Phys Chem Minerals (1997) 25: 70–78 C Springer-Verlag 1997

OR IG I NA L P A P E R

G.W. Watson ? P.M. Oliver ? S.C. Parker

Computer simulation of the structure and stability

of forsterite surfaces

Received: 28 October 1996 y Revised, accepted: 7 April 1997

Abstract The aim of this paper is to demonstrate that

atomistic simulations can be used to evaluate the struc- Introduction
ture of mineral surfaces and to provide reliable data for Atomistic simulation of the surfaces of polar solids was
forsterite surfaces up to a plane index of 2 using the code pioneered by the work of Tasker (1978, 1979a) and
METADISE. The methods used to calculate the surface Mackrodt and Stewart (1977). Initially, these simula-
structure and energy which have more commonly been tions concentrated on simple ceramics such as the cubic
used to study ceramics are briefly explained as is a com- halides (Tasker 1979b) and binary oxides, MgO (Col-
parison with experimental data, most notable the crystal bourn et al. 1983) and NiO (Tasker and Duffy 1984).
morphology. The predicted morphologies show that all Recent studies have considered more ternary oxides such
the methods (Donnay-Harker, Attachment energies and as MgAl2O4 (Davies et al. 1994), La2CuO4 and Nd2CuO4
equilibrium) show most of the surfaces that are ex- (Kenway et al. 1995) and where detailed experimental
pressed in observed crystals. The equilibrium morpholo- evidence is available, e.g. via Scanning Tunnelling Mi-
gy calculated from the relaxed surface energies is the croscopy (STM) on WO3, the simulations reproduce the
only method which expresses the {201} surfaces and the surface structure (Oliver et al. 1996).
{101} surfaces, which appear only upon relaxation. The The scope of simulations has been extended further
more stable surfaces are shown to be those which have by considering surface microfaceting of MgO (Watson
the highest surface density and more closely resemble et al. 1996) and the energies of the physical and chemical
close packed structures with highly coordinated surface adsorption of water onto the surfaces of MgO (de Leeuw
ions and silicon as far from the surface as possible. The et al. 1995, 1996).
most stable surfaces the {100} which has alternating lay- The calculation of surface structures and energies is
ers of MgO and SiO2 terminating with an MgO layer. The important step toward modelling and understanding the
structure is similar to the MgO {100} surfaces and has a behaviour of polycrystalline minerals and ultimately
similar energy (1.28 Jm22 compared to 1.20). The sec- rock assemblages, because it is the interfaces that modify
ond most stable are the {201} which have a stepped sur- the behaviour of the polycrystalline material. The aim of
face topology, but is also compact with a relaxed surface the work described in this paper is to apply the simula-
energy of 1.56 Jm22. The results indicate that atomistic tion methods that have been successful in modelling sim-
simulation is well suited to the prediction of surface ple ceramic oxide surfaces to a silicate mineral. This will
structure and morphology although care must be taken in allow us to investigate the extent to which potentials de-
choosing potentials which model the structure and elas- veloped and used for simulating bulk silicate crystal
tic properties accurately. properties can be used to model surfaces.
The first silicate considered was forsterite (Mg2SiO4)
because its bulk properties have been extensively mod-
G.W. Watson1 (✉) ? P.M. Oliver ? S.C. Parker elled (Price et al. 1987a, b; Parker and Price 1989) and
Computational Solid State Chemistry Group, the THB1 potential is considered a reliable potential for
School of Chemistry, University of Bath, Claverton Down, predicting the bulk properties of this system.
Bath, BA2 7AY. U.K.
Fax: 144(0)1225 826 231
email: g.watson @liv.ac.uk
Present address:
Surface methodology
1
Leverhulme Centre for Innovative Catalysis,
Department of Chemistry, University of Liverpool, The basic approach, used within the METADISE pro-
Oxford Road, Liverpool, L69 3BX, U.K. gram (Watson et al. 1996), is to consider the crystal as a
 71

The specific surface energy is defined as the energy

per unit area required to transform a bulk region into a
surface region. The surface energy is thus given by;
Usurf2Ubulk
g5
A
where Usurf refers to the energy of region I of a surface
calculation, Ubulk refers to the energy of an equivalent
number of bulk ions, and A is the area.
A condition for calculating the surface energy is that
there is no dipole perpendicular to the surface in the
repeat unit, because such a dipole leads to a divergent
surface energy (Bertaut 1958). Tasker (1979a) defined
three types of surfaces, I, II and III. Type I surfaces are
composed of stoichiometric layers and thus have no
dipole perpendicular to the surface (Fig. 1a). Type II
surfaces are composed of multi-layer repeat units which
have no dipole (Fig. 1b). Type III surfaces have a multi-
layer repeat unit and have a dipole perpendicular to the
surface (Fig. 1c). Type III surfaces would therefore be
predicted to be unstable, however recent work on NiO
(Oliver et al. 1993), MgO (Watson et al. 1996) and
MgAl2O4 (Davies et al. 1994) has shown that these sur-
faces can be stabilised by oxidation or reconstruction to
remove the dipole.
The calculated surface energies of a crystal are very
difficult to compare with experiment, because of the
lack of experimental data. However, one strategy for
comparing the relative stability of the surfaces is to eval-
uate the morphology which can be either equilibrium or
kinetic growth forms.

Crystal morphology

Fig. 1a–c Illustration of the three types of surfaces. a Type I, b Gibbs in 1878 (see Gibbs 1928) first proposed that the
Type II, c Type III equilibrium form of a crystal should, for a given volume,
possess minimal surface energy i.e.
block consisting of a stack of planes periodic in two di- g5Si gi Ai is a minimum for constant volume
mensions and parallel to the defect (surface). An as-
sumption is made that only those ions close to the defect where gi and Ai are the surface energy and surface area of
will need to relax from their lattice sites and thus the the ith crystallographic face.
blocks are split into an inner region (region 1) which Wulff (1901), starting from the Gibbs proposal, sug-
contains those ions close to the defect and an outer region gested that the shape thus defined would be such that hi,
(region 2) incorporating ions more distant from the de- the normal vector to the face from a point within the
fect. The ions in region 1 are allowed to relax to their crystal, would be proportional to the surface energy of
mechanical equilibrium while those in region 2 are fixed that face, gi;
at their bulk lattice sites. The energy and forces on the hi5l gi
ions are calculated within the Born model of solids (Bom
and Huang 1988) using a potential model which includes where l is a constant that depends on the absolute size of
long range electrostatic terms and short range interac- the crystal resulting in the crystal morphology as shown
tions. We have used the THB1 potential (Price et al. in Fig. 2.
1987a, b; Parker and Price 1989), which also includes This expression will only hold for crystals grown with
oxygen polarisability via the shell model (Dick and Over- all faces in equilibrium and takes no account of kinetic
hauser 1958) and directional bonding via three body factors such as growth rate. The difficulty is that it is
forces (Sanders et al. 1984) and has been shown to repro- often unclear when a crystal exhibits its equilibrium
duce successfully the structure, elastic properties and form. However, for small crystals where rearrangement
phonon frequencies of forsterite. of the crystal at each stage of the growth process is possi-
72

the growth rates. In energetic terms the most stable sur-

faces prefer least new material attached to their surface.
As the resulting morphology is equivalent to the equi-
librium form this approach does provide a explanation
for the good agreement between calculated equilibrium
forms and large grown crystals.
In practice these simple models can only be used as a
guide because the growth rate will depend on many fac-
tors that are not included, for example, the free energy of
formation of steps and kinks on each surface and the
properties of the growth medium including saturation,
Fig. 2 Schematic representation of the calculation of crystal mor- pH and ionic strength.
phology

The low index surfaces of forsterite (Mg2SiO4)

ble, due to the small distances over which material has to
travel, or if the crystal habit is formed over geological Forsterite (Mg2SiO4) is an end member of the mineral
time scales, a morphology close to the equilibrium form olivine, which shows a complete solid solution from
might be expected. However, in the latter case the surface forsterite, the magnesium end member, to fayalite
energy may be dominated by impurities, the concentra- (Fe2SiO4), the iron end member, with the typical sample
tion of which will be significant in geological materials. of the mineral olivine (Mg, Fe)2SiO4 containing approxi-
In an attempt to access kinetic factors we have also mately 10% Fe (Fo90). It has orthorhombic symmetry
calculated the growth morphology. One phenomenologi- (space group Pnma with a510.20 Å, b55.98 Å,
cal approach for estimating the growth morphology of c54.75 Å), and consists of independent SiO4 tetrahedra
crystals was introduced by Donnay and Harker (1937). In linked by divalent cations in octahedral co-ordination.
this, the morphological importance of a surface is in- The divalent cations occupy two distinct crystallographic
versely proportional to the interplanar repeat distance sites with the oxygen sub-lattice deviating significantly
dhkl, i.e. from the ideal close packing arrangement, resulting in
variation in Mg-O bond distance (2.05–2.21 Å).
1 The large deviation of the oxygen sub-lattice results
hi5l
dhkl in forsterite having a very complex set of crystal planes.
For the simplest surfaces, those containing plane indices
Thus low index surfaces with large interplanar spacing
of either 0 or 1 (termed the 1 index surfaces), there are
are predicted to dominate the morphology, which is in
seven distinct sets of planes ({100}, {010}, {001},
general true.
{110}, {101}, {011}, {111}) because of the crystal’s
A second approach is to use attachment energies,
orthorhombic symmetry. Figure 3 shows the first non
which are defined as the interaction of a layer of thick-
dipolar repeats found for each surface with indications of
ness dhkl with the rest of the crystal, assuming a bulk
addition planes which may be cut to form non dipolar
termination (Hartman and Bennema 1980). A large at-
repeats and include both type II and type III (or dipolar
tachment energy indicates that ions coming onto the sur-
surfaces). The type III surfaces can in all cases be recon-
face are strongly bound and so the attachment energy can
structed with a zero dipole by moving half of the ions
be used as a measure of the growth rate of the surface. By
from the top to the bottom of the repeat (e.g. {010a}).
finding the slowest growing cut for each surface, a pseu-
The repeat units were generated assuming that the
do-kinetic morphology can be predicted by relating the
tetrahedra would remain intact, as previous calculations
inverse of the surface’s morphological importance (i.e.
(Davies 1992) have indicated that the breaking of the
the height of the normal) to the growth rate growth rate
Si-O interaction always results in less stable surfaces.
via the attachment energy.
This causes some of the cuts not to be simple cleavage
hi5l E(att)i planes as the oxygens are cut according to the position of
the silicon they are bonded to. For example, the {111a}
Essentially the attachment energy approach is related to surfaces show an O1 and O2 above the Mg2 which was
the Donnay-Harker scheme (Hartman and Bennema on the plane of the cut due to them forming part of the
1980). The use of the attachment energies adds atomistic tetrahedron of the silicon below this magnesium.
information by adjusting the morphology to account for Some of the surface cuts are complicated by cation
faster growth of those surfaces which have strong bonds arrangement at the surface. For example, the {101b} sur-
perpendicular to the surface. faces cut a plane of four magnesium ions in half resulting
A less favoured phenomenological approach is to ar- in 50% occupancy of the surface magnesium layer.
gue that as the most stable surfaces are well known to Therefore there are six possible cation arrangements at
grow the most slowly then the surface energies, which the surface, but in general this can be reduced, in this
are a measure of the relative stability are proportional to case to three, by considering surface symmetry.
 73

Fig. 3 Schematic representa-

tion of the surface stacking se-
quences for the 1 index sur-
faces of forsterite, (100, 010,
001, 110, 101, 011, 111). The
figure shows the sequence of
ions perpendicular to the the
surface, and the possible cuts
for each surface that result in
a zero dipole repeat unit

A general problem of evaluating the surface energy is is surface-dependent and continues over 6 Å into re-
ensuring its convergence and is achieved by simulating a gion 1. The second is that because of forsterite’s ortho-
sufficiently large crystal region. Table 1 shows the varia- rhombic nature, the area of the surface unit cells vary
tion in the relaxation displacement of the first 28 ions as according to the orientation and therefore the same num-
a function of depth for the {100a} and {010b} surfaces ber of ions results in varying region 1 depths for different
with region 1s containing 112 ions (176 species). This surfaces. This combination means that great care must be
table illustrates two points. The first is that the relaxation taken to ensure that enough ions are considered to ensure
 74
Table 1 The displacement on
relaxation (x, y, z in Å) of the 100a 010b
top 28 cores from their bulk
positions in the METADISE Ion Depth Displacement (Å) Ion Depth Displacement (Å)
minimised (region 15112
ions) 100a and 010b surfaces (Å) X Y Z (Å) X Y Z
as a function of depth
Mg2 0.00 0.28 0.00 20.09 Mg1 0.00 0.39 0.58 20.02
O1 0.54 20.04 0.06 0.11 O1 0.21 0.36 20.98 0.60
O1 0.54 20.04 20.06 0.11 O1 0.21 0.51 20.69 20.40
Si 1.22 20.06 0.00 20.02 O3 1.50 20.63 20.17 20.47
O2 1.28 0.15 0.00 20.06 Mg2 1.50 0.29 20.15 0.23
O3 1.74 0.05 0.00 0.07 O2 1.50 0.14 20.44 0.25
Mg1 2.22 0.02 20.04 20.03 O3 1.50 20.17 20.38 0.43
Mg1 2.22 0.02 20.04 20.03 Si 1.50 0.01 20.39 0.29
O3 2.71 20.05 0.00 20.07 Mg2 1.50 20.16 20.26 20.15
O2 3.17 20.04 0.00 0.02 Si 1.50 0.00 20.12 20.23
Si 3.23 0.05 0.00 0.00 O2 1.50 20.09 20.03 20.28
O1 3.91 0.05 0.00 20.01 O1 2.79 0.07 20.14 0.11
O1 3.91 0.05 0.00 20.01 O1 2.79 0.24 0.50 20.37
Mg2 4.45 20.04 0.00 20.02 Mg1 2.99 0.28 20.51 0.04
Mg2 5.12 0.02 0.00 0.00 Mg1 2.99 20.24 20.17 0.20
O1 5.66 20.02 0.00 20.01 O1 3.20 20.01 20.36 0.05
O1 5.66 20.02 0.00 20.01 O1 3.20 20.07 0.22 20.17
Si 6.34 20.01 0.00 0.00 O2 4.49 0.07 20.15 0.04
O2 6.40 0.00 0.00 0.00 Si 4.49 0.06 20.22 0.04
O3 6.86 0.00 0.00 0.00 Mg2 4.49 0.00 20.11 0.09
Mg1 7.35 20.01 0.00 0.00 O2 4.49 20.15 20.05 0.01
Mg1 7.35 20.01 0.00 0.00 O3 4.49 20.23 0.00 0.08
O3 7.83 20.01 0.00 0.00 Si 4.49 20.16 20.04 0.03
O2 8.29 20.01 0.00 0.00 Mg2 4.49 0.13 20.09 20.01
Si 8.35 0.00 0.00 0.00 O3 4.49 0.31 20.16 20.04
O1 9.03 0.00 0.00 0.00 O1 5.78 20.03 20.02 0.05
O1 9.03 0.00 0.00 0.00 O1 5.78 20.10 20.11 0.01
Mg2 9.57 0.00 0.00 0.00 Mg1 5.99 20.15 20.11 0.02

Table 2 Lowest unrelaxed and

relaxed surface energies for the Plane Index Miller Index Unrelaxed surface Relaxed surface Attachment
1 index surfaces of forsterite energy (Jm22) (cut) energy (Jm22) (cut) energy (eVySi)

a
100 200 2.23 (a) 1.28 (a) 21.98
The 110b surface has a re- 010 020 3.84 (b) 1.61 (b) 27.20
laxed surface energy of 1.92 Jm22. 001 002 4.61 (b) 2.02 (a) 28.58
b
The 3 indicates this to be the 110 220 6.56 (a) 1.90 (a)a 214.06
third of three possible cation 101 101 3.51 (b3)b 1.70 (b3)b 27.02
arrangements. 011 011 4.88 (a) 1.81 (a) 25.47
c
The 111a surface has a re- 111 111 4.01 (b) 1.80 (b)c 25.21
laxed surface energy of 1.90 Jm22.

convergence of the surface energy. This is further illus-

trated by the surface energy of the {100a} surface which The 2 index surfaces
converges for a region size of 1 unit cell (44 species,
On inclusion of surfaces with plane indices of up to 2,
10.24 Å) while the {101b} requires 3 unit cells (132 spe-
the total number of structurally unique orientations in-
cies, 13 Å) and the {010a} requires 4 unit cells (176
creases from 7 to 19. The additional 12 have much larger
species, 23.95 Å).
surface areas, and thus for a given number of ions pro-
Minimisation of the 1 index surfaces gave rise to the
duce surface unit cells of less depth. This greatly in-
lowest surface energies for each index before and after
creases the computational effort for three reasons.
minimisation shown in Table 2. The relaxation results in
(i) The number of ions required to ensure convergence
only a small change in the order of stability of the sur-
of the surface energy increased. For example, the {012}
faces with the order before minimisation of,
surfaces for a cell of 44 species results in a depth perpen-
100 ,101 ,010 ,111,001,011,110 dicular to the surface of only 2.54 Å requiring 308 spe-
cies (a depth of 17.8 Å) to ensure convergence in
and after minimisation,
METADISE.
100 ,010,101,111©011,110,001. (ii) The occurrence of many type III cuts which on
reconstruction to form a zero dipole repeat results in 50%
 75

occupancy of the surface cation sites and hence a number stacking sequences are found they are composed of two
of possible cation rearrangements need to be considered. non-dipolar units alternating and thus a mirror image of
(iii) The occurrence of asymmetric stacking se- the stacking sequence can be formed (termed asymmet-
quences. This occurs when a zero dipole repeat unit is not ric inverse). Thus allows growth of both surfaces cut at
symmetric which results in different surfaces at each end both ends of the crystal, and thus the surfaces can be
of the repeat unit. Fortunately for forsterite where such treated independently (Fig. 4).
Minimisation of all the 2 index surfaces was achieved
although some of the more unstable surfaces showed a
number of high energy minima which was taken to be
an indication of the surface instability. For example, the
{210a} before minimisation have a surface energy of
7.5 Jm22 and relaxed to surface energies at both
2.39 Jm22 and 2.60 Jm22. This type of multiple mini-
mum was investigated by varying the number of steps
between updating the second derivative matrix and ad-
justing the maximum displacement an ion can move in a
single step. In general the more stable surfaces achieve
minimisation to a global minimum while some of the
unstable surfaces showed multiple minima. In each case
we believe the lowest energy surface has been identified.
The lowest surface energies for each index are shown
in Table 3 indicating that in general the symmetric sur-
faces are more stable than the asymmetric.

Crystal morphology

The order of surface stability including all of the one and

two index surfaces changes from
100,201,101,121,021,221,010,111,211,210
,212,011,122,110,001,012,120,112,102
before relaxation to
100,201,221,211,010,101,111,011,121,021
,120,110,210,001,212,122,122,012,102
after relaxation. This change in stability has an effect on
the calculated equilibrium morphology. Figure 5 shows a
comparison of the Donnay Harker (Fig. 5a), attachment
energy (Fig. 5b) and equilibrium morphologies, before
Fig. 4a–d Illustration of asymmetric and asymmetric inverse
stacking sequences for the {221}b surface. a 221bas repeat unit, b and after relaxation (Fig. 5c, 5d), with two observed mor-
221basi repeat unit c and d the two alternating non dipolar repeats phologies for natural olivine (Figs. 5e, 5f) (Dana and
within the surface repeat Ford 1958; Deer et al. 1982).

Table 3 Lowest unrelaxed and

relaxed surface energies for Plane Indexy Unrelaxed surface Relaxed surface energy Attachment energy
the 2 index surfaces of forster- Miller Index energy (Jm22) (cut) (Jm22) (cut) (eVySi)
ite
012 7.00 (b) 2.29 (a) 216.20
021 3.70 (a) 1.83 (a) 27.41
102 8.15 (c) 2.30 (a) 231.93
112 5.85 (a) 1.99 (a) 216.06
120 7.18 (b) 1.89 (b) 215.78
121 3.69 (a) 1.82 (a) 27.75
122 5.11 (c) 2.12 (c) 216.03
201 2.56 (a) 1.56 (a) 23.38
210 4.67 (b) 1.95 (b) 25.69
211 4.29 (a) 1.59 (a) 26.87
212 4.75 (b) 2.02 (b) 29.88
221 3.72 (b1) 1.58 (b1) 28.30
76

absent. In addition the {210} surfaces are also absent

which has a large impact on the observed morphologies.
The unrelaxed equilibrium morphology (Fig. 5c)
shows a number of features consistent with those ob-
served. The {100}, {201}, {021} and {010} are all
present, however the {121} and {221}, which are not
observed, also appear. This time the Donnay Harker and
attachment energy schemes the {201} is present however
the {101} is now absent.
The relaxed equilibrium morphology expresses the
{201} in agreement with observations unlike the Donnay
Harker and attachment energy and the {101} unlike the
unrelaxed equilibrium morphologies and indicates the
importance of surface relaxation on surface stability. It
does however, include a number of additional surfaces,
namely the {221} and {211} in contradiction to the ob-
served morphologies.
In summary, none of the morphologies match the ob-
served samples in all respects. The Donnay-Harker
scheme is the only one to match those observed with
respect to the appearance of the {210} surface while the
relaxed equilibrium morphology is the only scheme with
correctly predicts the appearance of both the {201} and
{101} indicating the relaxation is an important factor.

Surface relaxation and structure

Table 2 clearly shows that the {100a} surfaces are the
Fig. 5a–f Comparison of calculated a Donnay Harker, b attach- most stable before (2.23 Jm22) and after (1.28 Jm22) re-
ment energy, c unrelaxed, d relaxed and observed, e Dana (1958), laxation. The small relaxation energy implies that the
and f modified after Deer et al. (1982) crystal morphologies of
forsterite surface cut does not remove a large number of bonds and
this is seen in Fig. 6a, which shows the {100a} surface
before and after relaxation. The high surface density
Matching the predicted morphology with those grown with alternate SiO2 and MgO layers terminating in MgO
in the laboratory can be difficult despite the ideal condi- gives rise to a surface structure and energy very similar
tions. Kinetic factors play a major role in influencing the to that of the {100} surface of MgO (1.20 Jm22). Re-
morphology because of fast growth rates. However, even laxation is very minor with the surface magnesium layer
for such crystals thermodynamics may play a part espe- relaxing slightly (0.3 Å) towards the bulk, with the sec-
cially if the crystals selected for study are those that keep ond layer (SiO2) relaxing by a very small amount
the same morphology with time. For mineralogical sam- (,0.1 Å) away from the bulk, resulting in a slight undu-
ples the factors controlling crystal growth are much lating surface topology. This is in agreement with the
harder to determine. Additionally there are increase only experimental study of the surface structure of
amounts of impurities which can segregate and hence olivine (Hochella 1990), which looked at the {100} sur-
modify the growth characteristics of different surfaces to face using Low Energy Electron Diffraction (LEED).
different extents. They observed a 131 surface unit cell with only small
Comparison of the different models shows that the relaxations.
Donnay-Harker morphology (Fig. 5a) gives good agree- The next three lowest relaxed surface energies, the
ment with those observed (Figs. 5e, 5f) and has similari- {201a}, {221bas1} and {211a} are all within 0.03 Jm22
ties with the relaxed equilibrium morphology (Fig. 5d). with the {201} surfaces the most stable before and after
It expresses all of the required surfaces except the {201}, minimisation and is the result of, like the {100a}, a com-
which is absent because of the dominance of the low pact surface with magnesium in the top layer and the first
index {101} surface causing the crystal to appear tabu- silicon 1.2 Å below the surface. This results in a very
lar. stable surface with only minor relaxations of the surface
The attachment energy morphology (Fig. 5b) shows magnesium and oxygens (Fig. 6b) and gives rise to a
some agreement with observed morphologies although definite stepped surface topology.
the agreement is not as good as that for the Donnay Hark- The {221bas1} is an asymmetric surface which initial-
er scheme. The {100} surfaces are too dominant leading ly has a single oxygen pointing straight up as a result of
to a plate like crystal, with the {201} surface once again keeping the tetrahedra complete. The initial surface en-
 77

Fig. 6a–f The structure on the three lowest energy surfaces of out from the bulk. On relaxation one magnesium ion
forsterite before and after relaxation. a 100a, b 201b, c 221bnsl, d moves across the surface by 1.5 Å to coordinate to an
211b, e 010b and f 101b3 isolated oxygen (Fig. 6d). This appears to distort the
structure but only results in slight rotation of some of the
surface SiO4 tetrahedra with the oxygen ions moving to
ergy is surprisingly low (3.72 Jm22) and is a result of the increase their interactions. Relaxation of up to 0.1 Å
rest of the surface displaying a generally flat termination continues for up to 7 Å into the surface and the whole
with predominantly magnesium rather than silicon at the result is a large stabilisation of the surface from
surface. Relaxation of this surface is quite dramatic with 4.29 Jm22 to 1.59 Jm22.
surface tetrahedra rotating to allow the isolated oxygen Figure 6e shows the surface structure of the {010b}
ion to co-ordinate to two surface magnesium atoms (Fig. which is a type III surface formed by cutting a layer of
6c). The surface relaxes to accommodate this with the two magnesium ions in half. This results in a surface
surface oxygens moving as much as 2.00 Å resulting in with an isolated magnesium ion which on relaxation
an undulating surface structure. moves both toward the bulk by 0.4 Å and laterally within
The next surface, the {211a}, has a much less com- the plane of the surface by 0.6 Å moving away from one
pact jagged structure with oxygen ions sticking directly oxygen and toward another. This type of lateral move-
78

ment is observed up to 6 Å into the surface and together Colbourn EA, Mackrodt WC, Tasker PW (1983) The segragation
with rotation of the SiO4 tetrahedra gives rise to a flatter of calcium ions at the surface of magnesium oxide: theory and
calculation. J Mater Sci 18: 1917–1924
and more compact, but distorted, surface structure. Dana ES, Ford WE (1958) A textbook in mineralogy. Wiley, New
The {101b3} surface (Fig. 6f) is the result of cleaving York
a plane of four magnesium ions. This leads to three inde- Davies MJ (1992) Computer simulation of oxide surfaces and
pendent surfaces for which the surface energy varies grain boundaries. PhD thesis, University of Bath
Davies MJ, Parker SC, Watson GW (1994) Atomistic simulation of
from 7.00 Jm22 to 3.51 Jm22 for the unrelaxed and the surface structure of spinel. J Mater Chem 4: 813–816
2.28 Jm22 to 1.70 Jm22 for the relaxed surfaces. The fig- Deer WA, Howie RA, Zussman J (1992) An introduction to the
ure shows that the initial cut is quite jagged giving rise to rock forming minerals. Longman, London
exposed magnesium ions at the surface. On relaxation de Leeuw NH, Watson GW, Parker SC (1995) Atomistic simula-
tion of the effect of dissociative adsorption of water on the
these magnesium ions move towards the bulk occupying surface structure and stability of calcium and magnesium ox-
positions between oxygen ions increasing the number of ide. J Phys Chem 99: 17219–17225
nearest neighbour oxygens. This again results in an al- de Leeuw NH, Watson GW, Parker SC (1996) Atomistic simula-
most flat surface structure which is consequently rela- tion of adsorption of water on three-, four-, and five-coordinat-
tively stable. ed surface site of magnesium oxide. J Chem Soc Faraday Trans
92: 2081–2091
Dick BG, Overhauser AW (1958) Theory of the dielectric con-
stants of alkali halide crystals. Phys Rev 112: 90–103
Donnay JD, Harker G (1937) A new law of crystal morphology
Conclusions extending the law of Bravais. Am Mineral 22: 446–467
Gibbs JW (1928) Collected works, Longman, New York
This aim of this paper was to demonstrate the scope and Hartman P, Bennema P (1980) The attachment energy as a habit
limits in using computer simulation to model the surface controlling factor. J Cryst Growth 49: 145–156
Hochella MF Jr (1990) Atomic structure, microtopography, com-
energies and structures of a silicate mineral. The results position and reactivity of mineral surfaces. In: Ribbe PH (ed)
of these simulations have provided predictions of the sur- Mineral-water interface geochemistry (Reviews in Mineralogy
face structures which we hope will be verified by exper- vol 23). Mineralogical Society of America, Washington DC,
imental study. Indeed, the recent success of atomic force pp 87–132
Kenway PR, Parker SC, Mackrodt WC (1995) The calculated
microscopy (AFM) on a range of mineral surfaces sug- structure, stability and composition of the low index surfaces
gests this is now possible and that with reliable models of La2CuO4 and Nd2CuO4. Surf Sci 326: 301–310
the surface from simulation studies interpretation of Mackrodt WC, Stewart RF (1977) Defect properties in ionic
these images may be easier. The basis for suggesting we solids: I point defects at the surfaces of face centred cubic
crystals. J Phys C 10: 1431–1446
have a reliable model is first, that the potential model Oliver PM, Parker SC, Mackrodt WC (1993) Computer simulation
chosen has previously been shown to predict accurately a of the crystal morphology of NiO. Modelling in Materials Sci-
wide range of bulk crystal properties (Price and Parker ence and Engineering 1: 755–760
1987a, b; Parker and Price 1989). Secondly, there is good Oliver PM, Parker SC, Edgell RG, Jones FH (1996) Computer
accord with the crystal morphology of natural olivine, simulation of the surface structures of WO3. J Chem Soc Fara-
day Trans 92: 2049–2056
although for reasons discussed above an exact match is Parker SC, Price GD (1989) Computer modelling of phase transi-
not yet possible because of the limitations of the kinetic tions in minerals. Advances in Solid State Chemistry 1: 295–
growth models. Our future work in this area will include 327
the application and development of more sophisticated Price GD, Parker SC, Leslie M (1987a) The lattice dynamics of
forsterite. Mineral Mag 51: 157–170
growth models allowing not just qualitative comparison Price GD, Parker SC, Leslie M (1987b) The lattice dynamics and
with morphology but predictions of the influence of sur- thermodynamics of Mg2SiO4 polymorphs. Phys Chem Miner-
face defects and impurities. The latter will be particular- als 15: 181–190
ly important when understanding the defect properties of Sanders MJ, Leslie M, Catlow CRA (1984) Interatomic potentials
for SiO2. J Chem Soc Chem Commun:1271–1273
geological materials. Tasker PW (1978) U.K.A.E.A Report, AERE-R-9130, United
Kingdom Atomic Energy Authority, Harwell, UK
Acknowledgements We would like to thank the EPSRC and the Tasker PW (1979a) The stability of ionic crystals. J Phys C
NERC for financial support and Molecular Simulations Inc. for 12: 4977–4983
the provision of INSIGHT II. Tasker PW (1979b) The surface energies, surface tensions and
surface structure of the alkali halide crystals. Philo Mag
39: 119–136
Tasker PW, Duffy DM (1984) The structure and properties of the
stepped surfaces of MgO and NiO. Surf Sci 137: 91–102
References Watson GW, Kelsey ET, de Leeuw, Harris DJ, Parker SC (1996)
Atomistic simulation of dislocations, surfaces and interfaces in
Bertaut F (1958) The electrostatic term of the surface energy. CR MgO. J Chem Soc Faraday Trans 92: 433–438
Acad Sci 246: 3447 Wulff G (1901) Zur Frage der Geschwindigkeit des Wachstums
Bom M, Huang K (1988) Dynamical theory of crystal lattices. und der Auflosung der Krystallflachen. Z Kristallogr Kristall-
Clarendon Press, Oxford geom 34: 949