Académique Documents
Professionnel Documents
Culture Documents
Institute for Space and Nuclear Power Studies, University of New Mexico, Albuquerque, NM, USA
Nuclear Engineering Department, University of New Mexico, Albuquerque, NM, USA
c
Mechanical Engineering Department, University of New Mexico, Albuquerque, NM, USA
d
Chemical and Biological Engineering Department, University of New Mexico, Albuquerque, NM, USA
b
a r t i c l e
i n f o
Article history:
Received 29 May 2015
Received in revised form 10 September
2015
Accepted 16 September 2015
Available online 18 October 2015
Keywords:
Bond-order
Variable-charge reactive force fields
MD simulation of silica polymorphs
Alphabeta transition
Accuracy and transferability
a b s t r a c t
This paper investigates the applicability of the bond-order, variable-charge (BOVC) force fields of the
H2 O
, and ReaxFFGSI
Charge-Optimized Many-Body (COMB10), ReaxFFSiO
SiO , for molecular dynamics (MD) simulations of crystalline SiO2. The calculated lattice constants and densities of the four SiO2 polymorphs,
quartz, cristobalite, coesite, and stishovite, are compared to experimental values. Additionally, the calculated pair distribution and bond-angle distribution functions and the ab transition for quartz, the most
stable low-energy polymorph, are compared to experimental results. The simulations with the COMB10
force field accurately predict the properties of the SiO2 polymorphs, except the a-cristobalite, and the
GSI
2O
quartz ab transition. The results with ReaxFFH
SiO and ReaxFFSiO accurately predict the properties of
2O
the SiO2 polymorphs, except the stishovite, but those with ReaxFFH
SiO inaccurately predict the quartz
ab transition.
2015 Elsevier B.V. All rights reserved.
1. Introduction
Characterizing the interface of silica (SiO2) with other materials,
such as silicon (Si), water (H2O), and oxygen (O), is important in a
wide range of applications. Examples are the interaction of SiO2
with biomolecules, such as H2O, in chromatography, geochemistry,
biomedical devices, and the encapsulation of biomolecules within
silica gels [13]. The Si/SiO2 interface is encountered in the manufacturing of semiconductor transistors [4,5]. The O/SiO2 interface is
encountered in the dry-etching and degradation of silica surfaces,
commonly used in the aerospace industry, due to the collision by
atomic oxygen [6,7]. The challenging characterization of these
interfaces experimentally has stimulated a growing interest in
using molecular dynamics (MD) simulations for that purpose.
The MD simulations employ force fields for modeling atom
atom interactions in the materials of interest and the force fields
are fitted to high-level ab initio or experimental results at specified
temperatures and pressures [6,7]. Some force fields are rather simple, such as the Lennard-Jones pair potential [8], originally developed to model noble gases, while others are parameterized for
modeling complex systems and interfaces. Examples are the twobody force fields with Coulomb interactions (e.g. BKS [6], Pedone
Corresponding author at: Institute for Space and Nuclear Power Studies,
University of New Mexico, Albuquerque, NM, USA.
http://dx.doi.org/10.1016/j.commatsci.2015.09.042
0927-0256/ 2015 Elsevier B.V. All rights reserved.
[7], TTAM [8], CHIK [9]), the three-body force fields with no Coulomb interactions (e.g. Munetoh [10]), and the bond-order,
variable-charge (BOVC) force fields [1113]. The BOVC force fields
are the most suitable for modeling complex chemical reactions and
interfaces.
Two of the most popular BOVC force fields are the ChargeOptimized Many-Body (COMB) and ReaxFF. Since the introduction
of the original COMB for simulating the Si/SiO2 interface [14], and
the original ReaxFF for simulating hydrocarbons [15], several
materials have been parameterized in forms of these powerful
force fields. For instance, the capabilities with COMB have been
successfully extended for modeling the interfaces of hafnium/
hafnium oxide [16] and Cu/SiO2 [17], as well as zirconium [18].
Similarly, the capabilities of ReaxFF have been extended for modeling the aluminummolybdenum alloys [19], zinc oxide [20], and
the silicon/silica system [21].
The COMB10 [11] force field is an extension of COMB07 [14],
developed as an extension of the Tersoff-based potential by Yasukuawa [22] for semiconductors. The COMB10 force field, parameterized for modeling the Si/SiO2 interface and amorphous silica,
accurately predicts the structural and elastic properties of crystalline SiO2 and the trend of defect formation in amorphous silica
[11].
The ReaxFF is the foundation of two other parameterizations
H2 O
involving silica. The first is that of Fogarty et al. [12] (ReaxFFSiO
),
270
B.J. Cowen, M.S. El-Genk / Computational Materials Science 111 (2016) 269276
ReaxFFGSI
SiO
(COMB10 [11],
[12],
[13]) for modeling crystalline silica, developed specifically for characterizing complex
interfaces and reactions. Investigated are the accuracy and transferability of these force fields for calculating the material properties, including the lattice constants, density, and the pair and
bond-angle distribution functions. The transferability is examined
by comparing the simulation results to the experimental values
for the various silica polymorphs; namely, quartz, cristobalite, coeH2 O
site, and stishovite. The COMB10 [11] and ReaxFFSiO
[12] are also
investigated for predicting the ab phase transition for quartz.
The MD simulations with the BOVC force fields require, as an
input, the frequency of updating the atom charges. Generally, it
is beneficial to update the atom charges as frequently as possible,
without sacrificing either accuracy or computational efficiency.
Updating the atom charges every time step, although computationally expensive, ensures that the charges are at their energy minima. Updating less frequently decreases the computational time,
The first four terms in Eq. (2) are basically the same for the
COMB10
[11]
force
field,
Eq.
(1).
Other
terms:
Eangle ; Etorsion ; Eover ; Eunder ; Econjugation
implicitly in the Ebond term of the COMB10 force field (1). The
Eself term represents the energy difference in the various states of
the atom charges, while Eothers consists of several terms that correct
for allene-type terminal triple bond, and C2 species [32]. For more
details on the physics of the various terms in Eqs. (1) and (2), the
reader is referred to the original papers on the BOVC force fields
[1113].
2.2. Quartz ab transition
In the performed MD simulations of the ab transition for
H2 O
H2 O
quartz using COMB10-1, ReaxFFSiO
-1, and ReaxFFSiO
-10, the temperature in the ranges of 300700 K and 9001500 K is increased
in increments of 100 K. In the intermediate range of 700900 K,
the temperature increased in small increments of 10 K. The
271
B.J. Cowen, M.S. El-Genk / Computational Materials Science 111 (2016) 269276
3.1. Quartz
Quartz, the most abundant and stable SiO2 polymorph at ambient conditions, is used in many applications due to its superior
optical and piezoelectric properties and the formation of SiO2 at
the SiO2/Si interface [3335]. Table 1 compares the calculated
properties (standard deviation shown in parenthesis) and the atom
charges for a-quartz, using MD simulations with the BOVC force
fields to experimental values. The simulations results with the
ReaxFFGSI
SiO -1 accurately predict the lattice constants and density of
H2 O
vite, using MD simulations with COMB10, ReaxFFSiO
; and
ReaxFFGSI
SiO , with experimental results. These material properties
are the lattice constants, density, and the pair and bond-angle
H2 O
material properties in the simulations with the ReaxFFSiO
-1 are
within 3.6%, and those with COMB10 are within 7.3%, of the
Table 1
Comparison of calculated properties (standard deviation shown in parenthesis) and atom charges for a-quartz, using MD simulations with the BOVC force fields to experimental
values.
Parameter
a ()
c ()
q (g/cm3)
qavg (O)
qavg (Si)
qrange (O)
qrange (Si)
Experimental/force field
Exp.[36]
COMB10-1
COMB10-10
H2 O
-1
ReaxFFSiO
H2 O
ReaxFFSiO
-10
ReaxFFGSI
SiO -1
4.916
5.405
2.646
4.79 (4e3)
5.30 (6e3)
2.84 (3e3)
1.45 (1e4)
2.91 (3e4)
1.38 to 1.54
2.882.93
4.79 (3e3)
5.28 (5e3)
2.85 (2e3)
1.45 (5e5)
2.91 (1e4)
1.39 to 1.53
2.892.92
4.95 (3e3)
5.52 (3e3)
2.55 (1e3)
0.81 (2e3)
1.62 (4e3)
0.69 to 0.91
1.401.80
4.93 (1e2)
5.48 (3e2)
2.60 (3e3)
0.81 (1e3)
1.62 (4e3)
0.69 to 0.91
1.581.64
4.95 (3e3)
5.46 (3e3)
2.59 (1e3)
0.46 (9e4)
0.92 (2e3)
0.41 to 0.50
0.880.96
H2 O
Fig. 1. Comparison of calculated pair distribution functions for a-quartz, at ambient conditions (300 K and 1 atm), in MD simulations with COMB10 and ReaxFFSiO
to
experimental results.
272
B.J. Cowen, M.S. El-Genk / Computational Materials Science 111 (2016) 269276
H2 O
H2 O
COMB10-1, COMB10-10, ReaxFFSiO
-1, and ReaxFFSiO
-10 force
fields, with experimental values. The pair distribution functions
are important for showing the structural order of crystalline materials. There are differences in the calculated distributions with
updating the atom charges every time step and every 10-time
steps. The distribution functions curves, produced by evaluating
the atom charges every time step are smoother and more accurate,
compared to experimental values. The most accurate curves for the
OSi and OO pairs are those calculated in the simulations with the
COMB10-1 force field. They are accurate to within 1% of the
H2 O
the calculated curves with ReaxFFSiO
-10, which disappear when
updating the atom charges every time step. The MD simulations
with the COMB10 force field underestimate the experimental
results of the SiOSi BAD by 4%, while the simulations with
H2 O
ReaxFFSiO
over-predict the experimental results by as much as
10% (Fig. 2). For judicious comparison to experimental results,
Table 2 lists the ranges of the calculated bond-angles at the base
and half-width, the average values of the bond-angles, and the
bond-angles at the peaks of the BAD functions. The results in this
table show close agreements with the experimental values of the
average bond angles.
The equilibration frequency of the atom charges clearly affects
the simulations results, as demonstrated in Figs. 1 and 2. In
addition to not conserving energy, when updating the atom
charges each 10 time steps, the peaks of the calculated pair and
H2 O
experimental results. The simulation results with ReaxFFSiO
-1 for
the SiSi pairs are the most accurate; they are within 1% of experimental values. The MD simulations with the COMB10 force field
underestimate the SiSi distribution by 3%, and that with the
H2 O
ReaxFFSiO
underestimate both the OSi and OO pairs by 3%,
compared to the experimental values of the pair distribution functions (Fig. 1). For an accurate characterization, Table 2 lists the
ranges of the calculated bond-lengths at the base and half-width,
the average values of the bond-lengths, and the bond-lengths at
the peaks of the pair distribution functions.
Fig. 2 compares the calculated OSiO and SiOSi bond-angle
distribution (BAD) functions for a-quartz at ambient conditions,
using MD simulations with the COMB10-1, COMB10-10,
H2 O
H2 O
ReaxFFSiO
-1, and ReaxFFSiO
-10 to the experimental values. The
results with all BOVC force fields predict the OSiO BAD to within
1% of the experimental values. However, there are slight kinks in
Table 2
Comparison of the bond-lengths at the 1st peak of the pair distribution function, and the bond-angles for a-quartz, using MD simulations with the BOVC force fields, to
experimental values.
Force field
Atom Pairs
MinMax @ Half-Width
@ Peak
Average value
Calculated
Exp. [36]
COMB10-1
OSi
OO
SiSi
1.461.86
2.292.87
2.763.41
1.571.64
2.542.72
2.953.05
1.61
2.63
2.98
1.62
2.63
3.01
1.61
2.612.64
3.06
COMB10-10
OSi
OO
SiSi
1.511.80
2.332.82
2.853.38
1.591.64
2.562.70
2.953.02
1.61
2.61
2.99
1.62
2.62
3.02
1.61
2.612.64
3.06
H2 O
ReaxFFSiO
-1
OSi
OO
SiSi
1.461.78
2.192.94
2.943.31
1.531.61
2.472.66
3.043.13
1.57
2.55
3.08
1.57
2.56
3.09
1.61
2.612.64
3.06
H2 O
ReaxFFSiO
-10
OSi
OO
SiSi
1.521.65
2.162.90
3.023.23
1.561.59
2.442.67
3.063.10
1.57
2.52
3.08
1.57
2.57
3.09
1.61
2.612.64
3.06
Interface
Bond angles ()
COMB10-1
OSiO
SiOSi
89.9130.5
99.2153.5
103.6115.2
133.6141.5
108.9
136.7
109.3
137.7
109.5
143.7
COMB10-10
OSiO
SiOSi
94.4128.5
128.6155.3
104.4115.1
134.1140.1
108.7
137.2
109.3
138.0
109.5
143.7
H2 O
ReaxFFSiO
-1
OSiO
SiOSi
91.2132.6
137.5179.6
103.6114.6
151.9163.5
108.3
156.6
109.4
158.0
109.5
143.7
H2 O
ReaxFFSiO
-10
OSiO
SiOSi
89.6134.2
146.6172.0
102.6115.5
154.2160.8
107.0
157.0
109.3
157.7
109.5
143.7
B.J. Cowen, M.S. El-Genk / Computational Materials Science 111 (2016) 269276
273
3.1.1.2. c/a. The c/a ratio at the ab transition for quartz is difficult
to model and has eluded many force fields. Herzbach et al. [39]
indicated that the polarizable TS [40] is the only force field, among
those they investigated, that is capable of modeling the sudden
drop in c/a ratio at the ab transition for quartz. The present MD
simulation results with the COMB10-1 and COMB10-10 force fields
H2 O
Fig. 3. Comparison of calculated cell volume expansion at the ab transition of quartz in MD simulations with COMB10 and ReaxFFSiO
to experimental results.
274
B.J. Cowen, M.S. El-Genk / Computational Materials Science 111 (2016) 269276
H2 O
Fig. 4. Comparison of calculated c/a ratio at the ab transition of quartz, in MD simulations with COMB10 and ReaxFFSiO
to experimental results.
H2 O
performed simulations with the ReaxFFSiO
-10 show a similar
H2 O
simulations with both ReaxFFSiO
and ReaxFFGSI
SiO are within 3%
and 5% of the lattice constants and density, respectively, compared
to experimental values (Table 3). These results demonstrate the
transferability of these force fields to the a-cristobalite bonding
environment.
3.2. Cristobalite
3.3. Coesite
3.4. Stishovite
The crystal structure of stishovite resembles that of rutile
(TiO2), with each of the octahedrally coordinated silicon atoms
bound to six oxygen atoms [44]. Therefore, the transferability of
275
B.J. Cowen, M.S. El-Genk / Computational Materials Science 111 (2016) 269276
Table 3
Comparison of calculated properties (standard deviation shown in parenthesis) and atom charges for a-cristobalite in MD simulations with the BOVC force fields, to experimental
values.
Parameter
a ()
c ()
q (g/cm3)
qavg (O)
qavg (Si)
qrange (O)
qrange (Si)
Experimental/force field
Exp. [41]
COMB10 [11]
H2 O
[12]
ReaxFFSiO
ReaxFFGSI
SiO [13]
4.978
6.948
2.318
4.81 (1e2)
6.13 (1e2)
2.81 (7e3)
1.45 (1e4)
2.90 (3e4)
1.39 to 1.51
2.882.92
5.03 (7e3)
7.11 (1e2)
2.22 (3e3)
0.87 (4e3)
1.73 (7e3)
0.76 to 0.97
1.581.90
5.04 (9e3)
7.12 (1e2)
2.20 (4e3)
0.47 (1e3)
0.94 (3e3)
0.43 to 0.51
0.900.98
Table 4
Comparison of calculated properties (standard deviation shown in parenthesis) and atom charges for coesite, in MD simulations with the BOVC force fields, to experimental
values.
Parameter
a ()
b ()
c ()
q (g/cm3)
b ()
qavg (O)
qavg (Si)
qrange (O)
qrange (Si)
Exp. [43]
7.136
12.369
7.174
2.921
120.34
Force fields
COMB10 [11]
H2 O
[12]
ReaxFFSiO
ReaxFFGSI
SiO [13]
7.04 (1e2)
12.40 (2e2)
7.22 (9e3)
2.97 (6e3)
121.12 (1e1)
1.46 (2e4)
2.91 (4e4)
1.37 to 1.54
2.892.93
7.29 (2e2)
12.33 (1e2)
7.12 (7e3)
2.88 (3e3)
119.75 (8e2)
0.79 (3e3)
1.59 (5e3)
0.71 to 0.88
1.481.76
7.20 (1e2)
12.19 (2e2)
7.01 (1e2)
2.96 (4e3)
119.41 (1e1)
0.46 (1e3)
0.91 (2e3)
0.41 to 0.50
0.870.95
Table 5
Comparison of calculated properties (standard deviation shown in parenthesis) and atom charges for stishovite, in MD simulations with BOVC force fields to experimental values.
Parameter
a ()
c ()
q (g/cm3)
qavg (O)
qavg (Si)
qrange (O)
qrange (Si)
Exp. [44]
4.180
2.666
4.283
Force field
COMB10 [11]
H2 O
[12]
ReaxFFSiO
ReaxFFGSI
SiO [13]
4.09 (7e3)
2.97 (2e3)
4.02 (4e3)
1.47 (1e4)
2.94 (3e4)
1.43 to 1.51
2.932.96
4.56 (4e3)
3.04 (2e3)
3.16 (3e3)
0.53 (2e3)
1.07 (3e3)
0.48 to 0.59
1.021.11
5.68 (4e2)
2.54 (2e2)
2.38 (2e2)
0.41 (3e3)
0.82 (6e3)
0.23 to 0.51
0.600.96
ReaxFFGSI
SiO , respectively. This could be because the density and the
cohesive energies of various polymorphs were determined with
single point energies at different densities, which is more of an
evaluation than an optimization. Thus, despite the fact that MD
H2 O
simulations with both ReaxFFSiO
and ReaxFFGSI
SiO accurately predict
the properties of the other silica polymorphs, these force fields
are not necessarily suitable for all phases of silica, and need to be
optimized more effectively for stishovite. The MD simulation
results with the COMB10 force field are accurate, to within 11%
of the experimental values for the c lattice parameter, and
2.2% for the a lattice parameter (Table 5). Thus, the COMB10
force field is a more appropriate choice for modeling stishovite.
276
B.J. Cowen, M.S. El-Genk / Computational Materials Science 111 (2016) 269276
the BAD function for quartz are within 10% of the experimental
values.
H2 O
The simulation results with ReaxFFSiO
and ReaxFFGSI
SiO
inaccurately predict the properties of the high-pressure stishovite
polymorph; they are within 26% and 44% of the experimental
values, respectively. In addition, the results of the MD simulations
H2 O
with the COMB10 and the ReaxFFSiO
inaccurately model the ab
transition for quartz. The results of the cell volume expansion at
the ab transition are quite different from experimental values
and do not predict the experimentally measured plateau, caused
by the rotation of the SiO4 tetrahedra.
Acknowledgements
The financial support for this research is provided by the Institute for Space and Nuclear Power Studies at the University of New
Mexico (ISNPS-UNM) and by a U.S. Nuclear Regulatory Commission
(NRC) Graduate Fellowship under Grant # NRC-38-09-931 to
ISNPS-UNM.
The authors thank Dr. Susan Atlas of the Center for Advanced
Research Computing (CARC), at the University of New Mexico for
her help and suggestions, acknowledge the valuable support
provided by the CARC staff and are grateful for using the
supercomputer clusters at the center.
References
[1] J. Livage, T. Coradin, C. Roux, J. Phys.-Condens. Matter 13 (2001) R673.
[2] A. Cimas, F. Tielens, M. Sulpizi, M.P. Gaigeot, D. Costa, J. Phys.-Condens. Matter
26 (2014) 244106.
[3] A. Rimola, D. Costa, M. Sodupe, J.F. Lambert, P. Ugliengo, Chem. Rev. 113 (2013)
4216.
[4] M. Schulz, Nature 399 (1999) 729.
[5] C.J. Cochrane, P.M. Lenahan, Appl. Phys. Lett. 103 (2013) 053506.
[6] B.W. van Beest, G.J. Kramer, R.A. van Santen, Phys. Rev. Lett. 64 (1990) 1955.
[7] A. Pedone, G. Malavasi, M.C. Menziani, A.N. Cormack, U. Segre, J. Phys. Chem. B
110 (2006) 11780.
[8] S. Tsuneyuki, M. Tsukada, H. Aoki, Y. Matsui, Phys. Rev. Lett. 61 (1988) 869.
[9] A. Carre, J. Horbach, S. Ispas, W. Kob, EPL-Europhys. Lett. 82 (2008) 17001.
[10] S. Munetoh, T. Motooka, K. Moriguchi, A. Shintani, Comput. Mater. Sci. 39
(2007) 334.
[11] T.R. Shan, B.D. Devine, J.M. Hawkins, A. Asthagiri, S.R. Phillpot, S.B. Sinnott,
Phys. Rev. B 82 (2010) 255.
[12] J.C. Fogarty, H.M. Aktulga, A.Y. Grama, A.C.T. van Duin, S.A. Pandit, J. Chem.
Phys. 132 (2010) 174704.
[13] A.D. Kulkarni, D.G. Truhlar, S.G. Srinivasan, A.C.T. van Duin, P. Norman, T.E.
Schwartzentruber, J. Phys. Chem. C 117 (2013) 258.
[14] J.G. Yu, S.B. Sinnott, S.R. Phillpot, Phys. Rev. B 75 (2007) 085311.
[15] A.C.T. van Duin, S. Dasgupta, F. Lorant, W.A. Goddard, J. Phys. Chem. A 105
(2001) 9396.
[16] T.R. Shan, B.D. Devine, T.W. Kemper, S.B. Sinnott, S.R. Phillpot, Phys. Rev. B 81
(2010) 125328.
[17] T.R. Shan, B.D. Devine, S.R. Phillpot, S.B. Sinnott, Phys. Rev. B 83 (2011) 115327.
[18] M.J. Noordhoek, T. Liang, Z.Z. Lu, T.R. Shan, S.B. Sinnott, S.R. Phillpot, J. Nucl.
Mater. 441 (2013) 274.
[19] W.X. Song, S.J. Zhao, J. Mater. Res. 28 (2013) 1155.
[20] D. Raymand, A.C.T. van Duin, M. Baudin, K. Hermansson, Surf. Sci. 602 (2008)
1020.
[21] A.C.T. van Duin, A. Strachan, S. Stewman, Q.S. Zhang, X. Xu, W.A. Goddard, J.
Phys. Chem. A 107 (2003) 3803.
[22] A. Yasukawa, JSME Int. J. A-Mech. M 39 (1996) 313.
[23] T.T. Trinh, A.P.J. Jansen, R.A. van Santen, J. Phys. Chem. B 110 (2006) 23099.
[24] B.J. Cowen, M.S. El-Genk, Comput. Mater. Sci. 107 (2015) 88.
[25] S. Grazulis, D. Chateigner, R.T. Downs, A.F.T. Yokochi, M. Quiros, L. Lutterotti, E.
Manakova, J. Butkus, P. Moeck, A. Le Bail, J. Appl. Crystallogr. 42 (2009) 726.
[26] S. Grazulis, A. Daskevic, A. Merkys, D. Chateigner, L. Lutterotti, M. Quiros, N.R.
Serebryanaya, P. Moeck, R.T. Downs, A. Le Bail, Nucleic Acids Res. 40 (2012)
D420.
[27] R.T. Downs, M. Hall-Wallace, Am. Mineral. 88 (2003) 247.
[28] S. Plimpton, J. Comput. Phys. 117 (1995) 1.
[29] H.M. Aktulga, J.C. Fogarty, S.A. Pandit, A.Y. Grama, Parallel Comput. 38 (2012)
245.
[30] A.K. Rapp, W.A. Goddard, J. Phys. Chem.-Us 95 (1991) 3358.
[31] E. Polak, G. Ribiere, Rev. Fr. Inform. Rech. Oper. 3 (1969) 35.
[32] Y.K. Shin, T.R. Shan, T. Liang, M.J. Noordhoek, S.B. Sinnott, A.C.T. van Duin, S.R.
Phillpot, MRS Bull. 37 (2012) 504.
[33] I. Tajima, O. Asami, E. Sugiura, Anal. Chim. Acta 365 (1998) 147.
[34] J. Fontanel, C. Andeen, D. Schuele, J. Appl. Phys. 45 (1974) 2852.
[35] E.K. Chang, M. Rohlfing, S.G. Louie, Phys. Rev. Lett. 85 (2000) 2613.
[36] L. Levien, C.T. Prewitt, D.J. Weidner, Am. Mineral. 65 (1980) 920.
[37] M.A. Carpenter, E.K.H. Salje, A. Graeme-Barber, B. Wruck, M.T. Dove, K.S.
Knight, Am. Mineral. 83 (1998) 2.
[38] Y.H. Kim, M.S. Hwang, H.J. Kim, J.Y. Kim, Y. Lee, J. Appl. Phys. 90 (2001) 3367.
[39] D. Herzbach, K. Binder, M.H. Mser, J. Chem. Phys. 123 (2005) 124711.
[40] P. Tangney, S. Scandolo, J. Chem. Phys. 117 (2002) 8898.
[41] D.R. Peacor, Z. Kristallogr. 138 (1973) 274.
[42] E.C.T. Chao, E.M. Shoemaker, B.M. Madsen, Science 132 (1960) 220.
[43] L. Levien, C.T. Prewitt, Am. Mineral. 66 (1981) 324.
[44] L.G. Liu, W.A. Bassett, T. Takahashi, J. Geophys. Res. 79 (1974) 1160.