber algorithm, which instead uses the transition rate [16] 1 1 W = 1 + S tanh βE , (3.7) n→m 2 2 i n where Si = 1 is the spin being flipped and En is the energy of the original configuration.± The Glauber transition rate tends to 1/2 for large temperatures, avoiding the problem of the Metropolis algorithm at high temperatures. In other aspects, the Metropolis and Glauber algorithms do not differ significantly and they are both special cases of a more general transition rate [16]. Chapter 4
Magnetricity
In this chapter we treat direct and alternating currents of magnetic monopoles. We derive expressions for the monopole current and the AC susceptibility in the nearest neighbour model and calculate the AC susceptibility numerically when the field is applied in a 001 direction. Finally, we present a scheme to find the magnetisation and monopoleh currenti for an arbitrary time dependent magnetic field without using Monte Carlo simulations.
4.1 Magnetic monopole current
Due to the symmetry of Maxwell’s equations with respect to electric and magnetic fields, one may na¨ıvely expect magnetic monopoles in spin ice to behave like electric charges. This might lead us to believe that it is possible to drive a current of magnetic monopoles through a spin ice material. If we were to apply a constant magnetic field to a sample with periodic boundary conditions, we could get a steady state current through the sample – a magnetic equivalent of electricity, known to some as ”magnetricity” [19]. However, as can be seen from the Monte Carlo simulations of the monopole current presented in Figure 4.1, it turns out that the monopole current that emerges upon applying a magnetic field decays to zero exponentially with a characteristic relaxation time τ, both for periodic and open boundary conditions. It is important to keep in mind that while we in some respects indeed can view the elementary excitations in spin ice as magnetic monopoles, these excitations are not free particles. Every time a monopole passes through a plane, a spin in that plane is flipped in the direction of the current [20] as in Figure 4.2. Thus all monopole current contributes to magnetising the sample according to [21]
∂M j = , (4.1) mp ∂t
17 18 CHAPTER 4. MAGNETRICITY
0.012
0.008
0.004 Monopole current [a.u.]
0.000
0 20 40 60 80 100 Monte Carlo time
Figure 4.1: Monte Carlo simulation of the current of magnetic monopoles that appears upon applying a constant magnetic field at t = 0 at a temperature T/J = 0.2. Note that both in the case of open boundary conditions (red) and periodic boundary conditions (blue), the current decays to zero. For precise simulation parameters, please refer to Appendix A.
where jmp is the monopole current and M is the magnetisation of the sample. If we were to have a steady state current, the magnetisation of the sample would tend to infinity, which is obviously unphysical. A persistent direct current is thus out of the question. However, alternating currents with a large enough frequency, ω such that ωτ 2π, should remain relatively unaffected by the relaxation of the monopole current, provided that the system is able to respond swiftly enough to a change in the magnetic field.
4.2 Expression for the monopole current
Let us consider a system of non-interacting magnetic monopoles, equivalent to a nearest neighbour spin ice model. In the general case, we can write the entropy of a system slightly out of equilibrium as a quadratic form
1 X TS( αj ) = TS( 0 ) gikαi αk (4.2) { } { } − 2 · i,k 4.2. EXPRESSION FOR THE MONOPOLE CURRENT 19
Figure 4.2: An excited spin ice state projected on the (100) plane with a magnetic field applied along the [001] direction. Spins aligned with the field are black and spins anti-aligned are white. When a positive monopole (blue) moves in the direc- tion of the field, a spin is aligned with the field in the plane marked by the red rectangle, contributing to increasing the magnetisation of the sample. Eventually, the marked plane will be fully polarised and no positive monopoles can pass in the field direction. This effectively makes any direct monopole current transient.
where gik is a positive definite form [22] – that is, the eigenvalues of gik are all positive. In our case, the thermodynamic quantities, αj, are the magnetic field α1 = H and the magnetisation α2 = M. The form of Equation 4.2 is due to the fact that the entropy has a maximum at thermal equilibrium, meaning that the first derivatives with respect to αj must be zero and the second derivatives must be negative. Introducing the magnetic field and magnetisation into Equation 4.2, we get
2 2 TS(H, M) = TS(0, 0) C1H C2H M C3M , (4.3) − − · − where C1,C2,C3 are suitable constants. On differential form, taking dM as the differential, this becomes
T dS = C2H dM 2C3M dM. (4.4) − · − · As stated above, we require the entropy to have a maximum at equilibrium. Thus
T dS = (C2Heq + 2C3Meq) dM (4.5) − · must be zero for any infinitesimal change in magnetisation, dM,
C2Heq + 2C3Meq = 0. (4.6) 20 CHAPTER 4. MAGNETRICITY
Since we know that at equilibrium, Meq = χT Heq, where χT is the isothermal susceptibility, we get the relation 2C 1 3 = (4.7) C2 −χT for the constants. Let us now take the time derivative of the entropy in order to arrive at an expression for the entropy production [21]
∂S ∂M −1 T = C2 H χ M . (4.8) ∂t − ∂t · − T Returning to the formulation of the entropy in terms of the general state vari- ables, αj [22], we define the generalised currents as the time derivative of the state ∂αi variables, Ji , and the generalised forces as the linear combination of the ≡ ∂t P state variables, Xi k gikαk. If we now take the time derivative of Equation 4.2, we get ≡ −
∂S 1 X ∂αk 1 X ∂αi X ∂αi T = gikαi gik αk = gik αk. (4.9) ∂t −2 · ∂t − 2 ∂t · − ∂t · i,k i,k i,k This means that we can always write the entropy production as a product of gen- eralised currents and forces in the following manner ∂S X T = Ji Xi. (4.10) ∂t · i ∂M −1 In Equation 4.8, we identify J1 = jmp = ∂t and X1 = C2 H χT M . We will now make the assumption that the generalised forces− and currents obey the phenomenological equations [22] X Ji = LikXk (4.11) i where Lik are phenomenological constants. Since we only have one current, jmp, Equation 4.11 becomes −1 jmp = L11C2 H χ M . (4.12) − T Introducing the monopole mobility, u, the monopole concentration, c, and the monopole charge qm = 2µ/ad, where µ is the magnetic moment of the ions and ad is the lattice parameter of the diamond lattice dual to the pyrochlore lattice, we can use these to express the constants in Equation 4.12 as L11C2 = ucqmµ0 κ [23] and the final expression for the monopole current in the case of non-interacting≡ monopoles reads −1 jmp = κ H χ M . (4.13) − T The first term of Equation 4.13 looks rather familiar if we make an analogy with the transport of electric charges. The Drude model of electric transport tells 4.2. EXPRESSION FOR THE MONOPOLE CURRENT 21
us that the electric current density, jel, is proportional to the applied electric field, E, by the following relation jel = neueqeE (4.14) where ne is the carrier concentration, ue is the carrier mobility and qe is the charge of the carrier. Exchanging electric field for magnetic field, E µ0H, electric → current for magnetic current, jel jmp and electric carrier concentration, mobility → and charge for their magnetic equivalents, ne c, ue u, qe qm, Equation 4.14 is equivalent to the first term of Equation 4.13.→ → → The origin of the second term in Equation 4.13 cannot be inferred from an electric equivalent as easily as the first term. Whereas the magnetic Drude force works to increase the current, this entropic force works to reduce the current, thus effectively making any direct current of magnetic monopoles transient (see Section 4.2.1). The entropic contribution arises from the attempt to minimise the free energy of the system. Since the Helmholtz free energy is given by F = U TS, where U is the internal energy, T is the temperature and S is the entropy,− an increase in entropy serves to reduce the free energy of the system. When the system is magnetised, the internal energy is reduced because of the Zeeman term of the Hamiltonian (Equation 2.10). However, the system also gets more ordered with increasing magnetisation, reducing the entropy. In the fully polarised state, all the residual entropy of the system has vanished – the ground state lacks degeneracy. Equation 4.13 can thus be seen as the competition between internal energy and entropy.
4.2.1 Magnetisation
∂M Since jmp = ∂t , Equation 4.13 is actually a first order linear differential equation for the magnetisation ∂M = κ H(t) χ−1M(t) , (4.15) ∂t − T with the general solution
t Z 0 M(t) = e−κt/χT eκt /χT κH(t0, )dt0 + M(0) (4.16) 0 which gives the current
t κ Z 0 −κt/χT κt /χT 0 0 jmp = κH(t) e e κH(t )dt + M(0) . (4.17) − χT 0
If we look at the special case of a constant magnetic field H0 applied to a sample at thermal equilibrium, i.e. M(0) = 0, at t = 0, we get −κt/χT M(t) = χT H0 1 e (4.18) − 22 CHAPTER 4. MAGNETRICITY
−κt/χT jmp = κH0e . (4.19) We see that the monopole current relaxes and the magnetisation rises exponentially with a relaxation time τ χT /κ. ≡ 4.2.2 AC susceptibility In order to investigate the behaviour of spin ice in an alternating magnetic field, we take interest in the AC susceptibility, χ(ω), defined via
M(ω) = χ(ω)H(ω). (4.20)
To obtain an expression for χ(ω), we transform Equation 4.15 to Fourier space
iωM(ω) = κ H(ω) χ−1M(ω) . (4.21) − T Using M(ω) = χ(ω)H(ω) and rearranging terms, we get the following expression for the AC susceptibility [21] χ χ(ω) = T (4.22) 1 + iωτ where the relaxation time again is given by τ χT /κ. Extracting the real (in- phase) and imaginary (out-of-phase) parts of χ(≡ω) = χ0(ω) iχ00(ω), we have − χ χ0(ω) = T (4.23) 1 + ω2τ 2 χ ωτ χ00(ω) = T . (4.24) 1 + ω2τ 2 This form of the susceptibility is known as the Debye model, which was originally proposed for modelling the permittivity of dielectrics.
4.2.3 Analogy to an RL-circuit Equation 4.18 for the magnetisation resembles the expression for the current in a −t/τ RL-circuit biased with a constant voltage switched on at t = 0, i(t) = i0(1 e ). Thus we turn to a model of a RL-circuit biased with an alternating voltage− in order to see if we can model spin ice by a simple circuit model. The current-voltage relationship in this circuit is v = Z(ω)i (4.25) where i is the current, v is the voltage and Z(ω) is the complex impedance. If we now make the substitution v(ω) H(ω) I(ω) → M(ω) (4.26) Z−1(ω) → χ(ω) → 4.3. SPIN-LATTICE RELAXATION 23 we get back M(ω) = χ(ω)H(ω). This leads us to investigate whether or not we can model spin ice susceptibility data with the expression for the impedance of the RL-circuit. The expression for the impedance is
Z(ω) = R + iωL (4.27) where R is the DC resistance and L is the circuit inductance. Introducing the time constant of the circuit as τ = L/R, we can write
Z(ω) = R (1 + iωτ) . (4.28)
The inverse impedance is now given by
1 1 1 iωτ = − . (4.29) Z(ω) R 1 + (ωτ)2
If we identify the inverse resistance with the isothermal susceptibility, χT , we get the following expressions for the real and imaginary parts of the susceptibility χ χ0(ω) = T (4.30) 1 + ω2τ 2 χ ωτ χ00(ω) = T (4.31) 1 + ω2τ 2 which are identical to Equations 4.23 and 4.24.
4.3 Spin-lattice relaxation
A refinement of the above model can be made by considering the coupling between the lattice vibrations and the spins [24]. When an external magnetic field is applied to the system, this results in an almost instantaneous magnetisation [23, 25], resulting in a shift in the population of spin up and spin down states. However, this shift requires the release of energy. In a thermally isolated system, there is no way to release this energy and thus, instead of shifting the populations, the temperature of the spin system would increase from T1 to T2 such that H/T1 = (H + ∆H)/T2, where H is the initial external field and ∆H is the change in the field [24]. Now, the spins are not thermally isolated, but the heat transport from the spin system to the lattice is slower than the magnetisation of the sample, giving rise to a slightly higher temperature of the spin system. We model the spin-lattice relaxation by assuming that the exchange of heat is proportional to the temperature difference between the spin system and the lattice [23–25], i.e. dQ = α (T Tl) = αθ, (4.32) dt − − − 24 CHAPTER 4. MAGNETRICITY
where T is the temperature of the spin system, Tl is the temperature of the lattice and α is a proportionality constant. Let us now express the entropy in terms of temperature and magnetisation ∂S ∂S T dS = T dT + T dM. (4.33) ∂T M ∂M T We now make use of the identity dQ ∂U ∂S CM = = = T (4.34) dT M ∂T M ∂T M and the Maxwell relation ∂H ∂S = (4.35) ∂T M − ∂M T and write Equation 4.33 as ∂H ∂H T dS = CM dT T dM = CM dT T χdH. (4.36) − ∂T M − ∂T M For a reversible process, dQ = T dS. Using Equation 4.32, we get ∂H αθdt = CM dT T χdH. (4.37) − − ∂T M iωt Let now the magnetic field vary sinusoidally, H = H0e . As a consequence, the iωt temperature difference will vary in the same manner: θ = θ0e . Dividing Equation 4.37 by dt and inserting the explicit time dependences of the state variables, we arrive at ∂H αθ0 = iω CM θ0 T χH0 . (4.38) − − ∂T M In order to get an expression for the susceptibility, χ, we set out to eliminate θ0 in Equation 4.38. This is done by writing the magnetisation in terms of the magnetic field and the temperature ∂M ∂M dM = dH + dT (4.39) ∂H T ∂T H and differentiate with respect to time ∂M ∂M χH0 = H0 + θ0. (4.40) ∂H T ∂T H
Eliminating θ0 from Equation 4.38, we get ∂M ∂H ∂M (α + iωCM ) χ = iωT χ (4.41) − ∂H T ∂T M ∂T H 4.3. SPIN-LATTICE RELAXATION 25 where we can use the thermodynamic relation [24]
∂H ∂M CH CM = T (4.42) − − ∂T M ∂T H resulting in
∂M (α + iωCM + iω (CH CM )) χ = (α + iωCM ) (4.43) − ∂H T