The Core Composition of Terrestrial Planets:
A Study of the Ternary Fe-Ni-Si System
Elizabeth Wann
UCL
A thesis submitted to University College London for the degree of
Doctor of Philosophy.
August 2015 2 I, Elizabeth Wann, confirm that the work presented in this thesis is my own.
Where information has been derived from other sources, I confirm that this has been
indicated in the thesis.
3 Abstract
The exact composition of the cores of terrestrial planets is not known, but it is gen- erally agreed that they are composed of iron alloyed with a fraction of nickel plus a small percentage of a light element, likely Si, S, O, C or H. Silicon has long been a popular choice and is still regarded as a very likely candidate, based on density deficit and cosmochemical arguments. Although much work has been carried out on the Fe-Si system, studies on the Fe-Ni-Si system have only recently been carried out. The major- ity of studies have concentrated on specific candidate core compositions, based on core formation models or matching the observed density deficit. This can be problematic when core formation models depend on core composition. In this thesis, the Fe-Ni-Si system is investigated as a whole, starting with the end-member binary systems, FeSi and NiSi. This provides a more methodical approach to solving the core composition problem.
Both ab initio calculations and high-pressure, high-temperature experiments have been used in this work. Ab initio calculations at 0 K were used to find the transition pressure of the "-FeSi to CsCl-FeSi phase transition, and also to test the stability of newly discovered NiSi-structured phases in FeSi. Lattice dynamics calculations at high temperatures and pressures have been carried out to determine the Clapeyron slope of the "-FeSi to CsCl transition, in both FeSi and NiSi systems. Laser-heated diamond anvil cell experiments were used to measure the melting curves of NiSi and the
Fe-FeSi eutectic, and in-situ neutron di↵raction experiments were used to determine the equation of state of MnP-structured NiSi at high-pressure and high-temperature.
Finally, X-ray di↵raction experiments were used to measure the thermal expansion of a range of (Fe,Ni)Si alloys.
4 Acknowledgements
This PhD would not have been possible without the support and advice of my super- visors, Prof. Lidunka Voˇcadlo and Prof. Ian Wood. Their help and expertise were invaluable and I feel extremely grateful to have had such excellent supervisors. I am particularly grateful to Lidunka for her kindness and patience and whose encourage- ment pushed me forward to accomplishing this work.
I would also like to thank Dr. Oliver Lord, for all his guidance with the diamond anvil cell experiments, and Dr. Benjam´ıMartorell for his helpful advice with VASP. Thank you also to Prof. John Brodholt, Prof. David Dobson and Prof. Dario Alf´e,as well as Dr. Simon Hunt and Dr. Alex Lindsay-Scott for their help and input during my
PhD.
Finally, my thanks go to my friends and family for their support and encouragement.
To Becky, Adam, Jo and Amanda for sharing the highs and lows of my PhD and to
Amy for making each conference more fun and enjoyable. To my family, for their words of encouragement and for always having faith in me. And lastly to Stephen, for being a constant source of support and for being there for me always.
This PhD was funded by the Science & Technology Facilities Council, and made use of the facilities of HECToR and ARCHER (http://www.archer.ac.uk), the UK’s national high-performance computing service, which was provided by UoE HPCx Ltd at the
University of Edinburgh, Cray Inc and NAG Ltd, and funded by the O ce of Science and Technology through EPSRC’s High End Computing Programme.
5 6 Table of Contents
Abstract 4
Acknowledgements 5
Table of Contents 7
List of Figures 13
List of Tables 19
1 Introduction 21
1.1 Introduction ...... 21
1.2 The Earth’s Inner Core ...... 23
1.3 Seismology ...... 25
1.3.1 Anisotropy and Layering of the Inner Core ...... 26
1.4 The Terrestrial Planets ...... 28
1.4.1 The Moon ...... 28
1.4.2 Mars ...... 29
1.4.3 Mercury ...... 33
1.4.4 Venus ...... 37
1.4.5 Summary of Terrestrial Planetary Cores ...... 38
1.5 The Case For Silicon ...... 38
1.5.1 Silicon in the Earth’s Inner Core ...... 39
1.5.2 Seismological Evidence for Silicon ...... 40
1.5.3 The E↵ect of Si on the Structure of Iron ...... 41
1.6 E↵ect of Nickel ...... 45
1.7 Investigations on the Fe-Ni-Si System ...... 47
2 Computational Methods 49
2.1 Ab initio ...... 49
2.1.1 The Schr¨odinger Equation ...... 50
2.1.2 The Born-Oppenheimer Approximation ...... 50
7 2.1.3 The Many Electron Problem ...... 51
2.1.4 Independent Electrons ...... 51
2.1.5 Indistinguishable Electrons ...... 52
2.1.6 Self-Consistency ...... 53
2.1.7 The Hartree Method ...... 54
2.1.8 Hartree-Fock Theory ...... 55
2.2 Density Functional Theory (DFT) ...... 56
2.2.1 Hohenberg-Kohn Theorems ...... 56
2.2.2 Kohn-Sham Electrons ...... 57
2.2.3 LDA and GGA ...... 58
2.3 Plane Waves and Basis Sets ...... 58
2.4 k-point Sampling ...... 60
2.5 Pseudopotentials ...... 60
2.6 Projector Augmented Wave Method ...... 61
2.7 Ab initio Packages ...... 62
2.8 Static Calculations ...... 63
2.8.1 Geometry Optimisations ...... 63
2.8.2 Birch-Murnaghan Equation of State ...... 65
2.9 High Temperature Calculations ...... 66
2.9.1 Molecular Dynamics ...... 66
2.9.2 Lattice Dynamics ...... 67
2.9.3 Gibbs Free Energy ...... 68
2.9.4 Phase Diagram ...... 71
2.10 Summary ...... 72
3 Experimental Methods 73
3.1 High Pressure Experiments ...... 73
3.1.1 Shock Experiments ...... 74
3.1.2 Compression Experiments ...... 74
3.1.3 The Multi-Anvil Press (MAP) ...... 75
8 3.1.4 The Diamond Anvil Cell (DAC) ...... 77
3.2 O✏ine DAC Melting Experiments ...... 78
3.2.1 Sample Preparation ...... 79
3.2.2 Pressure Determination ...... 82
3.2.3 Heating the Cell ...... 83
3.3 X-Ray and Neutron Di↵raction ...... 85
3.3.1 Bragg’s Law ...... 86
3.3.2 Di↵use Scattering ...... 88
3.3.3 Analysis of Powder Di↵raction Data ...... 89
3.3.3.1 Rietveld Method ...... 89
3.3.3.2 Le Bail Method ...... 90
3.3.3.3 GSAS ...... 91
3.4 Thermal Expansion ...... 91
3.5 Synchrotron Experiments ...... 92
3.5.1 Introduction ...... 92
3.5.2 In situ Experiments ...... 93
3.5.3 ESRF DAC Experiments ...... 94
3.5.4 NSLS MAP Melting Experiments ...... 95
3.5.5 ISIS Equation of State Measurements ...... 95
3.6 Summary ...... 97
4 The Calculated FeSi Phase Diagram 99
4.1 Introduction ...... 99
4.2 The FeSi Phase Transition at 0 K ...... 100
4.3 Static FeSi Calculations ...... 102
4.3.1 VASP Calculations ...... 103
4.3.2 CASTEP Calculations ...... 104
4.3.3 Abinit Calculations ...... 104
4.4 Di↵erences in Transition Pressure ...... 105
4.5 The FeSi Phase Transition at High Temperatures ...... 107
9 4.6 Experiments on the FeSi Phase Transition ...... 107
4.7 Lattice Dynamics Calculations of FeSi ...... 109
4.8 FeSi Phase Diagram ...... 113
4.9 Conclusions ...... 115
4.10 Further Work ...... 115
5 Calculated Stabilities of NiSi-structured Phases in FeSi 117
5.1 Introduction ...... 117
5.2 NiSi Phases ...... 119
5.2.1 The MnP Phase ...... 119
5.2.2 The ‘anti-MnP’ Phase ...... 120
5.2.3 The Pbma-I Phase ...... 120
5.2.4 The WC Structure ...... 121
5.2.5 The Pmmn Phase ...... 121
5.3 VASP Calculations ...... 122
5.4 Conclusions ...... 124
6 The Calculated "-FeSi CsCl Phase Transition in NiSi 127 ! 6.1 Introduction ...... 127
6.2 The "-FeSi CsCl Phase Transition ...... 127 ! 6.3 Lattice Dynamics Calculations ...... 129
6.4 Conclusions ...... 134
7 NiSi Melting 137
7.1 Introduction ...... 137
7.2 Methods ...... 137
7.2.1 O↵-line LH-DAC Melting Experiments ...... 138
7.2.2 In situ LH-DAC Melting Experiments ...... 138
7.2.3 In situ MAP Experiment ...... 139
7.3 The Melting Curve of NiSi ...... 140
7.4 Conclusions ...... 143
10 8 Fe-FeSi Eutectic Melting 145
8.1 Introduction ...... 145
8.2 The Fe-Si Phase Diagram ...... 145
8.3 The Melting Curves of Fe and FeSi ...... 148
8.4 Experimental Methods ...... 150
8.5 The Fe-FeSi Eutectic Melting Curve ...... 152
8.6 Conclusions ...... 156
9 Equation of State for MnP-NiSi 159
9.1 Introduction ...... 159
9.2 The MnP Phase ...... 159
9.3 Neutron Di↵raction Experiments ...... 160
9.4 Birch-Murnaghan Equation of State ...... 164
9.5 Cell Parameters of MnP-NiSi ...... 167
9.6 Fractional Co-ordinates of Ni and Si ...... 170
9.7 Conclusions ...... 172
10 Thermal Expansion of (Fe,Ni)Si Alloys 173
10.1 Introduction ...... 173
10.2 Experimental Details ...... 173
10.3 Results ...... 174
10.4 Conclusions ...... 183
11 Conclusions 185
11.1 Summary of Results ...... 185
11.2 Implications for the Cores of the Terrestrial Planets ...... 186
Appendix A Convergence Tests 191
A.1 VASP Convergence Tests ...... 191
A.2 CASTEP Convergence Tests ...... 192
A.3 Abinit Convergence Tests ...... 193
11 Appendix B Further Work on FeSi 194
B.1 Crystal Structure of the "-FeSi Phase ...... 194
B.2 Equations of State ...... 194
References 199
12 List of Figures
1.1 Comparison of the terrestrial planets ...... 22
1.2 Three-shell model of the Earth ...... 23
1.3 Sound velocities as a function of density for metallic elements ...... 24
1.4 The PREM model ...... 26
1.5 Diagrams of upper and lower inner core of the Earth showing ...... 27
(a) non-uniform layering of inner core ...... 27
(b) observed seismic anisotropy ...... 27
1.6 Data collected by MGS of Mars’ magnetic field ...... 30
1.7 Three possible core models for Mars ...... 31
1.8 Areotherm of Mars ...... 32
1.9 Di↵erentiation model of Mars ...... 33
1.10 Comparison of internal structures of the Earth and Mercury ...... 34
1.11 Melting curves of Fe-S alloys ...... 35
1.12 Two possible snowing core models for Mercury ...... 36
1.13 ‘Hit and run’ simulation of the formation of Mercury ...... 37
1.14 V values against density for a range of iron alloys ...... 41
1.15 Premelting e↵ects of VP and VS observed in hcp-Fe ...... 42 1.16 Structure of "-FeSi ...... 43
1.17 Phase diagram of Fe-7.9Si ...... 44
1.18 Phase diagram of Fe-3.4Si ...... 44
1.19 Phase diagrams of Fe, Fe-16Si, Fe-9Si and FeSi ...... 45
1.20 Seismic wave velocities for a variety of Fe-Ni-Si alloys ...... 46
1.21 Phase diagrams for Fe-5Ni, Fe-15Ni and Fe-20Ni ...... 47
2.1 Representation of the Aufbau principle ...... 53
2.2 Di↵erence between a pseudopotential and an all-electron potential wave-
function...... 61
2.3 Potential energy as a function of atomic separation ...... 64
13 2.4 Phase transition of iron (bcc hcp) ...... 65 ! 2.5 Di↵erences between lattice dynamics and molecular dynamics ...... 67
2.6 Total Helmholtz free energies of CsCl-FeSi ...... 70
2.7 Pressure against volume values for CsCl-FeSi ...... 70
3.1 Schematic of a double-stage light-gas gun ...... 74
3.2 The first multi-anvil press ...... 75
3.3 Diagram of truncated cubic anvils around octahedral sample cell . . . . 76
3.4 Schematic of D-DIA multi-anvil press ...... 77
3.5 Schematic of a diamond anvil cell ...... 77
3.6 Diagram of a screw-type diamond anvil cell ...... 79
3.7 Princeton-type diamond anvil cell; used for melting experiments . . . . 79
3.8 Fe-FeSi disc; sample used for melting experiments ...... 81
3.9 Image showing cut-out Fe-FeSi discs ...... 81
3.10 Plot showing ruby fluorescence peak ...... 82
3.11 Setup for measuring ruby fluorescence at Bristol ...... 83
3.12 Setup of melting equipment at Bristol ...... 83
3.13 Diagrammatic representation of spectroradiometry ...... 85
3.14 Laser power versus temperature graph ...... 85
3.15 Diagrammatic representation of Bragg’s law ...... 87
3.16 Diagram showing systematic absences of fcc structure ...... 88
3.17 X-ray di↵raction pattern showing di↵use scattering ...... 89
3.18 Neutron di↵raction pattern of epsomite analysed by the Rietveld method 90
3.19 Screenshot of graphical user interface for GSAS ...... 91
3.20 Diagram of synchrotron at ESRF ...... 93
3.21 Set-up of anvils in the MAP, showing X-ray transparent anvils . . . . . 94
3.22 Image of hydraulic press installed at beamline X17B2, NSLS ...... 96
4.1 Structure of ...... 100
(a) CsCl-FeSi ...... 100
(b) "-FeSi ...... 100
4.2 Plots showing the "-FeSi CsCl-FeSi transition of ...... 101 !
14 (a) Voˇcadlo et al. (1999) ...... 101
(b) Caracas and Wentzcovitch (2004) ...... 101
4.3 Plot of enthalpy against pressure showing phase transition of "-FeSi to
CsCl-FeSi as calculated by VASP ...... 103
4.4 Plot of enthalpy against pressure showing phase transition of "-FeSi to
CsCl-FeSi as calculated by CASTEP ...... 104
4.5 Plot of enthalpy against pressure showing phase transition of "-FeSi to
CsCl-FeSi as calculated by Abinit ...... 105
4.6 Phase diagram of FeSi from LH-DAC experiments ...... 108
4.7 Phase diagrams of FeSi determined experimentally by ...... 109
(a) Fischer et al. (2013) ...... 109
(b) Geballe and Jeanloz (2014) ...... 109
4.8 Graphs of thermodynamic properties for CsCl-FeSi and "-FeSi at 200 K
ofthe ...... 112
(a) Helmholtz free energy ...... 112
(b) Pressure ...... 112
(c) Gibbs free energy ...... 112
4.9 Phase diagram of FeSi with calculated "-FeSi CsCl boundary . . . . . 114 ! 4.10 Calculated phonon dispersion curve of "-FeSi ...... 114
5.1 Calculated phase diagram of NiSi at 0 K ...... 118
5.2 Structure of the MnP phase ...... 119
5.3 Structure of the P bma-I phase ...... 120
5.4 Structure of the WC phase ...... 121
5.5 Structure of the P mmn phase ...... 122
5.6 Plots of enthalpy against pressure for di↵erent FeSi structures ...... 123
5.7 Phase diagram at 0 K of di↵erent transition metal silicides ...... 125
6.1 Phase diagram of NiSi at 0 K, with highlighted "-FeSi to CsCl-FeSi phase
transition ...... 128
6.2 Graphs of thermodynamic properties for "-FeSi and the CsCl phase at
500 K of the ...... 132
15 (a) Helmholtz free energy ...... 132
(b) Pressure ...... 132
(c) Gibbs free energy ...... 132
6.3 Phase diagram of NiSi with calculated phase boundary of the "-FeSi ! CsCl phase transition ...... 134
7.1 Screencaps of X-ray videography ...... 139
(a) before melting ...... 139
(b) after melting ...... 139
7.2 Temperature versus laser power plot of NiSi and corresponding XRD
patterns showing appearance of di↵use scattering ...... 140
7.3 Di↵use scattering of NiSi seen in XRD pattern from NSLS ...... 141
7.4 Melting curve of NiSi ...... 142
7.5 XRD patterns showing disappearance of P mmn peaks ...... 143
8.1 Phase diagrams of Fe-Si ...... 146
(a) at ambient pressure ...... 146
(b) at 21 GPa ...... 146
8.2 Phase diagrams of ...... 147
(a) Fe-9Si and Fe-16Si ...... 147
(b) Fe-18Si ...... 147
8.3 Fe-Si phase diagrams at 50, 80, 125 and 145 GPa ...... 148
8.4 Melting curve of FeSi ...... 149
8.5 Melting curve of Fe ...... 150
8.6 Laser power versus temperature plot of the Fe-FeSi eutectic at 44 GPa . 151
8.7 Screenshot of 2D analysis software used in melting experiments . . . . . 152
8.8 The Fe-FeSi eutectic melting curve ...... 155
9.1 Fitted neutron di↵raction pattern of MnP-NiSi ...... 161
9.2 The Equation of State of MnP-NiSi ...... 165
9.3 Comparison of computational and experimental EoS of MnP-NiSi . . . . 167
9.4 Relative cell parameters of MnP-NiSi ...... 169
16 (a) as a function of pressure ...... 169
(b) as a function of temperature ...... 169
9.5 Development of the MnP-NiSi structure with pressure ...... 170
9.6 Experimental and calculated fractional co-ordinates of Ni and Si of MnP-
NiSi for ...... 171
(a) Ni x ...... 171
(b) Ni z ...... 171
(c) Si x ...... 171
(d) Si z ...... 171
10.1 X-ray di↵raction pattern of (Fe0.9Ni0.1)Si at 300°C ...... 174
10.2 Thermal expansion of (Fe0.9Ni0.1)Si, (Fe0.8Ni0.2)Si and FeSi ...... 177 10.3 Comparison of unit cell volumes of FeSi with that of Voˇcadlo et al. (2002)178
10.4 Thermal expansion coe cient by numerical di↵erentiation ...... 180
(a) for (Fe0.9Ni0.1)Si, (Fe0.8Ni0.2)Si and FeSi ...... 180 (b) for FeSi, from neutron di↵raction ...... 180
10.5 XRD patterns of FeSi ...... 181
(a) showing the appearance of additional peaks at 900°C ...... 181
(b) indexed to identify extra peaks ...... 181
10.6 Volume ratios of (Fe0.9Ni0.1)Si and (Fe0.8Ni0.2)Si to FeSi as a function of temperature ...... 182
11.1 Adiabats of Mercury and Mars plotted alongside the Fe-FeSi eutectic
melting curve ...... 188
A.1 VASP convergence tests for the ...... 191
(a) k-point grid of CsCl-FeSi ...... 191
(b) k-point grid of "-FeSi ...... 191
(c) plane wave cut-o↵ energy ...... 191
A.2 CASTEP convergence tests for the ...... 192
(a) k-point grid of CsCl-FeSi ...... 192
(b) k-point grid of "-FeSi ...... 192
17 (c) plane wave cut-o↵ energy ...... 192
A.3 Abinit convergence tests for the ...... 193
(a) k-point grid of CsCl-FeSi ...... 193
(b) k-point grid of "-FeSi ...... 193
(c) plane wave cut-o↵ energy ...... 193
B.1 Graphs of experimentally measured EoS for ...... 196
(a) CsCl-FeSi ...... 196
(b) "-FeSi ...... 196
B.2 Graphs of calculated EoS for ...... 198
(a) CsCl-FeSi ...... 198
(b) "-FeSi ...... 198
18 List of Tables
2.1 The pressure and Gibbs free energy of CsCl-FeSi at selected temperatures
and volumes ...... 71
4.1 The calculated transition pressures for the "-FeSi CsCl-FeSi phase ! transition ...... 106
4.2 Calculated transition pressures for "-FeSi CsCl-FeSi ...... 113 !
5.1 Sizes of k-point grids used for the NiSi-structured phases in FeSi . . . . 122
5.2 Fitting parameters for the third-order BMEoS fits of the NiSi-structured
phases in FeSi ...... 124
6.1 NiSi phases found by Wood et al. (2013) ...... 129
6.2 Calculated transition pressures for the "-FeSi CsCl phase transition ! inNiSi...... 133
7.1 Fitting parameters for the Simon-Glatzel equation of NiSi ...... 141
8.1 Melting temperatures from Fe-FeSi melting experiments ...... 153
8.2 Melting Points Attributed to Fe-FeSi Eutectic Melting ...... 154
9.1 Room temperature lattice cell parameters of NaCl and NiSi ...... 162
9.2 The high-temperature, high-pressure lattice cell parameters of NaCl and
NiSi ...... 163
9.3 Fitting parameters for the room temperature Birch-Murnaghan equation
of state of MnP-NiSi ...... 165
9.4 Fitting parameters for the thermal Birch-Murnaghan equation of state
of MnP-NiSi ...... 166
10.1 Unit Cell Volumes of FeSi, (Fe0.9Ni0.1)Si and (Fe0.8Ni0.2)Si ...... 174 10.2 Thermal expansion co-e cients for the (Fe,Ni)Si alloys ...... 179
B.1 Fractional co-ordinates of "-FeSi at room pressure ...... 194
19 B.2 The calculated fitting parameters for the third-order BMEoS fits of CsCl-
FeSi and "-FeSi ...... 197
B.3 The experimental fitting parameters for the third-order BMEoS fits of
CsCl-FeSi and "-FeSi ...... 197
20 CHAPTER 1
Introduction
The stars began to burn through the sheets of clouds
— Mary Oliver
The study of the planets has been driven by a desire to understand the natural world, and the planets closest to our own have held a particular interest. These terrestrial planets are very similar to the Earth in composition and structure, and research into them can inform research into the Earth. Despite this, a lot of questions about Earth and the terrestrial planets are still unanswered. In particular, there is still a lot un- known about the cores of these planets. A relative lack of data on the core, combined with the extreme conditions under which the cores are found, have hindered advances, and the exact composition of the cores of terrestrial planets is still unknown. Deter- mining the composition, structure and dynamics of the planetary cores would aid our understanding of many other processes of the planets, answering questions about the core dynamo and formation of the planets.
1.1 Introduction
The terrestrial planets are so named because of their likeness to the Earth. Mars, Venus and Mercury have a similar composition to the Earth, having formed at roughly the same time nearly four and a half billion years ago. Sharing a similar bulk composition and conditions of formation mean that the internal structure of the terrestrial planets share similarities. As Figure 1.1 shows, each planet possesses a dense metallic core surrounded by a silicate mantle and a thin crust making up the surface of the planet.
The core formed when the much denser metal sank to the centre, separating from the
21 1.1. Introduction lighter silicate material. From consideration of the bulk composition, it is known that the Earth’s core is made up of mainly iron alloyed with a fraction of nickel along and a small percentage of a light element (Birch, 1952). The identity (or identities) of the light element, however, is still unknown.
Figure 1.1: Although the terrestrial planets vary in size they all have the same structural features, with the core making up a di↵erent proportion of each planet (from NASA (2015)).
Research on the composition of the cores of the terrestrial planets has mainly focused on binary Fe-X systems, where the X is a light element such as S, Si, O or C (Lord et al.,
2009; Mookherjee et al., 2011; Asanuma et al., 2008). However, recent experiments and calculations have found that under some conditions relevant to planetary cores, nickel has a significant e↵ect on the physical properties of the iron system (Mao et al.,
2006; Voˇcadlo et al., 2006; Martorell et al., 2013b), and therefore cannot be left out of investigations on the properties of the core. Nickel has long been accepted to be present in the cores of terrestrial planets, from consideration of the Fe/Ni ratio in chondritic meteorites (Birch, 1952; McDonough and Sun, 1995). There have been a number of studies investigating the ternary Fe-Ni-X systems (Antonangeli et al., 2010; Narygina et al., 2011; Sakai et al., 2011). However, these studies have tended to focus on specific candidate core compositions within the ternary system. Since the precise proportions in the planetary cores are not known definitively, these studies have an inherent limitation.
More useful, instead, would be a fuller understanding of the ternary system, ideally with data and calculations mapping the entire phase diagram of the ternary system. This is the approach adopted in this work, used to study the Fe-Ni-Si system. Silicon has long been considered a likely light element in the Earth’s core, having first been suggested as a candidate by Birch (1952) due to its abundance in the Earth and the fact that only a small percentage is required to lower the density su ciently (see Section 1.2).
22 Chapter 1. Introduction
Silicon is still widely considered as a possibility for the light element present in the cores of the terrestrial planets (see Section 1.5) and it is for this reason that this work focuses on the Fe-Ni-Si system.
1.2 The Earth’s Inner Core
Of all the terrestrial planets, the Earth’s core is most understood, primarily because of the ability to directly observe the Earth’s core using seismic waves. The Earth’s inner core was first discovered by Inge Lehmann in 1936 (see Figure 1.2). Lehmann noticed abnormally large amplitudes in the P-wave shadow zone, which could only be explained by the presence of a solid inner core. Assuming a simple three-shell model of the Earth, Lehmann calculated travel times for these waves, finding that they matched extremely well with those observed (Bolt, 1987), proving her hypothesis.
Figure 1.2: Lehmann’s theory and proof of the three-shell model of the Earth explaining the apparently anomalous seismic waves. Shown are the PKIIKP and PKiKP seismic waves. Figure taken from Bolt (1987).
After the initial discovery of the Earth’s inner core, the focus turned to what the core was made of. Measurements carried out by Birch (1937) revealed that the velocity of seismic waves was a↵ected by the density of the material. A comparison of the seismic wave velocities of the Earth’s core and that of iron showed a mismatch (see Figure 1.3) meaning that the Earth’s core cannot be made of just iron. Further investigation re- vealed that the density of the core was approximately 10 wt.% lower than that expected for pure iron (Birch, 1964). This density deficit could only be explained by the presence
23 1.2. The Earth’s Inner Core of light elements, such as silicon, sulphur or oxygen, in the core.
Figure 1.3: Sound velocities as a function of density for metallic elements and the Earth’s mantle and core from Birch (1937). The closest match for the density of the core is iron, but a core consisting only of iron would not produce the correct sound velocity.
Confirming the identity of the light element is di cult since there is no way to directly sample the Earth’s core. However, there are some pieces of information that can aid in identification such as, for example, details about core formation. It is generally agreed that the core formed during early accretion of the Earth, therefore the light element in the core must have been able to partition into liquid iron at relatively low pressures.
In addition, the light element must have been su ciently abundant during this time so that even after volatilisation by the heat generated during accretion it did not escape completely. The current status of the Earth’s core must also be taken into account. For example, the continuing crystallisation of the inner core is thought to drive convection in the outer core, creating the geodynamo responsible for the Earth’s magnetic field.
For this to occur some of the light element must be released into the melt during crystallisation. Details such as these all provide clues as to the identity of the light element in the core, but for direct observables, we must turn to seismology.
24 Chapter 1. Introduction
1.3 Seismology
Other than satellite data, seismology o↵ers the only way of directly observing the
Earth’s core. Earthquakes produce seismic waves that travel through the Earth, known as body waves. Measuring these waves gives information about the material they have travelled through. The two most useful waves for observing the deep interior of the
Earth are the P- and S-waves. P-waves can travel through both solids and liquids and are observed first, while the secondary S-waves can only travel through solids.
The velocities of the P- and S-waves are dependent on the density of the material. The velocity of the P-wave, vp, is given by
4 K + 3 µ vp = (1.1) s ⇢ where K is the bulk modulus, µ is the shear modulus and ⇢ is the density of the material.
The S-wave velocity, vs, is dependent only on the shear modulus and density
µ v = (1.2) s ⇢ r
Therefore, measurements of the seismic velocities can yield information about impor- tant physical properties of the material the waves are travelling through. In addition to the body waves, seismic activity also causes the whole Earth to oscillate. These free oscillation waves are known as normal modes and can provide invaluable information on the whole Earth, including the Earth’s core.
From the seismic wave data collected, a seismic model can be generated. One of the first such models, the Preliminary Earth Reference Model (PREM), created by Dziewonski and Anderson (1981), is still widely used today. PREM combined a large data set, gen- erating a reference model for the Earth of bulk elastic properties as a function of depth
25 1.3. Seismology
(see Figure 1.4). Although such a model does not give direct information on the com- position of the Earth, the model can be used for comparison with candidate materials since it is derived from observed data. Suggested compositional models for the Earth can therefore be tested for robustness against seismic models such as PREM.
Figure 1.4: The PREM model of Dziewonski and Anderson (1981).
1.3.1 Anisotropy and Layering of the Inner Core
One of the key results to come from seismological studies of the Earth’s core is the observed anisotropy of the inner core. It has been observed that P-waves travel about
3 % faster along the polar axis than in the equatorial plane (Poupinet et al., 1983;
Beghein and Trampert, 2003; Creager, 1992; Oreshin, 2004). Originally, this anisotropy was thought to extend across the whole of the inner core, with cylindrical symmetry that has its symmetry axis aligned with the Earth’s rotational axis. However, recent seismological observations have shown that the inner core is not uniformly anisotropic
(Morelli et al., 1986; Woodhouse et al., 1986; Song and Helmberger, 1998; Ishii and
Dziewoski, 2003). Instead, the inner core is split into two parts, consisting of an irregularly shaped central region, termed the lower inner core, which exhibits seismic anisotropy, and an upper layer, which is isotropic or weakly anisotropic, surrounding this (see Figure 1.5a). The most recent seismic studies (Wang et al., 2015) indicate
26 Chapter 1. Introduction that, contrary to previous work, the fast direction in the lower inner core lies in the equatorial plane (see Figure 1.5b). Hemispherical variations in anisotropy have also been observed seismically (Wang et al., 2015; Sun and Song, 2008) with the western region exhibiting stronger anisotropy at greater depths compared to the eastern region
(see Figure 1.5b).
(a) (b)
6*$
-*$ "YJT -*$
Figure 1.5: Illustrations of the upper inner core and lower inner core showing (a) the non- uniform layering seen in the inner core and (b) the seismic anisotropy observed. Blue lines depict the magnitude of anisotropy and alignment of the fast direction. The upper inner core (UIC) is only weakly anisotropic, with the fast direction aligned North-South, but showing hemispherical variations – stronger anisotropy is observed at greater depths in the western part. Recent seismic studies by Wang et al. (2015) suggest that the lower inner core (LIC) has the fast direction aligned equatorially. Figure (a) after Song and Helmberger (1998) and Figure (b) after Wang et al. (2015).
The isotropic, or weakly anisotropic, upper layer is seen to vary in thickness from between 100 to 400 km (Shearer, 1994; Tanaka and Hamaguchi, 1997; Song and Helm- berger, 1998; Ishii and Dziewoski, 2003; Ouzounis and Creager, 2001; Sun and Song,
2008). This observed layering raises the question as to whether there are two composi-
tionally or structurally distinct areas in the inner core. The presence of seismic isotropy
and anisotropy could give information about the structure of the material in the core.
Seismic isotropy would be exhibited by a phase with an intrinsically low anisotropy, or
alternatively it could be explained by having a random orientation of crystals of any
phase, regardless of any intrinsic anisotropy, such as the body-centred cubic (bcc), face-
27 1.4. The Terrestrial Planets centred cubic (fcc) or hexagonally-close packed (hcp) structures. On the other hand, anisotropy may come from the inherent structure of a phase, or when there is preferred orientation of aligned crystals, either of the same phase as in the upper isotropic layer, or an entirely di↵erent phase of iron. Therefore the correct composition for the inner core must also have the right structure to match the anisotropy seen in the seismic data.
Since the addition of light elements and nickel to iron a↵ects the structure adopted (see
Section 1.5) it is important to understand the Fe-Ni-Si system as a whole rather than just the pure iron system.
1.4 The Terrestrial Planets
Other than the Earth and the Moon, seismic data are unavailable for the terrestrial planets, although there have been plans for seismic stations on Mars and Venus. Despite this, exploratory missions to the planets, such as the MESSENGER mission to Mercury and the Mars Global Surveyor, have yielded data from direct observations of the planets.
In addition, since formation for the terrestrial planets followed a similar process as the
Earth (Morbidelli et al., 2012), our understanding of the deep interior of the Earth and its core can inform our understanding of the terrestrial planets and can be used to provide a preliminary model for their composition and structure.
1.4.1 The Moon
Although there is no seismic data available for the terrestrial planets, there is seismic data available for the Moon. This data was gathered from the four-station seismic network established on the moon during the Apollo lunar landing missions (Nakamura et al., 1982). Using the collected seismic data, along with other measurements from the
Apollo missions such as remote sensing, surface exploration and sample return, some key observations about the Moon have been made. Anderson et al. (1970) first proposed that a large part of the Moon was initially molten, from which mafic cumulates crys- tallised out. These then make up the mantle, and plagioclase flotation cumulates rise to form the crust. This hypothetical structure of the Moon is now generally accepted as correct (Wieczorek, 2006). However, the structure of the lunar core is controversial.
28 Chapter 1. Introduction
Although indirect geophysical observations support the existence of a metallic core in the Moon (Wieczorek, 2006; Konopliv, 1998; Williams et al., 2001), key features such as the radius, composition and whether the core is partially molten or solid, remain unclear. Additionally, there has only been one seismic study on the lunar core (Weber et al., 2011). This study suggested that the lunar core consists of a solid inner and
fluid outer core, much like Earth, with the liquid outer core making up approximately
60 % by volume of the total core. As with the Earth, a precise determination of the composition and structure of the lunar core is necessary for an understanding of the
Moon as a whole, and is especially important for understanding the dynamo.
1.4.2 Mars
Out of all the terrestrial planets, Mars has been subject to the most missions, many of which have received significant media coverage. These have included rovers, including the 1997 Mars Pathfinder Sojourner and NASA’s current Curiosity rover (Bell, 2012), as well as several orbiters of the planet. Much media attention has focused on the rovers, which are on the lookout for evidence of extra-terrestrial life, but the orbiters provide invaluable data and insight into the deep interior of the planet.
The Mars Global Surveyor (MGS) was an orbiter launched by NASA in 1996, staying in orbit four times longer than originally planned and becoming one of the most productive missions to Mars. In addition to the much reported discovery of water on Mars – and therefore the possibility of life – the MGS also provided magnetic field observations of the whole planet. It was found that Mars does not have a global magnetic field like the
Earth. Instead, localised areas of strongly magnetised ancient crust was detected by the
MGS (see Figure 1.6). The crustal magnetic field detected is much stronger than that of the Earth, with the strongest magnetisations an order of magnitude larger than the
Earth’s. The large areas of magnetised crust suggest that the magnetisation measured is thermal remanence of a global magnetic field obtained during cooling through the blocking temperature.
For a global magnetic field to have been present on Mars, a dynamo must have once operated, that has now ceased to be active. Since Mars was geologically active only
29 1.4. The Terrestrial Planets
Figure 1.6: Data collected by the MGS of Mars’ magnetic field. The Southern hemisphere contains strong localised anomalies. Figure from Stevenson (2001).
during its early history, the early chemical and isotopic signatures of its di↵erentiation process remain and therefore research into Mars can shed light onto the di↵erentiation process for the Earth. Using data from the MGS, Yoder et al. (2003) calculated the tidal Love number, k ,tobe0.153 0.017, indicating at least a partially liquid core 2 ± since the Love number is greater than 0.10, the value required for a liquid core to be ⇠ present. However, this does not answer the question of whether a solid inner core is present. For core convection to occur, and therefore generate the core dynamo, there either has to be large heat flow, which can be di cult to obtain, or the presence of a solid inner core. There are three possible core models that satisfy the original presence of a core dynamo that has since ceased to operate (see Figure 1.7). The first is of a completely liquid core, where the large heat flow is obtained from the monotonic cooling of the Martian mantle. Core convection eventually stops when mantle cooling stops. The second model is one of a solid inner core inside a liquid outer core, such as in the Earth, but with a stagnant lid. Convection then occurs as in the Earth, with the solid inner core growing to the point where the outer core is too small to sustain a core dynamo. The third model is the plate tectonic model, exactly as on the Earth, but with the core dynamo ceasing when plate tectonics ceases.
30 Chapter 1. Introduction
Figure 1.7: Illustrations of three possible core models that fit the current data obtained about Mars. Figure from Stevenson (2001).
The presence of at least a partially liquid core indicates that the Martian core contains a proportion of a light element. The estimated temperature profile of Mars, known as the areotherm, is lower than the melting curve of pure iron (see Figure 1.8), and since light elements are known to depress the iron liquidus (Fei, 1997; Fischer et al.,
2012; Terasaki et al., 2011), it follows that the Martian core must be made up of some percentage of a light element. This would have implications for the state of the core, since di↵erent light elements a↵ect the melting curve by di↵erent amounts. For example, sulphur depresses the melting point of iron very strongly (Fei, 1997), whereas
silicon has a much smaller e↵ect (Fischer et al., 2013). Which light element is present,
and in what quantities, would influence core formation and whether a solid inner core
is present. Nickel, too, is also likely to a↵ect the melting curve of iron, although the
e↵ect may not be large (Lord et al., 2014).
Other than the Earth, Mars is the only planet for which material is available for direct
analysis in the laboratory, although data from Martian missions shows that meteorites
do not sample all the rocks on Mars. Martian meteorites, named SNC meteorites for
the three areas where these meteorites were found – Shergotty, Nakhla and Chassigny
31 1.4. The Terrestrial Planets
Figure 1.8: A graph showing the estimated temperature profile of Mars, known as the areotherm, determined through consideration of thermal evolution models of Mars based on data captured by the Mars Global Surveyor (Fei and Bertka, 2005).
– all have iron contents less than that of chondritic meteorites, implying that the core consists primarily of iron, as for the Earth. However, SNC meteorite studies also indicate that the Martian mantle is far richer in iron than the Earth’s mantle, in the form of FeO in the mantle (Halliday et al., 2001). This implies that Mars has a smaller metallic core than the Earth, with recent estimates putting the core radius at 1794
65 km (Rivoldini et al., 2011). The smaller metallic core and higher Fe content in ± the Martian mantle indicate that core formation occurred in oxidising conditions. A possible model of core formation in Mars is shown in Figure 1.9.
Finally, being much smaller than the Earth, the pressure within Mars at any given depth is only about one third that of the Earth. This has implications for the mineral assemblages found in Mars, which may be quite di↵erent to those found in the Earth.
For example, the presence of light elements, even in small amounts, have a greater e↵ect at lower temperatures and pressures than at the core conditions of the Earth (see
Section 1.5).
32 Chapter 1. Introduction
Figure 1.9: A possible model of the main di↵erentiation events occurring on Mars, including the evolution of a Martian magma ocean. The metallic core, shown in black, is formed by segregation from the silicate portion of the mantle. The magma ocean then crystallises from the bottom up (shown in the middle image). The dense Fe-rich cumulates that result are found on top of the less dense, Mg-rich cumulates, which results in mantle overturn as seen in the picture on the right. Diagram from Mezger et al. (2012).
1.4.3 Mercury
There have been only two spacecraft sent to Mercury, the recent MESSENGER mission and the previous Mariner 10 mission. The Mariner 10 spacecraft flew past the planet three times in 1974 and 1975, gathering data on less than half the surface of the planet.
The MESSENGER mission however, orbited Mercury 4105 times and has returned over
10 terabytes of data on the planet. The flybys of Mariner 10 established the existence of an intrinsic magnetic field in Mercury; data from MESSENGER confirmed the dipolar nature of the magnetic field (Anderson et al., 2008). The dipolar nature of Mercury’s magnetic field suggests that a core dynamo, such as that in the Earth, is present in the planet. However, Mercury has a much weaker magnetic field than the Earth, which is di cult to explain with an Earth-like dynamo, and various dynamo models have been proposed (Heimpel et al., 2005; Wicht et al., 2007; Schubert and Soderlund,
2011). These models have reproduced the weak magnetic field but not the observed asymmetry of the magnetic field. Recent work suggests that in order to produce the asymmetry observed in the magnetic field, a dynamo driven by thermal buoyancy, in
33 1.4. The Terrestrial Planets addition to local excess equatorial heat flow at the core-mantle boundary, is required
(Cao et al., 2014). This is in contrast to the Earth’s dynamo, which is primarily driven by chemical buoyancy as well as the release of latent heat of fusion due to inner core growth.
Mercury also has a very large inner core (see Figure 1.10). Moment of inertia data measured by MESSENGER reveal that Mercury is extremely dense (Smith et al., 2012).
Most recent analysis of MESSENGER data puts the radius of the core of Mercury at
2020 30 km, over 80% of the radius of the planet, with an average core density of ± 6980 280 kg m-3 (Hauck et al., 2013). If such a large core were made entirely of iron ± then it would currently be solid, based on thermal history considerations (Siegfried and Solomon, 1974), which is at odds with the core dynamo model of Mercury and observations of at least a partially liquid core from libration measurements (Margot et al., 2007). For a liquid core to exist, the iron must be alloyed with a proportion of light element, although, as in the Earth, the identity of this light element is unknown.
Figure 1.10: A comparison of the internal structure of the Earth and Mercury from Stevenson and David (2012). Note that Mercury’s core makes up the majority of the interior, unlike the Earth, which has a mantle consisting of two-thirds of the total planetary mass.
A lot of work has focused on sulphur being the light element in Mercury’s core due to its cosmochemical abundance and the fact that sulphur strongly depresses the melting curve of iron. However, the highly reducing conditions suggested by the high abundance
34 Chapter 1. Introduction of sulphur and low abundance of iron in the silicate part of Mercury would indicate that silicon is more likely to alloy with iron in the core either with or instead of sulphur
(Malavergne et al., 2010). A Fe-Si-S alloy would also work with the buoyancy dynamo model suggested by Cao et al. (2014). Experiments have shown that the Fe-S-Si alloy exhibits liquid immiscibility at pressures less than 15 GPa (Morard and Katsura, 2010).
A core made of a Fe-S-Si alloy would therefore segregate into a S-rich liquid which floats to the top of the core, giving the necessary buoyancy to generate the core dynamo.
Research on core composition and theories of core formation and evolution are mutually dependent on each other – understanding core formation, as well as how the core is currently evolving, can aid in determining the core composition and vice versa. As with the other terrestrial planets, it is believed that the metallic core formed from gravitational separation of the more dense iron from the silicate part. One suggested model for core evolution in Mercury is the ‘snowing core’ model. Chen et al. (2008) found that the melting curves of Fe-S alloys exhibit two inflection points – at 14 GPa and 21 GPa – that coincide with phase transitions to Fe3S2 and Fe3Srespectively (see Figure 1.11). The presence of these inflections mean that the the cooling adiabat
Figure 1.11: Melting curves measured by Chen et al. (2008) for a range of Fe-S alloys. All exhibit two inflections (at 14 GPa and 21 GPa) allowing the adiabat to cross the liquidus more than once for di↵erent sulphur concentrations. of Mercury’s core may intersect the melting curve at multiple points. It is therefore
35 1.4. The Terrestrial Planets possible for solid to precipitate at a range of depths within Mercury’s core, depending on the sulphur content. At shallow depths, solid precipitation is in the form of ‘snow’
– solid precipitates out, sinks and accumulates at the centre to form the core. Chen et al. (2008) estimate the present-day temperature of Mercury’s core to be between
1700 and 1900 K, with an adiabatic gradient of 11 K/GPa. Assuming this, there are three possible core states for Mercury – a Ganymede-like state, a double-snow state
(see Figure 1.12) or an Earth-like state. If the core contains 7 – 8 wt. % sulphur, solid forms at shallow depths (Ganymede-like state); a core containing 8 – 10 wt. % sulphur would have solid forming at low and high pressures simultaneously (double-snow state); if the core contains 12 wt. % sulphur, freezing occurs at high pressure only (Earth-like state). These core formation models, however, are dependent on Mercury’s core being solely composed of a binary Fe-S alloy. Other light elements, as well as the addition of nickel, are likely to a↵ect the melting curve.
Figure 1.12: Two of the possible suggested states of Mercury’s core, as proposed by Chen et al. (2008). The double-snow state has two distinct zones where solid precipitation (in yellow) occurs, whereas the Ganymede-like state only has solid precipitation occurring at shallower depths.
More recently, it has been suggested that the ‘hit and run’ model which is commonly accepted for Moon formation, can also be used to explain the formation of Mercury
(Asphaug and Reufer, 2014). This theory suggests that proto-Mercury had its mantle stripped in a high-speed collision with a larger target planet, which could possibly explain the large iron core found in Mercury. Numerical hydrocode simulations show that this type of collision can produce a Mercury-like planet (see Figure 1.13).
36 Chapter 1. Introduction
Figure 1.13: One solution of the numerical hydrocode simulations carried out by Asphaug and Reufer (2014) of two planetary bodies after a collision, showing a Mercury-like remnant many hours after colliding.
1.4.4 Venus
Unlike Mercury, Venus has been explored by over 40 spacecraft and the Magellan mission has managed to map approximately 98 % of the planet. However, although there have been multiple orbiters, atmospheric probes and landers that have explored
Venus (Sengupta, 2010), the lack of a bulk of comprehensive data on Venus has inhibited the research e↵ort into understanding Venus, particularly its deep interior.
Venus is very similar to the Earth in terms of size, density, mass and bulk composition during formation. However, there are some key di↵erences between Venus and the
Earth. For example, Venus lacks plate tectonics and also has no measurable magnetic
field. It has been suggested that the lack of plate tectonics is the cause of the lack of magnetic field. Nimmo (2002) suggests that in the absence of plate tectonics, the mantle cannot cool fast enough to provide a driving force for core convection and generate a dynamo. The lack of plate tectonics could also point to the absence of a solid inner core, since the inability to take heat away from the core quickly enough
37 1.5. The Case For Silicon would result in a higher core temperature at the same pressures of the Earth, meaning the inner core cannot solidify. Calculations of the tidal Love number, k2, using data from Magellan and the Pioneer Venus Orbiter also reveal that the core must be at least partially molten (Konopliv and Yoder, 1996). However, problems with the slow spin rate of Venus combined with its dense atmosphere means that measurements which would shed light on the interior of Venus are not well constrained (Mocquet et al.,
2011). Therefore the most conservative model for the deep interior of Venus is still only a scaled version of the Earth’s deep interior. Further data on Venus is required to draw deeper conclusions on its history and the state of its core.
1.4.5 Summary of Terrestrial Planetary Cores
It can be seen from these discussions that there are still many aspects that remain un- known about the cores of terrestrial planets. It is also clear that a deeper understanding of core composition would help resolve a lot of the mysteries surrounding the cores of these planets. Although the current focus has been on sulphur as the light element in the cores of the terrestrial planets, there has been no evidence to definitively rule out other light elements. Indeed, silicon is now considered likely in Mercury’s core (see
Section 1.4.3). In any case, a larger database of knowledge on the ternary Fe-Ni-X sys- tems likely to make up the cores of terrestrial planets would be highly useful, especially since current theories may need to be adapted as newer data on the planets becomes available.
1.5 The Case For Silicon
From the discussions so far, it is clear that despite major advances in research on the terrestrial planets, a lack of data makes it di cult to firmly establish the identity of the light element in the planetary cores. In this section, the evidence for silicon as the light element is considered. Although this has been discussed with regards to the Earth’s core – since there is greater availability of data on the Earth – the similarities between the Earth and the terrestrial planets, as well as the Moon, mean that the evidence presented here is also widely applicable to the other terrestrial planets.
38 Chapter 1. Introduction
1.5.1 Silicon in the Earth’s Inner Core
Silicon has long been considered a potential light element in the Earth’s core. Since the core is surrounded by a silicate mantle, the required abundance of silicon for it to be in the core is satisfied. In addition, the mantle is ‘missing’ a small percentage of silicon if the Earth is to have a chondritic composition (MacDonald and Knopo↵ , 1958;
Ringwood, 1961). Having silicon in the core would solve this problem.
In addition to being abundant enough, the light element must be congruent with core formation. However, a core formation model is still not agreed upon. There is still de- bate surrounding whether the core formed in reducing or oxidising conditions. Previous work has suggested that core formation could only occur with little or no oxygen present
(Wood et al., 2006; Huang et al., 2011). A highly reducing environment would result in a significant proportion of silicon being incorporated into the core (Malavergne et al.,
2004). More recent work has suggested more oxidising conditions for core formation
(Siebert et al., 2013). However, this does not rule out silicon being in the core; it has been found that the presence of oxygen influences silicon partitioning, and causes more silicon to partition into the metal (Tsuno et al., 2013). Therefore it seems likely that silicon would be present in the core regardless of whether the core formed in reducing or oxidising conditions.
Another method for determining the composition of the inner core is to consider the partitioning of Fe-X alloys between the inner and outer core. If the solid inner core and liquid outer core are in thermodynamic equilibrium at the inner core boundary (ICB), then the chemical potentials of each species will be equal at that point. The ratio of the concentrations of elements in the liquid and solid phase is then fixed, which in turn fixes the density ratios. Since the inner core and outer core densities must match that determined by seismic data, the proportion of the light element in the core can be determined. Chemical potential calculations carried out by Alf`eet al. (2002) show that both silicon and oxygen partitions preferentially from solid to liquid, but only weakly so for silicon and much more strongly for oxygen. A binary system of either Fe/Si or Fe/O is ruled out because the calculated percentages of either light element does
39 1.5. The Case For Silicon not match the ICB density discontinuity seen seismically – the percentage of silicon is too low and the percentage of oxygen too high. However, a mixture of silicon and oxygen together can account for the seismic data, but since oxygen partitions strongly to liquid, this requires less than 1 % oxygen in the solid inner core, and 8 % in the liquid outer core. In contrast, there would be about 8 % silicon in the inner core and 10
% in the outer core. Previously, the presence of both silicon and oxygen was thought to be incongruous since silicon is a known reducing agent. However, the work of Takafuji
(2005); Ricolleau et al. (2011); Tsuno et al. (2013) suggests that this incompatibility may not be as problematic as once thought. Their work indicates that a small increase in temperature, from 3000 K to 3500 K, is enough to raise the levels of silicon and oxygen in liquid iron to above that required to compensate for the density deficit seen seismically.
1.5.2 Seismological Evidence for Silicon
In addition to inferred density measurements, the velocities of seismic waves can be matched to those of mineralogical models in order to identify the light element in the core. Seismic wave velocities can be measured experimentally or calculated us- ing computational techniques for di↵erent candidate compositions. Experimental and computational work both show that the shear wave velocity of pure hcp-Fe is higher than that seen seismically (Antonangeli et al., 2004; Voˇcadlo, 2007). Calculations on
Fe0.875Si0.125 show that a small amount of silicon reduces the shear wave velocity by 15 % in the hcp phase of iron at 360 GPa (Tsuchiya and Fujibuchi, 2009). These cal- culations found a linear relationship between seismic velocity and density – behaviour which is also observed in FeSi experimentally and computationally (Badro et al., 2007;
Voˇcadlo, 2007). However, recent work by Mao et al. (2012) suggests that a power-law function may be more suitable for describing the velocity-density relationship. Mao et al. (2012) also analysed the V ⇢ data for a wide range of iron alloys (see Fig- ure 1.14). From this, it can be seen that there are many possible compositions that would satisfy the seismically measured velocity data, meaning that using this method alone is not enough to determine the composition of the Earth’s core.
40 Chapter 1. Introduction
Figure 1.14: Graph showing V , the bulk sound velocity, reproduced from Mao et al. (2012). Solid lines, except the black and grey lines, are the V of Fe alloys calculated from static compression results at 300 K. Blue line indicates hcp-Fe, orange line indicates Fe3S, violet line indicates Fe0.85Si0.15, dark cyan line indicates Fe3C, red solid and dashed lines indicate FeO in the B1 and B8 phase, respectively, navy blue line indicates FeSi, purple line indicates FeS in the IV phase, magenta and olive circles stand for V of Fe0.9O0.08S0.02 and Fe0.925O0.053S0.022 from shock-compression studies respectively. For references and further details see Mao et al. (2012).
Very recent work has suggested that pre-melting behaviour in pure iron is su cient to explain the discrepancy in VS between mineralogy and seismology. Calculations carried out by Martorell et al. (2013a) show that above 6600 K, some of the elastic constants of hcp-Fe decrease with temperature, resulting in both VP and VS decreasing to values that match those seen seismically (see Figure 1.15). The density deficit still requires a small percentage of light element, which would likely decrease the melting temperature, bringing the softening e↵ect to a temperature more reasonable for core conditions. In any case, it is clear that although much work has been done to pin down the seismic behaviour of iron and iron alloys, more research is necessary to fully understand the behaviour of seismic waves in mineralogical models.
1.5.3 The E↵ect of Si on the Structure of Iron
It has been found, in both experimental and computational studies, that the addition of a light element increases the stability field of the bcc structure of iron (Cˆot´eet al., 2008;
Lin et al., 2002; Sata et al., 2008; Voˇcadlo et al., 1999). The stable high pressure phase
41 1.5. The Case For Silicon
Figure 1.15: Calculated VP and VS velocities for hcp-Fe as a function of T/TM and simu- lation temperature at 360 GPa. The grey band represents minimum and maximum melting temperatures for hcp-Fe. Figure from Martorell et al. (2013a). of FeSi is known to have the CsCl structure, which itself is topologically equivalent to the bcc structure. The low-pressure phase of FeSi is the "-FeSi structure, a slightly modified form of an idealised seven-fold co-ordinated structure (see Figure 1.16). The pressure at which the "-FeSi structure transforms to the high-pressure CsCl-FeSi structure has been investigated experimentally and computationally. In computational studies, the phase transition is found to occur at 13 GPa (Voˇcadlo et al., 1999), 20 GPa (Zhang and Oganov, 2010) and 40 GPa (Caracas and Wentzcovitch, 2004). The reason for the discrepancy between these studies is not known definitively, and has been investigated in Chapter 4. Experimentally, the transition to CsCl-FeSi has been seen at 24 GPa, at temperatures greater than 1950 50 K (Dobson et al., 2002), although the experimental ± phase boundary of this transition is also in dispute (see Chapter 4).
As noted above, experiments have found that even a small percentage of silicon increases the stability field of the bcc phase of iron. Lin et al. (2002) found that 7.9 wt.% Si can stabilise the bcc phase up to about 40 GPa, forming a large two phase region of bcc
+ hcp (see Figure 1.17). This two phase region, consisting of a silicon-rich bcc phase and a silicon-poor hcp phase, suggests that the inner core may consist of a mixture of
42 Chapter 1. Introduction
Figure 1.16: A diagram of the "-FeSi structure that is stable at low pressures. Silicon is represented by small light grey spheres and iron by darker spheres. The structure is formed from small displacement of the Fe and Si atoms from the ideal seven-fold co-ordinated structure, resulting in one slightly shorter Fe-Si bond, three intermediate length bonds and three slightly longer bonds. From Voˇcadlo et al. (1999).
two iron phases. Further work with an Fe0.85Si0.15 sample found the same two phase mixture persists to 150 GPa and 3000 K, but between 170 GPa and 240 GPa, only the hcp phase is stable (Lin et al., 2009). However, experiments carried out by Asanuma et al. (2008) with a lower percentage of silicon – 3.4 wt.% Si – found that the fcc phase is stabilised instead at high temperatures and pressures (see Figure 1.18), creating a two phase fcc and hcp stability field.
Computer simulations have also found that silicon stabilises certain structures in iron.
Calculations carried out at 0 K by Cˆot´eet al. (2008) found that even small amounts of silicon (3.2 % Si) stabilises the bcc phase over the hcp phase. An increasing amount of silicon has the e↵ect of increasing the stabilising e↵ect of bcc. However, lattice dynamics calculations on iron alloyed with silicon found that the addition of about 7 wt.% Si to iron stabilises the fcc phase at high temperature (Cˆot´eet al., 2010), in agreement with
Asanuma et al. (2008); whereas calculations by Zhang and Oganov (2010) on a range of FexSi(1-x) phases, using a structure searching method, find that only the CsCl phase of FeSi is stable at high pressures, with all other silicides decomposing to Fe + FeSi.
More recent experiments by Fischer et al. (2012, 2013) have found that iron with 9 wt.% Si creates a two phase CsCl-FeSi and hcp stability region at high pressures and
43 1.5. The Case For Silicon
Figure 1.17: Phase diagram measured by Lin et al. (2002) showing the di↵erent stability fields for bcc, hcp and fcc structured iron in the presence of 7.9 wt.% Si.
Figure 1.18: The phase diagram measured by Asanuma et al. (2008) of iron with 3.4 wt.% Si showing the fcc phase persisting at high temperatures and pressures compared to the pure Fe phase diagram.
44 Chapter 1. Introduction temperatures, and increasing the amount of Si to 16 wt.% increases this two phase stability field (see Figure 1.19).
Figure 1.19: Phase diagrams for a range of compositions in the Fe-FeSi systems showing the stability field of the B2 (CsCl-FeSi phase) and hcp phase increasing with an increasing amount of silicon added. Figure from Fischer et al. (2013).
1.6 E↵ect of Nickel
Previously it was generally assumed that since the amount of nickel alloyed to the iron in the core was small it wouldn’t have a significant e↵ect on the properties of the core.
Additionally, nickel has a similar density to iron, making the two indistinguishable seismically. Indeed, experiments carried out by Antonangeli et al. (2010) to measure the seismic velocities of Fe0.89Ni0.04Si0.07 found that the addition of a small percentage of nickel has negligible e↵ects on the seismic velocities (see Figure 1.20).
Experiments on Fe0.78Ni0.22 also show no significant deviations from the elastic prop- erties of pure iron (Kantor et al., 2007). Computational studies have shown that the negligible e↵ect of nickel is due to the high temperature of the system rather than chemical similarities between iron and nickel as previously thought (Martorell et al.,
2013b), indicating that more work needs to be done to fully understand the e↵ects of nickel. Experiments have also shown that nickel influences the structure adopted by
45 1.6. E↵ect of Nickel
Figure 1.20: Seismic wave velocities for a range of Fe-Ni-Si alloys. Adding a significant percentage of Ni to both pure-Fe and Fe-Si appears to have a negligible e↵ect on the seismic wave velocity. From Antonangeli et al. (2010). iron. Mao et al. (2006) carried out laser heated DAC experiments on Fe-Ni alloys and found that nickel has the e↵ect of stabilising the fcc phase to lower temperatures and higher pressures (see Figure 1.21). Experiments by Komabayashi et al. (2012) on Fe-Ni alloys with up to 15 wt.% Ni show the same e↵ect on stability of fcc. In addition, experiments that reached higher temperatures and pressures – 340 GPa and 4700 K – on iron alloyed with 10 wt.% Ni found that the hcp phase was most stable throughout, including up to core conditions (Tateno et al., 2012).
Ab initio calculations, too, have shown that a small amount of nickel stabilises the hcp structure over the bcc structure by about 20 GPa (Voˇcadlo et al., 2006). More recent calculations have shown that the fcc phase is stabilised by the addition of nickel at core conditions (Cˆot´eet al., 2012).
46 Chapter 1. Introduction
Figure 1.21: Phase diagram for a range of Fe-Ni alloys, showing that an increasing amount of nickel has the e↵ect of increasing the stability of the fcc phase to higher pressures and lower temperatures Mao et al. (2006).
1.7 Investigations on the Fe-Ni-Si System
Although a large body of work exists on the Fe-Ni-Si system, more research is required to fully define the e↵ects of nickel and silicon on the properties of iron. One common aspect of the work carried out so far is the focus on specific compositions that the cores of terrestrial planets are thought to consist of. The basis for picking out candidate compositions is either from core formation models or matching the density deficit from seismic data. Core formation models are in part reliant on mineralogical models, which introduces circularity to the problem of determining composition. Using the density
47 1.7. Investigations on the Fe-Ni-Si System deficit is also problematic since there is not a unique core composition which satisfies this (see Section 1.5.1). Therefore as more data is collected from exploratory missions to the terrestrial planets, as well as continued analysis of seismic data on the Earth, it seems likely that the arguments for these specific core compositions studied so far may change.
Rather than focus on specific compositions, this work has instead focused on under- standing the Fe-Ni-Si ternary system as a whole. This approach provides a more me- thodical way of researching how nickel and silicon a↵ects iron and its properties. The two endmembers, FeSi and NiSi, have been studied using both computational and experimental techniques. The phase transition from "-FeSi to CsCl-FeSi has been in- vestigated computationally at 0 K (Chapters 4) with additional calculations performed to ascertain whether any of the numerous structures recently found to be stable in
NiSi (Voˇcadlo et al., 2012) are also stable in FeSi (Chapter 5). In addition, lattice dynamics calculations were carried out to determine the phase boundary of the "-FeSi
CsCl transition at high temperatures for both FeSi and NiSi (Chapters 4 and 6). ! Experimentally, the melting curve of NiSi (Chapter 7) and the the eutectic melting curve of the Fe-FeSi system (Chapter 8) have both been investigated, as have the equa- tion of state of MnP-NiSi (Chapter 9) and the thermal expansions of di↵erent Fe-Ni-Si alloys (Chapter 10). The melting curves have been determined by laser-heated DAC experiments, and the thermal expansions and equation of state measurements measured using di↵raction techniques. The next two chapters explain the methodology behind this work, with computational techniques discussed in Chapter 2 and experimental methods in Chapter 3.
48 CHAPTER 2
Computational Methods
Tonight the giant galaxies outside Are tiny, tiny on my windowpane
— Gjertrud Schnackenberg
As seen in the previous chapter, computational methods are an important tool in inves- tigating the cores of terrestrial planets, and are a necessary complementary technique to experimental methods, providing an alternative method of investigating materials at high temperatures and pressures – conditions that are di cult to reach and main- tain in experiments. In this work, ab initio calculations have been used to investigate the phase transition from "-FeSi to CsCl-FeSi at 0 K and at high temperatures and pressures using lattice dynamics, as well as to calculate the stability of NiSi structures in the FeSi system at 0 K. The "-FeSi CsCl phase transition has also been studied ! using ab initio lattice dynamics calculations in the NiSi system.
2.1 Ab initio
Ab initio simulations employ quantum-mechanical theory in order to determine the energy of a system accurately from its wavefunction. The wavefunction, (r), of a system contains all the information of that system. It is not observable directly, but the wavefunction multiplied by its complex conjugate gives the probability of an electron at a point, r. The following sections outline the basics of an ab initio calculation, starting with the Schr¨odinger equation.
49 2.1. Ab initio
2.1.1 The Schr¨odinger Equation
The Schr¨odinger equation links the total energy of the system with the wavefunction through the Hamiltonian, Hˆ
Hˆ =E (2.1)
where is the wavefunction and E is the energy. The Hamiltonian describes the potentials in terms of the nuclei and electrons
1 1 Z 1 Z Z Hˆ = 2 2 µ + + µ ⌫ (2.2) 2M 5µ 2 5i r r R µ µ i,µ ij µ,⌫>µ µ⌫ X Xi Xi,µ i,j>iX X
The Greek labels, µ and ⌫, refer to the nuclei and labels i and j refer to the electrons, with M being the mass, R being the distance and Z being the atomic number of the nucleus, all in atomic units. Thus the first two terms make up the kinetic energy of the nuclei and the electrons respectively with the last three terms corresponding to the potential energy.
Solving Schr¨odinger’s equation gives the total energy of the system, from which all rele- vant physical properties may be calculated. The computational power required to solve
Schr¨odinger’s equation increases exponentially with the number of electrons, making it impossible to calculate exact solutions for systems with more than two electrons.
However, with a number of simplifications, it is possible to find the total energy to a very high degree of accuracy.
2.1.2 The Born-Oppenheimer Approximation
The first simplification is the Born-Oppenheimer approximation, which is an adiabatic approximation and provides a way of decoupling the dynamics of the atomic nuclei and the electrons. The Born-Oppenheimer approximation exploits the huge di↵erence in mass between the nucleus and the electrons. Even considering the lightest element,
50 Chapter 2. Computational Methods the hydrogen nucleus is about 1800 times heavier than an electron. It is therefore safe to assume that the nuclei move much more slowly compared to the electrons, and thus the dynamics of the nuclei can be decoupled from that of the electrons. This immediately simplifies the problem since the nuclei can be fixed at certain positions in space and Schr¨odinger’s equation solved just for the electrons in a potential generated by the stationary nuclei. The nuclear dynamics can be solved classically; however, the electronic problem requires more thought.
2.1.3 The Many Electron Problem
The Schr¨odinger equation can be solved exactly when only one or two electrons are considered. However, the complexity of the problem grows exponentially as the number of electrons increases, since the wavefunction exists in 3N dimensions, where N is the number of electrons. In a first-principles simulation, the wavefunction requires an infinite space of several thousand dimensions. As a first step, we can consider the electrons as independent. From this first consideration, a density related method can be developed (see Section 2.2).
2.1.4 Independent Electrons
One of the most obvious simplifications is to consider electrons that do not interact with each other, instead seeing an external potential generated by the other electrons. For a non-interacting system, the electronic Hamiltonian is the sum of the Hamiltonians of the single particles, hˆ
Hˆ = hˆ + hˆ + + hˆ (2.3) 1 2 ··· N where the Hamiltonian for the single particle is given by
1 hˆ = 2 +V ext(r) (2.4) 2 5 and V ext(r) is the external potential felt by the electron at position, r.
51 2.1. Ab initio
Solving Schr¨odinger’s equation for a single particle, with wavefunction n(r), gives the energy, ✏n
hˆ n(r)=✏n n(r) (2.5)
The energy of the whole system, E, is therefore
E = ✏ni (2.6) Xi since the many-particle wavefunction is
(r r )= (r ) (2.7) 1 ··· N ni i Yi
2.1.5 Indistinguishable Electrons
In quantum formalism, each particle is required to have a label even though they are indistinguishable. Because of this, the wavefunction used must contain the same physical information when the labels are exchanged. This can be achieved by using an antisymmetric wavefunction. Any wavefunction may be antisymmetrised in the following way
1 (r1,r2) (r1,r2) (r2,r1) (2.8) ! p2
However, this produces a wavefunction that is not physical. This can easily be seen when considering the wavefunction for a system consisting of two electrons
= (r1) (r2) (2.9)
52 Chapter 2. Computational Methods
Antisymmetrising this produces
1 (r1) (r2) (r2) (r1)) (2.10) p2
This equals zero since (r1)= (r2) which is not a physical result. In order to pro- duce a physically meaningful result, the single particle wavefunction must be di↵erent.
However, so far the spin of the electron has been ignored, if this is taken into account now then two electrons may share the same function if they are of di↵erent spin. This explanation is now easily recognisable as the Pauli Exclusion Principle, which states that two electrons cannot occupy the same energy state if they have the same spin. This also leads to the well-known Aufbau picture of electronic energy levels, with electrons occupying the lowest possible energy levels first (see Figure 2.1).
Figure 2.1: Electronic energy levels filled according to the Aufbau principle, so that the lowest energy orbitals are filled first and electrons fill an orbital with the same spin number until it is full before beginning to fill with the opposite spin number.
2.1.6 Self-Consistency
To solve the Schr¨odinger equation, we require the wavefunction to define the Hamil- tonian; however the wavefunction is obtained as a result of solving the Schr¨odinger equation. This produces a self-consistency issue. The equation must therefore be solved iteratively by
i. First proposing trial wavefunctions
ii. Building the Hamiltonian
iii. Calculating the eigenfunctions by solving the Schr¨odinger equation
53 2.1. Ab initio
iv. Iterate using the output of the previous cycle as input
This loop is then repeated until convergence is reached.
2.1.7 The Hartree Method
The previous section has only considered independent electrons within an external potential generated by the nuclei but neglects electron-electron repulsion. Hartree introduced a method to include this repulsion term using a mean field approximation.
The electrons are regarded as independent but each feeling an electrostatic repulsion of electron density from the other electrons
1 hˆ = 2 +V ext(r)+V H (r) (2.11) i 2 5 i where hˆ is the one particle Hamiltonian, 1 2 is the kinetic energy term, V ext(r)is i 2 5 H the potential due to the nuclei and Vi (r) is the Hartree potential – the potential due to other electrons. The potential terms are linked to the charge density by the following equation
⇢(r0) V (r)= dr0 (2.12) r r Z | 0| where ⇢(r) is the charge density and is given by
(r) (2.13) | i | Xi
The Hartree potential is then given by
2 H nj(r0) 3 V (r)= | | d r0 (2.14) i r r j=i Z 0 X6 | |
Hartree’s method only partially used the indistinguishability of electrons – in this
54 Chapter 2. Computational Methods method, the electronic states are filled according to Pauli’s exclusion principle, which is a direct result of the indistinguishability of electrons. However, this method does not take into account the requirement for the wavefunction to be antisymmetric.
2.1.8 Hartree-Fock Theory
Given a trial, many-electron wavefunction of the form
( r , )= ( r , ) (2.15) { i i} i { i i} Yi where is the total wavefunction and is the wavefunction of a single electron, with ri being the position of the electron and i being the spin, the Variational Principle applies
( r , ) Hˆ ( r , ) = ( r , ) Hˆ ( r , ) (2.16) h { i i} | | { i i} i h i { i i} | | i { i i} i Yi Yi Yi Yi
This can be generalised to take into account the need for an anti-symmetric wavefunc- tion if the above trial wavefunction is replaced by the Slater determinant
1 ( ri, i )= 1 2 N (2.17) { } pNi k ··· k
This is the Hartree-Fock method, and it results in the following single particle Hamil- tonian
1 hˆ = 2 +V ext(r)+V H (r)+V X (r) (2.18) i 2 5 i i
The only di↵erence between this and the Hamiltonian from Hartree’s original method
X is the addition of the exchange term, Vi (r). The Hartree-Fock method lacks accuracy because it still uses independent particles. The electron correlation is the deviation of the Hartree-Fock approximation from the exact solution for both the energy and the
55 2.2. Density Functional Theory (DFT) wavefunction. For a very small number of electrons, the electron correlation may be calculated. Again, however, as the number of electrons increases, the complexity of the problem increases very quickly and the computational power required to calculate the electron correlation increases exponentially. Using a density-related method simplifies this problem.
2.2 Density Functional Theory (DFT)
DFT is a method of approximating the electron correlation, and is one of the most common methods used when implementing ab initio technique in calculations today.
Two assumptions are made
1. The ground state properties of a many-electron system are uniquely deter-
mined by an electron density that depends on only 3 spatial co-ordinates.
2. The non-interacting electrons are in an e↵ective, self-consistent potential
(all other electrons are seen not as individual particles but a collective
potential)
2.2.1 Hohenberg-Kohn Theorems
Density Functional Theory was formalised through the two Hohenberg-Kohn theorems.
The first theorem states that for a system of interacting particles, the total energy of the system is a unique functional of the electron density and is always larger or equal to the ground state energy. This may be written as
E[⇢(r)] = d3rV ext(r)⇢(r)+F [⇢(r)] (2.19) Z where F [⇢(r)] is a universal functional. The second theorem states that the ground state energy, EGS, can be obtained via the Variational Principle
E[⇢(r)] = d3rV ext(r)⇢(r)+F [⇢(r)] E (2.20) GS Z
56 Chapter 2. Computational Methods and therefore the ground state energy corresponds to the ground state density. Thus, by minimising the density, the ground state energy may be found. Finding the ground state energy therefore becomes a trivial problem were it not for the fact that F [⇢(r)] is unknown.
2.2.2 Kohn-Sham Electrons
In order to put the theory of DFT into practice, we require a mathematical form for the functional F [⇢(r)]. Kohn and Sham rewrote the unknown functional, F [⇢(r)], and therefore the functional E[⇢(r)], by re-introducing the idea of auxiliary electrons (also known as Kohn-Sham electrons). These are electrons that do not interact with each other but with the e↵ective potential, giving the correct density and energy. Rewritten, this gives the following equation
1 E [⇢(r)] = T [⇢] d3r⇢(r) V ext(r)+ (r) + E [⇢] (2.21) 0 2 XC Z where
E[⇢(r)] is the energy as a function of density
T0[⇢] is the kinetic energy for a system with no electron-electron interactions
⇢(r)isthedensity
V ext(r) is the external potential