JOURNAL OF GEOPHYSICAL RESEARCH, VOL. 112, B04206, doi:10.

1029/2005JB003685, 2007
Click
Here
for
Full
Article

Pressure effect of self-diffusion in periclase (MgO)

by molecular dynamics
Yosuke Ito1 and Mitsuhiro Toriumi2
Received 21 February 2005; revised 11 August 2006; accepted 27 October 2006; published 27 April 2007.

[1] Pressure effect of self-diffusion in periclase (MgO) was investigated by molecular

dynamics (MD), and viscosity variation of lower mantle periclase was predicted by
employing the Nabarro-Herring model. Self-diffusion coefficients were derived
directly from recorded trajectories of atoms. The MD calculations were performed at
2400–6000 K and 0–140 GPa. Migration energy (Hm*) and formation energy (Hf*) of
ordinary pressure (0 GPa) are consistent with experimental data. Observed pressure
dependence of self-diffusion coefficients was turned from negative with increasing
pressure. The behavior of self-diffusion coefficients predicts softening of lower mantle
periclase with increasing depth. We predict 1.3 times reduction of the viscosity
(0.3 K/km geothermal gradient) or 2.0 times (0.6 K/km geothermal gradient).
Citation: Ito, Y., and M. Toriumi (2007), Pressure effect of self-diffusion in periclase (MgO) by molecular dynamics, J. Geophys.
Res., 112, B04206, doi:10.1029/2005JB003685.

1. Introduction to reproduce the spontaneous vacancy migration by MD

[2] Diffusion limits many kinetic phenomena, which are
important in Earth and planetary interiors, such as creep of
solid phases and chemical mass transportation. Periclase
(MgO), which is thought to be a minor component of
Earth’s lower mantle (
20%), may have a strong influence
on its rheological property because periclase is weaker than
the major component perovskite (MgSiO3) [Yamazaki and
Karato, 2001]. Diffusion in periclase has been studied
extensively in the materials science community, but the
pressure dependence has hardly been considered. Recently,
Van Orman et al. [2003] measured self-diffusion coeffi-
cients of periclase at 15– 25 GPa pressures, though the
pressures in the lower mantle exceed 25 GPa. Because of
experimental difficulties theoretical approaches have been
employed to derive self-diffusion coefficients, though those
lead to other difficulties. Since self-diffusion is an activated
mechanism, the self-diffusion coefficient can be separated
into two elements; the height of the energy barrier (i.e.,
difference between the maximum and minimum free ener-
gies during atom migration) and the attempt frequency of
atom migration. The difficulty is that no method to derive
the attempt frequency is known. Ita and Cohen [1997]
measured the heights of energy barriers by free energy
integrations, and 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. Another approach is
1
Department of Earth and Planetary Science, Graduate School of
Science, University of Tokyo, Tokyo, Japan.
2
Department of Complexity Science and Engineering, Graduate School
of Frontier Science, University of Tokyo, Tokyo, Japan.
Copyright 2007 by the American Geophysical Union.
0148-0227/07/2005JB003685$09.00
(molecular dynamics) and to derive self-diffusion coeffi-
cients directly from recorded trajectories of atoms. Although
this approach does not need to treat the attempt frequency
explicitly, it is known that it is difficult to be precise about
the self-diffusion coefficient because frequencies of atom
migrations are truncated due to the short (e.g., 100 ps) MD
running time. Akamatsu and Kawamura [1999] derived
self-diffusion coefficients of noble gas atoms in MgO, but
they also reported that their results have large errors, up to 3
orders of magnitude.
[3] In this paper, we report on molecular dynamics
(MD) calculations, which simulate self-diffusion in peri-
clase. We selected the latter approach (reproducing sponta-
neous vacancy migrations) and found that one can achieve a
certain precision of the self-diffusion coefficient through
long (20 ns) running times. The Buckingham model was
adopted as an interatomic potential. A pair of vacancies
(Mg and O atoms) was initially introduced into the MD
system. MD calculations were performed at 2400 –6000 K
temperatures and 0– 200 GPa pressures. Results were com-
pared with past experimental and theoretical studies. The
lack of seismic anisotropy in the lower mantle indicates
diffusion creep [Karato et al., 1995]. We derived the
viscosity of lower mantle periclase by employing the
Nabarro-Herring model of diffusion creep.
2. Methods
[4] In our MD simulations, the Buckingham model was
adopted to define the interatomic potential. Formulation of
the model is
 

 qq
 ðai þ aj
rij Þ CC
Uij rij ¼ riij j þ f bi þ bj exp b þ b
ri 6 j ð1Þ
ð i jÞ

B04206 1 of 7
B04206 ITO AND TORIUMI: SELF-DIFFUSION IN PERICLASE (MgO)
Figure 1. Thermal expansions of molar volumes under
ordinary pressure (0 GPa).
where Uij is the interatomic potential, rij is the distance
between ith and jth atoms, and q, a, b, c are parameters. The
parameters were those of Akamatsu and Kawamura [1999].
The Buckingham model is simple. It requires little
calculation and successfully reproduces the equation of
state (EOS) of periclase [Akamatsu and Kawamura, 1999].
Because details of the reproduction are not given by
Akamatsu and Kawamura [1999], we reexamined the EOS.
Three different sets of EOS are compared with those of
experiments in Figures 1, 2, and 3: (1) thermal expansions
under ordinary pressure (0 GPa), (2) compression under
ordinary temperature (300 K), and (3) compression under
high temperature (2000 K) of molar volumes, respectively.
In the first two cases the molar volumes (Figures 1 and 2)
are in excellent agreement with those of the experiments.
The last case the molar volumes (Figure 3) are a little
smaller than those of experiments, though the difference is
small (
1% at 50 GPa). Therefore we consider that the
reproduction of the EOS of the model is good and self-
diffusion coefficients calculated by the model are reliable
enough to provide base data for geophysics. It is known that
ab initio electronic state calculations (reviewed by Karki et
al. [2001]) successfully reproduce well the elasticity at high
pressures, but they require much more calculation. As
mentioned here below, the calculation of self-diffusion
coefficients needs an ensemble of many (107) steps, so it is
considered that the electronic state calculations are not
appropriate for this study.
[5] The MD system prepared was composed of 5  5 
5 unit cells of periclase (rock salt structure), which contain
4 Mg atoms and 4 O atoms per unit cell, with a pair
vacancies (998 atoms) and three-dimensional periodic
boundary conditions. The potential, which an atom feels,
was given by infinite summation of potentials between the
atom and the other atoms. The Coulomb term (first term of
the equation of Buckingham model) was summed by the
Ewald method [Ewald, 1921]. Summations of short-range
(second and third) terms were cut off at 7.5 Å. Newton’s
equations of motions were integrated by the Verlet Algo-
rithm with 2 fs intervals. Isothermal and isobaric (NPT)
ensembles were realized by the scaling method [Woodcock,
1971]. Thermodynamically rigorous methods such as the
Nose-Hoover method [Nose, 1984] or the Andersen method
[Andersen, 1980] are more preferable. However, when the
2 of 7
B04206
Nose-Hoover method or Andersen method was used, the
equations of motions of heat baths (Nose-Hoover method)
or boundaries of basic cells (Anderson method) sometimes
diverged when atoms migrate into vacancies. We consider
that the divergence is due to the rapid increase of system
energy when atoms reach saddle points of their migration
path. In order to avoid the divergence, we selected the
scaling method. The scaling method is more stable because
it does not solve any additional equations of motion. The
MD program MXDTRICL (K. Kawamura, 1996, available
at http://www.sccj.net/download/JCPE/) was used for cal-
culations.
[6] The self-diffusion coefficient (D) was derived by
Einstein equation of which the formation is
 
D ¼ ðd=dtÞ < jRn j2 > =6 ð2Þ
where Rn indicates the positional vector of nth atom, while
the angle brackets indicate the average of all atoms. The
positional vectors were given by the trajectories of atoms
recorded during MD calculations. Initial relaxation was
performed for 2 ns (106 steps) to achieve thermal
equilibrium. 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. To study
the precision of the averaged result of D, we calculated
standard deviations of the 10 self-diffusion coefficients (D)
at the lowest temperature conditions (2400 K, 3400 K) and
several (0 GPa, 50 GPa, 100 GPa, and 140 GPa) pressure
conditions. We confirmed that the differences are smaller
than 25% of the averages. We also confirmed that the ratios
between the standard deviations and the averages mono-
tonously decreased with increasing temperatures. Therefore
the standard deviations of the 10 self-diffusion coefficients
are considered to be always smaller than 25%, which is
reliable enough.
[7] We also calculated the formation enthalpy (Hf*) of the
vacancy pair. In this study, the Hf* were derived from
differences of total enthalpies between perfect and vacancy
containing MD systems. Initial relaxation was performed
for 0.2 ns (105 steps). After the initial relaxation, MD
calculation were performed for 0.2 ns (105 steps) recording
total enthalpies of each step and the average was adopted as
Figure 2. Compressions of molar volumes under ordinary
temperature (300 K).
B04206 ITO AND TORIUMI: SELF-DIFFUSION IN PERICLASE (MgO)
Figure 3. Compressions of molar volumes under high
temperature (2000 K).
result. We are able to use larger system than the calculations
of the self-diffusion coefficient, because the calculation for
Hf* needs shorter ensemble time (0.2 ns) than that for the
self-diffusion coefficient (2 ns). The size of the MD systems
is 10  10  10 unit cells containing 8000 and 7998 atoms
for the perfect system and for the one with a pair of
vacancies, respectively.
[8] In ionic crystals such as MgO, the dominant mecha-
nism of the self-diffusion is vacancy hopping. The vacancy
hopping is classified in two types; extrinsic and intrinsic.
The ‘‘extrinsic’’ indicates that vacancies are formed by
external sources and the ‘‘intrinsic’’ means that vacancies
are thermally formed. Typical external sources are impurity
atoms, which have different charges than host atoms. For
keeping charge neutrality, in MgO two trivalent impurity
atoms produce one Mg vacancy or two monovalent impu-
rity atoms produce one O vacancy. If there is no external
sink or source of the impurities, the extrinsic vacancy
concentrations do not depend on temperature or pressure.
In this case, the self-diffusion coefficient is proportional to
the self-diffusion coefficient of the MD calculations because
in the both cases the vacancy concentrations are constant.
The proportional relation can be written as D = (C/C0)*DC0
where D and C are the self-diffusion coefficient and
vacancy concentration of the extrinsic system and DC0
and C0 is that of MD calculations, respectively. In addition,
this proportional relation postulates that the activation
energies of the extrinsic system and the MD calculations
can be considered to be equal. This relation is expressed as
D ¼ ðC=C0 Þ*DC0;0 expð
Hm*=RT Þ ð3Þ
where DC0,0 is the prefactor, Hm* are the activation energies
of the MD self-diffusion coefficients. The Hm* are often called
the migration energies. When intrinsic, the O and Mg vacancy
concentrations (Cint,O, Cint,Mg) depend on Hf* as Cint,O Cint, Mg
/ exp(
Hf*/RT ). The vacancy concentration of O and Mg is
considered to be the same because of charge neutrality.
Therefore this relation can be written as

 
Cint;0 ¼ Cint;Mg / exp
H*f =2 =RT ð4Þ
Equation (3) can be interpreted as (1) the diffusion coefficients
D depend exponentially on
Hm* and (2) the D also depend
3 of 7
B04206
proportionally on the vacancy concentrations C. Intrinsic
vacancy concentrations are given by equation (4). Therefore, by
replacing Cint,O (or Cint,Mg) for C in equation (3), the following
relation is derived:

 
D / exp
Hm* þ H*f =2 =RT ð5Þ
Equation (5) means that the activation energies of the intrinsic
regime are interpreted as Hm* + Hf*/2. Up to now, the
introduction of the equation (3) or (5) is the same as in the
work by Ando [1989].
[9] In this simulation, the natural system is approximated
by three-dimensional infinite repetitions of the small sys-
tems (5  5  5 unit cells with a pair vacancies (998 atoms)
for self-diffusion or 10  10  10 unit cells with a pair
vacancies or without any vacancy (7998 or 8000 atoms)).
The vacancy concentrations of the small systems (1/500 for
5  5  5 unit cells or 1/4000 for 10  10  10 unit cells)
are much higher than that of natural systems. This high
vacancy concentrations may affect the results of self-
diffusion coefficients D or vacancy formation energy Hf*
because of interactions between vacancies. When there is no
size effect, the self-diffusion coefficient D is inversely pro-
portional to the vacancy concentration as of equation (3).
In order to examine whether D is inversely proportional
with the vacancy concentration or not, we calculated D at
the small size (5  5  5 unit cells with a pair vacancies;
998 atoms) and the large size (8  8  8 unit cells with a
pair vacancies; 4094 atoms) at several P-T conditions and
compared D normalized by vacancy concentrations. The
ratio of the normalized D998/D4094 (where D998 indicates the
usual size (998 atoms) and D4094 indicates the large size
(4094 atoms)) is distributed from 77% to 144%. The wide
distribution seems to come from the error of D (25%) and it
suggests that the size effect of the D is below the margin.
We also studied the size effect for the vacancy forma-
tion energy Hf* by comparing Hf* of the normal system
(10  10  10 unit cells with a vacancy pair, 7998 atoms)
with that of the large system (16  16  16 unit cells with a
vacancy pair, 32,766 atoms). Note that the Hf* does not
depend on the system size because we adopt kJ/mol as the
unit. The observed Hf* of the large system is 2% smaller
than that of the normal system at 300 K and 0 GPa, which
Figure 4. Trajectories of atoms observed by MD simula-
tions under 2500 K and 0 GPa.
B04206 ITO AND TORIUMI: SELF-DIFFUSION IN PERICLASE (MgO)
Figure 5. Self-diffusion coefficients (D) under 2400 –
3000 K temperatures at ordinary pressure (0 GPa).
seems to be small. Finally, we examined the effect of the
integration interval by comparing the self-diffusion coeffi-
cient D for the normal integration interval (2 fs) with that
for the short integration interval (1 fs). The ratio of the D2fs/
D1fs (where D2fs and D1fs indicate normal integration
interval (2 fs) and shorter integration interval (1 fs), respec-
tively) is distributed from 86% to 125%, which seems to be
within the margin of error (25%).
3. Results
[10] First, we performed MD calculations under various
(2000 – 3400 K) temperatures at ordinary pressure (0 GPa)
with 100 K intervals. Figure 4 shows trajectories of atoms
of MD calculation at 2500 K. In Figure 4, an O atom and an
Mg atom are migrating to O and Mg vacancies, respectively.
Such migrations were observed at 2400 –3000 K. Above
3100 K, the MD systems melted. The melting above 3100 K
is consistent with the melting point of periclase (3100 K).
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 coef-
ficients were calculated for 2400– 3000 K.
[11] Figure 5 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. We compare the derived
activation enthalpies to migration energies (Hm*) because the
vacancy concentration does not vary during the simulation.
In the MD simulation, vacancies are not created or
4 of 7
B04206
destroyed because the system does not have any external
sources of vacancies. There are also no dislocations or grain
boundaries in the system because the system consists of 5 
5  5 MgO unit cells with a pair of vacancies and their
infinite three-dimensional repetitions.
[12] Table 1 compares Hm* with the activation enthalpies
of three experimental [Sempolinsky and Kingery, 1980;
Ando et al., 1983; Yang and Flynn, 1994] and one theoret-
ical study [Ita and Cohen, 1997]. Table 1 also compares
formation enthalpies (Hf*) given by the total energy differ-
ences of perfect and vacancy containing crystals. In Table 1,
we interpret the activation enthalpies derived from experi-
ments of Mg and O self-diffusions as Hm* and Hm* + Hf*/2,
respectively, because the Mg and O self-diffusions are
thought to be extrinsic and intrinsic, respectively. We will
discuss the interpretation in the next section. 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 inter-
atomic potentials because the two studies gave formation
enthalpies by the same methods (differences of total ener-
gies between perfect and vacancy containing systems).
[13] We observed self-diffusion coefficients under various
temperatures (2400 – 6000 K) and pressures (0 –140 GPa).
Like under ordinary pressure (0 GPa), vacancies did not
migrate when temperatures were too low and the systems
melted when temperatures were too high. At each pressure,
self-diffusion coefficients were fitted to the Arrhenius
equations. Table 2 shows the fitted results of preexponential
factors (D0) and Hm* over the temperature range at each
pressure. We calculated squared multiple correlation coef-
ficients (s2) to estimate how well they fit. The correlation
coefficients s of the two arrays x = {x1, x2, . . . , xn} and y =
{y , y2, . . . , yn} are given by s = (nSxy
SxSy)/
r1ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
h P
P ih P P i
n x2
ð xÞ2 n y2
ð yÞ2 . In this case, x
and y correspond to temperature and ln(D), respectively.
Results of s2 are always close to 1 (over 0.99), indicating
that the self-diffusion coefficients at high pressures were
also linear in the Arrhenius plot. Pressure dependences
of self-diffusion coefficients under several temperatures
(3000 K, 3400 K, 4400 K, and 5400 K) were drawn at
Figure 6. We found that the 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 the
previous study [Ita and Cohen, 1997]. Their self-diffusion
coefficients did not increase with increasing pressure. In
order to examine the difference of these two results, we
compared two components (preexponential factors D0 and
migration energies Hm*) of self-diffusion coefficients in
Table 1. Comparison of H *m and H *m + Hf*/2
H *m, kJ/mol H *m + H f*/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
B04206 ITO AND TORIUMI: SELF-DIFFUSION IN PERICLASE (MgO)
Figure 6. Pressure variations (0 – 140 GPa) of self-
diffusion coefficients (D) under several (3000, 3400,
4400, 5400 K) temperatures.
Figures 7 and 8, respectively. We also drew pressure
derivatives of Hm* (that is, activation volumes of migrations
(Vm*), Vm* = dHm*/dP) in Figure 9 because Vm* can be com-
pared with experimental data [Van Orman et al., 2003]).
Although Ita and Cohen’s [1997] D0 is constant due to
the constant attempt frequency which they assumed, the
observed D0 decreases considerably with increasing pres-
sure (1.5 times from upper (25 GPa) to lower (135 GPa)
mantle pressures). This observation indicates that the
dependence of attempt frequency should not be ignored
during calculations of self-diffusion coefficients. Regarding
Hm* and Vm*, our Vm* went from negative to positive values at
60 GPa, indicating a 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 to
the difference between the self-diffusion coefficients shown
by Figure 6. At 20 GPa, Vm* of our study and of Ita and
Cohen [1997] were indistinguishable and both Vm* were
consistent with Vm* of Van Orman et al. [2003].
[14] Figure 10 compares the formation enthalpies (H*f )
with that of Ita and Cohen [1997]. Both Hf* increased with
pressure, but our increase rate was larger than that of Ita and
Cohen [1997]. The cause of the difference is considered to
come from the interatomic potentials because both H*f are
Figure 7. Comparisons of preexponential terms (D0) of
Mg.
5 of 7
B04206
Figure 8. Comparisons of migration enthalpies (Hm*) of
Mg.
given by the total enthalpy difference of perfect and defect
containing systems.
4. Discussion
[15] Our calculations predict that pressure dependencies
of self-diffusion coefficients (Figure 6) turn from negative
to positive with increasing pressure. This behavior may
seem strange because it is not consistent with previous
defect analyses [Karato, 1981] or theoretical calculations
[Ita and Cohen, 1997]. However, there are several studies
that are consistent with this behavior. For example,
Gregoryanz et al. [2005] report a negative pressure depen-
dence of the sodium melting point at high pressure. A
negative pressure dependence of melting points would
imply a positive pressure dependence of the self-diffusion
coefficient because self-diffusion coefficients D and the
melting point Tm are related through the homogeneous
relation D = D0 exp (
g Tm/T ), where D0 is preexponential
factor, g is a coefficient, and T is temperature. Although data
of MgO melting experiments are available up to 35 GPa
[Zerr and Boehler, 1994], MD calculations of MgO melt-
ing are performed by employing the two-phase method
[Belonoshko and Dubrovinsky, 1996] under whole lower
mantle pressures (up to 140 GPa). Increasing of their
melting points dramatically slows with increasing pressure
(increment is only 300 K between 60 and 140 GPa, though
3000 K between 0 and 60GP), which is a trend similar to
that of small positive activation volumes of self-diffusion
Figure 9. Comparison of activation volumes of migration
(Vm*) of Mg.
B04206 ITO AND TORIUMI: SELF-DIFFUSION IN PERICLASE (MgO) B04206

Table 2. Fitted Results of Self-Diffusion Coefficients to Arrhenius Equation D = D0exp(
H */RT )a

D0, m2/s H *m , kJ/mol s2
Pressure, GPa Temperature 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 0.9930
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 0.9990
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 0.9990 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
a
D is self-diffusion coefficients, D0 is the preexponential factor, R is the gas constant, T is temperature; s2 is squared multiple correlation coefficient of
the least squares fitting. Temperature range is also shown at each pressure. Intervals of temperatures are 100 K (0 – 10 GPa) or 200 K (20 – 140 GPa). Read
9.05E-08 as 9.05  10
8.

coefficients reported in this manuscript. In addition, Karato
[1978] reports local minima of defect formation energies
with increasing pressures, supporting the observed pressure
dependence of self-diffusion coefficients.
[16] We discuss a mechanism of MgO self-diffusion. In
ordinary pressure, it is considered that the Mg diffusion is
extrinsic, because Mg vacancies are formed by trivalent
impurities, while O diffusion is intrinsic [Ando, 1989]. It is
known that the Mg self-diffusion coefficient increases with
increasing trivalent impurity concentrations [Sempolinsky
and Kingery, 1980] and that the O self-diffusion coefficient
does not depend on the trivalent impurity concentration
[Ando et al., 1983]. At high pressure, the mechanism of
MgO self-diffusion has hardly been known. Van Orman
et al. [2003] suggested that Mg diffusion is extrinsic at 15–
25 GPa because the self-diffusion coefficients of Mg depend
on Al concentrations. It is considered that Mg diffusion in
lower mantle periclase is extrinsic because the pyrolite
model [Ringwood, 1975] indicates that the lower mantle
has a certain (
5%) Al content. O self-diffusion coefficients
at 15– 25 GPa [Van Orman et al., 2003] are larger than that
at ordinary pressure [Ando et al., 1983; Yang and Flynn,
1994]. Van Orman et al. [2003] indicated that the O self-
diffusion is extrinsic where vacancies are created by dis-
locations generated during compression. Ita and Cohen
[1998] speculated that O self-diffusion is extrinsic because
O vacancies are introduced through monovalent impurities.
We considered that O diffusion of lower mantle periclase is
not intrinsic but extrinsic, because Hf* rapidly increases
with increasing pressure (Figure 10) [Ita and Cohen, 1997].
Equation (4) says that the intrinsic vacancy concentration
exponentially decreases with increasing Hf*. Monovalent
impurities may be possible as a source of O vacancies
because the pyrolite model [Ringwood, 1975] postulates
some (
1%) sodium.
[17] The lack of seismic anisotropy indicates that the
lower mantle flows by diffusion creep [Karato et al.,
1995]. In a material flowing by diffusion creep, viscosity
and self-diffusion coefficients are related according to the
Nabarro-Herring model


h ¼ A*ð RT=V Þ* G2 =Deff ð6Þ
where h is viscosity, A is a constant, R is the gas constant, T
is temperature, V is the molar volume of the material, G is
the grain size, and Deff is the effective self-diffusion
coefficient, which is that of the most slowly diffusing
species in the mineral. Here we try to estimate the viscosity

Figure 10. Comparison of vacancy formation energies Figure 11. Relative viscosity (h/h0) of the lower mantle
(Hf*). periclase stimated from Nabarro-Herring equation.

6 of 7
B04206 ITO AND TORIUMI: SELF-DIFFUSION IN PERICLASE (MgO)
of the lower mantle periclase by employing the Nabarro-
Herring model. We assume that (1) the most slowly
diffusing species in periclase is O, (2) the O diffusion of
lower mantle periclase is extrinsic, and (3) the vacancy
concentration is always uniform. At ordinary pressure, the
Mg diffusion is 2– 3 times faster than the O diffusion [Ando,
1989]. If the sources of the O vacancies are monovalent
cations, the uniform assumption may be appropriate when
neglecting partitioning of the monovalent cations between
periclase and other lower mantle minerals such as
perovskite. We can apply self-diffusion coefficients of the
MD calculations for the Deff of equation (6) by introducing
these three assumptions. This gives us three unknown
parameters (A, G in equation (6) and C in equation (3)). The
coefficient A in the Nabarro-Herring equation is difficult to
estimate, so we eliminate A by using a proportional relation
and the variable C is eliminated by introducing a
proportional relation. We already assume that the vacancy
concentration is uniform. Assuming that the grain size G is
constant in the lower mantle, we can derive the relationship
between viscosities of lower mantle periclase h and self-
diffusion coefficients derived by the MD simulations D as
h / ðV =RT Þ*ð1=DÞ
where T is lower mantle geotherm, V is molar volume. The
grain growth rate of periclase is very small under high pres-
sure [Yamazaki et al., 1996], indicating uniform grain size
of periclase in the lower mantle. The temperature at top of
the lower mantle is 1900 K [Ito and Katsura, 1989]. We con-
sider adiabatic (0.3 K/km) and superadiabatic (0.6 K/km)
temperature gradients. We derive the relative viscosity of
lower mantle periclase by using equation (7). Figure 11
shows the relative viscosity. We found that the viscosity of
lower mantle periclase decreases with increasing depth. The
rate of the softening is 1.3 times (0.3 K/km) or 2.0 times
(0.6 K/km). The softening of the lower mantle periclase
comes from the self-diffusion coefficients decreasing at low
pressure, though the decrease slows down and turns into an
increase with increasing pressure (Figure 6). Yamazaki and
Karato [2001] suggested that the viscosity of the lower
mantle is near that of the weaker phase, periclase, in the
place most highly strained. The softening may be one of the
causes of dynamic geological activities of the lower mantle,
such as plume generation.
[18] Acknowledgments. We thank K. Kawamura and T. Tsuchiya for
their helpful discussion. K. Kawamura kindly offered the MD program
MXDTRICL. Some computations were performed by Earth Simulator, at
Earth Simulator Center, Japan Agency for Marine-Earth Science and
Technology (JAMSTEC).
References
Akamatsu, T., and K. Kawamura (1999), Molecular dynamics of solid
solution and coexisting liquid, Mol. Simul., 21, 387 – 399.
Andersen, H. C. (1980), Molecular dynamics simulations at constant pres-
sure and/or temperature, J. Chem. Phys., 72, 2384 – 2393.
Ando, K. (1989), Self-diffusion in oxides, in Rheology of Solids and of the
Earth, pp. 57 – 82, Oxford Univ. Press, New York.
B04206
Ando, K., Y. Kurokawa, and Y. Oishi (1983), Oxygen self-diffusion in Fe-
doped MgO single crystals, J. Chem. Phys., 78, 6890 – 6892.
Belonoshko, A., and L. Dubrovinsky (1996), Molecular dynamics of NaCl
(B1 and B2) and MgO (B1) melting: Two-phase simulation, Am. Mineral.,
81, 303 – 316.
Dewaele, A., G. Fiquet, D. Andrault, and D. Hausermann (2000), P-V-T
equation of state of periclase from synchrotron radiation measurements,
J. Geophys. Res., 105, 2869 – 2877.
Dubrovinsky, L., and S. Saxena (1997), Thermal expansion of periclase
(MgO) and tungsten (W) to melting temperatures, Phys. Chem. Mineral.,
24, 547 – 550.
Ewald, P. P. (1921), Die Berechnung Optischer und Electrostatischer Git-
terpotentiale, Ann. Phys., 64, 253 – 287.
Fei, Y. (1999), Effects of temperature and composition on the bulk modulus
of (Mg, Fe)O, Am. Mineral., 84, 272 – 276.
Gregoryanz, E., O. Degtyareva, M. Somayaulu, R. J. Hemley, and H. K.
Mao (1997), Melting of dense sodium, Phys. Rev. Lett., 94, 185502.
Ita, J., and R. E. Cohen (1997), Effects of pressure on diffusion and vacancy
formation in MgO from non-empirical free-energy integrations, Phys.
Rev. Lett., 79, 3198 – 3201.
Ita, J., and R. E. Cohen (1998), Diffusion in MgO at high pressure: Im-
plications for lower mantle rheology, Geophys. Res. Lett., 25, 1095 –
1098.
Ito, E., and T. Katsura (1989), A temperature profile of the mantle transition
zone, Geophys. Res. Lett., 16, 425 – 428.
Karato, S. (1978), The concentration minimum of point defects in solid at
high-pressures and the plasticity of the lower mantle (in Japanese), Seismol.
Soc. Jpn. Programme Abstr., 1, D31.
Karato, S. (1981), Pressure dependence of diffusion in ionic solids, Phys.
Earth Planet. Inter., 25, 38 – 51.
ð7Þ Karato, S., S. Zhang, and H. Wenk (1995), Superplasticity in Earth’s lower
mantle: Evidence from seismic anisotropy and rock physics, Science,
270, 458 – 461.
Karki, B., and G. Khanduja (2006), Vacancy defects in MgO at high pres-
sure, Am. Mineral., 91, 511 – 516.
Karki, B., L. Stixrude, and R. Wentzcovitch (2001), High-pressure elastic
properties of major materials of Earth’s mantle from first principles, Rev.
Geophys., 39, 507 – 534.
Mao, H., and P. Bell (1979), Equations of state of MgO and epsilon-Fe
under static pressure conditions, J. Geophys. Res., 84, 4533 – 4536.
Nose, S. (1984), A unified formulation of the constant temperature mole-
cular-dynamics methods, J. Chem. Phys., 81, 511 – 519.
Ringwood, A. E. (1975), Composition and Petrology of the Earth’s Mantle,
McGraw-Hill, New York.
Sempolinsky, D. R., and W. D. Kingery (1980), Ionic conductivity and
magnesium vacancy mobility in magnesium oxide, J. Am. Ceram. Soc.,
63, 664 – 669.
Van Orman, J. A., Y. Fei, H. Hauri, and J. Wang (2003), Diffusion in MgO
at high pressures: Constraints on deformation mechanisms and chemical
transport at the core-mantle boundary, Geophys. Res. Lett., 30(2), 1056,
doi:10.1029/2002GL016343.
Woodcock, L. V. (1971), Isothermal molecular dynamics calculations for
liquid salts, Chem. Phys. Lett., 10, 257 – 261.
Yamazaki, D., and S. Karato (2001), Some mineral physics constraints on
the rheology and geothermal structure of Earth’s lower mantle, Am.
Mineral., 86(4), 385 – 391.
Yamazaki, D., T. Kato, E. Ohtani, and M. Toriumi (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, Phys. Rev. Lett., 73, 1809 – 1812.
Zerr, A., and R. Boehler (1994), Constraints on the melting temperature of
the lower mantle from high-pressure, experiments on MgO and magne-
siowustite, Nature, 371, 506 – 508.























Y. Ito, Department of Earth and Planetary Science, Graduate School of
Science, University of Tokyo, 7-3-1 Hongo, Bunkyo-ku, Tokyo 113-0033,
Japan. (ito@gaea.k.u-tokyo.ac.jp)
M. Toriumi, Department of Complexity Science and Engineering, Graduate
School of Frontier Science, University of Tokyo, 5-1-5 Kashiwanoha,
Kashiwa, Chiba, Tokyo 277-8581, Japan. (tori@k.u-tokyo.ac.jp)
7 of 7