You are on page 1of 149

学位論文

下部マントルレオロジイの計算科学的研究

A computational study of the lower mantle rheology

平成19年12月 博士(理学)申請
東京大学大学院理学系研究科
地球惑星科学専攻
伊藤 洋介
Abstract

Self-diffusion coefficients of the two major lower mantle minerals (MgO

periclase and MgSiO3 perovskite) were investigated with molecular

dynamics (MD) by reproducing spontaneous migrations of vacancies under

lower mantle pressure-temperature (P-T) conditions. It was clarified that

self-diffusion coefficients of MgO periclase have significant non-linear

pressure dependence (initial decrease turns to be increase with increasing

pressure), while pressure dependence of self-diffusion coefficients of MgSiO3

perovskite is relatively modest, which is not distinct with linear increase.

Pressure dependence of self-diffusion coefficient of the solid neon (Ne) is also

investigated for comparisons. Results show good linear relationship with

pressure variation so that long-range force of MgO periclase is indicated as

its possible cause because interatomic potential of solid neon does not

include it. Self-diffusion coefficients of the two minerals were implicated

with the diffusion creeps of the corresponding two phases of the lower mantle.

Derived depth variations of viscosities are modest for MgSiO3 perovskite and

are significantly (102) decreasing for MgO periclase. Modest variation of

MgSiO3 perovskite is consistent with geophysically observed radial profiles

of viscosity of the lower mantle. From significantly ( ≈ 102) decreasing

viscosity of MgO periclase, possibility of large ( ≈ 104.5) viscosity contrast at

the bottom of the lower mantle between the two phases was pointed out.
Index

1 Introduction 1

2 Molecular Dynamics 5

2.1 Introduction 5

2.2 Fundamental Approximations of the Molecular Dynamics 6

2.3 Interatomic potentials 7

2.4 Ewald method 11

2.5 Equations of motions 13

2.6 Temperatures, Pressure, and their controlling 15

2.7 Thermodynamic properties 18

2.8 Melting Temperatures 20

2.9 Elastic Constants 22

2.10 Self-Diffusion 25

2.11 Precisions of Self-Diffusion Coefficients 31

2.12 Vacancy formation energies 33

3 Periclase (MgO) 35

3.1 Introduction 35

3.2 Equation of State 36

3.3 Melting Temperatures 39

3.4 Elasticity 45

3.5 Self-Diffusion 51

3.5.1 Introduction and methodology 51

3.5.2 Results of ordinary pressure 55


3.5.3 Results of ambient pressures 60

3.6 Discussion 67

4 Noble Gas Crystals 69

4.1 Introduction 69

4.2 Equation of State 72

4.3 Self-diffusion coefficients 75

4.4 Discussion 79

5 MgSiO3 perovskite 82

5.1 Introduction 82

5.2 Equation of state 96

5.3 Orthorhombic distortions 90

5.4 Melting 97

5.5 Self-diffusion 102

5.6 Discussion 115

6 Lower mantle rheology 117

6.1 Introduction 117

6.2 Lower mantle viscosity 121

6.2.1 Viscosity of the each mineral 121

6.2.2 Two-phase mixture 126

6.3 Discussion 129

7 Conclusion 132

Acknowledgements 133

References
Chapter 1

Introduction

Plastic deformation gives a strong effect on many processes in the

long-term dynamics of the Earth's interior. A number of studies have tried to

clarify rheology of the Earth. The most fundamental property of plastic

deformation is described by stress (σ) –strain ( ε ) relationship, that is

viscosity ( η = ε& / σ ) (Newtonian). Studies for clarifying viscosity could be

classified into two types: observations and experiments. Rheological

properties of the earth's interior can be deduced by several methods.

Postglacial rebounds [e.g. Mitrovica and Forte, 2004] and geoid anomalies

(e.g. Tamisiea et al. [2007], Latychev et al. [2005]) observed in the earth's

surface can be decomposed by inversion technique to obtain the viscosity

profile in the earth's interior. On the other hand, experiments are intended

to reproduce the plastic deformation of the earth's interiors in the laboratory

scale. Reproductions can be performed by high pressure and temperature

containable apparatus. History of the progress in rheology was reviewed in

detail by Karato and Wenk [2003], recently. Although experimental studies

have roughly clarified viscosity of the earth's materials of the upper mantle,

viscosity in the lower mantle has poorly known compared with those of upper

mantle, though lower mantle should play an important role in the Earth's

whole dynamics because of its large volume (<50 % in the whole earth).

Karato [1981] showed theoretical approach of the rheology of the lower

mantle from diffusive properties of composed minerals with the idea that the

1
lower mantle is diffusion creep. This idea has been continuously examined

and supported. Karato [1995] shows this idea highly likely from lack of

seismic anisotropy (Tanimoto and Anderson [1984]). Li et al. [1996]

investigated dominated mechanism of plastic deformation mechanism of

polycrystallineCaTiO3 perovskite, which is an analogue of the MgSiO3

perovskite, with ambient stress(σ) –strain rate( ε& )-grain size(d) conditions

and inferred that MgSiO3 perovskite is diffusion creep by extrapolating the

conditions to those of the lower mantle. The possibility of other mechanisms

was also checked. Based on elastic data given by electronic state calculations

(reviewed in Karki et al. [2002]), Yamazaki and Karato [2002] reproduced

experimental seismic anisotropy derived for deformation by power-law creep

of (Mg,Fe)O and showed that significant seismic anisotropy should be

observed in case (Mg,Fe)O of the lower mantle is power-law creep.

In case diffusion creep dominate in the lower mantle, the viscosity η is

related with diffusion coefficient D as following Nabarro-Herring

formulation

η = (d2/D)*(RT/AV)), (1.1)

where d is the grain size, R is the gas constant, T is temperature, A is a

constant, and V is the molar volume, respectively. This equation provides us

to obtain the lower mantle viscosity from diffusion coefficients of composed

minerals. We have two ways to obtain diffusion coefficients; experimental

and theoretical procedures. Numbers of experimental approaches have been

attempted. Although pressures of even recent experiments (15-25GPa, Van

Orman et al. [2003] for Mg in MgO, 25GPa for Yamazaki et al [2000] for Si in

MgSiO3) have extended to those of the lower mantle, pressure range of the

2
lower mantle is from 25GPa to 130GPa and it is impossible to get rheological

nature of the lower mantle from experimental studies in the near future.

On the other hand, theoretical approaches have a great advantage for

obtaining viscosity in the very high pressures. There is still difficulty: it

requires the diffusion based on atomistic scales. First, Karato [1981]

proposed nearest neighborhood interatomic interaction model to obtain the

lattice diffusion coefficient and he suggested the negative viscosity change

from pressure increase. Important progress of theoretical approaches was

molecular dynamics (MD) because it gives a direct simulation of moving

atoms in crystals. The Einstein relation D = (d/dt)<Rn^2>/6 where Rn is the

position of nth atom, while the angle bracket indicate the average of all

atoms. This relation is widely employed for providing diffusion coefficients

from MD simulations. Tsuchiyama and Kawamura [1994] simulated

diffusion of noble gas in MgO crystal. They reported large ( ± 3times)

expected errors in their data because of their short MD running times. There

is a technique to avoid the difficulty: adding an artificial force to an atom

surrounding a vacancy to migrate the atom into the vacancy. Although

simulation of diffusion by the method with short times becomes possible, we

cannot use time information of the MD method with this technique. This

vacancy hopping is too artificial even natural vacancy hopping is

spontaneous. Ita and Cohen [1997] employed this technique and they did not

use the Einstein equation directly. Instead, they used Arrhenius formulation

of D=D0exp(-H*/RT), where D0 is pre-exponential term, H* is activation

enthalpy, R is gas constant, and T is temperature, respectively. By using this

equation , diffusion coefficient obtained under two assumptions (1) that

3
activation enthalpy H* is same with energy barrier of the migration, and that

the pre-exponential term D0 does not have pressure dependence. Although

assumption (1) is widely accepted, assumption (2) has not been examined

yet.

The goal of this thesis is to get insight of rheological property of the lower

mantle from studies of lower mantle minerals' diffusion. To achieve the goal,

MD method is employed because of the great advantage mentioned below.

Regarding the precision of calculating thereafter considered that MD with

spontaneous vacancy migration might give diffusion coefficients with certain

precision by increasing running time. Increasing running time require much

more computations so that a supercomputer (Earth Simulator, Japan Agency

for Marine-Earth Science was used. By using this, MD runs with 20nsec

(corresponding to 107 steps integrations) became possible and it was found

that these long MD runs meet enough exact calculations of diffusion. The

author targets major minerals of the lower mantle (periclase [MgO] and

perovsktie [MgSiO3]). In addition, because strange behaviour of MgO

diffusion of pressure dependence was found reported in Ito and Toriumi

[2007], diffusion of noble gas (Ne) were examined in order to infer the origin

of the strange behaviour by comparisons. Results were tried to apply for

lower mantle rheology to evaluate the depth variation of the lower mantle.

4
Chapter 2

Molecular Dynamics

2.1 Introduction

Possibilities of theoretical calculations to give information of lower mantle

diffusions were mentioned at previous chapters. There are several

theoretical methods used for simulating atomic scale phenomenon and

Molecular dynamics is one of them. Besides molecular dynamics, other

method such as energy calculations Karki and Khanduja [2006, 1], [2006, 2]

and Monte Carlo Alfe et al [2005] have been used for study of lower mantle

minerals. Although these calculations successfully applied for their interests,

for example, structures of vacancies in MgO periclase and MgSiO3 perovskite,

B1-B2 transitions of MgO periclase, respectively, Molecular Dynamics (MD)

is considered to be more appropriate for investigating diffusion because it

gives trajectories of vacancies/atoms, which directly related with diffusion

coefficient by Einstein relation, as mentioned in the previous section. This

chapter describes MD method of diffusion. The description includes how this

model is applied to natural system, how calculations are performed, and how

physical properties are given from results.

5
2.2 Approximations of the natural system

The MD models makes a method in the natural system by (1) particles

with mass, (2) these interactions, (3) movement of the particles following

Newton’s equations of motions. Figure 2.1 represents whole system of

molecular dynamics. These MD simulations target reproducing materials of

the natural system. One of the difficulties at simulation of natural material

is to be required huge number of atoms contained. When we consider

materials with visible scale (e.g. 1mm) grain size, number of atoms is order

of the Avogadro number [6.02*1023]. This number is apparently too huge

simulated by computes so MD approximates them by small number of atoms

and its 3D infinite repetitions. We have to become careful for applying this

approximation because it might introduce unexpected effect in the results of

simulation. This effect can be examined based on the assumptions that it

becomes significant when size of basic cell become smaller and that it decay

with increased basic cell size. Based on the assumptions, this effect is

considered to be ignorable if we find conversions of interested physical

properties under running of MDs of ambient basic cell sizes.

6
2.3 Interatomic potentials

In the bulk material, major forces that act in the material are

electron-electron and electron-nucleus interactions. These interactions forces

are given by Schrödinger equations of n-electrons system. Common

techniques to solve them are Hartee-Fock method and density functional

theory [Hohenberg and Korn, 1965]. Based on the density functional theory,

Car-Parrinello method [Car and Parrinello, 1985] realizes the molecular

dynamics, in which all the interactions are given by the solution of

Schrödinger equations. Although Car-Parrinello method is commonly used to

relax a crystal structure, its successful applications to reproduce thermal

vibration are very difficult because it requires much calculation than

relaxation. Considering realistic simulation of atoms by MD with increasing

numbers of steps, these methods are not available for actual simulation of

diffusion.

Instead of these ab-initio approaches, there are an alternate ways to give

the interactions: introducing the simple model to describe interatomic

potentials. Although accuracy of MD is governed by adopted interatomic

potential type and their parameters in the case, these models are well

studied with good accuracy. Regarding lower mantle materials, Matsui

[1988] reported that pairwise potential effectively reproduces equation of

state (EOS) of thermal expansions and compressibility of MgSiO3 perovskite.

Matsui [1989] indicated that heat capacity under Debye temperature

improves by using quantum corrections obtained from Wigner-Kirkwood

Expansion [Landau and Lifshitz, 1958] in a pairwise potential model.

7
Recently, Matsui et al. [2000] used more complicated model (pairwise

potential + breathing shell model for O replusion) improves EOS of MgO at

high temperature and pressure conditions. It has been pointed out that

simple pairwise potential model satisfy Cauchy relation

c11 − c44 − 2 P = 0 , (2.a)

,where c11 and c44 are elastic coefficients and P is pressure. Tsuchiya [2000]

reported that elasticity of MgO is improved by introducing more complicate

charge equilibrium model [Rappe and Goddard III, 1991]). However,

successes of these complicate models are limited in material with simpler

crystal structure (MgO), and applications for materials with more complex

structures have not been carried out. imple pair potentials is considered to be

available for aim of this study because (1) they have been already confirmed

to work well for both MgO and MgSiO3 (2) main idea of this thesis is to get

accurate diffusion coefficients with increasing number of steps. Simulations

of lower mantle minerals' (MgO periclase and MgSiO3 perovskite) properties

have been successfully performed by using Born-Mayer-Huggins type

potential [Matsui 1988]:

qi q j ai + a j − rij ci c j
U ij (rij ) = + f (bi + b j ) exp( )− , (2.1)
rij bi + b j r6

where first (1/r), second (exp(-r) ), and third (1/r6) terms represent the

Colomb, the Pauli repulsion, and the van der Waals energy, respectively. rij

shows the interatomic distance between i-th and j-th atoms, and f is the

standard force constant 4.184kJ/mol. The effective charge q, the repulsive

radius a, the softness parameter b and the van der Waals coefficient c are

parameters listed later. Considering pressure-dependency of diffusivity Ito

8
and Toriumi [2007] derived, the author compares the diffusivity of Ne crystal

of which the pairwise is governed by Lenard-Jones type as follows:

⎡⎛ σ ⎞12 ⎛ σ ⎞ 6 ⎤
U ij (rij ) = 4ε ⎢⎜ ⎟ − ⎜ ⎟ ⎥ (2.2)
⎣⎢⎝ r ⎠ ⎝ r ⎠ ⎦⎥

where parameters are ε and σ . Two ways for obtaining interatomic

potential parameters have been considered, one is empirical way and the

other is given by electronic state calculations. Empirical way is usually based

on fitting of EOS of varied P-T conditions. Further, non-empirical (called "ab

inito") approaches are more preferable because of the potential in the system

multi-body interactions. Electronic state calculations usually have been

studied to obtain the non-empirical interatomic potential parameters.

Kamiya [1996] reported that parameters of MgO given by electronic state

calculations transit B1-B2 phase transitions at the middle of the lower

mantle pressures (90GPa), though shock-wave experiments [Duffy et al.

1995] reported that B1 phase of MgO is stable even in higher pressures of

the bottom of the lower mantle (135GPa). Discrepancy between them seems

that the simple model does not provide enough accuracy of approximations.

9
Table 2.1 Parameter sets used in this study. Upper MgO (Akamatsu and

Kawamura [1998])), and middle MgSiO3 (Oganov et al. [2000]) are

parameters of Born-Mayer-Huggins formula (eq. 2.1). Lower Ne (Kittel 1972)

are based on Lennard-Jones model (eq. 2,2)

Ion z a/Å b/Å c/[(KJ/mol)1/2 Å3]


O -1.56 1.8137 0.172 45.001
Mg 1.56 1.2391 0.075 6.136

z a(Å) b(Å) c/[(KJ/mol)^1/2 Å3]


Si 2.904 1.620 0.145 0.000
O -1.605 1.614 0.138 36.532
Mg 1.910 1.635 0.149 0.000

ε(10^-16erg) σ(Å)
Ne 50 2.74
Ar 167 3.40

10
2.4 Ewald method

In the pairwise model used in this work, total pontentials of an atom are

given by contributions of all other atoms as follows;


U i = ∑ U ij , (2,3)
j =0

where term Uij is the pairwise potential, which is given by equations (2.1)

and (2.2) with their parameters (table 2.1). Here, we must consider the ways

to obtain values of the infinite summation by finite sum. Forms of distance r

dependencies of each term of (2,1) and (2.2) are follows: 1/r (first of (2.1)),

exp(-r) (second of (2.1)), 1/r6, (third of (2.1) and second of (2.2)), 1/r12 first of

(2.2)). Firstly, excluding 1/r term, it is found that sum is quickly converged

with increasing r. This feature can be used for introducing certain cut-off

radius so that the summation is considered to be converged by the sum of all

pair, which interested atom is center and its radius is the cut-off distance. In
& ) by Hirao and
this study, empirically derived cut-off distance (7.5 A

Kawamura [1994]) is adopted.

It is known that this approach does not work well for long-range 1/r

forces because simple summation of 1/r terms starting from the short to the

long distance does not converge; contributions are still effective when

distance r becomes very long. However, the way to calculate converged value

is also known as Ewald method [Ewald 1921]. In the method, Fourier

transform is applied to the infinite summation (eq. 2.1) and it is split as

follows:

11
U = U1 + U 2 + U 3
erfc(αrij )
U 1 = qi ∑ q j
j rij

q ⎧ − π 2 ⎣n'⎦ 1 ⎡
U2 = i
πV
∑n ⎨exp( α 2 ) ⎣n'⎦ ⎢cos(2πn ⋅ ri ' )∑j q j cos(2πn ⋅ r j ')
⎩ ⎣ (2.4),
⎤ ⎫⎪
− sin( 2πn ⋅ ri ' )∑ q j sin( 2πn ⋅ r j ')⎥ ⎬
j ⎦ ⎪⎭
qi α
2
U3 =
π

where U is total energy, which is corresponding to pairwise potential

energy Uij in eq. (2.3). U1, U2, U3 are split components and usually called

real space term, reciprocal space term, and self energy term, respectively.

The vector n = (n1, n2, n3) in the reciprocal term is called wave number

vector, which elements n1, n2, n3 are zero or positive integers. From

formulations of real and reciprocal terms, terms (erfc(x) and exp(-x^2)),

which rapidly converged to zero with increasing distance, are found so that

whole energy U is converged. Actually, it is known that this splitting gives

certain converged value of the Colomb term (1/r). The Ewald method is

applied in this study. In the Ewald method, there are two elements increased

with increasing summation; distance r of real space term and the wave
& for
number vector n. In this study, empirical cut-off thresholds of 15 A
2
distance r and n <= 23 for wave number vector n (Hirao and Kawamura

[1994]) is used.

12
2.5 Equations of motions

When potentials of all atoms are given, their positions are given by

integrating Newton’s equations of motions. Newton’s equations of motions

can be discretized for numerical integrating by computers. The author uses

following Verlet forms for discretiation:

F (t )
r (t + ∆t ) = r (t ) + ∆r (t ) + (∆t ) 2 ,
m
∆t (2.5)
v (t + ∆t ) = {F(t + ∆t ) + F(t )}.
2m

where r, v, F, t, and, m are position, velocity, applied force, time, and mass,

respectively. Empirically it is reported that appropriate length of Δt is 2fs

(Hirao and Kawamura [1994]). Effects of Δt to the total energies of the

system are calculated in order to explore the appropriateΔt as listed in

Table 2.2. In the table 2.2, we would find the effect of Δt over 6ps. It is

confirmed that Δt = 2fs is appropriate one so that this is used here.

∆t(fs) E(KJ/mol)
1 -2487.606
2 -2487.614
3 -2487.614
4 -2487.611
5 -2487.605
6 -2487.579
10 -2487.282
15 diversed

Table 2.2 Total energy E dependencies of the time step of integrations. The

MD calculations are that of MgO. Temperature and pressure (300K and

13
0Gpa) are controlled by scaling method and the running time is 30ps. At 15fs,

the time integrations are diversed during MD calculation.

2.6 Temperatures, Pressure, and their controlling

Temperatures (T) of MD system is given by following relations of statistical

mechanics

1 3

2
mi vi = Nk bT (2.3)
2 i 2

and pressure is derived from Virial theorem, which is expressed as follows:

Nk BT 1
P=
V

3V
∑∑ (−F ) ⋅ r
i< j
ij ij (2.4)

where N is the number of atoms and V is volume of the system. We might

⎡ ⎡ Pxx Pxy Pxz ⎤ ⎤


⎢⎢ ⎥⎥
consider tensor expression of the pressure P = ⎢ ⎢ Pxy Pyy Pyz ⎥ ⎥ when
⎢ ⎢ Pxz Pyz Pzz ⎥⎦ ⎥⎦
⎣⎣

considering non-hydrostatic pressures. Tensor expressions of the Virial

theorem as follows:

14
1 ⎛ Fij ⎞
⎜ Nk T − ⎟
Pxx =
V ⎜ B yy ∑∑
i< j rij
rijx
2

⎝ ⎠
1 ⎛ Fij ⎞
⎜ Nk T − ⎟
∑∑
2
Pyy = rijy
V ⎜ B yy i< j rij ⎟
⎝ ⎠

1 Fij ⎞
⎜ Nk T − ⎟
∑∑
2
Pzz = rijz
⎜ B zz
V i< j rij ⎟
⎝ ⎠
1 ⎛⎜ Fij ⎞

Pxy =
V⎜
Nk B T xy − ∑∑
i< j rij
rijx rijy

⎝ ⎠
1 ⎛⎜ Fij ⎞

Pxz =
V⎜
Nk B T xz − ∑∑
i< j rij
rijx rijz

⎝ ⎠
1⎛ Fij ⎞
Pyz = ⎜ Nk B T yz − ∑∑ rijy rijz ⎟,
V⎜ i< j rij ⎟ (2.5)
⎝ ⎠

where tensor elements of temperatures Txx,Txy,Tzz,Txy,Txz, Tyz are given by

∑m v ∑m v ∑m v
2
i ix v
i ix iy v
i ix iz
Txx = i
, Txy = i
, Txz = i
,
Nk B Nk B Nk B
. (2.6)
∑m v ∑m v ∑m v
2
i iy v
i ix iz v
i iy iz
Tyy = i
, Txz = i
, Tyz = i
,
Nk B Nk B Nk B

Controlling temperature conditions are simply realized by scaling velocities

of atoms and lengths of basic cells. Scaling factors are empirically given and

amount of this factor would be controlled with monitored temperature and

pressures. The scaling method was firstly reported by Woodcock [1976] and

15
recent examinations (Hirao and Kawamura [1994]) also reported these

methods effectively works for ambient material models. More complex

controlling methods (Nose [1984] for temperatures and Anderson [1980] for

pressures) realizes thermodynamically rigorous ensembles, though they

were not used because equations of motions of heat baths (Nose-Hoover

method) or boundaries of basic cells (Anderson method) sometimes diversed

when atoms migrate into vacancies. Instead, we used simple scaling method

because of its stability.

16
2.7 Thermodynamic properties

Results directly given by MD contain positions, velocities, and forces

of each atom and steps so that all results should be extracted from them. The

way to calculate temperature T and pressure P were shown in the previous

section. Internal energy U and kinetic energy K are the summation of

potential energies and kinetic energies of all atoms. Relations of E=U+K and

H=E+PV give total energy E, and enthalpy H. Temperature derivatives of

the enthalpy H under isobaric of isothermal conditions give the isobaric heat

capacity CP and isothermal heat capacity CT, respectively. Calculating

entropy S by MD is difficult. Its calculation principally requires reproduction

of macroscopic disorder of the natural system, though it is prevented by

periodic boundary conditions of MD. Gibbs energy G=H-TS) is also difficult

to derive because it requires information of entropy S.

These thermodynamic properties are derived from microscopic MD

systems. Consistency of results of MD systems and natural macroscopic is

not naturally assured so that we consider meaning of microscopic MD

systems from a viewpoint of statistical dynamics. In this study, all MD

calculations are performed under thermal equilibrium attained by

relaxations as mentioned before. In the case, temperatures and pressures are

not controlled but no energy inflow/outflow is kept, the probability

distribution of MD system corresponds to the micro-canonical ensemble of

statistical mechanics, whereas the MD system corresponds to ground

canonical ensemble in the case that temperatures and/or pressures are

controlled. Another names of ensembles (NVE for adiabatic and isochoric

17
ensemble, NVT for isothermal and isochoric ensemble, NPT for isothermal

and isobaric ensemble) are often used and these names are adopted in this

thesis. In any ensembles (NVE, NVT, NPT), macroscopic thermodynamic

amounts are given by simple average of all microscopic thermodynamic

amounts given along the trajectory of topological space as follows:

1 t
Amacro = ∑ Amicro (τ )
t τ =1
(2.7)

where Amacro is macroscopic thermodynamic amounts, which is our interest,

and Amicro is microscopic phenomenon. Ensemble averages in relating

macroscopic and microscopic thermodynamic quantities are based on ergodic

hypothesis, and thus equation (2.7) is assumed to be converged after enough

long time. With the assumption, we can apply equation (2.7) for obtaining

thermodynamic quantities by means of MD.

18
2.8 Melting Temperatures

MD calculations can reproduce not only solid states but also liquid

states of the material successfully so that properties of melts of which bulk

compositions are approximately mantle materials have been investigated by

Molecular Dynamics (e.g. Kubiki and Lasaga 1991, MgSiO3). Temperatures

of solid -> liquid transition (melting temperature) has been interested

physical properties because melting has played important role in Earth’s

material. For example, ocean plates are generated by chemical depletion

from mantle material caused by partial melting. However, it is known that

MD does not reproduce solid -> liquid transition (e.g. melting) well. The

simplest method we would consider is increasing temperature of perfect

lattice so that trying to observe the solid -> liquid transition of the MD

system, though it is known that transitions by this way cause unrealistically

high temperature, indicating significant overheating in the MD systems.

Chaplot et al. [1998] try to avoid this by inserting vacancies in the MD

system based on idea that in the actual material vacancies have higher

energies than around parts so that they might become a nucleus of melt and

prevents the overheating. They compares melting temperatures of perfect

and defect contained systems of MgSiO3 perovskite and reported that the

overheating are reduced in defect contained systems.

Belonoshko [1994], Belonoshko and Dubrovinsky [1995,1996] gives a new

method to give melting temperatures of MD system. Fundamental idea of

the method is to (1) building MD system by both solid and liquid systems (2)

two phases are contacted and separated (3) keeping system in a temperature

19
and pressure and trying to observer growth of more stable phases. Figure 2.5

shows snapshot of the two-phase method. If we observe solid part grows in a

temperature, this shows that the temperature is under melting temperature

and over melting temperature in case liquid part growths. Considered

advances of the two-phase method are (1) this method is fundamentally free

from overheating because liquid phase always exists (2) this method

separates given temperature is under or over melting temperature clearly.

Without this method, we have to run MD for a long time in order to confirm

solid -> liquid transition does not happen, though the way to decide running

time is not clear. In this thesis, two-phase method is performed to give

melting temperatures for targeting to clarify relations between melting and

diffusion and examinations of potential models. Results will be reported in

later chapters later.

Figure 2.5 The Snapshot of initial structure of two-phase method MD

simulation. The left side is in solid phase and the right side is in liquid phase.

The growth of which phase will be observed with running.

20
2.9 Elastic Constants

Elastic constants are the proportional coefficients of the generalized

Hook’s law,

3 3
σ ij = ∑∑ cijkl ε kl , (2.16)
k =1 k =1

where σ is stress tensor, ε is strain tensor, and c is elastic constants. Since

σ and ε are both two-dimensional 3 × 3 tensor, c is the four-dimensional

tensor, which have 81 elements. Elastic constants cijkl have the symmetry of

cijlk=cjikl=cijlk=clkij. Therefore, by introducing brief expression of indexes 11→1,

22→2, 33→3, 23→4, 31→5, 12→6, equation (2.16) is expressed as

⎛ σ 1 ⎞ ⎛ c11 c12 c13 c14 c15 c16 ⎞⎛ ε 1 ⎞


⎜ ⎟ ⎜ ⎟⎜ ⎟
⎜σ 2 ⎟ ⎜ c22 c23 c24 c25 c26 ⎟⎜ ε 2 ⎟
⎜σ ⎟ ⎜ c33 c34 c35 c36 ⎟⎜ ε 3 ⎟
⎜ 3⎟=⎜ ⎟⎜ ⎟. (2.17)
⎜σ 4 ⎟ ⎜ c44 c45 c46 ⎟⎜ ε 4 ⎟
⎜ ⎟ ⎜ ⎟⎜ ⎟
⎜σ 5 ⎟ ⎜ c55 c56 ⎟⎜ ε 5 ⎟
⎜σ ⎟ ⎜ c66 ⎟⎠⎜⎝ ε 6 ⎟⎠
⎝ 6⎠ ⎝

The half of the elastic constants matrix is omitted for symmetry. The

number of independent components of elastic constants, which are in

equation (2.17), is 21. In a cubic crystal as MgO, crystallographic symmetry

reduces the independent elastic constants as

⎛ σ 1 ⎞ ⎛ c11 c12 c12 0 0 0 ⎞⎛ ε 1 ⎞


⎜ ⎟ ⎜ ⎟⎜ ⎟
⎜σ 2 ⎟ ⎜ c11 c12 0 0 0 ⎟⎜ ε 2 ⎟
⎜σ ⎟ ⎜ c11 0 0 0 ⎟⎜ ε 3 ⎟
⎜ 3⎟ = ⎜ ⎟⎜ ⎟. (2.18)
⎜σ 4 ⎟ ⎜ c44 0 0 ⎟⎜ ε 4 ⎟
⎜σ ⎟ ⎜ c44 0 ⎟⎜ ε 5 ⎟
⎜ 5⎟ ⎜ ⎟⎜ ⎟
⎜σ ⎟ ⎜ c44 ⎟⎠⎜⎝ ε 6 ⎟⎠
⎝ 6⎠ ⎝
21
In the case of cubic crystal, the number of the independent components of

elastic constants are only three: c11, c12, and c44.

Although unit cells are cubic (MgO and Ne) and orthorhombic (MgSiO3

perovskite) are considered, basic cells of MD could be parallelepiped so that

we can apply available strains to the crystals by distorting basic cells

artificially. Resultant from strains are measured by the generalized

expression of pressure (equation 2.6). Therefore, we can give the elastic

constants from relations with observing strains by employing equation (2.18).

Two ensembles (NVE and NVT) are available for MD: NVE and NVT

ensembles are corresponding to adiabatic and isothermal elastic constants,

respectively. Hook’s law between stresses and strains is only avaiabler for

small stains and so the law would be checked for application of MD studies.

In addition, effects of ensemble differences (NVE and NVT) can be checked

by comparing MD results under the two ensembles. Figure 2.5 shows the

relations of strain and elastic constants of MgO. Results of c11 have certain

strain dependence, whereas no relation betweens results of c12, c44 and strain

are observed. However the strain dependence of c11 are less than 5% when

the strain is 0.01. The difference of results from the kind of elastic constants

(adiabatic one or isothermal one) is hardly observed. In this study, the

isothermal elastic constants are mainly simulated because many

experimental studies have given elastic constants under isothermal

conditions.

22
400
c11(adiabatic)
c11(isothermal)
c12(adiabatic)
elastic constants (GPa) c12(isothermal)
300 c44(adiabatic)
c44(isothermal)

200

100

0
0 0.01 0.02 0.03
strain

Figure 2.5 Strain dependencies of elastic constants. The pressure and

temperature are 0GPa 300K.

23
2.10 Self-Diffusion

In case that there is a constant in concentration of random by walking

particles, it is observed that the contrast becomes relaxed with time by

exchanges of momentum. This phenomenon is called diffusion Statistical

mechanics enables us to give with time t and position x evolution of particle

concentration c(t,x) with the equation called diffusion equation as follows:

∂c ∂D ∂c ∂ 2c
= +D 2 . (2.18)
∂t ∂x ∂x ∂x

One parameter, D is diffusion coefficient and it defines how the system

evolves with varying time t and position x. In this study, middle term of

above diffusion equation, ∂D / ∂x = 0 so that the diffusion equation

(equation 2.18) is simplified as follows:

∂c ∂ 2c
=D 2 . (2.19)
∂t ∂x

Diffusion is free from external forces is specially called as self-diffusion and

it is characterized by the self-diffusion coefficient. Crystals always have

point defects because of conditions of thermally induced defect concentraion

[Shewmon 1973] or point defects doped ions induced. Interstitial atoms are

possibly induced but energy calculations as reviewed by Ando [1989] indicate

that vacancies are more likely to be introduced because of lower formation

energies. Self-diffusion of simple ionic crystals including MgO has been

observed as migrations of atoms into vacancies in MD (e.g. Tsuchiyama and

Kawamura [1994]). Figure 2.5 draws trajectories of atoms around and is

migration into vacancies as an example.

24
Figure 2.5 Trajectories of atoms during vacancy migration. Initial structure

of defect is a pair of neighboring Mg and O vacancies. MD simulations are

performed at 2500K and 0GPa during 20ps (10,000 steps).

25
Einstein’s theory of Brownian motion includes a formula, which relates

macroscopically defined self-diffusion coefficient D with microscopic

trajectory of atoms, which is observable by MD. The formula (called Einstein

relation) is as follows:

2
Rn = 6 Dt , (2.20)
atom

where Rn is the displacement vector of each atom and the mean < > is of all

atoms. The term |Rn|2 is called mean squared displacement (MSD)

[Shewmon 1963]. Einstein relation (2.20) indicates that relation between

MSD |Rn|2 and times t would be linear, so that we can decide identical

self-diffusion coefficient D in case linear relation is found by plotting them in

a 2D plot. Figure 2.6 shows two examples of relaxations of 5000K-80GPa and

4000K-80GPa. The 5000K is considered to reproduce self-diffusion with

enough accuracy, and it is hardly detected in the case of 4000K and 80GPa

and self-diffusion coefficients seems to difficult to estimate. Poor MD steps

cause this fail.

26
1.5

1
MSD 5000K, 80GPa
(Å2)
0.5
4000K, 80GPa

0
0 500 1000 1500 2000
running times (ps)

Figure 2.6 Examples of relations between MSD (|Rn|2) - running time t ,

materials are MgSiO3 perovskite modeled by Oganov et al. [2000].

27
In materials such as periclase and perovskite dominant mechanism of the

self-diffusion is vacancy hopping. The vacancy hopping is classified in two

types; extrinsic and intrinsic. The 'extrinsic' case indicates that vacancies are

formed by external sources and the 'intrinsic' case means that vacancies are

thermally formed. Typical external sources are impurity atoms, which have

different charges from host atoms. In the MgO crystal, two trivalent

impurity atoms form one Mg vacancy but two monovalent impurity atoms

form one O vacancy for keeping charge neutralities. If there is no external

sink and source of the impurities, the vacancy concentrations do not depend

on temperatures or pressures. In this case, the self-diffusion coefficient is

proportional with vacancy because vacancy concentration keeps constant.

The relationship can be written as D=(C/C0)*DC0 where D and C is

self-diffusion coefficient and vacancy concentration of the extrinsic case and

DC0 and C0 is those of MD calculations, respectively. In addition, this

relationship gives that activation energies of the extrinsic case and the MD

calculation are equal with each other. This relationship is expressed as

D=(C/C0)*DC0,0exp(-Hm*/RT) (2.23)

where DC0,0 is pre-exponential factor, Hm* is the activation energy of the

MD self-diffusion coefficients. The Hm* is often called migration energy of

vacancy. When intrinsic, O and Mg vacancy concentrations (Cint,O, Cint,Mg)

depends on Hf* as
*
Cint ,O Cint ,Mg ∝ exp(− H f / RT ) , (2.b)

The vacancy concentration of O and Mg is considered to be the same

because of charge neutrality. This relation can be written as

D=(C/C0)*DC0,0exp(-Hm*/RT) (2.24)

28
where DC0,0 is pre-factor, Hm* is activation energies of the MD self-diffusion

coefficients. By inserting the Cint,O (or Cint,Mg) to the C of equation 3, the

following relation
* *
D ∝ exp[−( H m + H f / 2) / RT ] (2.25)

is derived

Now two equations [(2.23) and (2.25)] are available for both mechanisms

(extrinsic and intrinsic) to implicate MD results. Ando et al. [1983] measured

self-diffusion coefficients D of Fe3+ doped MgO and reported that higher and

lower temperature regimes are considered to be intrinsic and extrinsic,

respectively because self-diffusion coefficients of low T regime depend on

impurity Fe3+ concentrations but in higher regime, they do not depend on the

concentration of Fe3+.

29
2.11 Accuracy of MD Self-Diffusion Coefficients

Chapter 1 mentioned the author's idea that accuracy of self-diffusion

coefficients MD obtains is improved by increasing MD running times. This

idea was examined by MD calculations. The MD calculations were performed

for MgO (Akamatsu and Kawamura [1998]). In the MD, two different [2ns

and 0.4ns] running times were set. Self-diffusion coefficients were derived by

Einstein equation (equation 2.20). Temperature and pressure were kept in

2500K and 0GPa, respectively, and both MD calculations were repeated in

ten times in order to get amount of scattering of the data. Figure 2.7 shows

the self-diffusion coefficients (D). It was found that scatting of D of 2nsec

[1011.5-1011.3(m2/sec)] is smaller than that of 0.4nsec [1011.6-1011.3(m2/sec)] and

the idea is supported. In case average of the repeated results is adopted as a

result, amount of error is estimated from standard deviation 1σ . In the case

of 2ns with ten times repeats, 1σ =11% is estimated as an amount of error.

The 11% seems small enough for quantitative discussion. Therefore we adopt

the procedure of 2nsec for all calculations of self-diffusion coefficients in this

study. Number of steps are 2nsec*10 / 2fs = 107 per one

pressure-temperature condition.

30
-10.5

log10(D(m /sec))
-11
2

-11.5
4ns 20ns

-12

Figure 2.7 Results of calculation time dependencies of self-diffusion

coefficients. Error bars show standard deviations. Both 0.4ns and 2ns

calculation times are repeated in ten times. Numbers of steps are 2 ×105

(0.4ns) and 1 ×106 (2ns), respectively.

31
2.12 Vacancy formation energies

In the case of intrinsc vacancies, formation of vacancy is the enthalpy of

the system, ∆H as

∆H = nH *f (2.24)

where n is number of vacancies and H *f is vacancy formation energy.

When considering entropy, formation of vacancies increases configuration

entropy. Configuration entropy is expressed by number of possible states w

as follows:

∆S = k b ln(w) . (2.25)

Number of possible configurations of vacancies in the system composed of N

atoms is

( N + n)!
w= . (2.25)
N !n!

Therefore, increase of entropy with introducing n vacancies in N crystal is

( N + n)!
∆S = k b ( ) ≈ k b [( N + n) * ln( N + n) − N ln( N ) − n ln(n)] (2.26)
N !n!

Here, Stirling's approximation ln( x!) = x ln( x ) − x is used. Thermal

equilibrium achieves when system's free energy G = H − TS is minimized

and the minimum regarding vacancies consists ∂G / ∂n = 0 . Therefore, we

introduce equilibrium vacancy concentration c = n / N as

∂G ∂H ∂S *
= − T * ( ) = H f − T * k b [ln( N + n) − ln(n)] = 0 .
∂n ∂n ∂n
*
n n Hf
≈ = exp( − ). (2.27)
N +n N k bT

32
We might consider unit of vacancy as mole. In this case, Boltzmann's

constant will be replaced to Gas constant R, therefore,


*
Hf
c[/ mol ] = exp( − ) (2.28)
RT
Stoichiometric oxide such as MgO requires charge neutrality as follows:
''
[VMg ] = [VO'' ]

Therefore, concentrations of both vacancies are:

*
Hf
[V ] * [V ] = exp( −
''
Mg O
''
)
RT
* . (2.29)
H f /2
''
[VMg ] = [VO'' ] = exp( )
RT
*
In this case, vacancy formation enthalpy H f is defined for a pair vacancy
''
[VMg ] − [VO'' ] .

33
Chapter 3

Periclase (MgO)

3.1 Introduction

Periclase(MgO) is a mineral important in Earth's lower mantle as a major

phase, which is thought to be 20% volume fraction. Crystal structure of it is

simple NaCl-type [B1]. This structure is stable even in ordinary pressure

and higher pressures. Under the lower mantle conditions (25-135GPa),

transition of NaCl-type structures to CsCl-type structures [B2] has been

confirmed in several materials (NaCl at 29GPa [Y. Sato-Sorensen 1983],

CaO at 68GPa [Richet et al. 1988]). However, MgO is the NaCl-type

structure even at 225GPa (shock compression experiment [Duffy et al.

1995]).

Figure 3.1 NaCl-type Structure

34
3.2 Equation of State

Although MgO is considered to be one stable phase (B1, NaCl type) in the

lower mantle, its equation of state (EOS) has been important. MgO's thermal

expansion at ordinary pressures has been reported by Dubrovinsky and

Saxena [1997] up to 3000K and compression at ordinary temperature is

reported by Mao and Bell [1979] up to 100GPa, so they are considered to be

good as an first step test of the MgO's interatomic potential. Comparisons

are shown in Figure 3.3 and Figure 3.4. Figure 3.4 also shows data of Fei

[1999], which examined affects of non-hydrostatic effects of uniaxial

compression of diamond anvil cell (e.g. Mao and Bell [1979]) with effects of

improved hydrostatic pressure conditions by using solid neon as a

pressure-transmitted material. Effect of non-hydrostatic conditions seems

not to be so significant, and MD's data for both thermal expansion at

ordinary pressure (Figure 3.3) and compression at ordinary temperature

(Figure 3.4) show well agreements with those of experiments.

Dewaele et al. [2000] reported compression of MgO in the wide range of

pressure (-50GPa) around high (2000K) temperatures so that this data was

compared with MD's results (shown in Figure 3.5). This figure also includes

data of breathing shell model [Matsui 2000]. MD's results seem to

overestimate the molar volume of high pressure. Relative difference ∆V / V of

the molar volume is estimated to be 1.1% at 50GPa by adopting MD's results

as 9.62cm3/mol at 2000K-50GPa and results of Dewaele et al. [2000] as

9.51cm3/mol at 2050K, 48.2GPa. Result of breathing shell model

[9.45cm3/mol, 2000K 50GPa, Matsui et al. 2000] are better agreements with

35
experiments (Dewaele et al. [2000]). However, results of temperature

estimations of (Dewaele et al. [2000]) includes large uncertainness ( ± 200K

according to their estimates) have preventing more certain discussion of

accuracy of these data. Aside of the detailed comparison, estimated amount

of difference ∆V / V is small 1.1% so that physical properties investigated by

molecular dynamics described in chapter 2 is considered to be reasonable

even in high-temperature and high-pressure conditions.

14
Molar Volume (cm /mol)

This Study
Dubrovinsky and Saxena (1997)
13
3

12

11
1000 2000 3000
T(K)

Figure 3.3 Temperature dependence of molar volume. Experimental

measurements [Dubrovinsky and Saxena, 1997] was performed for

periclase's powder holded by a hole of tangsten (W) wire and temperature is

measured by gray body radiation of the tangsten.

36
12

This Study

Molar Volume (cm3/mol)


11 Fei(1999) hydrostatic
Fei(1999) Non-hydrostatic
Mao and Bell (1979)
10

0 100 200
Pressure (GPa)

Figure 3.4 Pressure dependence of molar volume. Pressures determinations

were by ruby fluorescene technique (Mao and Bell [1979]) and NaCl and

Gold's equation of states (Fei [1999]).

13
Pair Potential (This study)
Experimental (Dewaele et al., 2000)
molar volume (cm /mol)

12 Breathing shell model (Matsui, 2000)


3

11

10

8
0 50 100
Pressure (GPa))

Figure 3.5 Pressure dependence of molar volume at 2000K. Temperature

range of experimental values (Dewaele et al., 2000) is from 1950K to 2050K.

37
3.3 Melting Temperatures

The melting temperatures are fundamental physical properties for earth's

materials because solid loses rigidity and starts flowing as liquid so that

dramatic macroscopic dynamics are expected. Because possibility of

large-scale melting in the lower mantle is believed to be unlikely, the primal

geophysical significance of melting of lower mantle materials is to give upper

constraints of geotherm at lower mantle. Next, melting temperatures have

been used for estimations of activation enthalpies of self-diffusions because

detection of melting is experimentally easier than direct measurements of

self-diffusion coefficients. Following van Liempt relations

H * = gTm (3.1)

where H * is activation enthalpies of self-diffusion, g is a derived constant,

Tm is melting temperature, relates self-diffusion and melting has been

investigated. This relation was firstly found in metallurgy and known to be

well consistent with metals (Figure 3.6). Because of large unknown of

variation of lower mantle's self-diffusion coefficients, the van Liempt relation

(equation 3.1) has used for estimating lower mantle rheology. By inserting

equation 3.1 to the following Arrhenius temperature dependency form of

self-diffusion coefficient

D = D0 exp(− H * / RT ) , (3.2)

relations between self-diffusion coefficients and melting temperature are

derived as follows:

D = D0 exp(− gTm / T ) . (3.3)

Experimental melting temperature of lower mantle materials is partially

38
measured (up to 62.5GPa for MgSiO3 perovskitre, Zerr and Boehler [1993],

up to 31.5GPa for MgO periclase, Zerr and Boehler [1994]). Equation 3.3 has

been used to model lower mantle rheology (Yamazaki and Karato [2002],

Steinberger and Calderwood [2006]) with these experimental data. However,

consistency of van Liempt relation has not been enough confirmed yet.

Poirier [2000] plotted relations between melting temperature Tm and

activation enthalpy for the diffusion based on data compiled by Freer [1980]

(quoted below as Figure 3.7) and reported this plot yield only scattered cloud,

indicating possibility that the van Liempt relation is not held. Poirier [2000]

indicated different of impurity concentrations of each experimental sample

quoted by Freer [1980] as a cause. This report notifies us to be careful for

applying empirical van Liempt relation to the materials of earth's interior

without enough examinations. However, comparison of melting temperature

Tm with H * might be useful as a rough check of MD simulation results

because holded van Liempt relation is thought to be likely by assuming

similarity with metal and mantle materials (silicates and oxides).

39
Figure 3.6 The van Liempt relation (equation 3.1) for some metals. Lines

shows relations with g = 34/R (kcal/K). The figure is quoted from Poirier

[2000] and originally Bocquet et al. [1983].

40
Figure 3.7 Activation enthalpy of diffusion at high temperature 18O vs.

melting temperatures in binary oxides. Data were from Freer [1980] and

pictures were quoted from Poirier [2000].

41
As discussed in section 2.8.1 two-phase method (Figure 2.5) effectively

avoids superheating problem of molecular dynamics so that it was used for

investigating melting temperatures in this thesis. Molecular dynamics

calculations were performed via 100K intervals temperatures at ambient

pressures. In all conditions, detections of growth of either (solid or liquid)

phases are clearly observed so that amount of error expected is under ± 50K.

Error estimation of experimental MgO melting temperature measured by

black-body emission [Zerr and Boehler, 1994] is up to ± 300K in the

literature so that MD's precision in this thesis is considered to be enough for

investigating melting of materials in lower mantle's P-T conditions.

Figure 3.8 compares melting temperatures of ambient pressure conditions

up to 200GPa, which include those of the lower mantle (25-135GPa).

Ordinary pressure's melting temperature by molecular dynamics was

measured as 2750 ± 50K, showing –10% underestimation from experiments

(3040K ± 100K, Zerr and Boehler, 1994]). Comparing with models of this

thesis and previous MD works [Belonoshko and Dubrovinsky, 1996], latter

includes terms of former and additionally Morse potential form, which is

intended to expressed the covalent bonding. Latter model consider smaller

colomb terms (q = 1.251 / e, where e is amount of unit charge) than that of

this study 1.56 / e (Table 2.1). In spite of these differences, no large difference

of results of MD melting temperatures is found, especially lower mantle

pressure range (25-135GPa), indicating that dependence of MD results with

potential model is not so significant. However, Clapeyron slope of MD

(dTm/dP = 60K/GPa) are twice bigger than that of Zerr and Boehler [1994]

(30K/GPa) and resulted melting temperature difference between MD and

42
extrapolated experiments are -2000K (-7000K for MD and -5000K for

extrapolations of experiments, Zerr and Boehler, 1994]). This significant

inconsistence between MD and experiments has remained yet even when

using two-phase model. We consider two possibilities as causes of the

inconsistency (1) MD's overestimation caused by higher configuration

entropies due to three-dimensional infinite repetitions of the small basic cell

(2) underestimation of experiments (Zerr and Boehler, [1994]) due to

temperature gradient in experimental sample spot heating of laser. Idea (1)

MD's overestimation would be examined by increasing MD sizes because

underestimation of configuration entropy decreases with increasing basic

cell size.

10000
Melting Temperature (K)

8000

6000

4000 This study


Cohen and Gong (1994)
Zerr and Boehler (1994)
Belonoshko and
Dubrovinsky (1996)
2000
0 100 20
Pressure (GPa)

Figure 3.8 Comparison of melting temperatures of MD and experimental

results. The triangles and reverse triangles of Belonoshko and Dubrovinsky

(1996) show the ceiling and floor of the results.

43
3.4 Elasticity

Importance of elasticity for earth's dynamic is primly as for providing

insight of fundamental process of seismic wave propagation in earth' interior.

Following formulation of velocities of seismic wave propagations ( V p , Vs ) in

case assuming isotropy of materials have been used as a fundamental

equation by using two independent elastic constants bulk modulus

K = (C11 + 2C12 )/3 and rigidity µ = c 44 where c11, c12, and c44 is the elastic

constants, which express generalized Hook's law (equation 2.17):

K + (3 4) µ
Vp = (3.4)
ρ

µ
Vs = . (3.5)
ρ

Composed crystals of the polycrystalline have anisotropies even the most

symmetrical structure (cubic) so that assumption of isotropy is based on idea

of perfectly untidy distribution of crystallographic orientation of composed

crystal in polycrystalline of earth's interior. Amount of asymmetry, which

examines this assumption, has observed by analyzing surface waves

[Tanimoto and Anderson, 1984] or free oscillations of the whole Earth

[Montagner and Kennett 1996, Figure 3.9]. Important result of the analysis

regarding the lower mantle is its lack of seismic anisotropies, as found in

Figure 3.9. This lack of seismic anistropy is considered to be remarkable.

Sliding of dislocation, which is the most common mechanism of plastic flow,

causes rotations of crystallographic orientations and should have seismic

anisotropies except in case each crystal is perfectly isotropic. Anisotropies of

44
corresponding single crystal can be estimated from elastic constants of

Hook's law (equation 2.17) by using Thomsen's model (Thomsen [1986]).

Although experimental measurements have been limited up to 50GPa, which

corresponds to surface of lower mantle, elastic constants of lower mantle

materials have been well studied by electronic state calculations (reviewed in

Karki [2001]). Because of good consistency between calculations and existing

experimental data, calculated results of significant anisotropies (MgO

periclase and MgSiO3 perovskite, shown in Figure 3.10) have widely

accepted and it is supposed that lower mantle's crystallographic orientation

distributions of grains are perfectly untidy (i.e. there is no preferred

orientations). Based on the idea, Karato [1995] pointed out possibility of

diffusion creep in lower mantle as a cause of lack of seismic anisotropy.

Figure 3.9 Amount of anisotropy ξ = (VSH / VSV ) 2 where VSH and VSV are

horizontal and vertical directions of share (S) wave velocities, respectively.

(this figure is quoted from Karato [2000])

45
Figure 3.10 Amount of anisotropies derived by LDA (local density

approximation) electronic state calculations [Karki et al. 2001]. Three

anisotropy factors azimuthal anisotropies of P waves AP = (VP max − VP min ) / VP , S

waves AS = (VS max − VS min ) / VS , and polarization anisotropy ASPO = (VS1 − VS 2 ) / VS

are used for expressing anisotropies.

46
It is pointed out that Cauchy relation (equation 2.a) is unexpectedly filled

for two-body central interatomic potential used in this thesis. However, we

calculate elastic properties of MD models in order to examine reproducibly of

the model by rough level. Three independent elastic constants (c11, c12 and

c44) of two temperature conditions (300K and 2000K) were calculated under

ambient pressures up to 100GPa. Temperature dependences are shown in

Figure 3.11. Temperature dependence is found for components c11, which

indicates influence of temperature is mainly for resistance of perpendicular

to compression/ tension, as the thermal pressure. Figure 3.12 compares 300K

MD results with those of experiments (Sinogekin and Bass 1999, Zha et al.

2001). Relations between c12 and c44 of MD is inconsistency with those of

experiments. MD's relation is c12=c44 at 0GPa and always c12 < c44. In

experiments, c12 > c44 at 0GPa and c44 increases more rapidly than c12

with increasing pressure so that relation inverts to c12<c44 at 45GPa.

Behaviors of each components (1) increase of c12 with increasing pressure

slows down and turns to be decrease (2) c44 increases more rapidly than c12

(3) c11 is far more drastically than c12 and c44 and approximately linearly

increase with increasing pressure experimentally observed seems to be

similar with MD results. Breaks of Cauchy relation is reported to increase

with increasing pressure (Tsuchiya and Kawamura [2001]). However,

because MD calculations of self-diffusion is performed under nearly

hydrostatic pressures and topic of C12/C44 would be more important at

existence of share stresses, afefct of simplicity of pairwise model used in this

study is expected to be not significantly for calculations of self-diffusion

coefficients even in lower mantle's high-pressures.

47
1000
300K
2000K
800
Elastic constants (GPa) c 11
600

400
c 44
200

c 12
0
0 20 40 60 80 100
Pressure (GPa)

Figure 3.11 MD results of temperature dependences of elastic constants,

c11, c12, and c44.

48
1000
c 11
800 Sinogeikin and Bass 1999

Elastic Constants (GPa)


600
Zha et al. 2001
400
c44

200

c 12
0
0 20 40 60 80 100
Pressure (GPa)

Figure 3.12 Comparisons of MD studies with experimental results, shapes

of marks (circle, square, and triangle) indicates species of elastic constants

(c11, c12, and c44), respectively. Styles (filled, unfilled big, unfilled small) of

marks indicates source of data (this study, Sinogekin and Bass [1999], Zha et

al. [2001]), respectively.

49
3.5 Self-Diffusion
3.5.1 Introduction and methodology

Mechanism of MgO self-diffusion has been well studied in ordinary

pressure, reviewed in Ando [1989]. Mackrodt [1982] compared required

energies of several mechanisms of diffusions (interstitial atoms, vacanies,

and ring diffusions) and indicates vacancy as dominant in self-diffusion

because of much smaller formation energies of vacancies than others. Yang

and Flynn [1994] shows temperature dependence of oxygen (O) diffusion

drawed in an Arrhenius plot (Figure 3.13) has two regions divided by kink of

line. They also showed impurity concentration (Fe3+) dependence of lower

part, indicating extrinsic process (i.e. dominating vacancies formed to

compensate external charge of impurity ions) and independence of higher

part, indicating intrinsic process (i.e. dominating thermally formed

vacancies). Oxygen self-diffusion coefficient are also measured by Yang and

Flynn [1994]. Magnesium [Mg] diffusion is investigated by ionic conduction

(Sempolinski and Kingery [1980]) related with diffusion coefficient by

Nernst-Einstein relationship

z 2 F 2c
σ= D (3.6)
RT

where σ is electronic conductivity, z is valence of drifted ions, F is

Faraday constant, R is gas constant, T is temperature. They reported ionic

conductivities depend on trivalent impurity concentrations related with

extrinsic process.

50
Figure 3.13 Arrhenius Plot of oxygen diffusion, which has intrinsic and

extrinsic regimes, after Yang and Flynn [1994].

51
Understanding of self-diffusion at lower mantle, which this thesis

interests, requires knowledge of affects of huge pressure (25-135GPa) of

lower mantle, though experimental studied had been lacked. Van Orman et

al. [2003] tried to measure self-diffusion coefficients at high pressure

(15-25GPa), which is near those of the lower mantle. Although their

experiments were unexpectedly affected by Al ion diffused from alumina

gasket, they analyzed Mg diffusions by assuming Mg vacancies formed by Al

ion. Experiments were performed under three (15, 16, and 25 GPa) pressures

with same (2273K) temperatures so that activation volume of migration

(V*m) of Mg diffusion was reported. They also tried to measure diffusions of O

ions. Measured values of self-diffusion coefficients were higher (10-15-16

m2/sec) than experiments at ordinary pressures (10-16-17 m2/sec, Ando et al.

[1983], Yang and Flynn [1994]) and they consider diffusion enhanced by

dislocation formed by share stress. Challenges of high-pressure experiments

should be admired, though unexpected affects reported (diffusion of Al ions

and enhances by dislocations) indicate difficulties of such the experiments.

Due to the difficulties, we would consider simulations as a main method to

investigate pressure affect of self-diffusion. Information of formation

enthalpies (Hf*) gives insight of diffusion mechanism because concentration

of vacancies thermally formed are given by the Hf* as equation (2.24).

Pressure dependence of Hf* were calculated by Variational induced breathing

[VIB] model potential (Ita and Cohen [1997]) and LDA (local density

approximation) electronic state calculations (Karki and Khanduja [2006])

and migration energies were also calculated by breathing [VIB] model

potential (Ita and Cohen [1997]). Ita and Cohen [1997] also tried to derive

52
self-diffusion coefficients by modeled defect jumping form, which is composed

of jumping length l, frequency atoms try to jump into vacancies (called

attempt frequencies, ν ) as follows:

Zm 2 - ∆G f /W - ∆G m
D = Zf ν l exp( ) (3.6)
6 k bT

Zf is the number of equivalent ways of forming a vacancy type, Zm is the

number of equivalent diffusion paths, respectively. They considered Gibbs

free energy as thermodynamic expressions of vacancy formations and

migrations. Although this approach does not require direct calculation of

self-diffusion coefficients, the difficulty is that no method to derive the

attempt frequency is known. They estimated the attempt frequency as being

equal to the lowest peak of the frequency spectrum, which is obtained from

the Fourier transform of trajectories of diffusing atoms, though the relation

between this peak and the attempt frequency is not clear.

MD performed for periclase is system composed of 5x5x5 The MD system

prepared was composed of 5x5x5 unit cells of periclase (rock salt structure),

which contain 4 Mg atoms and 4 O atoms per unit cell, with a pair of

vacancies (998 atoms) and three-dimensional periodic boundary conditions.

After the initial relaxation, 2 ns (106 steps) MD was repeated 10 times and

the average of the 10 self-diffusion coefficients (D) was adopted as result.

Precisions of self-diffusion coefficients given by this procedure are discussed

section 2.8.7 and examined as shown in Figure 2.7. Because model case

shown in the section is small enough (11%), self-diffusions coefficients given

by this procedure are considered to be reliable.

53
3.5.2 Results of ordinary pressure

First, we performed MD calculations under various (2000–3400 K)

temperatures at ordinary pressure (0 GPa) with 100 K intervals.

Spontaneous atom migrations, shown in Figure 2.5 were observed at

2400–3000 K. Above 3100 K, the MD systems melted. Under 2300 K, no

vacancy migrations were observed. This is because the probability of vacancy

migration was too small within the time span of the MD simulation. Note

that the running time of our MD simulations (2 ns) was much smaller than

the typical experimental timescale of self-diffusion measurements (over 1

hour). Self-diffusion coefficients were calculated for 2400–3000 K. Figure

3.14 shows calculated self-diffusion coefficients at temperatures of

2400–3000 K at ordinary (0 GPa) pressure. The self-diffusion coefficients D

lie on a straight line in the Arrhenius (log(D)-1/T ) plot, where T is

temperature. This feature is consistent with experimental self-diffusion

coefficients. Activation enthalpies (H*) were given by the Arrhenius equation

D = D0exp(H*/RT ), where D are the self-diffusion coefficients, D0 is the

preexponential factor, R is the gas constant, and T is the temperature,

respectively. Derived activation enthalpies were similar; 202 kJ/mol for Mg

and 215 kJ/mol for O. The similar activation enthalpies suggest that the

diffusion of Mg and O occurs as a pair. Melting point observed by two-phase

simulations is 2750K so that MD at 2800 - 3000K are considered to be

performed with overheating, though the overheating does not seem to be

affect significantly because these results does not deviate from the straight

line.

54
Activation energies of the extrinsic system and the MD calculation can be

considered to be equal one activation energies of the intrinsic regime is

interpreted as Hm*+Hf*/2 of which MD derives, as section 2.8.3 describes.

Formation energies are also calculated for three conditions (300K, 1000K,

and 2000K). Data beyond 2000K do not be included because we concern

high-energy state during vacancy migration unexpectedly contaminate to the

result so that we do not include them. Observed temperature dependence is

small (-5%, 729KJ/mol for 300K and 767 KJ/mol for 2000K) so that we

adopted value of 2000K for comparing with experiments. Table 3.1 compares

activation enthalpy for extrinsic process (Hm*) and with the activation

enthalpies with three experimental [Sempolinsky and Kingery, 1980; Ando et

al., 1983; Yang and Flynn, 1994] and one theoretical study [Ita and Cohen,

1997]. Table 1 also compares activation enthalpy for intrinsic process Hm*+

Hf*/2. Observed Hm* and Hf* were also consistent with those of experimental

studies. Hm* of this study and Ita and Cohen [1997] are indistinguishable,

though Hf* of our study is larger than that of Ita and Cohen [1997]. We

suppose that the difference between the Hf* comes from the differences in

the interatomic potentials because the two studies gave formation enthalpies

by the same methods (differences of total energies between perfect and

vacancy containing systems).

55
Figure 3.14 Self-diffusion coefficients (D) under 2400–3000 K temperatures

at ordinary pressure (0 GPa).

56
1000
800
Hf* (kJ/mol)

600
400
200
0
0 1000 2000 3000
Temperature (K)

Figure 3.16 Vacancy formation enthalpies of shottky vacancy pair


''
[VMg ] − [VO'' ] ( H f * ) under ambient temperatures at ordinary pressure (0

GPa).

57
Table 3.1. Comparison of activation enthalpy for extrinsic process (Hm* ) and

intrinsic process Hm*+ Hf*/2.

Hm*(KJ/mol) Hm*+Hf*/2(KJ/mol)

This Study 202 586

Ita and Cohen (1997) 190 427

Sempolinsky and Kingery

(1980) 220 -

Ando et al., (1983) - 536

Yang and Flynn (1994) 243 667

58
3.5.3 Results of ambient pressures

We observed self-diffusion coefficients under various temperatures

(2400K-6000K) and pressures (0-140GPa). Like ordinary pressure (0GPa),

vacancies did not migrate when temperatures were too low and systems

melted when temperatures were too high. At each pressure, self-diffusion

coefficients were fitted to the Arrhenius equations. Table 3.2 shows the fitted

results of pre-exponential factors (D0) and Hm* with temperature range of

each pressure. We calculate squared multiple correlation coefficient ( σ 2 ) to

estimate how well they fit. The correlation coefficient σ of the two arrays

x={x1, x2, ... , xn} and y={y1, y2, ... , yn} are defined as
σ = ( n ∑ xy − ∑ x ∑ y ) / [n ∑ x 2 − ( ∑ x ) 2 ][n ∑ y 2 − ( ∑ y ) 2 ] . (3.7)

In this case the x and y correspond with temperature and ln(D), respectively.

Observed results of σ 2 are always near 1 (over 0.99), indicating that the

self-diffusion coefficients of high pressures were also linear on the Arrhenius

plot. Pressure dependences of self-diffusion coefficients under several

temperatures (3000K, 3400K, 4400K, and 5400K) were drawn at Figure 3.17.

We found that self-diffusion coefficients decreased at low pressure, though

the decrease slowed down and rose with increasing pressure. The behavior of

self-diffusion coefficients was different from that of previous study [Ita and

Cohen 1997]. Their self-diffusion coefficients did not increase with increasing

pressure. We consider that this difference comes from pressure effect of the

“attempt frequency” because is considered to be independent with

temperature in Ita and Cohen [1997]. In case we accept idea that two

migration enthalpies Hm* , which this study gives as temperature derivative

59
of self-diffusion coefficients as equation (2.23) and Ita and Cohen [1997] gives

from height of energy barrier, respectively, are same, we would seek cause

this difference in the pre-exponential term D0. Figure 3.18 shows decrease of

pre-exponential term D0 with increasing pressure and we would find that

pre-exponential term linearly decreases with increasing pressure, which the

ratio DP=0/DP is about 15 at bottom of lower mantle (135GPa). Elements,

which would give of pressure dependence in equation [3.6] used in Ita and

Cohen [1997], are square jumping distance l 2 and the attempt frequency ν .

By approximating jumping distance l as proportional with length of unit

lattice, we would estimate decrease of the l 2 . Estimated l 2 by data of molar

volume, which this study gives [7.6607[cm3/mol] at P = 160GPa, where

11.2151[cm3/mol] at P = 100GPa], is l 2 ≈ (7.6607/11.2151)2/3=0.78. Variation

of square jumping distance l 2 is apparently much smaller than the

observed variation of D0 so that we consider attempt frequency as a cause of

observed different behavior of self-diffusion coefficients.

Pressure dependence of self-diffusion coefficients (D) are expressed by

activation volume (V*) as

V*=dH*/dP. (3.8)

When H* corresponds with Hm*, the V* is interpreted as activation volume

of migration (Vm*). Figure 4 compares Vm* of Mg derived from data of Table 2

with that of Ita and Cohen [1997] and Van Orman et al., [2003]. Our Vm*

went positive from negative at 60GPa, indicating rise of self-diffusion

coefficients with increasing pressure. On the other hand, Vm* of Ita and

Cohen [1997] did not turn negative. The difference corresponds with the

difference of self-diffusion coefficients shown by Figure 3. At 20GPa, Vm*of

60
this study and Ita and Cohen [1997] were indistinguishable and both Vm*

were consistent with Vm*of Van Orman et al., [2003].

Figure 3.20 compares formation enthalpies (Hf*) with that of Ita and

Cohen [1997]. Both Hf* increased with pressure, and our increase rate was

larger than that of Ita and Cohen [1997]. The cause of the difference is

considered to come from interatomic potentials because both Hf* is given by

total enthalpy difference of perfect and defect containing systems. As

describing section 2.8.5, Hf* is related with equilibrium vacancy

concentration formed by intrinsic process by equation 2.29. Because formula

of vacancy concentration (equation 2.29) does not contain any other pressure

dependence terms, we can compare vacancy concentrations with ordinary an

d high pressure. Although different models result different Hf*, all Hf* shows

monotonic increases with increasing pressure. Due to Arrhenius type

dependencies of vacancy concentration with Hf*, this monotonic increase

cause estimated vacancy concentration under high-pressure significantly

smaller than under ordinary one. Even in case we adopt uppermost (so

lowest pressure) conditions of lower mantle (1900K, 25GPa, Ito and Katsura
'' ''
[1989]), estimated ratio [VO ] P = 25GPa /[VO ] P =0GPa is 7*10-14, (with Hf* of this

study), 3*10-8 (with Hf* of Karki and Khanduja [2005]), 9*10-9 (with Hf* of Ita

and Cohen [1997]). Due to the significant reduction of intrinsic vacancy

concentration under high pressure, diffusion dominated by intrinsic vacancy

is considered to be impossible so that we would assume extrinsic process

always dominates self-diffusion of lower mantle P-T conditions.

61
Table 3.2. Fitted results of self-diffusion coefficients to Arrhenius equation

D=D0exp(-H*/RT) where D is self-diffusion coefficients, D0 is the

pre-exponential factor, R is the gas constant, T is temperature, respectively.

σ 2 is squared multiple correlation coefficient of the least-square fitting.

Temperature range is also shown at each pressure. Intervals of temperatures

are 100K (0-10 GPa) or 200K (20-140GPa).

temperature D 0 (m2/sec) H m * (KJ/mol) σ2


Pressure (GPa) range (K) O Mg O Mg O Mg
0 2400-3000 9.05E-08 6.90E-08 215 202 0.9948 0.9961
10 2700-3400 3.44E-08 2.80E-08 230 215 0.9957 0.993
20 2800-4200 3.52E-08 3.57E-08 257 250 0.9981 0.9978
30 3400-5000 2.77E-08 1.96E-08 269 252 0.9962 0.9983
40 3400-5000 2.17E-08 1.94E-08 277 266 0.9991 0.9987
50 3400-5200 2.33E-08 2.36E-08 292 287 0.9991 0.999
60 3400-6000 2.04E-08 1.84E-08 297 288 0.9984 0.9968
70 3400-6000 1.55E-08 1.21E-08 294 279 0.9978 0.9987
80 3400-6000 1.24E-08 9.93E-09 290 276 0.9996 0.9983
90 3400-6000 7.34E-09 7.07E-09 272 265 0.9984 0.9976
100 3400-6000 5.39E-09 4.16E-09 261 245 0.9969 0.9989
110 3400-6000 4.82E-09 3.36E-09 257 236 0.999 0.9977
120 3400-6000 3.46E-09 2.56E-09 243 226 0.9984 0.9978
130 3400-6000 2.35E-09 2.09E-09 226 215 0.9981 0.9979
140 3400-6000 2.19E-09 1.68E-09 223 205 0.9987 0.9994

62
Figure 3.17. Pressure variations (0-140GPa) of self-diffusion coefficients (D)

under several (3000, 3400, 4400, 5400K) temperatures.

63
Figure 3.18 Decrease of pre-exponential term with pressure.

Figure 3.19. Comparison of activation volumes of migration (Vm*) of Mg

64
Figure 3.20. Comparison of vacancy formation energies (Hf*)

65
3.6 Discussion

Reproductivity of modeled lower mantle of MgO by Buckingham type pair

potential and parameters (Akamatsu and Kawamura, 1998) has been widely

examined by comparing physical properties (Equation of state, elasticity,

melting points, and self-diffusion coefficients) with those of experimental

data. Activation energies self-diffusion coefficients are well consistent with

existing experimental data. Limitations of pair potential (e.g. unexpected

Cauchy relation equation 2.a) have been reported and inconsistencies [EOS

at 2000K, Figure 3.5, elasticity Figure 3.12] are also found during the

examinations. Examination of high-pressure self-diffusion data with

experiments is limited only with those of Van Orman et al. [2003] of

relatively low pressure (15-25GPa), which appears to be not enough.

However, due to excellent agreements with those of ordinary pressure (Table,

3.1), it is expected that high-pressure data with same model does not contain

significant discrepancy, is applicable to the actual self-diffusion and rheology

of the lower mantle. Reported and obtained increase of formation enthalpies

with increasing pressure (Figure 3.20) causes significant reduction of

intrinsic vacancy concentration. Reported different formation enthalpies is

considered to be disappeared due to reduction of vacancy concentration

vacancy formation enthalpy (Hf*) causes, so examination of the difference

with experimental study would be difficult even in the future. In the

extrinsic process likely dominant in high pressure, vacancies are formed by

some external process, which might be possible to treat as independent with

pressure, so well simulated by this MD, which does not treat vacancy

66
formation and extinctions. Result of self-diffusion coefficients with ambient

pressures (Figure 3.17) seems strange (turns to increase from decrease with

increasing pressure), which is never reported in previous studies. Cause of

this behavior was examined by comparing with those of previous study (Ita

and Cohen [1997]) and Ita and Cohen [1997]'s assumption of constant

attempt frequency was found to be likely as a cause. Under the lower

mantle's high pressure and temperature, atoms vibrate significantly and

have high energies. Atoms' energy is ununiform as Maxwell-Bortzmann

principle gives. In case atom around vacancy occasionally becomes

high-energy state, the atoms would migrate in.to vacancy and sites atom sit

turns to be a vacancy instead. This process is fully spontaneous one so that

understanding of lower mantle diffusion is considered to require whole

reproduction of the spontaneous migration process on the computer.

Observed behaviors of pressure dependence of self-diffusion coefficients

(negative -> positive transitions of pressure dependences) are far from any

reported melting points (Figure 3.8). Van Liempt relations (equation 3.1)

seems to be inconsistent so that we cannot satisfy depending only it and

direct studies of self-diffusions are needed for more rigorous understanding

of lower mantle rheology.

67
Chapter 4

Noble Gas Crystals

4.1 Introduction

Previous chapter reports successful reproduction of MgO self-diffusion of

lower mantle temperature and pressure conditions and reported behavior of

pressure dependence of self-diffusion coefficients are strange. Observed

decrease of self-diffusion coefficients with increasing pressure slows dowand

turns to be increase with increasing pressure [Figure 3.17]. This reverses are

observed in middle of lower mantle (70GPa at 4000K). This results have

significant geophysical importance because MgO is a major end-member of a

component of [Mg,Fe]O magnesiowustite and self-diffusion is an atom-scale

elemental process of diffusion creep, which is considered as a mechanism of

lower mantle plastic deformation. So this behavior is worth additional

examinations. One of the examination considered is comparing with other

materials, which has different interatomic interactions. Especially, affects of

interatomic potential types to the behavior are expected to be clarified by

comparing other materials, which has different type of interatomic

potentials. Noble Gas Crystals, (Helium [He], Neon [Ne], Argon [Ar], etc) are

considered to be a good candidates for the first target of the comparisons.

First, it is known that their physical properties are well approximated by

simple Lennard-Jones model (equation 2.2), which is easy to treat. Second,

noble gas's diffusion itself has geophysical importance as a tracer of rock's

68
origin and history. Electron captures reaction of 40K -> 40Ar is widely used for

geochronometries (K-Ar, and Ar-Ar dating). Specific isotopes are considered

to be generated only interactions of surface with cosmic rays so its

concentrations are used for deriving surface exposure timescales (21Ne in

quartz, Shuster and Farley 2005). These using of isotopes as tracers also

requires knowledge how well corresponding isotopes are retained because all

atoms contained in minerals outflows from them by these self-diffusions. If

dating by measurements of isotropic concentrations overlooks their

unexpected outflows, then the dating would contain significant

overestimations. Physical properties of rare gas under ambient pressures

have another importance relating with experimental techniques. Rare gas

freezes high pressure corresponding to earth's interior (1.5GPa at 300K for

Ar [Datch et al. 2000], 4.9GPa at 300K for Ne [Vos et al., 1991], 13GPa at

300K for He [Vos et al., 1990]), it is chemically stable, and they are

significantly soft even solid phases. Therefore, in case they were used

medium pressure translation materials between compressions and

compressed sample chambers, it is expected that non-hydrostaticity of the

compressions are reduced. Problem is more serious for diamond anvil cell

experiments because its compression is uniaxial, which might cause

significant non-hydrostaticity in high pressure. Fei [1999] compared MgO's

compressions with unfilled filled chambers with Neon by diamond anvil cells

in order to examine affect of non-hydrostaticity caused by the uniaxial

compression.

In this chapter, we report results of pressure-dependence of self-diffusion

of Noble gases (Neon and Argon). First, their reproductions of molecular

69
dynamics are examined by comparing EOS (compression) with those of

experiments. Second, pressure dependence of self-diffusion coefficients are

tried to be calculated. Trials partially succeeded, Ne self-diffusion

coefficients are calculated. Resulted behaviors of the pressure dependence is

compared with those of MgO (previous chapter) and discussed.

70
4.2 Equation of State

Adopted parameters are listed in Table 2.1. Because they are empirically

derived from viscosities and second-order derivatives of gas phases, their

reproducivities of crystals are not clear and it was examined. Both Argon and

Neon melts under lower pressures so that molar volumes were measured

from 10GPa to 100GPa. All initial positions of atoms of MD was set in order

to consists FCC structures, which are observed as experiments. Spontaneous

phase transitions did not observed for any temperature/pressure conditions.

Experimental study of Errandonea et al. [2007] reported coexistence of

FCC/HCP phases of Argons over 49.5GPa, so FCC/HCP phase boundaries

possibly examined with calculation, though this was not treated.

Figure 3.1 compares results with experiments [Anderson et al. 1973 and

Ross et al. 1986] and LMTO (linear muffin-tin-orbital) electronic state

calculations [Tsuchiya and Kawamura 2002]. Although determination of

parameters does not include physical properties of solids, compressions of

neon up to 100GPa agrees well with those of experiments and electronic

state calculations. Ross et al. [1986] also examined potentials, which is

composed of exp(-1/r) type repulsion and r-6 type terms by Monte Carlo

calculations, and reported it good agreements up to 90GPa (Figure 4.3).

Because repulsion terms become more important under high pressures,

differences of repulsion terms are considered as a main cause this difference.

Better agreement of exp(-1/r) model would indicate that Argon more

electrons than Neon would cause strong repulsions, which is expressed

exp(-1/r) better than r-12 type form of Lennard-Jones model.

71
Tsuchiya and Kawamura
2002 (LMTO)

This Study

Tsuchiya and Kawamura


2002 (LMTO)

This Study

Figure 4.1 Comparison of compressions of solid noble gases (Neon and Argon)

with Experiments (Anderson and Swenson [1973] and Ross et al.

72
Figure 4.2 Comparisons of Argon [Ar] diamond anvil compression

experiments with exp(-1/r) + r-6 type pairwise interaction models reporeted

by Ross et al. [1986]. Experimental and model calculation data are shown by

triangles and line, respectively. This figure is partially quoted from Ross et al.

[1986].

73
4.3 Self-diffusion coefficients

Self-diffusion coefficients had been tried to calculate at ambient pressures

up to 3000K temperatures. In the simulations, one vacancy is introduced in

the basic cell, which contains 5x5x5 FCC unit cells (499 atoms). Because of

weak long term force (var der Waals force well modeled by r-6 term) of noble

gas atoms, size affect is considered not to be likely to be significant

comparing with ionic materials such as MgO. Like MgO's simulations,

systems melt with too high temperatures and atom migration does not

observed (shown in Figure 2.26) with too low temperatures. Neon's diffusion

was successfully observed at 90-200GPa and 1500-3000K temperatures and

results are shown in Figure 3. In lower temperatures (500K, 1000K) of Neon,

atom's migrations have not been observed. Systematic diffusion of Argon (Ar)

were failed to be observe in method in targeted temperatures and pressures.

This failure would be caused by heavy (40 a.m.u.) atomic mass of (Ar) and/or

resulted larger repulsive forces, which would become more important at high

pressure.Atomic masses of successfully observed diffusion species (Mg:

24.a.m.u, O: 16 a.m.u, Ne: 20 a.m.u) are about twice lighter than those of Au.

Except 170-200 GPa at 1500K, behaviors of pressure dependences are

basically log-linear decrease with increasing pressure. At 1500K,

self-diffusion coefficients are discontinuously decreased between 160GPa to

170GPa (2.0*10-14m2/sec to 3.8*10-16m2/sec) and pressure dependence is

disappeared over 170GPa. Because small self-diffusion coefficients (under

10-15m2/sec) are possibly caused only thermal vibrations without any atoms'

migrations, this discontinuity is interpreted as disappearance of atom's

74
migration so linear fitting for 1500K data were only done for 100-160GPa.

Pressure dependence of self-diffusion is characterized by activation volume

(V*) defined by equation 3.8 and it can be also derived directly from

self-diffusion coefficient. Pressure dependence of self-diffusion coefficients

are expressed by using equation 3.18 and Arrhenius form (eqation 3.2) as:

D = D0 exp([− H 0 * + PV * ] / RT ) . (4.1)

By applying natural logarithm ln(x) and partial derivative with pressure P,

following relations

∂ ln D
V * = − RT . (4.2)
∂P

become available for deriving V* from D of ambient pressure. Table 3.1 shows

derived activation volumes (V*) at ambient temperatures (1500-3000K).

Derived activation volumes are scattered from 0.6-0.8 cm3/sec, though

systematic temperature dependence are not found.

75
Pressure (GPa)
50 100 150 200 250
-8

-9

-10
1500K
-11 2000K
D
2
(m /sec)
-12 2500k
3000k
-13

-14

-15

-16

Figure 4.3 Pressure dependence of self-diffusion coefficients of neon (Ne)

76
Table 4.1 Observed activation volume (V*) of Neon (Ne) at ambient

(1500-3000K) temperatures
3
P(GPa) T(K) Activation volume (cm /mol)
90-160 1500 0.62
90-200 2000 0.78
90-200 2500 0.76
90-200 3000 0.74

77
4.4 Discussion

Pressure dependence of self-diffusion of solid neon (Ne) and argon (Ar)

were tried to be observed and partially succeeded (Ne for 90-200GPa

pressures, 1500-300K temperatures). Observed pressure dependence is

interpreted as linear, which does not consistent with non-linear behavior of

MgO of which examination is the first motivation to study. Cause of this

inconsistency would be sought to difference of interatomic potential forms.

There are two difference terms between MgO's potential form (Backingham

form, equation 2.1) and that of solid Ne (Lennard-Jones form, equation 2.2),

existence of long-range Colomb (1/r) term of that of MgO and repulsion term

(exp(-1/r) for that of MgOI and 1/r12 for that of solid Ne). Firstly considered

candidate of the cause is long-range Colomb (1/r) term because it is strongly

dominates behaviors of ionic materials such as MgO. As mentioned at section

2.4, it is known that natural Colomb term still affects in a long range and

deriving it needs special way in order to accelerate its convergent. The Ewald

method realized the acceleration by introducing three dimensional Fourier

transforms. Pressure dependences of silicon (Si) diffusion olivine have

experimentally reported by Bejina et al. 1999 (quoted as Figure 4.4). By

taking into account partial pressure of oxygen (pO2) with increasing pressure,

weak (0.7 cm3/mol) pressure decrease and non-linearity cannot be found up

to 9GPa. Although olivine is stable with more pressure (16GPa, with pyrolite

mantle composition, Weidner [1986]) so we cannot exclude non-linearity with

wider pressure ranges, olivine's non-linear pressure dependence seems

unlikely. 4-coordinate silicate, including olivine transits to 6-coordinate

78
silicate perovskite at uppper-lower mantle boundary and silicon's ionicity

discontinuously increases with this transition. Therefore, in case we accept

the assumption that long-range force as an origin of the non-linear behavior,

we would suppose that the non-linearity would be increased with increasing

ionicity, it is imagined that this transition would cause non-linearity of

silicion diffusion. It is also possible latter repulsion terms become a cause of

the non-linearity. Improvement of Argon's reproduction of experimental

compression by using exp(-1/r) repulsion term (Ross et al. 1986) is reported.

Neon's compression is also reproduced well by using 1/r12 term (Figure 4.1).

However, it would also be examined with alternatively developed parameters

exp(-1/r) + form like Argons.

79
Figure 4.4 Pressure dependence of Si self-diffusion coefficients of olivine

quoted form Bejina et al. [1999]. Upper data is original one and lower data is

adjusted data by taking into account of increase of pO2 with increased

pressure.

80
Chapter 5

MgSiO3 perovskite

5.1 Introduction

MgSiO3 perovskite is a mineral which has a perovskite (RMO3) type

crystal structure (Figure 5.1) and is considered a major end member (-70%)

of the lower mantle. A MgSiO3 single crystal has an orthopyroxene structure

(i.e. Enstatite) at ordinary pressure, and transits to clinopyroxene, ganent,

ilmenite, and perovskite structures at upper mantle P-T (-2500K, -25GPa)

conditions. The perovskite structure of MgSiO3 is stable at over 23GPa (at

MgSiO3 bulk) and for almost all P-T conditions of the lower mantle. At

lowermost pressure (115-125GPa), the transition of MgSiO3 perovskite into

more dense CaIrO3 (post-perovskite) structures was recently observed by

DAC (diamond anvil cell) experiments (Murakami et al. 2004, Figure 5.2).

The amount of this new phase in the lower mantle has been examined and

discussed, though these studies have not caused any doubt in the idea that

almost all silicate in the lower mantle exists as MgSiO3 perovskite. Because

of its major role in the lower mantle, a number of studies model MgSiO3

perovskite and investigate it through molecular simulation. Developed

models include pair-potentials (Matsui 1988, Oganov et al. 2000), using

additional angle terms (Wright and Price 1993), anion shell model (Stuart

and Price [1996], Watson et al. [2000]), and the more complex Gordon-Kim

type model (Marton and Cohen [2002]). Models are more than just potential;

81
direct calculations of electronic states of MgSiO3 perovskite were performed

and successfully applied to derive its crystal structures and elasticity

(reviewed in Karki et al. 2002). Regarding diffusion, formation energies and

microscopic structures of vacancies were studied using ab inito LDA

electronic state calculations (Karki and Khanduja [2006](2)). Migration

pathway and height of energy surface were also studied with fixed

surrounding atoms (Wright and Price [1993], Karki and Khanduja [2007]).

These calculations do not treat finite temperature, though the effects of

temperature should be significant, especially in results relating to the lower

mantle because the temperature of lower mantle is high (over 1900K). In

addition, simulation with spontaneous migration is desired in order to

evaluate self-diffusion coefficients more precisely.

This section intends to give properties of self-diffusions of MgSiO3

perovskite by reproducing spontaneous migration of atoms with molecular

dynamics (MD). By considering the difficulties of keeping precisions as

discussed in section 2.8.4, we used the simple pair-wise Buckingham model

(equation 2.1) with parameters (Table 2.1, Oganov et al. [2000]). Firstly, we

examine the reproduction of MgSiO3 perovskite by comparing EOS (equation

of state) with those of experiments. Melting temperatures were also

calculated by using a two-phase method (Belonoshko and Dubrovinsky 1996).

MD simulations of spontaneous migrations of vacancies are performed in

order to derive self-diffusion coefficients. Especially, this work focuses on

deriving Si self-diffusion coefficients. Si is likely to be the slowest diffusion

species of MgSiO3 perovskite so it is expected to dominate diffusion creep of

lower mantle MgSiO3 perovskite [Yamazaki et al. 2000].

82
Figure 5.1 Unit cell of ideal cubic perovskite structures (RMO3). Mg, Si of the

MgSiO3 Peroskite are positioned to R and M sites, respectively.

83
Figure 5.2 DAC (diamond anvil cell) observations of MgSiO3 Peroskite –

post-perovskite transitions after Murakami et al. [2004].

84
5.2 Equation of state

Figure 5.3 compares isothermal compression at ordinary temperature.

EOS was calculated by this model and those of experiments. Experiments of

MgSiO3 perovskite were for pure synthetic MgSiO3 perovskite (Yagi et al.

1982) and MgSiO3 perovskite of Fe/Mg = 0 – 0.2 (Mao et al. 1991), done by

DAC (diamond anvil cells). Another experiment (Gong et al. 2004) consisted

of the shock compression of natural Enstatite. The data of DAC experiments

are considered more precise than those of shock compressions because the

compression of DAC is stable, though wider pressure data of shock

experiments are available. Differences of molar volumes between this study

and DAC experiments (Yagi et al. [1982], Mao et al. [1991]) are under 2% at

25-30GPa, while at 40-120GPa, differences between this study and

experiments (Gong et al. [2004]) are smaller than 5%. The 40GPa molar

volume of Gong et al. [2004] is bigger than the 30GPa molar volume of Mao

et al. [1991], indicating that some overestimation of Gong et al. [2004] came

from the instability of shock compression. We consider that the 5%

inconsistency with Gong et al. [2004] and this study is likewise due to the

instability of shock compression.

85
24

molar 21
volume
(cm3/mol)

18

15
0 30 60 90 120 150
Pressure (GPa)

Figure 5.3 Comparison of isothermal ordinary temperature compressions of

MgSiO3 MgSiO3 perovskite with experimental data. Filled square indicates

results of this study, and unfilled marks are those of experiments.

Experiments were performed by DAC (diamond anvil cell) (Yagi et al. 1982

shown as open squares, Mao et al. 1991 shown as plus mark) and shock

compression (Gong et al. 2004) of natural Enstatite shown as open triangles.

86
Figure 5.4 shows comparisons of isobaric (25GPa) thermal expansion of

MgSiO3 perovskite with the experimental data of Funamori et al. [1996]

performed by a multianvil apparatus. We selected multianvil experiments as

a reference of comparison of thermal expansion because experiments with a

multianvil apparatus can measure temperature precisely with

thermocouples. Black-body emission temperature measurements of laser

heated DAC is reported with large (over ± 100K) errors (e.g. Shen and Lazor

[1995]) and transition of diamond -> graphite with increasing temperature

limits temperature of externally heated DAC experiments of up to 1100K (e.g.

Fei [1999]). Better agreements (within 3% agreements values itself and their

first derivatives [i.e. thermal expansibility]) between the data of this study

and those of other experiments (Funamori et al. [1996]) were confirmed.

87
24

molar 21
volume
(cm3/mol)
18

15
0 1000 2000 3000

Temperature (K)

Figure 5.4 Comparison of isobaric (25GPa) thermal expansion of MgSiO3

perovskite with experiments. Filled squares show data of this study and plus

marks show data of Funamori et al. [1996]. Data of Funamori et al. [1996].

were selected from 25 ± 0.5 GPa.

88
5.3 Orthorhombic distortions

Although the ideal shape of perovskite is cubic (unit cell lengths a = b = c

and axes α = β = γ = 90 o ), it is known that many perovskites have various

lower symmetries. Even CaTiO3 perovskite, which is the origin of the

mineral name “perovskite”, has an orthorhombic ( a ≠ b ≠ c , α = β = γ = 90 o )

structure. This symmetry could be changed with variation of

pressure/temperatures discontinuously as phase transitions. For example,

BaBiO3, reported by Cox et al. [1979], is monoclinic ( a ≠ b ≠ c , α ≠ β = γ = 90 o )

at room temperature, transits to higher symmetry of rhombohedral structure

( a = b = c , α = β = γ ≠ 90 o ) at 400K, and finally transits to cubic structure at

740K. In situ high-pressure observation (ex. Asahara et al. [2005]) confirmed

that [Ca,Mg]SiO3 perovskite decomposes into its end members with

increasing temperatures (over 1300K), so two perovskite phases (MgSiO3,

CaSiO3), are considered to be separated in the lower mantle. The two

perovskites (MgSiO3, CaSiO3) are well characterized by their symmetries;

orthorhombic MgSiO3, and cubic CaSiO3. Therefore, not only EOS measured

by molar volume but also the reproduction of the symmetry could well be a

checkpoint of reproductions of the materials of MD simulations.

The symmetry of MgSiO3 is known to be orthorhombic at the uppermost

pressure of the lower mantle (25GPa, (Liu, 1976)). Because of the wide

pressures and temperatures (1900K-3000K) of the lower mantle (25-135GPa),

the possibility of its discontinuous transitions to the more symmetric phase

is considered. Previous Molecular Dynamics studies (Matsui and Price [1991],

Chaplot et al. [1998]) reported orthorhombic->cubic transition at pre-melting

89
high-temperatures. Cubic MgSiO3 perovskite is also reported in-situ

synchrotron radiation experiments (Meade et al. 1995) at 70GPa, though

later in situ observations of laser heated DAC with 40-140GPa pressures

(Fiquet et al. 2000, Shim et al. 2000, 2004, Murakami et al. 2004) did not

reproduce it. Zero-temperature ab initio energy calculations (Stixrude and

Cohen 1991, Wentzcovitch et al 1993, 1995) show orthorhombic MgSiO3 has

lower energies than cubic MgSiO3 perovskite up to 150GPa, so pressure

induced transition of orthorhombic->cubic is considered to be unlikely.

Observed temperatures of transitions in MD simulations reports are quite

high (5000K at 30GPa [Matsui and Price, 1991], 6000K for 40GPa, 7000K for

60GPa, [Chaplot et al. 1998]), indicating this transition as a temperature

induced one. Belonoshko [2001] considers the transitions which these MD

studies observe as artificial thermal instability, because these transition

points are higher than melting temperatures observed from two-phase MD

simulations (Belonoshko 1994) and experiments (Zerr and Boehler 1993).

Through these examinations, orthorhombic structure has been confirmed as

MgSiO3's preference in wide pressure and temperature conditions including

those of the lower mantle, so it would be a good point for examining MD

models of reproductivity.

Orthorhombic distortions are well characterized by ratios of orthorhombic

( a ≠ b ≠ c ) unit cell lengths (b/a and c/a). Symmetry increases to cubic when

b/a=1 and c/a = 2 ≅ 1.414 , so the amount of differences of b/a and c/a from

these values correspond to the amount of orthorhombic distortions. In this

MD simulation, a basic cell is set to match an orthorhombic unit cell of

MgSiO3 perovskite, with the ratios being given by observed basic cell lengths

90
of constant pressure and temperature (NPT) MD. Reproductions of MD's

orthorhombic distortions were examined at wide temperatures (0-3000K)

and pressures (0-120GPa) and compared with experiments. In Figure 5.5,

pressure dependences of ratios (b/a, c/b) were compared with those of

experiments (Mao et al. 1991, Fiquet et al. 1998) and other theoretical data

(Marton et al. 2001, Wentzcovitch et al 1995). Our b/a distortion is less than

those of experiments. Experiments of b/a are around 1.03 and results of this

study are smaller than that (1.02 – 1.015). In order to derive the cause of

inconsistency, results of VIB model potential (Marton et al. 2001), and

zero-temperature ab initio energy calculations (Wentzcovitch et al 1995)

were also compared. The VIB model includes many-body effects explicitly,

though results of the VIB model is consistent with this study's two-body

results and improvement of this mismatch is not observed. On the other

hand, b/a distortions of LDA ab initio calculations (Wentzcovitch et al 1995)

are more significant (1.047-1.081) than experiments, which seem to have

been somewhat overestimated. Data of Wentzcovitch et al [1995] also

overestimates c/b. In summary, (1) model potential calculation tends to

underestimate b/a (2) introduction of more complex many-body effects does

not simply solve the problem (3) distortion might be observed by ab initio

calculations, though LDA approximation tends to overestimate it.

Temperature variations of distortions are also compared with experiments

(Funamori et al. 1996) in Figure 5.6. Temperature variations of experiments

(Funamori et al. 1996) are not detected, though temperature variations of

distortions of MD results decrease with increasing temperature. Especially,

decrease of c/a with increasing temperature is bigger than those of a/b,

91
indicating decrease of orthorhombic distortions with increasing temperature.

Because predicted variation of distortion is small (under 1% while in the

800-1600K temperature range (Funamori et al. 1996)) and zero-temperature

ab initio calculations are difficult to apply to isobaric expansions, we would

need more experimental data in order to examine the variation.

1.1

b/a

1
0 20 40 60 80 100 120
Pressure (GPa)

92
1.514

c/a

1.414
0 20 40 60 80 100 120
Pressure (GPa)

Figure 5.5 Pressure dependences of orthorhombic distortions (ratio of

basic cell length c/a and b/c) of MgSiO3 perovskite at ordinary temperatures.

Rotated and un-rotated crosses show experimental data (Mao et al. 1991 and

Fiquet et al. 1998), and open circles and squares theoretical data

(Wentzcovitch et al 1995, Marton et al 2001), respectively. Pressures of

Fiquet et al. [1998] are given by averages of multiple pressure scales (Pt and

Ruby fluorescence). 50GPa data of Marton et al. [2001] are given by 900K

data because of lack of 300K data. Temperatures of energy calculations

(Wentzcovitch et al. [1995]) were corresponding to 0K because this

calculation does not treat finite temperatures.

93
1.1

b/a

1
0 1000 2000 3000
Temperature (K)

1.514

c/a

1.414
0 1000 2000 3000
Temperature (K)

94
Figure 5.6 Temperature dependences of orthorhombic distortions (ratio of

basic cell length c/a and b/c) of MgSiO3 perovskite at ordinary temperatures.

Filled circles show data of this study and open squares show experimental

data of Funamori et al. [1996].

95
5.4 Melting

As mentioned in section 3.3, the melting of mantle minerals is an

interesting topic for geophysics. Especially, as MgSiO3 perovskite is a

dominant phase of lower mantle, it is expected to give strong information for

understanding current and past Earth's dynamics. Experimental data on the

melting of MgSiO3 (Figure 5.7) is controversial. Earlier experiments (Knittle

and Jeanloz [1989], Heinz and Jeanloz [1989], and Sweeney and Heinz

[1993]) report lower melting temperatures (3000-3500K at 40, 60GPa) and

small melting slope (dTm/dP = 0 - 20[K/GPa]). Zerr and Boehler, [1993],

which detects melting by texture observations of heated samples with

keeping pressures, reports higher temperature (3800K at 40GPa, 5200K at

62GPa) and high temperature slope (dTm/dP = 40[K/GPa]). Heinz et al.

[1994] and Boehler and Zerr [1994] discussed and the difference of ways to

detect melting is supposed as a cause. Shen and Lazor [1995], who detect

melts by discontinuous behavior of laser reflections, reproduced results the

Zerr and Boehler [1993] up to 40GPa. Belonoshko [1994] gives melting points

by two-phase MD simulations with Matsui [1988] pair potential, and results

are well consistent with experimental data of higher

temperature/temperature slope (Zerr and Boehler, [1993], Zerr and Boehler,

[1993]). Other molecular dynamics works (Matsui and Price [1991], Chaplot

et al. [1998]) report higher melting points (5000K at 30GPa) with

orthorhombic -> cubic transitions near melting temperatures. However,

Belonoshko [2001] argued that the transition was an artificial one

introduced by thermal instability with overheating.

96
Here, we also performed two-phase MD simulations for examining the

interatomic potential model. Results are compared in Figure 5.8 with

experimental data and previous two-phase MD simulation data (Belonoshko

[1994]). Our calculations were performed at whole lower mantle pressures

(25-120GPa) with 5GPa intervals. Two MD systems of which basic cells were

composed of 5x5x5 unit cells were prepared for separate MD simulations.

Initially, each MD system was kept for cold (300K) and hot (7000K) at 25GPa,

then both systems were heated or cooled to closer temperatures (3500K and

4000K, respectively). After both systems reached thermal equilibrium by

initial running, they were contacted and temperatures / pressures were

scaled to keep target ones. Running time of two-phase MD is 105 steps

(corresponding to 0.2ns). Trajectories of atom's coordinates during MD

running were recorded. Stable phases (solidus/liquidus) were judged by clear

observation of growth during either phase. Boundaries of stable phase

transitions were successfully observed with 100K temperature intervals so

widths of indeterminate errors of the results were estimated as ± 50K. Aside

from the resolution difference of both data, both dTm/dP of MD data

(Belonoshko [1994] and this study) are consistently decreased with

increasing pressure and dTm/dP of lower pressure (-50K/GPa) are well

consistent with those of experiments dTm/dP. Our results largely (–500K)

estimate higher melting higher temperatures than Belonoshko [1994]. This

difference is considered to come from potential parameters, especially

charges related with assumed degrees of partial ionizations. Parameters of

ratios of total positive (or negative) charges per MgSiO3 molecular are 3.894

[for Matui. 1988] / 4.814 [for Oganov et al.] = 0.79. The resulting difference of

97
melting temperatures would be explained by this charge differences.

98
Figure 5.7 Summary of experimental melting temperatures of MgSiO3

perovskite after Shen and Lazor[1995]. S&L, Z&B, K & J, H&J, S & H

indicates Shen and Lazor [1995], Z Zerr and Bohenler [1993], Knittle and

Jeanloz [1989], Heinz and Jeanloz [1989], and Sweeney and Heinz [1993],

respectively.

99
8000

7000

6000
This Study
Temp.
(K) 5000
Belonoshko [1994]

4000

3000

2000
0 20 40 60 80 100 120
Pressure (GPa)

Figure 5.8 Comparisons of melting temperatures of MgSiO3 perovskite.

Triangles, lines and cross bars are experimental data as same of Figure 5.7.

Upper and lower data of Belonoshko [1994] shows results of two-phase

simulations, which correspond to liquidus and solidus, respectively.

100
5.5 Self-diffusion

The importance of the self-diffusion of MgSiO3 perovskite is mainly in its

implication for diffusion creep. Diffusion creep is dominated by the diffusion

of the slowest diffusion species, and the slowest species is possibly the silicon

in MgSiO3 perovskite, because many silicates silicon are slow species

(reviewed in Bejina et al. 2003). Reported experimental measurements of

silicon diffusion (Yamazaki et al. [2000]) were extremely slow (10-21-10-20

m2/sec at 1773K, 25GPa and 10-19-10-18 m2/sec at 2073K, 25GPa), supporting

the idea of Si diffusion as the slowest one. Comparisons between the

diffusions of oxygen and silicon dopants (Dobson et al. [2003]) report that

oxygen diffusions are 5-50 times faster than silicon diffusions.

Measurements of Mg-Fe inter-diffusion of MgSiO3 perovskite (Holzapfel et al.

[2005]) show that it is comparably as slow as silicon diffusion (10-19-10-17

m2/sec at 2073K – 2273K). Application of this data to diffusion creep requires

information of the relations between Mg-Fe interdiffusion and self-diffusion

of each divalent cations Mg. However, by assuming that they are similar, the

possibility of divalent cations Mg as the slowest one can be considered.

Attempts to derive self-diffusion related properties with computational

methods have been performed. Schottky vacancy formation enthalpies (Hf*)

of MgSiO3 perovskite have been calculated. (Karki and Khanduja [2006](2)).

Likewise MgO periclase, Hf* of Schottky pairs (SiO2 and MgSiO3)

significantly increases with increasing pressure (Figure 5.9), indicating that

Si intrinsic diffusion of MgSiO3 perovskite at the lower mantle is unlikely.

Wright and Price [1993] derived migration enthalpies of vacancies by static

101
energy calculations with model potential (pair potential with angle terms of

Si-O-Si bonds). In Wright and Price [1993], migration enthalpies were given

through energy increases of climbing migrating atoms to the saddle points.

Reported migration enthalpies (Hm*) of Si diffusion (733kJ/mol [0GPa] and

458 kJ/mol [125GPa] with assuming one neighborhood oxygen vacancy) are

inconsistently higher than those of experiments (336kJ/mol at 25GPa,

Yamazaki et al. [2000]). Karki and Khanduja [2007] also investigated

migration enthalpies in a similar way with ab initio LDA pseudopotential

interatomic interactions. This method has been successfully applied to

structure/elasticity of Earth's deep interior minerals (reviewed in Karki et al.

[2001]), though Karki and Khanduja [2007] report that introducing LDA

interatomic interactions does not improve the inconsistency (657kJ/mol at

30GPa with assuming one neighborhood oxygen vacancy), indicating that the

problem is not solved by only introducing more precise interatomic

interactions such as ab initio LDA pseudopotential. In these previous

calculations, heights of energy barriers are calculated statically without

moving atoms so relaxations of total structures with increasing

temperatures are not treated, though it can cause overestimations of heights

of energy barriers. MD simulations of spontaneous migration of atoms,

which is reported in section 3. 4 and Ito and Toriumi [2007], explicitly treat

temperatures by atom motions. Moreover, section 3 and Ito and Toriumi

[2007] show that reproductions of spontaneous migrations of atoms are

needed in order to treat attempt frequencies of atom migrations (parameter

ν expressed in equation 3.6) through MD simulations of MgO self-diffusion.

In consideration of these points, it is expected that this approach is also

102
effective for the self-diffusion of MgSiO3 perovskite.

Here, we report our attempts, which apply direct reproduction approaches

to self-diffusion of MgSiO3 perovskite. Silicon diffusion is likely the slowest

species of MgSiO3 perovskite and it is considered that its contributions to

diffusion creep are the largest, so we mainly target observation of silicon

diffusion in the work. We compose MD's basic cell with of 5x5x5 unit cells of

MgSiO3 perovskite and one SiO2 Schottky vacancy was introduced in 5x5x5

(1997 atoms). Although silicon is the diffusion species we are most interested

in, two oxygen vacancies are additionally introduced. Applying the Ewald

method for calculating converged value of long-range 1/r Coulomb term

(Section 2.4) requires keeping the neutrality of the system's total charge.

Wright and Price [1993] calculated heights of energy barriers of <110>

orthorhombic directions of Si migration with and without surrounding an

oxygen vacancy and reported that heights of energy barriers significantly

decrease with oxygen vacancy, indicating that the existence of vacancy is

needed in order to reproduce actual Si self-diffusion of MgSiO3 perovskite by

MD simulations. Calculation steps of MD simulations are kept to 106 (2ns),

repeated 10 times and results of self-diffusion coefficients are averaged.

Width of temperature conditions is limited. At low temperature, atom's

migrations do not happen enough in order to derive self-diffusion coefficients.

The lack of number of migrations during MD simulations of self-diffusion

causes the reduction of precisions as shown in section 2.8.4. High

temperature causes melting of the systems. Because this simulation hopes to

clarify diffusion of solid MgSiO3 perovskite, we do not treat temperature

conditions which cause melting of the system. Calculations were repeated 10

103
times with the same pressure and temperature and averaged.

104
Figure 5.9 Formation enthalpies (Hf*) of Schottky vacancies MgSiO3

perovskite (after Karki and Khanduja, [2006] (2)).

At first, we examine temperature variations of self-diffusion coefficients at

100GPa. In this simulation, vacancies are initially introduced and do not

form/disappear during MD simulations because of MD's basic cell with

infinite repeated boundary conditions, therefore temperature dependence of

self-diffusion coefficients is expected to be characterized by Arrhenius

relation (Equation 2.23) and activation enthalpies derived from fitting of

observed self-diffusion coefficients and its temperatures onto corresponding

migration enthalpies (Hm*). Figure 5.10 shows observed temperature

dependencies of silicon and oxygen diffusions at 100GPa. Temperature range

is 4200-5900K and calculations were performed for 50K step intervals.

105
Results were characterized by (1) linear increase in low temperature range

(2) abnormal discontinuous rapid increase over high temperature range

5200K (3) kink at boundary of the two regions. These features are observed

for both (Si and O) diffusions. Observed ratios of self-diffusion coefficients

are Do /Dsi = 30 (min) –170(max). In basic cells of the MD, one set of SiO2

Schottky vacancies is introduced in 5x5x5 orthorhombic unit cell of MgSiO3

perovskite, which is composed of 400 Mg atoms, 400 Si atoms, and 1200 O

atoms. Vacancy concentrations of the Si and O are 1/400 and 2/1200 = 1/600,

respectively, so expected ratios of diffusion coefficients are Do /Dsi =

(1/600)/(1/400) = 1.5 in case same mobility of the two vacancies are assumed.

Observed ratios of self-diffusion coefficients (are Do /Dsi = 30 (min)

–170(max)) are at least 15 times larger when same mobility of the vacancies

are assumed. These high ratios indicate oxygen diffusions as much faster

than those of silicon. Faster diffusion of oxygen than silicon in MgSiO3

perovskite is also observed in experiments (Dobson et al. [2003]). For both

diffusions, boundaries are observed at 5200K and activation enthalpies were

calculated as 400(kJ/mol), at 4200K-5200K, 1488(kJ/mol), at 5200K-5900K

for silicon diffusion and 280 (kJ/mol) at 4200-5200K, 2032(kJ/mol) at

5200K-5900K for oxygen diffusion.

106
100GPa-Si 10000/T
1.5 2 2.5 3 3.5
-8

-9

D:
-10
Self
Diffusion
Coeff. -11
(m2/sec)
-12

-13

-14
5000 4000 3000
T:Temperature (K)

100GPa-O 10000/T
1.5 2 2.5 3 3.5
-8

-9

D:
Self -10
Diffusion
Coeff. -11
(m2/sec)
-12

-13

-14
5000 4000 3000
T:Temperature (K)

Figure 5.10 Temperature dependences of self-diffusion coefficients at

100GPa , 4200-5900K. Vertical lines indicate boundary of two (normal

[lower], abnormal [higher]) temperature regions.

107
Temperature dependences of Si self-diffusion coefficients were also

examined at other various (40, 60, 80 GPa) pressures. These results were

shown in Figure 5.11. At 80GPa and 60GPa, features observed at 100GPa

(linearity on low temperatures, abnormal increase at higher temperatures,

and kink as boundary) are also observed, while abnormal increase is

relatively indistinct and scattering at lowermost temperatures

(3800K-4050K) is observed at 40GPa. Slow self-diffusion coefficients [10-13.5

(m2/sec)] can cause the scattering. Section 2.8.4 discusses precision of

self-diffusion coefficients of MD simulations and shows that the insufficient

number of atom migrations can cause large errors. For 80GPa and 60GPa,

activation enthalpies for normal ranges are calculated as 376(kJ/mol) and

364(kJ/mol), respectively. Although the normal - abnormal boundary of

40GPa data is relatively indistinct, there is a small jump between

4400K-4450K and it might be considered as a boundary. To avoid large errors,

we do not adopt data from the scattered lower area (3800K-4050K). This

treatment derives activation enthalpies as 346(kJ/mol) for 40GPa (at 4050 –

4450K).

Now data of various pressures (40, 60, 80, and 100GPa) is available and

their pressure variations were examined next. Figure 5.12 plots observed

migration enthalpies (Hm*). Large ( ± 25%) error bar of 40GPa data show its

expected large uncertainty come from ambiguity of its treatments described

in the previous paragraph. Sections 3, 4 successfully examines linearity of

Hm* with pressure variations of self-diffusions of MgO periclase and solid

Neon, though it is difficult to examine linearity of pressure variations from

the data because number of available data is limited (four) and large

108
uncertainly is expected in 40GPa data. So we assume linearly of it and got

relationship by fitting as

Hm* (P) [kJ/mol] = 0.875 *P[GPa] + 310. (5.1)

This relationship gives extrapolated Hm* at 25GPa as 332 [kJ/mol], which

is well consistent with that of experiment (337 kJ/mol at 25GPa, Yamazaki et

al. [2000]). As shown at Figure 3.18, pre-exponential term D0, of MgO

diffusion also varies with pressures like Hm*. Pre-exponential terms largely

varies with adopted Hm* so we used linearity fitted Hm* (equation 5.1)

instead of observed Hm* in order to avoid errors observed Hm* reflects to the

D0. D0 were estimated with following procedures. For each pressure

condition, (1) Hm* is given by using equation (5.1) (2) D0,min, is found by

minimizing following square difference


*
[ Di − D0,min exp(− H m / RTi )]2 (5.2)
i

where Di and Ti are observed self-diffusion coefficient and corresponding

temperature, respectively. (3) D0,min, is adopted as estimated D0. The

estimated D0 (2.92*10-9[m2/sec] at 40GPa, 4.17*10-9[m2/sec] at 60GPa,

6.83*10-9[m2/sec] at 80GPa, 9.60*10-9[m2/sec] at 100GPa) log-linearly

increase with increasing pressure, as shown at Figure 5.13. Log-linear fitting

of the data derives pressure dependence of the D0 as

−3
P[ GPa ]−8.892
D0 (m 2 / sec) = 108.825*10 . (5.3)

109
80GPa-Si 10000/T
1.5 2 2.5 3 3.5
-8

-9
D:
Self -10
Diffusion
Coeff.
-11
(m2/sec)
-12

-13

-14
5000 4000 3000
T:Temperature (K)

60GPa-Si 10000/T
1.5 2 2.5 3 3.5
-8

-9

D:
Self -10
Diffusion
Coeff. -11
(m2/sec)
-12

-13

-14
5000 4000 3000
T:Temperature (K)

110
40GPa-Si 10000/T
1.5 2 2.5 3 3.5
-8

-9

D:
Self -10
Diffusion
Coeff. -11
(m2/sec)
-12

-13

-14
5000 4000 3000
T:Temperature (K)

Figure 5.11 Temperature dependences of silicon (Si) self-diffusion

coefficients at 80GPa (4200-5700K), 60GPa(3800K-5100K),

40GPa(3800K-4600K). Vertical lines indicate boundaries.

111
1000

800

600
Hm*
(kJ/mol)
400

200

0
0 20 40 60 80 100 120
Presssure (GPa)

Figure 5.12 Temperature dependences of migration enthalpies (Hm*) silicon

(Si) self-diffusion coefficients. Filled Diamonds, open diamonds, and filled

circles indicate this study, Karki and Khanduja [2007], and Wright and Price

[1993], respectively. Open circle shows experimental data (Yamazaki et al.

[2000]) at 25GPa.

112
Pressure (GPa)
-4
0 20 40 60 80 100 120
-5

-6
log10(D0)
(m2/sec) -7

-8

-9

-10

Figure 5.13 Pressure dependence of pre-exponential terms (D0) estimated

by minimizing of least – square difference (equation 5.2).

113
5.6 Summary and Discussion

Reproduction of crystal structure of modeled MgSiO3 perovskite by

Buckingham type pair potential and parameters (Oganov et al. [2000]) has

been examined and confirmed that this model well reproduces compressions

at ordinary temperature, thermal expansions at 25GPa pressure, which

corresponds to the top of the lower mantle. Its orthorhombic distortions,

which is characterized by ratios of unit cell length b/a and c/a, are compared

with those of reported experiments and calculations. Through this

comparison, results are summarized that (1) this model reproduces tendency

of distortion of the b/a and c/a (2) this model and the more complex VIB

model (Marton et al. [2001]) also tends to underestimate both b/a and c/a and

LDA ab initio (Wentzcovitch et al. [1995]) tends to overestimate them (3)

amount of overestimation of c/a is relatively smaller than b/a (Figure 5.5).

Significant inferiors of reproductions of model potential calculations

comparing with ab initio LDA (Wentzcovitch et al. [1995]) are not found. For

melting, it is shown that a two-phase method is needed in order to avoid

overestimations coming from thermal instability (Belonoshko [1994]).

Two-phase melting simulation requires direct reproductions of

high-temperature thermal motions of all atoms. Its applications with ab

initio molecular dynamics have been limited to only systems with few

valence electrons (Alfe [2003], Ogitsu et al. [2003]) so that model potential

appears to be better for applying more complex MgSiO3 perovskite.

Interatomic interaction models of this study also reproduce experimental

dTm/dP ≈ 50[K/GPa] during 20-60GPa (Zerr and Boehler [1993]). At higher

114
pressure, this model and previos MD calculations (Belonoshko [1994]) also

predicts reduction of dTm/dP, though the prediction has not been examined

with experiments.

Regarding self-diffusion, it is shown that Si migration enthalpies (Hm*)

derived by direct reproduction of spontaneous migrations are consistent with

those of experimental data (Figure 5.12). Static calculations of model

potential overestimate this even while introducing intermediate oxygen

vacancy (Wright and Price [1993]). Karki and Khanduja [2007] reported that

introducing ab initio LDA does not improve this inconsistency. Since it is

found that this inconsistency is improved by direct reproduction of

spontaneous migrations, elements which previous studies do not treat

(structural relaxation with high temperature and/or frequencies of

spontaneous jumps of atoms) are considered as causes. Pressure variations

of the pre-exponential term D0 was also estimated. Derived pressure

variations of D0 and Hm* enable us to estimate depth variations of Si

self-diffusion of MgSiO3 perovskite, which is considered to dominate

diffusion creep of lower mantle perovskite.

115
Chapter 6

Lower mantle rheology

6.1 Introduction

The lack of seismic anisotropy of the lower mantle is interpreted as

deformation by diffusion creep [Karato et al., 1995], and the model has been

said through the deformation mechanism of analogue CaTiO3 perovskite [Li

et al. 1996] and viscoelastic modeling in order to fit observed postglacial sea

level change [Wu and van der Wal 2003]. Yamazaki and Karato [2001]

estimated viscosity profiles of the lower mantle by employing the

Nabaro-Herring relation (equation 1.1). Because of the lack of available data

of self-diffusion coefficients, the viscosity profiles were estimated from

empirical relations between self-diffusion coefficients and other physical

properties (melting points, and elastic constants), though they have not been

examined enough at the high pressures of the lower mantle. Even at

ordinary pressure, Poirier [2000] pointed out the discrepancy of relations

between melting points and diffusion coefficients as mentioned at chapter

3.3. MD calculations of self-diffusions of two major minerals (MgO periclase,

MgSiO3 perovskite) were reported in previous chapters (chapters 3 and 5),

respectively. These results were compared with data of experiments

(formation enthalpies (Hf*) and migration enthalpies (Hm*) for MgO periclase

at ordinary pressure (Table 3.1), activation volumes of migrations (Vm*) for

MgO periclase at 15-25GPa (Figure 3.19), and migration enthalpies (Hm*) at

116
25GPa (Figure 5.12)). Effective diffusion coefficients of the Nabarro-Herring

form (equation, 1.1) are related with the slowest diffusion species (Frost and

Ashby 1982). Oxygen is the slowest diffusion species of MgO periclase

because of the reported 102–103 times slower O diffusion than Mg diffusion

at ordinary pressures [Ando, 1989], while Si is the slowest diffusion species

of MgSiO3 perovskite (chapter 5.1). Variation of grain size also affects the

viscosity. The Nabarro-Herring form suggests square linear dependence of

grain size and viscosities. However, due to lack of information of grain size

distribution in the lower mantle, we will assume that it is uniform. Yamazaki

et al. [1996] reports significantly slow grain growths of MgO periclase and

MgSiO3 perovskite aggregates (G ∝ t1/10.6 for MgSiO3 perovskite and G ∝ t1/10.8

for MgO periclase, where G and t is grain size and time, respectively) at

pressure-temperature (P-T) conditions near those of the lower mantle

(-1700K and 25GPa). Two regimes (intrinsic and extrinsic) for self-diffusions

should also be considered. The activation enthalpies of Si lattice diffusion of

Yamazaki et al. [2000] (337kJ/mol) at the lower mantle P and T conditions

(1700K-2000K, 25GPa) are much smaller than those of intrinsic diffusion of

oxygen in MgO periclase (586kJ/mol [Ando et al., 1983] and 667kJ/mol [Yang

and Flynn, 1994] and Si grain boundary diffusion (311 kJ/mol), supporting

extrinsic Si lattice diffusion for the experiments of Yamazaki et al. [2000].

However, the possibility of intrinsic lower mantle can still be considered

because of higher temperatures of the lower mantle (over 1900K) than the

experiments. The possibility can be examined by intrinsic vacancy

concentration (equation 2.28). For both MgO periclase and MgSiO3

perovskite, significant increases of Schottky vacancy formation enthalpies

117
with increasing pressures have been reported by Ita and Cohen [1997], Karki

and Khanduja [2006] for MgO periclase, and by Wright and Price [1993] and

Karki and Khanduja [2007] for MgSiO3 perovskite. Therefore, possibility at

the lower mantle is considered to be unlikely (Ita and Cohen [1998]). The

Nabarro-Herring model (equation 1.1) and the assumption of uniform grain

size relates effective diffusion coefficients Deff with the viscosity η as follows:

η ∝ (V / RT ) / Deff (6.1)

where V is molar volume, R is the gas constant, T is temperature, and Deff is

effective diffusion coefficient. Extrinsic Si lattice self-diffusion of MgSiO3

perovskite and extrinsic O lattice self-diffusion of MgO periclase, which

corresponds to the Deff are related with equation 2.23. The equation 2.23 and

equation 6.1 are connected as

η ∝ (V / RT ) /[(C/C 0 ) * D C0,0 exp(-H m * /RT))] (6.2)

where DC0,0 is pre-exponential term of vacancy concentration of MD

simulation (C0), Hm* is the migration enthalpy, and C is corresponding

vacancy concentration. Pressure dependences of Hm* and D0 can be derived

for both MgO periclase and MgSiO3 perovskite from Table 3.2, and Figures

5.14 and 5.15, respectively, so that their pressure variations are expressed as

Hm*(P) and D0 (P). Therefore, equation (6.2) is updated as follows:

η ∝ (V / RT ) /[(C/C 0 ) * D C0,0 (P)exp(-H m * (P)/RT))] . (6.3)

Vacancy concentration C is still unknown, though we assume C independent

on pressure and temperature. For MgO periclase, it is confirmed that Mg

and O ions are formed by trivalent (Sempolinski and Kingery, 1980) and

monovalent (Oishi et al. 1987) impurities in the extrinsic regime at ordinary

pressure. There are certain amounts of chemical species, which can become

118
trivalent and monovalent impurities at the lower mantle. Pyrolite model

(Ringwood 1991) estimates bulk concentrations of Al3+ and Na+ as ≈ 3% and

≈ 0.5 %, respectively. Chemical species, which form Si vacancy of MgSiO3

perovskite, have not been clarified. Dobson et al. [2003] report enhancement

of electronic conductions with concentrations of doped Na+ ions in MgSiO3

perovskite. Formations of oxygen vacancies are interpreted as a cause of the

enhancements. Yamamoto et al. [2003] compare enthalpy changes of two Al

substitution mechanisms (Tschermak substitution [Mg2+ + Si4+ -> 2Al3+]

and substitution with forming of oxygen vacancies [2Si4+ -> 2Al3+ + VO'']) and

report that the Tschermak substitution, which does not form any vacancy, is

more energetically favored. The assumption of constant vacancy

concentrations eliminate the term of C/C0 of (6.3) as follows:

η ∝ (V / RT ) /[D 0 (P)exp(-H m * (P)/RT))] , (6.4)

Here, we will show estimations of the two major minerals (MgO periclase

and MgSiO3 perovskite) of the lower mantle by equation (6.4) with data

derived from MD calculations (D0(P) and Hm*(P)). Assumptions made for the

lower mantle are summarized as follows: (1) diffusion creep described by

Nabarro-Herring model (equation 1.1) (2) uniform grain size (3) effective

diffusion coefficients dominated by extrinsic Si and O lattice diffusions for

MgO periclase and MgSiO3 perovskite, respectively. (4) uniform extrinsic Si

and O vacancy concentrations.

119
6.2 Lower mantle viscosity

6.2.1 Viscosity of the each mineral

MD calculations of diffusion coefficients, which are covered with previous

sections, will be applied. Although radial pressures of the lower mantle

derived from PREM (Dziewonski and Anderson [1981]) are widely accepted,

a large amount of uncertainty concerning the geotherm remains (e.g.

Deschamps and Trampert, [2004]). However (1) P-T of the top of the lower

mantle as 1900K at 25GPa (Ito and Katsura [1989]) and (2) adiabatic

temperature gradient (0.3[K/km]) with assuming advection as a dominant

mechanism of heat transfer of the lower mantle, are acceptable and they

were applied for geophysical implications of mineral physics studies (e.g.

Yamazaki and Karato [2001], Tsutsuya et al. [2004]). Derived linear depth-T

variation ((line of Figure 6.2) increases to 2571K at the core mantle boundary

(CMB). The assumption of the advection as a dominant heat transfer

mechanism also requires that the thermal boundary layer at the bottom of

the CMB comes from convection of the whole mantle. It is confirmed that

transition of perovskite -> post perovskite occurs at 2500K, 2700km depth

(corresponding to 125GPa) (Murakami et al. [2004]). Debye extrapolations

from zero temperature energy calculations (Tsuchiya et al. [2004]) predict a

large (7.5 MPa/K) Clapeyron slope of curve of the transition.The possibility of

double transition (perovskite -> postperovskite-> perovskite) has been

pointed out (e.g. Hernlund et al. [2005]) and Kawai et al. [2007] the double

transition from reduction of S-wave velocity at the lowermost mantle

120
(2800km depth). In the case that we assume the double transition is caused

between 2700km and 2800km, temperature at 2800K is estimated as 3207K.

Figure 6.2 shown summarized geotherm with the constraints.

5000

4000

Temp. 3000
(K)
2000

1000

0
660 1160 1660 2160 2660
depth (km)

Figure 6.2 Geotherm considered in this study. The constraints (1900K at

25GPa, Ito and Katsura [1989], 2500K at 2700km Murakami et al. [2004],

and 3207K at 2800km (Kawai et al. [2007]) are shown as open diamonds.

121
The author will treat only 660-2700 km depth range of the lower mantle.

This range is over 90% of the lower mantle and apparently it is the most

important for the entire rheology of the lower mantle. Figure 6.3 shows

depth variations of viscosities of the two minerals (MgO periclase and

MgSiO3 perovskite) derived from this MD. The viscosity of MgSiO3

perovskite increases with increasing depth up to 16 times at 2200km depth.

Behavior of viscosity of MgO periclase is clearly different from that of

MgSiO3 perovskite. It slightly (2.3 times) increases in the upper area of the

lower mantle (660-1160km) and begins to decrease for more depth

(1160-2700km) with increasing depth. Decrease rate from 1160 to 2700km is

estimated as 21 times. These results are comparable with previous

estimations (Yazamaki and Karato [2001]). The viscosity of the MgSiO3

perovskite is considered to come from linear pressure dependence of Hm*

(equation 5.1). Possibility of non-linear behavior is not removed, though

monotonous increase seems to be likely. Consequently, viscosity profiles of

this study and those of Yamazaki and Karato [2001] are indistinguishable.

On the other hand, significant decrease of viscosity profiles of MgO periclase

at deeper (1160-2700km) lower mantle depths cannot be interpreted as

consistent with uniform [Figure 6.4 (a)] or 30 times increase [Figure 6.4 (b)]

of Yamazaki and Karato [2001]. In the models (a) (b) of Yamazaki and Karato

[2001], activation volume V* and pre-exponential factor D0 are assumed as

constants. Although their assumed value of V* (2.2cm3/mol) is similar with

reported activation volume V* (3.0cm3/mol) at lower (15-25GPa) pressure

(van Orman et al. [2003]), our MD calculations show V* of MgO periclase

significantly decreases with increasing pressure (Figure 3.19). Yamazaki and

122
Karato [2001] estimate activation enthalpy H* in the deeper lower mantle

(2700km depth) as 420kJ/mol and 520kJ/mol for both their models,

respectively, while our results (Table 3.2) of H* are smaller (240kJ/mol). This

significant difference could be the main reason of the inconsistency. Also,

calculated D0 of MgO periclase (Figure 3.18) decreases with increasing

pressure. The amount of decrease ratio is D0 (P=120GPa)/ D0 (P=25GPa)

≈ 14. This ratio is not ignorable so it should be taken into account for more

reliable estimations.

1.5
MgSiO3 Perovskite
1

Relative 0.5
Viscosity
log10(η/η660km)0
660 1160 1660 2160 2660
-0.5
MgO Periclase
-1

-1.5
depth (km)

Figure 6.3 Depth variations of relative viscosities ( η / η 660 km ) of the lower

mantle minerals.

123
(a)

(b)

Figure 6.4 Previous estimations of viscosity variations by Yazamaki and

Karato [2001] estimated from melting points by van Liempt relations

(equation 3.1) [upper figure (a)] and from elastic (Gruneizen paramenter and

bulk moduls) parameters by Lindemann's model. (Poirier and Lierbermann

1984) [lower figure (b)].

124
6.2.2 Two-phase mixture

It is shown that depth variations of viscosities of the two minerals are

significantly different; monotonous increase for the MgSiO3 perovskite and

positive -> negative turning of pressure variation with increasing pressure

for the MgO periclase, so we would want to examine affects of the difference

for total rheology of the lower mantle. Yamazaki and Karato [2001] model

rheology of the lower mantle by a flow law for a mixture of two viscous

phases, which is developed by Handy [1994]. In the flow low, two basic kinds

of microstructures are distinguished in deformed two-phase aggregates: (1) a

load-bearing framework (LBF) of a strong phase containing isolated pockets

of a weak phase; (2) an interconnected layer of a weak phase (IWL)

separating boudins and clasts of a strong phase. Takeda [1998] proposed a

two-phase flow law model on the basis of the continuum physics of a

multiphase assemblage. He suggested two models: (1) linear relationship

between bulk viscosity and the volume fraction of the constituent minerals,

corresponding to case weaker phase isolated (i.e. LBF, named model 1) (2)

non-linear relationship corresponding to a case where a weaker phase forms

a connected film (i.e. IWL, named model 2). By using density contrast

a= γ 1 / γ 2 , viscosity contrast b= η1 / η 2 and volume fractions of less viscous

phase φ1 , normalized (from 1 to b) viscosity η * is expressed as following

equations (6.5) and (6.6), respectively:

η * = (1 − b)φ1 + b (6.5)

[a 2 − 2a(a − 1)φ1 + (a − 1) 2 φ1 ]b
2

η =
*
. (6.6)
a 2 + (b − a 2 )φ

125
Yamazaki and Karato [2001] considered viscosity contrast b > 10 2 at the top

of the lower mantle. They estimate viscosity of MgSiO3 perovskite from Si

self-diffusion coefficients of Yamazaki et al. [2000] and viscosity of MgO

periclase from creep experiments (Passmore et al. [1966]). Measurements of

uniaxial stress components in compressed diamond anvil cells under

ordinary temperature (Merket et al. [2002], [2003]) also supports the model

that MgSiO3 perovskite is stronger than MgO periclase. In the case of

significant (b > 102) viscosity contrasts of the two minerals, convergence of

η * of equations (6.5) and (6.6) by b >> (a, φ ) is suggested as follows,

respectively:

η * → η * max = (a 2 − 2a (a − 1)φ + (a − 1) 2 φ 2 ) / φ . (6.7)

η * → η * max = (1 − φ1 )b . (6.8)

In the case that we assume (1) volume fractions of minerals of the lower

mantle MgO periclase as 20% (e.g. Ringwood [1991]), MgSiO3 perovskite as

70%, and CaSiO3 perovskite as 10% and (2) CaSiO3 perovskite has same

viscosities with MgSiO3 perovskite so that it behaves as stronger phase as

MgSiO3 perovskite, these assumptions derive φ as 0.2. With φ = 0.2, η * max of

the equation (6.7) is derived as 5-8 under considerable range of the density

contrast (a=1.0-1.3) and η * of equation (6.8) is derived as 0.8*b. These

relative viscosities (η * max for the LBF and η * for the IWL) are quite similar

to viscosity of MgO periclase (η * = 1 ) and hard MgSiO3 perovskite (η * = b ),

respectively and that it is a good approximation of viscosity of the lower

mantle that (1) it is dominated by softer MgO periclase in case its texture is

corresponding to the IWL (2) it is dominated by harder MgSiO3 perovskite in

case its texture is corresponding to the LBF. Maximum amount of error of

126
the treatment is estimated as 8/0.8 = 10 (1 magnitude).

127
6.3 Discussion

Viscosity variations of minerals of the lower mantle are derived under

several assumptions (Figure 6.3). The result of major phase MgSiO3

perovskite has depth increasing pattern of viscosity with increasing depth,

which is consistent with Yamazaki and Karato [2001] and geodynamical

modeling of radial profiles of the lower mantle viscosity (e.g. Mitrovica and

Forte [1997]). This consistency supports Yamazaki and Karato [2001]'s

suggestions; (1) observation of geodynamical modeling is corresponding to

the stagnant core of a convection cell in the lower mantle (2) its rheology

controlled by dominated by strong LBF structure, when MgSiO3 perovskite is

dominant phase. Examination of IWL structure for the stagnant core is not

included in Yamazaki and Karato [2001]. Our result of viscosity of MgO

periclase does not support the possibility of IWL structure for the stagnant

core because its significant decrease appears with increasing pressure from

the middle to the bottom of the lower mantle. Instead, Yamazaki and Karato

[2001] suggested the IWL dominant rheology at boundary layer of convection.

The suggestion is based on their speculation that MgO periclase has

significantly smaller viscosity than MgSiO3 perovskite. This speculation

will be supported in the case that there is an evidence of softening of MgO

periclase. Under adiabatic geothermal gradient (0.3K/km), they predict

uniform (Figure 6.4 (a)) or slight hardening (Figure 6.4(b)) of viscosity of

MgO periclase, and softening of MgO periclase is not considered without

assuming superadiabatic (0.6K/km) geothermal gradient (Figure 6.4(a)). On

the other hand, the present result of viscosity of MgO periclase decreases

128
significantly with increasing depth on the adiabatic temperature gradient.

Olson and Singer [1985] experimentally studied affects of viscosity contrasts

of an upwelling plume and its surrounding in a fluid model. They showed

simplified relationship between ratios of radii of head (r) and tail (a) of the

plume (i.e. r/a) and the viscosity contrasts ( η / η 0 , where η and η 0 are

viscosities of the plume and its surrounding, respectively) as

r a ≈ 1.8 * (η / η 0 )1 / 4 . (6.9)

This relation would be applied to the whole-mantle upwelling plume of which

existence is accepted from whole earth tomography (e.g. Norte et al. 2007).

Applying this relation for the whole-mantle plume, we assume (1) its

surrounding has the LBF structure (2) both LBF and IWF structures are

possible in plume itself. In the case that both plumes and surroundings

have LBF structures, MgSiO3 perovskite governs both rheologies and thicker

tail of the plume is considered in that case. The ratio of the radii is estimated

as r / a ≈ 1.8 when ( η / η 0 )1 / 4 ≈ 1 and thermal effect of the viscosity is

neglected. However, in the case of the IWF structure, it causes thinner tail

plumes at the bottom of the mantle because of significant increase of

viscosity contrast η / η 0 . Figure 6.5 shows variations of relative viscosity

contrasts of the two phases derived from figure 6.3. From figure 6.5,

increases of viscosity contrast with increasing depth is estimated as

(η / η 0 ) 2700 km /(η / η 0 ) 660 km = 102.5 at 2700km depth. By using the viscosity

contrast (η / η 0 ) 660 km = 10 2 of Yamazaki and Karato [2001], the ratio of the

radii at 2700km is estimated as r a ≈ 24 at 2700km depth. Consequently, it

is estimated that a plume of IWF structure has 24/1.8 ≈ 13.3 times thicker

tails than a plume of LBF structure. Possible radii of plume tails are related

129
with scale of thermal boundary, which are the origins of the plumes.

Thin-tail plumes are probably stable under more small-scale thermal

boundary. Resolutions of seismic tomography are still rough (-200km, Norte

et al. [2007]), and then it would be insufficient to detect detailed shape of the

plume, though we would expect future technical progress to enable us to

detect variations of the tail radii of the whole-mantle plume.

Viscosity 1
contrast
log10([η2/η1]/ 0
[η2/η1]660km) 660 1160 1660 2160 2660
-1

-2

-3
depth (km)

Figure 6.5 Depth variations of viscosity contrasts of the two phases

(MgSiO3 perovskite

130
Chapter 7

Conclusion

Self-diffusions of lower mantle minerals (MgO periclase and MgSiO3

pervovskite) were simulated by reproducing spontaneous migrations of

atoms via vacancies with molecular dynamics (MD) and self-diffusion

coefficients were derived. In order to examine reliabilities of the results, not

only self-diffusion coefficients themselves but also other physical properties

(equations of states, melting points, and elastic properties) were compared

with those of experimental data. Through the examinations, significant

deficiencies, which let us consider the results as unreliable ones, were not

found. Especially, activation enthalpies and volumes of the results were well

consistent with those of experiments, supporting reliabilities of the results.

From these results, self-diffusion coefficients of MgO periclase and MgSiO3

perovskite are derived. In addition, long-range force of MgO periclase is

indicated as a possible cause of the strange pressure dependence of the

results of MgO periclase. Depth variations of viscosities of the lower mantle

were estimated as modest (MgSiO3 perovskite) and significant (MgO

periclase). Possibilities of (1) radial profile of viscosity derived from

geophysical observations corresponding to MgSiO3 perovskite (2) large

( ≈ 104.5) viscosity contrast at the bottom of the lower mantle were pointed out

from the results. Through the successes of the work, it is expected that

further implications of the MD would clarify other diffusive and rheological

properties of Earth's interior.

131
Acknowledgements

The author expresses his thanks for Prof. M Toriumi for his continuous

supervising and encouraging. He is also grateful to Profs. K. Kawamura and

A. Kageyama for providing molecular dynamics program (MXDTRICL) and

project support for using high-performance computers, respectively.

Discussions with Prof. T. Tsuchiya and many members of Toriumi

Laboratory turn to be fruitful for the author. Most calculations were

performed on Earth Simulator at Japan Agency for Marine-Earth Science

and Technology.
References

Akamatsu, K. and Kawamura, K. (1999)


Molecular dynamics of solid solution and coexisting liquid, Molecular Simulation, 21,
387-399

Alfè D. (2003)
First-principles simulations of direct coexistence of solid and liquid aluminum, Physical
Review B, 68, 064423

Alfe, D., Alfredsson, M., Brodholt, J., Gillan, J., Towler, M. D. and Needs, R. J. (2005)
Quantum monte carlo calculations of the structural properties and the B1-B2 phase
transition of MgO, Physical Review B, 72, 014114

Andersen, H. C. (1980)
Molecular dynamics simulations at constant pressure and/or temperature, J. Chem.
Phys., 72, 2384–2393

Anderson, M. S., Fugate, R. Q. and Swenson, C. A. (1973)


Equation of state for solid neon to 20 kbar, Journal of Low Temperature Physics, 10,
345-357

Ando, K. (1989)
Self-diffusion in oxides, "Rheology of Solids and of the Earth" edited by Karato, S. and
Toriumi, M.; Oxford University Press Tokyo, 57-82

Ando, K., Kurokawa, Y. and Oishi, Y. (1983)


Oxygen self-diffusion in Fe-doped MgO single-crystals, Journal of Chemical Physics, 78,
6890-6892

Asahara Y., E. Ohtani, T. Kondo, T. Kubo, N. Miyajima, T. Nagase, K. Fujino, T. Yagi,


and T. Kikegawa (2004)
Formation of metastable cubic-perovskite in high-pressure phase transformation of
Ca(Mg,Fe,Al)Si2O6, American Mineralogist, 90, 457-462, 2004

Béjina, F., Jaoul, O. and Liebermann, R. C. (1999)


Activation volume of Si diffusion in San Carlos olivine: implications for upper mantle
rheology, Journal Geophysical Research, B 104 (11), 25529–25542

i
Bejina, F., Jaoul, O. and Liebermann, R. C. (2003)
Diffusion in minerals at high pressure: a review, Physics of the Earth and Planetary
Interiors, 139, 3-20

Belonoshko, A. B. (1994)
Molecular-dynamics of MgSiO3 perovskite at high-pressures - equation of state,
structure, and melting transition, Geochimica et Cosmochimica Acta, 58, 4039-4047

Belonoshko, A. B. (2001)
Molecular dynamics simulation of phase transitions and melting in MgSiO3 with the
perovskite structure - Comment, American Minerallogist, 86, 193-194

Belonoshko, A. B. and Dubrovinsky, L. S. (1996)


Molecular dynamics of NaCl (B1 and B2) and MgO (B1) melting: Two-phase simulation,
American Mineralogist, 81, 303-316

Birch F. (1961)
The velocity of compressional waves in rocks to 10 kilobars, part 2, Journal of
Geophysical Research, 66, 2199-2224

Boehler, R. and Zerr, A. (1994)


High-pressure melting of (Mg,Fe)SiO3-perovskite, Science, 264, 280–281

Brown, J. M. (1999)
The NaCl pressure standard, Journal of Applied Physics, 86, 5801–5808

Car, R. and Parrinello, M. (1985)


Unified approach for molecular-dynamics and density-functional theory, Physical
review letters, 55, 2471-2474

Chaplot, S. L., Choudhury, N., and Rao, K. R. (1998)


Molecular dynamics simulation of phase transitions and melting in MgSiO3 with the
perovskite structure, American Mineralogist, 83, 937-941

Cox D. E. and A. W. Sleight (1979)


Mixed-valent Ba2Bi3+Bi5+O6: structure and properties vs temperature, Acta
Crystallographica section B, 35 (1), 1-10

Datchi F, Loubeyre P, LeToullec R (2000)


Extended and accurate determination of the melting curves of argon, helium, ice (H2O),
and hydrogen (H2), Physical Review B, 61, 6535 - 6546

Decker, D. L. (1971)
High-pressure equation of state for NaCl, KCl, and CsCl, Journal of Applied Physics, 42,
3239–3244

ii
Deschamps F. and J. Trampert (2004)
Towards a lower mantle reference temperature and composition, Earth and Planetary
Science Letters, 222, 161-175

Dewaele, A., Fiquet, G., Andrault, D. and Hausermann, D. (2000)


P-V-T equation of state of periclase from synchrotron radiation measurements, Journal
of Geophysical Research, 15, 2869-2877

Dobson, D., Dohmen, R. and Wiedenbeck, M. (2003)


Oxygen and silicon diffusion, and electrical conductivity in MgSiO3 perovskite, EGS -
AGU - EUG Joint Assembly, Abstracts from the meeting held in Nice, France, 6 - 11
April 2003, abstract #8774

Dubrovinsky L. S. and S. K. Saxena (1997)


Thermal Expansion of Periclase (MgO) and Tungsten (W) to Melting Temperatures,
Physics and Chemistry of Minerals, 24, 547–550

Duffy, T. S., Hemley, R. J. and Mao, H. K. (1995)


Equation of state and shear strength at multimegabar pressures: Magnesium oxide to
227 Gpa, Physical Review Letters, 74, 1371–1374

Dziewonski, A. M. and Anderson, D. L. (1981)


Preliminary reference Earth model, Physics of the Earth and Planetary Interiors, 25,
297-356
Errandonea, D., Boehler, R., Japel, S., Mezouar, M. and Benedetti, L. R. (2006)
Structural transformation of compressed solid Ar: An x-ray diffraction study to 114 GPa,
Physical Review B, 73 (9), 092106

Ewald, P. (1921)
"Die Berechnung optischer und elektrostatischer Gitterpotentiale", Annalen der Physik,
64, 253-287

Fei, Y. W. (1999)
Effects of temperature and composition on the bulk modulus of (Mg,Fe)O, Americal
Mineralogist, 84 (3), 272-276

Fei, Y. W. and Mao, H. K. (1994)


In Situ Determination of the NiAs Phase of FeO at High Pressure and Temperature,
Science, 266, 1678 – 1680

Fiquet, G., D. Andrault, A. Daweale, T. Charpin, M. Kunz, and D. H äussermann (1998)


P-V-T equation of state of MgSiO3-perovskite, Physics of the Earth and Planetary
Interiors, 105, 21-31

iii
Fiquet, G., Dewaele, A., Andrault, D., Kunz, M. and Bihan, T. (2000)
Thermoelastic properties and crystal structure of MgSiO3 perovskite at lower mantle
pressure and temperature conditions, Geophysical Research Letters, 27 (1), doi:
10.1029/1999GL008397

Freer, R. (1980)
Self-diffusion and impurity diffusion in oxides, Journal of Materials Science, 15,
803-824

Frost, H. J. and Ashby, M. F. (1982)


Deformation-mechanism maps, The plasticity and creep of metals and ceramics;
Pergamon Press, Oxford, p.44

Funamori, N., Yagi, T., Utsumi, W., Kondo, T., Uchida, T. and Funamori, M. (1996)
Thermoelastic properties of MgSiO3 perovskite determined by in situ X ray
observations up to 30 GPa and 2000 K [MgSiO3], Journal of Geophysical Research, 101,
8257-9269

Gong, Z., Fei, Y., Dai, Y., Zhang, L. and Jing, F. (2004)
Equation of state and phase stability of mantle perovskite up to 140 GPa shock pressure
and its geophysical implications, Geophysical Research Letters, 31, L04614,
doi:10.1029/2003GL019132

Handy MR (1994)
Flow laws for rocks containing two non-linear viscous phases: A phenomenological
approach, Journal of Structural Geology, 16, 287-301

Heinz, D. L. and Jeanloz, R. (1987)


Measurement of the melting curve of Mg0.9Fe0.1SiO2 at lower mantle conditions and its
geophysical implications, Journal of Geophysical Research, 92, 11437-11444

Heinz, D. L., Knittle, E., Sweeney, J. S., Williams, Q. and Jeanloz, R. (1994)
High-Pressure Melting of (Mg,Fe)SiO3-Perovskite, Science, 264, 279 - 280

Hernlund, J. W., Thomas, C. and Tackley, P. J. (2005)


A doubling of the post-perovskite phase boundary and structure of the Earth’s
lowermost mantle, Nature, 434, 882– 886

Hirao K. and K. Kawamura (1994)


Material Design using Personal Computer (in Japanese), Shokabo, Tokyo

Hoherberg, P. and Korn, W. (1965)


In homogeneous electron gas, Physical Review B, 136, 864-871

iv
Holzapfel, C., Rubie, D. C., Frost, D. J. and Langenhorst, F. (2005)
Fe-Mg Interdiffusion in (Mg,Fe)SiO3 Perovskite and Lower Mantle Reequilibration,
Science, 309, 1707-1710

Irifune, T., Sueda, Y., Sanehira, T., Yamazaki, D., Nishiyama, N. and Funakoshi, K.
(2003)
Generation of pressures exceeding 50 GPa in multianvil apparatus and some
applications to lower mantle mineralogy, American Geophysical Union, Fall Meeting
2003, abstract #V32E-04

Ita, J. and Cohen, R. E. (1997)


Effects of pressure on diffusion and vacancy formation in MgO from nonempirical
free-energy integrations, Physical Review Letters, 79 (17), 3198-3201, 1997

Ita, J. and Cohen, R. E. (1998)


Diffusion in MgO at high pressures: Implications for lower mantle rheology.
Geophysical Research Letters, 25, 1095–1098

Ito, E. and Katsura, T. (1989)


A temperature profile of the mantle transition zone, Geophysical Research letters, 16 (5),
425-428, 1989

Ito, Y. and Toriumi, M. (2007)


Pressure effect of self-diffusion in periclase (MgO) by molecular dynamics, Journal of
Geophysical Research, 112, B04206, doi:10.1029/2005JB003685

John, W. Hernlund1, Christine, T. and Paul J. T. (2005)


A doubling of the post-perovskite phase boundary and structure of the Earth's
lowermost mantle, Nature, 434, 882-886

Kamiya (1996)
Determination of interatomic potential by ab-initio periodic calculation for MgO;
Japanese journal of applied physics, 35, 3688-3694

Karato, S. (1981)
Rheology of the lower mantle, Physics of the Earth and Planetary Interiors, 24, 1–14

Karato, S., Zhang, S. Q. and Wenk, H. R. (1995)


Superplasticity in earths lower mantle - evidence from seismic anisotropy and rock
physics, Science, 270, 458-461

Karato, S. and Wenk, H. R. (2003)


Theoretical Analysis of Shear Localization in the Lithosphere, Plastic Deformation of
Minerals and Rocks edited by Karato, S. and Wenk, H. R., Mineralogical Society of
America; Washington DC, 387-420

v
Karato, S. and Wu, P. (1993)
Rheology of the Upper Mantle: A Synthesis, Science, 260, 771 - 778

Karki, B. B. and Khanduja, G. (2006)


Vacancy defects in MgO at high pressure, American Mineralogist, 91 (4), 511-516

Karki, B. B. and Khanduja, G. (2006)


Computer simulation and visualization of vacancy defects in MgSiO3 perovskite,
Modelling and Simulation in Materials Science and Engineering, 14, 1041–1052

Karki, B. B. and Khanduja, G. (2007)


A computational study of ionic vacancies and diffusion in MgSiO3 perovskite and
post-perovskite, Earth and Planetary Science Letters, 260, 201–211

Karki B. B., L. Stixrude and R, M. Wentzcovitch (2001)


High-pressure elastic properties of major materials of earth’s mantle from first
principles, Reviews of Geophysics, 39 (4), 507-534

Katsura, T., Yamada, H., Shinmei, T., Kubo, A., Ono, S., Kanzaki, M., Yoneda, A., Walter,
M., J., Ito, E., Urakawa, S., Funakoshi, K., and Utsumi, W. (2003)
Post-spinel transition in Mg2SiO4 determined by high P-T in situ X-ray diffractometry,
Physics of the Earth and Planetary Interiors, 136 (1), 11-24

Kawai, K., N. Takeuchi, R. J. Geller and N. Fuji (2007)


Possible evidencde for a double crossing phase transition in D” beneath Central
America from inversion of seismic waveforms, Geophysical Research Letters, 34,
L09314, doi:10.1029/2007GL029642

Kittel C (1972)
Introduction to Solid State Physics, Wiley, NJ, USA

Koci, L., Ahuja, R. and Belonoshko, A. B. (2007)


Ab initio and classical molecular dynamics of neon melting at high pressure, Physical
Review B, 75, 214108

Kubiki, J. D. and Lasaga, A. C. (1993)


Molecular dynamics simulations of interdiffusion in MgSiO3-Mg2SiO4, Physics and
Chemistry of Minerals, 20 (4), 255-262

Kumagai, I. (2002)
On the anatomy of mantle plumes: effect of the viscosity ratio on entrainment and
stirring, Earth Planetary Science Letters, 198, 211-224

Landau L. D. and E. M. Lifshitz (1958)


Statistical Physics, Pergamon Press, London, Section 33

vi
Lennard-Jones J. and J. E. Cohesion (1931)
Proceedings of the Physical Society, 43, 461-482

Leslie M. and M. J. Gillan (1985)


The energy and elastic dipole tensor of defects in ionic crystals calculated by the
supercell method, Journal of Physics: Condensed Matter, 18, 973-982.
Li P., S. Karato, and Z. Wang (1996)
High-temperature creep in fine-grained polycrystalline CaTiO3, an analogue material of
(Mg, Fe)SiO3 perovskite, Physics of the Earth and Planetary Interiors, 95, 19−36
Liu L. (1976)
Orthorhombic perovskite phases observed in olivine, pyroxene and garnet at high
pressures and temperatures, Physics of the Earth and Planetary Interiors, 11, 289-298

Mackrodt W. C. (1982)
Computer Simulations of Solids edited by C. R. A. Catlow and W. C. Mackrodt,
Springer-Verlag, Berlin

Mao H.K. and Bell P.M. (1979)


Equations of state of MgO and epsilon-Re under static pressure conditions, Journal of
Geophysical Research, 84, 4533-4536

Mao H. K., R. J. Hemley, Y. Fei, J. F. Shu, L. C. Chen, A. P. Jephcoat, and Y. Wu (1991)


Effect of pressure, temperature and composition on lattice parameters and density of
(Fe,Mg)SiO-perovskites to 30 GPa, Journal of Geophysical Research, 96, 8069-8080

Matsui, M. and Price, G. D. (1991)


Simulation of the pre-melting behavior of MgSiO3 perovskite at high-pressures and
temperatures, Nature, 351, 735-737

Marton F. C. and R. E. Cohen (2002)


Constraints on lower mantle composition from molecular dynamics simulations of
MgSiO3 perovskite, Physics of the Earth and Planetary Interiors, 134 (3-4), 239-252

Marton F. C., J. Ita and R. E. Cohen (2001)


Pressure-volume-temperature equation of state of MgSiO3 perovskite from molecular
dynamics and constraints on lower mantle composition, Journal of Geophysical
Research, 106, 8615-8627

Matsui M. (1988)
Molecular Dynamics Study of MgSiO3 Perovskite, Physics and chemistry of minerals,
16, 234-238

Matsui M. (1989)
Molecular dynamics study of the structural and thermodynamic properties of MgO
crystal with quantum correction, Journal of Chemical Physics, 91, 489-494

vii
Matsui M., S. C. Parker and M. Leslie (2000)
The MD simulation of the equation of state of MgO: Application as a pressure
calibration standard at high temperature and high-pressure, American Mineralogist, 85,
312-316

Matsui M. and G. D. Price (1991)


Simulation of the pre-melting behavior of MgSiO3 perovskite at high pressures and
temperatures, Nature, 351, 735–737
Meade C., H. K. Mao and J. Z. Hu (1995)
High-temperature phase-transition and dissociation of (Mg,Fe)SiO3 perovskite at lower
mantle pressures, Science, 268, 1743-1745

Merkel S., H. R. Wenk, J. Badro, G. Montagnac, P. Gillet, H. Mao and R. Hemley (2003)
Deformation of (Mg0.9,Fe0.1)SiO3 Perovskite aggregates up to 32 Gpa, Earth and
Planetary Science Letters, 209, 351-360

Merkel S., H. R. Wenk, J. Shu, G. Shen, P. Gillet, H. Mao and R. Hemley (2002)
Deformation of polycrystalline MgO at pressures of the lower mantle, Journal of
Geophysical Research, 107 (B11), 2271, doi:10.1029/2001JB000920

Mitrovica J. X. and A. M. Forte (1997)


The Radial Profile of Mantle Viscosity: Results From the Joint Inversion of Convection
and Post-Glacial Rebound Observables, Journal of Geophysical Research, 102,
2751-2769

Montagner J. P. and B. L. N. Kennett (1996)


How to reconcile body-wave and normal-mode reference earth models, Geophysical
Journal International, 125, 229–248

Murakami M., K. Hirose, N. Sata, Y. Ohishi, and K. Kawamura (2004)


Phase transition of MgSiO3 perovskite in the deep lower mantle, Science, 304, 855-858

Nolet G., R. Allen and D. Zhao (2007)


Mantle plume tomography, Chemical Geology, 241, 248‒263

Nose S. (1984)
A unified formulation of the constant temperature molecular-dynamics methods,
Journal of Chemical Physics, 81, 511 –519

Oganov A. R., J. P. Brodholt and G. D. Price (2000)


Comparative study of quasiharmonic lattice dynamics, molecular dynamics and Debye model
applied to MgSiO3 perovskite, Physics of the Earth and Planetary Interiors, 122, 277–288

viii
Oganov A. R., M. J. Gillan and G. D. Price (2003)
Ab initio lattice dynamics and structural stability of MgO, Journal of Chemical Physics,
118, 10174-10182

Ogitsu T., E. Schwegler, F. Gygi and G. Galli (2003)


Melting of Lithium Hydride under Pressure, Physical Review Letters, 91, 175502

Oguri K., N. Funamori, F. Sakai, T. Kondo, T. Uchida and T. Yagi (1997)


High-pressure and high-temperature phase relations in diopside CaMgSi2O6, Physics
of the Earth and Planetary Interiors, 104, 363-370
Ohno I. (1996)
High-pressure phase transitions and thermodynamics of the Earth component
materials, in "Earth and Planetary Material Science", edited by Mitsuhiro Toriumi,
Iwanami Shoten Publishers, Tokyo, Japan (in Japanese)

Oishi Y., K. Ando and K. Yasumura (1987)


Oxygen Self-Diffusion in Lithium-Doped Single-Crystal Magnesium Oxide, Journal of
the American Ceramic Society, 70 (11), C327-C329, doi:10.1111/j.1151-2916.1987.tb05649

Olson P. and H. Singer (1985)


Creeping plumes, Journal of Fluid Mechanics, 158, 511–531

Passmore E. M., R. H. Duff and T. Vasilos (1966)


Creep of dense, polycrystalline magnesium oxide, Journal of the American Ceramic
Society, 49, 594–600

Poirier J. P. and R. C. Liebermann (1984)


On the activation volume for creep and its variation with depth in the Earth’s lower
mantle, Physics of the Earth and Planetary Interiors, 35, 283–293

Porier J. P. (2000)
Introduction to the Physics of the Earth's Interior, edited by Poirier Jean-Paul,
Cambridge University Prress

Rappe A. K. and W. A. Goddard III (1991)


Charge Equilibration for Molecular Dynamics Simulations, Journal of Physical
Chemistry, 95, 3358-3363

Richet, P., H. K. Mao and P. M. Bell (1988)


Static compression and equation of state of CaO up to 1.35 MBar, Journal of
Geophysical Research, 93, 15279-15288

Ringwood A. E. (1991)
Phase transformations and their bearing on the constitution and dynamics of the
mantle, Geochimica et Cosmochimica Acta, 55, 2083-2110

ix
Ringwood A. E. (1975)
Composition and Petrology of the Earth’s Mantle, McGraw-Hill, New York

Ross, M., H.-K. Mao, P. M. Bell, and J. A. Xu (1986)


The equation of state of dense argon: a comparison of shock and static studies, Journal
of Chemical Physics, 85, 1028–1033

Sato-Sorensen Y. (1983)
Phase transitions and equations of state for the sodium halides: NaF, NaCl, NaBr, and
NaI, Journal of Geophysical Research, 88, 3543-3548

Sempolinsky D. R. and W. D. Kingery (1980)


Ionic conductivity and magnesium vacancy mobility in magnesium oxide, Journal of
American Ceramics Society, 63, 664-669

Shannon R. D. and C. T. Prewitt (1969)


Effective ionic radii in oxides and fluorides, Acta Cryst, B25, 925-946

Shen, G. and Lazor, P. (1995)


Measurement of melting temperatures of some minerals under lower mantle pressures,
Journal of Geophysical Research, 100, 17699-17713

Shewmon P. G. (1963)
Diffusion in Solids, McGraw-Hill, New York

Shim S. H., T. S. Duffy and G. Shen (2001)


Stability and Structure of MgSiO3 Perovskite to 2300-Kilometer Depth in Earth's
Mantle, Science, 293, 2437-2440

Shim S. H., T. S. Duffy, R. Jeanloz and G. Shen (2004)


Stability and crystal structure of MgSiO3 perovskite to the core-mantle boundary,
Geophysical Research Letters, 31, L10603, doi:10.1029/2004GL019639

Shuster D. L. and K. A. Farley (2005)


Diffusion kinetics of proton-induced 21Ne, 3He, and 4He in quartz, Geochimica et
Cosmochimica Acta, 69 (9), 2349–2359

Sinogeikin S. V. and J. D. Bass (2000)


Single-crystal elasticity of pyrope and MgO to 20 GPa by Brillouin scattering in the
diamond cell, Physics of the Earth and Planetary Interiors, 120, 43-62

Speziale S., C. S. Zha, T. S. Duffy, R. J. Hemley and H. K. Mao (2001)


Quasihydrostatic compression of magnesium oxide to 52 GPa: Implications for the
pressure-volume-temperature equation of state, Journal of Geophysical Research, 106,
515–528

x
Steinberger B. and A. R. Calderwood (2006)
Models of large-scale viscous flow in the Earth's mantle with constraints from mineral
physics and surface observations, Geophysical Journal International, 167 (3), 1461-1481

Stixrude L. and R. E. Cohen (1993)


Stability of orthorhombic MgSiO3 perovskite in the Earth's lower mantle, Nature 364,
613-616

Stocker R. L. and M. F. Ashby (1973)


On the rheology of the upper mantle, Reviews of Geophysics and Space Physics, 11,
391–426

Stuart J. A. and G. D. Price (1996)


Atomistic potentials for MgSiO3 with the orthorhombic perovskite structure,
Philosophical Magazine, 73, 41-48

Sweeney J. S. and D. L. Heinz (1993)


Melting of iron magnesium silicate perovskite, Geophysical Research Letters, 20,
855-858

Takeda Y. (1998)
Flow in rocks modelled as multiphase continua: application to polymineralic rocks,
Journal of Structural Geology, 20, 1569-1578

Tanimoto T. and D. L. Anderson (1985)


Lateral heterogeneity and azimuthal anisotropy of the upper mantle: Love and Rayleigh
waves 100-250 sec, Journal of Geophysical Research, 90, 1842- 1858

Thomsen L. (1986)
Weak elastic anisotropy, Geophysics, 51 (10), 1954-1966

Tsuchiya T. (2000)
Molecular dynamics of pressure-induced structural transformations and elastic
stability in oxide minerals, Ph D. thesis, Osaka Univ

Tsuchiya T. and K. Kawamura (2002)


First-principles study of systematics of high-pressure elasticity in rare gas solids, Ne, Ar,
Kr, and Xe, Journal of Chemical Physics, 117, 5859-5865

Tsuchiya T., J. Tsuchiya, K. Umemoto, and R. M. Wentzcovitch (2004)


Phase transition in MgSiO3 perovskite in the earth's lower mantle, Earth and
Planetary Science Letters, 224, 241-248

xi
Van Orman J. A., Y. Fei, E. H. Hauri and J. Wang (2003)
Diffusion in MgO at high pressures: Constraints on deformation mechanisms and
chemical transport at the core-mantle boundary, Geophysical Research Letters, 30 (2),
1056, doi:10.1029/2002GL016343

Vos W. L., M. G. E. van Hinsberg and J. A. Schouten (1990)


High-pressure triple point in helium: The melting line of helium up to 240 kbar,
Physical Review B, 42, 6106 - 6109

Wang Z. W. (1999)
The melting of Al-bearing perovskite at the core-mantle boundary, Physics of the Earth
and Planetary Interiors, 115, 219–228

Watson G. W., A. Wall and S. C. Parker (2000)


Atomistic simulation of the effect of temperature and pressure on point defect formation
in MgSiO3 perovskite and the stability of CaSiO3 perovskite, Journal of Physics:
Condensed Matter, 12, 8427-8438

Weidner D. J. (1986)
Mantle models based on measured physical properties of minerals, In “Chemistry and
Physics of Terrestrial planets” edited by S. K. Saxena, Springer, New-York, 251-274

Wentzcovitch R.M., José Luís Martins, and G. D. Price (1993)


Ab initio molecular dynamics with variable cell shape: Application to MgSiO3, Physical
Review Letters, 70, 3947-3951

Wentzcovitch R. M., N. L. Ross and G. D. Price (1995)


Ab initio study of MgSiO3 and CaSiO3 perovskites at lower-mantle pressures, Physics
of the Earth and Planetary Interiors, 90, 101-112

Woodcock L.V., Angell C.A., Cheeseman P. (1976)


Molecular dynamics studies of the vitreous state: Simple ionic systems and silica
The Journal of Chemical Physics, 65, 4, 1565-1577

Wright K. and G. D. Price (1993)


Computer simulations of defects and diffusion in perovskites, Journal of Geophysical
Research 98, 22245–53

Wu P. and W. van der Wal (2003)


Postglacial sealevels on a spherical, self-gravitating viscoelastic earth: effects of lateral
viscosity variations in the upper mantle on the inference of viscosity contrasts in the
lower mantle, Earth and Planetary Science Letters, 21, 57-68

xii
Yagi T., H. K. Mao and P. M. Bell (1977)
Crystal structure of MgSiO3 perovskite, Carnegie Institution of Washington Year Book,
76, 516-519

Yagi T., H. K. Mao, and P. M. Bell (1979)


Lattice parameters and specific volume for perovskite phase of orthopyroxene
composition (Mg,Fe)SiO3, Carnegie Institution of Washington Year book, 78, 612–613

Yamamoto T., D. A. Yuen and T. Ebisuzuki (2003)


Substitution mechanism of Al ions in MgSiO3 perovskite under high pressure
conditions from first-principles calculations, Earth and Planetary Science Letters, 206,
617–625

Yamazaki D. and S. Karato (2001)


Some mineral physics constraints on the rheology and geothermal structure of Earth's
lower mantle, American Mineralogist, 86 (4), 385-391

Yamazaki D., T. Kato, and H. Yurimoto (2000)


Silicon self-diffusion in MgSiO3 perovskite at 25 GPa, Physics of the Earth and
Planetary Interiors, 119, 299-309

Yamazaki D., T. Kato, and E. Ohtani (1996)


Grain growth rates of MgSiO3 perovskite and periclase under lower mantle conditions,
Science, 274, 2052-2054

Yang M. H. and C. P. Flynn (1994)


Intrinsic diffusion properties of an oxide - MgO, Physical Review Letters, 73, 1809-1812

Zerr A. and R. Boehler (1993)


Melting of (Mg,Fe)SiO3 perovskite to 625 kilobars: Indication of a high melting
temperature in the lower mantle, Science, 262, 553–555

Zerr A. and R. Boehler (1994)


Constraints on the melting temperature of the lower mantle from high-pressure
experiments on MgO and Magnesiowustite, Nature, 371, 506-508

Zha C. S., H. K. Mao and R. J. Hemley (2000)


Elasticity of MgO and a primary pressure scale to 55 GPa, Proceedings of the National
Academy of Sciences of the United States of America, 97 (25), 13494-13499

xiii

You might also like