2 Novemberc 2001

Chemical Physics Letters 348 (2001) 131±138

www.elsevier.com/locate/cplett

Hartree±Fock geometry optimisation of periodic systems

with the CR Y S T A L code
a,*
B. Civalleri , Ph. D'Arco b, R. Orlando c, V.R. Saunders d, R. Dovesi a,e

a
Dipartimento di Chimica IFM, Universit a di Torino, Via P. Giuria 7, 10125 Torino, Italy
b
Laboratoire de P etrologie, Mod elisation de Materiaux et Processus, Universit
e Pierre et Marie Curie,
4, Place Jussieu, 75232 Paris Cedex 05, France
c
Dipartimento di Scienze e Tecnologie Avanzate, Universit a del Piemont Orientale, C.so Borsalino 54, 15100 Alessandria, Italy
d
CLRC Daresbury Laboratory, Daresbury, WA4 4AD Warrington, UK
e
Unit
a INFM di Torino, Sezione F, Via P. Giuria 7, 10125 Torino, Italy
Received 9 August 2001; in ®nal form 6 September 2001

Abstract
Results are reported on the geometry optimisation of periodic systems with the Hartree±Fock analytical gradients
recently implemented in the CR Y S T A L code. Application to the structure optimisation of molecules, polymers, slabs
and crystals is presented. Ó 2001 Elsevier Science B.V. All rights reserved.

1. Introduction First-derivatives can be computed either nu-

The determination of equilibrium structure is of
primary importance in the modelling of chemical
systems and several methods for their geometry
optimisation have been proposed [1]. All of them
try to locate minima and saddle points on the
potential energy surface (PES). Basic tools to de-
scribe the PES are ®rst- and second-derivatives.
First-derivatives permit to ®nd stationary points
on the PES whilst second-derivatives allow to
characterize them. Among geometry optimisation
techniques, gradient-based algorithms have been
indicated as the methods of choice for most levels
of theory [2]. Thus, gradients play a twofold cen-
tral role in the geometry optimisation of chemical
systems.
*
Corresponding author. Fax: +39-011-6707855.
E-mail address: civalleri@ch.unito.it (B. Civalleri).
0009-2614/01/$ - see front matter Ó 2001 Elsevier Science B.V. All rights reserved.
PII: S 0 0 0 9 - 2 6 1 4 ( 0 1 ) 0 1 0 8 1 - 8
merically or analytically. Nevertheless, analytical
gradients have become an ever-increasing area in
quantum chemistry and, when available, provide
an important tool to facilitate the determination of
equilibrium structures.
In the ®eld of quantum chemistry, the Hartree±
Fock approach still remains a fundamental theory
for both molecular and solid-state chemistry and is
the basis for most of the wavefunction-based
methods. Analytical HF gradients are commonly
implemented in standard molecular codes and
routinely used in molecular calculations, but this
does not apply for solid-state calculations al-
though, signi®cant work has been done for one-
dimensional periodic systems [3±5]. Recently, a
general implementation of analytical Hartree±
Fock gradients has become available for zero, one,
two, and three-dimensional systems [6±8].
In this Letter we report on the geometry opti-
misation of periodic systems based on these ana-
132 B. Civalleri et al. / Chemical Physics Letters 348 (2001) 131±138
lytical Hartree±Fock gradients [6±8] as imple-
mented in the CR Y S T A L code [9]. Examples of
structure optimisation are reported ranging from
molecules to polymers, slabs and crystals. To the
best of our knowledge this is the ®rst time that
geometry optimisation of crystals adopting ana-
lytical Hartree±Fock gradients is reported.
2. Computational details
The calculation of the forces on the atoms has
been performed by using the recent implementa-
tion of analytical HF energy gradients [6±8] in the
CR Y S T A L program [9]. These have been used to
relax the atoms to equilibrium using a modi®ed
conjugate gradient algorithm as proposed by
Schlegel [10]. The optimisation algorithm evalu-
ates the gradients each time the energy is com-
puted and the second derivative matrix is updated
employing the gradients. At each step, a one-
dimensional minimisation using a quadratic
polynomial is carried out, followed by an
n-dimensional search using the Hessian matrix.
In the present work, a unit matrix is used as an
initial guess for the Hessian and symmetry is fully
exploited. Convergence is tested on the root-
mean-square (RMS) and the absolute value of the
largest component of both the gradients and the
estimated displacement of nuclei. In the present
calculations the threshold for the maximum and
the RMS forces and the maximum and the RMS
atomic displacements on all the atoms have been
set to (in a.u.) 0.00045, 0.00030 and 0.00180,
0.00120, respectively. The optimisation is consid-
ered complete when all four conditions are sat-
is®ed.
The choice of these conditions enables a tight
comparison with standard ab initio molecular
codes (vide infra) For the same reason the toler-
ances that rule the calculation of the Coulomb and
exchange integrals were set to TOL ˆ 15 15 15 15
20 in the molecular cases [9]. For periodic calcu-
lations the default CR Y S T A L tolerances [9] were
adopted, instead. The threshold for stopping the
SCF cycle was 1  10 8 and 1  10 6 Hartree for
molecular total energy and in the case of periodic
calculations, respectively.
For periodic systems, a cell ®xed unconstrained
optimisation of the atomic coordinates was carried
out [11].
3. Results and discussion
3.1. Preliminary calculations
Preliminarily, a detailed investigation has been
performed on the geometry optimisation of mo-
lecular structures. The main goal is to compare our
implementation with that of molecular ab initio
codes. We refer to the performance test on the
determination of equilibrium molecular structures
presented by Baker [12], where the geometry op-
timisations in Cartesian coordinates and natural
internal coordinates of 30 molecules at the HF
level and by using a STO-3G basis set are com-
pared. In particular, in the present work, we use as
a reference the set of results obtained in [12] for the
geometry minimisation in cartesian coordinates
with a unit Hessian as initial guess.
The geometry optimisation has been carried out
by adopting almost the same computational con-
ditions as Baker. Results are reported in Table 1.
Initial and ®nal energies of the molecules di€er by
less than 1  10 5 Hartree from Baker's determi-
nations, showing that the optimised geometries are
essentially the same in both cases. From the
number of iterations one can conclude that e-
ciency is comparable for the two implementations.
Although a few critical cases are observed such as
benzidine and menthone, the global sum of the
number of iterations needed for this set of mole-
cules is quite similar: 750 in our case and 765 in
Baker's. This small di€erence is probably due to
slightly di€erent computational conditions and the
di€erent optimisation algorithm adopted to per-
form the geometry relaxation. Nevertheless, results
in Table 1 clearly show that the present imple-
mentation is reliable and of comparable eciency
as standard molecular codes.
The application of this procedure to periodic
systems requires a careful preliminary analysis of
the in¯uence of computational conditions, such as
the truncation of the Coulomb and exchange series
and the de®nition of the sampling net in the irre-
 B. Civalleri et al. / Chemical Physics Letters 348 (2001) 131±138 133

Table 1
Comparison between present work (pw) and previous data on Baker's [12] set of molecules
Molecule Number of at- Symmetry Number of Number of cycles
oms variables pw Ref. [12]
Water 3 C2v 2 5 5
Ammonia 4 C3v 2 6 7
Ethane 8 D3d 3 7 7
Acetylene 4 D1h 2 6 7
Allene 7 D2d 3 7 10
Hydroxysulphane 4 C1 6 26 21
Benzene 12 D6h 2 5 6
Methylamine 7 Cs 10 8 10
Ethanol 9 Cs 13 19 18
Acetone 10 C2v 8 15 22
Disilyl ether 9 C2v 7 20 27
1,3,5-trisilacyclohexane 18 C3v 11 23 36
Benzaldehyde 14 Cs 25 14 17
1,3-di¯uorobenzene 12 C2v 11 7 8
1,3,5-tri¯uorobenzene 12 D3h 4 7 7
Neopentane 17 Td 3 6 10
Furan 9 C2v 8 12 10
Naphthalene 18 D2h 9 10 11
1,5-di¯uoronaphthalene 18 C2h 17 14 16
2-hydroxybicyclopentane 14 C1 36 44 45
ACHTAR10 16 C1 42 47 51
ACANIL01 19 Cs 34 35 36
Benzidine 26 D2 18 39 26
Pterin 17 Cs 31 20 23
Difuropyrazine 16 C2h 15 18 21
Mesityl oxide 17 Cs 28 38 38
Histidine 20 C1 54 92 102
Dimethylpentane 23 C1 63 31 29
Ca€eine 24 Cs 42 42 39
Menthone 29 C1 81 127 100

ducible Brillouin zone (IBZ) [9], on the accuracy of
geometry optimisation.
On this purpose, we performed a systematic in-
vestigation on the geometry optimisation of crys-
talline urea, as a test case. The experimental
structure determined by neutron low temperature
di€raction data [13] has been used as initial ge-
ometry. All the optimisations have been carried out
by employing a STO-3G basis set and a unit matrix
as initial Hessian. Results are shown in Table 2.
In Table 2, the number of optimisation cycles,
the initial geometry total energy, Ei and the energy
gain at convergence, DE, are reported as well a the
largest (Max DX10 ) and (RMS DX10 ) di€erence
between the optimised cartesian coordinates for
each case and case 10, considered as a reference.
Starting from the default tolerances for the inte-
grals screening (TOL ˆ 6 6 6 6 12), [9] we have
progressively increased the number of sampling
points (NKP) in IBZ (see cases 1, 3, 8, and 9) and
tested the convergence on Ei . Results in Table 2
show that with 18 k-points (case 3) Ei is already
stable. As regards the dependence of Ei on TOL
(see cases 2, 3, 4, 5, 6, and 7) one can see that Ei is
still varying in the explored range of TOL, though
by less than 1 mHartree. However, the number of
iterations is de®nitely insensitive to the computa-
tional conditions, being always about 15 cycles
and it is worth noting that the gain in the com-
puted total energy is almost constant in all cases,
indicating that the ®nal structure obtained in dif-
ferent conditions are all very similar. This is also
con®rmed when considering Max DX10 and RMS
DX10 which are in general below 1  10 3 and
134 B. Civalleri et al. / Chemical Physics Letters 348 (2001) 131±138

Table 2
Dependence of the geometry optimisation of crystalline urea on the number of sampling points in the IBZ, NKP, and on the integrals
screening tolerances, TOL [9]
Case NKP TOL Number Ei a DEa Maxb RMSb CPU
of cycles DX10 DX10 timec
1 6 6 6 6 6 12d 15 442:069367 0.015228 0.001021 0.000435 179
2 18 5 5 5 5 10 15 442:069659 0.015213 0.001310 0.000512 110
3 6 6 6 6 12d 15 442:069584 0.015265 0.000257 0.000083 180
4 7 7 7 7 14 15 442:069929 0.015326 0.000222 0.000118 299
5 8 8 8 8 16 15 442:070365 0.015330 0.000468 0.000185 443
6 5 5 5 10 10 14 442:069901 0.015291 0.000701 0.000262 132
7 6 6 6 12 12 14 442:069759 0.015330 0.000254 0.000108 228
8 40 6 6 6 6 12d 15 442:069584 0.015265 0.000653 0.000287 181
9 75 6 6 6 6 12d 14 442:069584 0.015265 0.000649 0.000253 185
10 8 8 8 8 16 16 442:070365 0.015330 0.000000 0.000000 439
a
Ei : initial geometry total energy; DE: gain in total energy at convergence (atomic units).
b
Max DX10 and RMS DX10 : maximum and RMS di€erence between the optimised cartesian coordinates of the current case and case 10,
as the reference.
c
Average CPU time for optimisation cycle in seconds.
d
Default values in CR Y S T A L .

5  10 4 A, respectively. In particular, for case 3
their values are 0.000257 and 0.000083 A,  well
below the accuracy of an experimental structure
determination.
In summary, good equilibrium structures can be
obtained by adopting well-converged HF solutions
and default integrals screening tolerances, as in
case 3. This also leads to a signi®cant saving in
computational time as shown in the last column of
Table 2. Indeed, the cost of case 10 is about 2.5
times larger than that of case 3.
As a further information on the cost of the
calculations, a few details on the CPU time (on an
IBM RS/6000 43P) for case 3 follow. For a con-
ventional SCF, i.e., storing all of the integrals on
disk the average CPU time for an optimisation
step takes 180 s, i.e., 37 s for the SCF cycle and 143
s for the gradient evaluation. For a direct SCF
calculation, the average CPU time for an optimi-
sation step is 346 s where the SCF cycle and the
gradient computation take 203 and 143 s, respec-
tively (it must be noticed that the analytical gra-
dient evaluation is always performed by means of
a direct algorithm).
3.2. Periodic test cases
On the basis of the previous calculations and in
order to illustrate the performances of the method
in the geometry optimisation of periodic systems,
we de®ned an appropriate test suite of 19 systems.
General information on the adopted set of test
cases is presented in Table 3.
Systems are grouped on the basis of their peri-
odicity: polymers, slabs, and crystals. For a sys-
tematic assessment, periodic systems are selected
to cover a wide range of structural types and
symmetries. They represent most of the di€erent
kinds of chemical bonds that can be found in solid-
state chemistry. Thus, ionic, semi-ionic, covalent
and molecular periodic systems are included.
Furthermore, the examined systems play an im-
portant role in materials science and most of them
have been the subject of previous studies in our
laboratory [14±24]. In particular, examples have
been reported on: oxides and their surface relax-
ation (a-Al2 O3 and a-Cr2 O3 from bulk to a slab
model [14,15]), silicates and zeolites [17±20], ad-
sorption on surfaces (MgO(0 0 1)/CO [21]) and in
zeolites (H±CHA(3:1)/NH3 [20]), defective systems
(carbon substitution in bulk silicon, C@Si [22])
and molecular crystals (e.g., ice [23] and urea [24]).
Initial starting structures were taken from well-
de®ned sources either from experimental data or
from previous quantum or molecular mechanics
calculations. For the sake of brevity, we do not
report all the references and structural details here,
however, full information about the starting ge-
 B. Civalleri et al. / Chemical Physics Letters 348 (2001) 131±138 135

Table 3
General information on the periodic systems adopted as test cases in the present work
Dim. System Typea Symmetryb Number Basis set
of atoms
One-dimensional In®nite water chain P P1 (1) 6 STO-3G
Polyglycine (C5 conformer) P P1m1 (2) 14 STO-3G
Polytetra¯uoroethylene P Pmcm (8) 6 STO-3G
Two-dimensional a-Al2 O3 slab (6 atomic layers) S P3 (6) 10 STO-3G
a-Cr2 O3 slab (6 atomic layers) S P3 (6) 10 Cr,O [14,15]
MgO(0 0 1)/CO (3 atomic layers) Ad P4=mmm (16) 10 Mg,O [21]
Three-dimensional a-Al2 O3 IC R3c (12) 10 STO-3G
a-Cr2 O3 IC R3c (12) 10 Cr,O [14,15]
a-Quartz IC P32 21 (6) 9 STO-3G
a-Boron CC R3m (12) 12 STO-3G
C@Si (32 atoms super-cell) CC I43m (24) 32 STO-3G
Ferroelectric ordered ice (Ice XI) MC Cmc21 (4) 12 STO-3G
a-Oxalic acid dihydrate crystal MC P21 =n (4) 28 STO-3G
Urea crystal MC P421 m (8) 16 STO-3G
Silica sodalite SOD(Si) Z P43m (24) 36 STO-3G
Silica faujasite FAU(Si) Z Fd3m (48) 144 STO-3G
Titano chabazite Ti±CHA(1:1)c Z R3 (6) 36 Si,O,Ti [18]
Acidic chabazite H±CHA(3:1)d Z R3 (3) 39 STO-3G
H±CHA(3:1)/NH3 d Ad R3 (3) 51 STO-3G
a
Type: P ˆ polymer, S ˆ surface, Ad ˆ adsorption, IC ˆ ionic/semi-ionic crystal, CC ˆ covalent crystal, MC ˆ molecular crystal,
Z ˆ zeolite.
b
Number of symmetry operations in parentheses.
c
Si:Al ˆ 1:1
d
Si:Al ˆ 3:1

ometries can be found at the CR Y S T A L web site
[25]. All optimisations were carried out employing
the STO-3G basis set, except for a few cases ex-
plicitly indicated in Table 3, and a unit matrix has
been used as initial guess for the Hessian.
Results are shown in Table 4. For each system,
the number of variables to be optimised and the
number of optimisation cycles are reported as well
as energetic data, namely: initial and ®nal energies,
and the gain in total energy at convergence.
Results in Table 4 show that geometry optimi-
sation performs equally well for systems with any
periodicity: polymers, slabs and crystals. The
quality of geometry optimisation appears to be
independent of the type of structure and symmetry
and, as a general trend, the number of iterations
ranges from one to ®ve times the number of vari-
ables to be optimised.
Indeed, molecular crystals are important test
cases for geometry optimisation due to the pres-
ence of soft modes and weak interactions. All of
the studied systems are formed by three-dimen-
sional networks of H-bonds. For instance, in the
a-oxalic acid dihydrate crystal, each molecule is
surrounded by six water molecules which are then
involved in three hydrogen bonds with nearby
oxalic acid molecules (Fig. 1). The crystal structure
results in a low symmetry complex network of
H-bonds. In spite of such a complicated structure,
the geometry optimisation of the oxalic acid crystal
(21 degrees of freedom) is achieved after 49 cycles
only, i.e., about twice the number of variables. In
general, results show that, even for weakly bound
systems, optimisation performs quite well.
Zeolites are another important class of materi-
als. Because of their relevance in materials science,
a few cases have been included in the test suite.
Among them, the geometry optimisation of H±
CHA(3:1)/NH3 (Fig. 2) is worth being remarked
upon. In fact, among the test cases here consid-
ered, H±CHA(3:1)/NH3 is the one with the largest
number of degrees of freedom and the speci®c Si/
136 B. Civalleri et al. / Chemical Physics Letters 348 (2001) 131±138

Table 4
Convergence data on the test suite of periodic systemsa
System Number of Number of cycles Ei Ef DE
variables
In®nite water chain 15 15 149:945729 149:960431 0.014702
Polyglycine 24 19 408:218296 408:268540 0.050243
Polytetra¯uoroethylene 3 6 466:968596 466:970283 0.001687
a-Al2 O3 slab 5 10 1399:799902 1400:114819 0.314917
a-Cr2 O3 slab 5 18 4622:595509 4622:823020 0.227511
MgO(0 0 1)/CO 4 12 1046:361591 1046:393148 0.031557
a-Al2 O3 2 4 1400:392826 1400:395110 0.002284
a-Cr2 O3 2 9 4623:109982 4623:110039 0.000057
a-Quartz 4 10 1300:382505 1300:383282 0.000777
a-Boron 4 5 292:609911 292:615488 0.005577
C@Si (SC32) 5 10 8892:597943 8892:655827 0.057884
Ice XI 10 15 299:957883 299:965335 0.007452
a-Oxalic acid 21 49 1042:387770 1042:637879 0.250109
Urea 8 15 442:069367 442:084595 0.015228
SOD(Si) 3 13 5201:458862 5201:472245 0.013383
FAU(Si) 10 23 20805:841975 20805:857697 0.015722
Ti±CHA(1:1) 18 39 8620:992411 8620:993034 0.000623
H±CHA(3:1) 38 70 5063:591308 5063:616329 0.025021
H±CHA(3:1)/NH3 50 169 5229:832874 5230:147140 0.314266
a
Ei and Ef : initial and ®nal total energies; DE: gain in total energy at convergence. For the initial geometry see [25] (atomic units).

Fig. 1. a Oxalic acid dihydrated structure.

Al loading of the framework considered allows the

interaction of three molecules of ammonia with the
acidic Brùnsted sites per unit cell of the zeolite.
Hence, geometry optimisation is inherently di-
Fig. 2. Structure of H±CHA(3:1) acidic zeolite supercage
cult due to structural complexity and presence of showing the interaction of the Brùnsted sites with NH3 . Atoms
hard and soft mode coupling. Nevertheless, the are displayed.
 B. Civalleri et al. / Chemical Physics Letters 348 (2001) 131±138
number of iterations required (169 cycles) appears
as reasonable.
It has been shown [12,29] that eciency in ge-
ometry optimisation can be improved by adopting
a good initial guess for the Hessian and an ap-
propriate coordinate system. On the one hand,
Baker [12] addressed that a reduction by a factor
up to six can be obtained for geometry optimisa-
tion in cartesian coordinates provided that a reli-
able estimated Hessian is available (e.g., from
molecular mechanics calculations or model Hes-
sians [26±28]). On the other hand, the choice of a
proper coordinate system, like natural internal
coordinates, has been shown to be very useful [29]
to deal with dicult structure optimisations. In-
ternal coordinates reduce the coupling between
hard and soft modes facilitating the location of
energy minima and could be very advantageous
for complex systems like H±CHA(3:1)/NH3 . In
this respect, geometry optimisation in either re-
dundant internal coordinates [30] or delocalised
internal coordinates [31], has been recently pro-
posed for periodic systems. Implementation of
such techniques is under study in our laboratory.
The use of cartesian coordinates, however, al-
lows partial geometry optimisation to be readily
carried out. Thus, geometry optimisation can be
limited to an atomic fragment instead of the whole
system. Accordingly, the symmetrised cartesian
coordinates are then generated on the basis of a list
of atoms to be optimised. This procedure leads in
general to a reduction in the number of cycles and
is useful in the case of large systems where a small
atomic region, only, plays the major role in de-
termining the chemical activity (e.g., acidic zeo-
lites). As a test case, a fragment geometry
optimisation has been carried out for H±
CHA(3:1). By starting from the same initial ge-
ometry as in the full geometry optimisation, but
limiting the optimisation to the Brùnsted site
(BSiO(H)AlB) only, the geometry relaxation takes
20 cycles instead of 70. The ®nal geometry of the
Brùnsted site is very similar in both cases, the
main structural features are (fragment optimisa-
tion results in parentheses): rSiOb ˆ 1:740 (1.746);
rAlOb ˆ 1:835 (1.849); rOb H ˆ 0:975 (0.976);
\SiOb Al ˆ 140:0 (141.2); sSiOb (H)Al ˆ 179.3
( 179:7).
137
4. Conclusions
Application to the structure optimisation of
periodic systems at the Hartree±Fock level has
been reported. We have shown that geometry
minimisations are now feasible for molecules,
polymers, slabs, and crystals. Further, the imple-
mentation of an automated geometry optimisation
in a standard periodic ab initio code like CR Y S T A L
[9] provides an important tool for solid-state
computational chemistry, as it has been demon-
strated for systems of interest in many ®elds of
materials science.
Acknowledgements
The present work is part of a project coordi-
nated by A. Zecchina and co-®nanced by the
Italian MURST (Co®n98, A03). The authors are
grateful to P. Ugliengo, C. Roetti and C.M. Zi-
covich-Wilson for suggestions and helpful discus-
sions.
References
[1] H.B. Schlegel, Adv. Chem. Phys. 687 (1987) 249.
[2] H.B. Schlegel, Geometry optimisation: 1, in: Encyclopedia
of Computational Chemistry, Wiley, New York, 1998.
[3] H. Terame, T. Yamabe, A. Imamura, J. Chem. Phys. 81
(1984) 3564.
[4] J.-Q. Sun, R.J. Bartlett, J. Chem. Phys. 104 (1996) 8553.
[5] D. Jacquemin, J.-M. Andre, B. Champagne, J. Chem.
Phys. 111 (1999) 5324.
[6] K. Doll, N.M. Harrison, V.R. Saunders, Int. J. Quantum
Chem. 82 (2001) 1.
[7] K. Doll, Comp. Phys. Comm. 137 (2001) 74.
[8] R. Orlando, V.R. Saunders, R. Dovesi, unpublished.
[9] V.R. Saunders, R. Dovesi, C. Roetti, M. Causa N.M.
Harrison, R. Orlando, C.M. Zicovich-Wilson, CR Y S T A L -
98 User's Manual (Universita di Torino, Torino, 1999).
[10] H.B. Schlegel, J. Comp. Chem. 3 (1982) 214.
[11] The generalisation of the code to gradients with respect to
cell parameters is in progress.
[12] J. Baker, J. Comp. Chem. 14 (1993) 1085.
[13] S. Swaminathan, B.M. Craven, R.K. McMullan, Acta
Crystallogr. B 40 (1984) 300.
[14] M. Catti, G. Sandrone, G. Valerio, R. Dovesi, J. Phys.
Chem. Solids 57 (1996) 1735.
[15] M. Catti, G. Valerio, R. Dovesi, M. Causa, Phys. Rev. B
49 (1994) 14179.
138 B. Civalleri et al. / Chemical Physics Letters 348 (2001) 131±138
[16] M. Catti, B. Civalleri, P. Ugliengo, J. Phys. Chem. B 104
(2000) 7259.
[17] B. Civalleri, C.M. Zicovich-Wilson, P. Ugliengo, V.R.
Saunders, R. Dovesi, Chem. Phys. Lett. 292 (1998) 394.
[18] C. Zicovich-Wilson, R. Dovesi, J. Phys. Chem. 102 (1998)
1411.
[19] P. Ugliengo, B. Civalleri, C.M. Zicovich-Wilson, R.
Dovesi, Chem. Phys. Lett. 318 (2000) 247.
[20] E.H. Teunissen, A.P.J. Jansen, R.A. van Santen, R.
Orlando, R. Dovesi, J. Chem. Phys. 101 (1994) 5865.
[21] A. Damin, R. Dovesi, A. Zecchina, P. Ugliengo, Surf. Sci.
479 (2001) 255.
[22] R. Orlando, P. Azavant, M.D. Towler, R. Dovesi, C.
Roetti, J. Phys.: Cond. Matter 8 (1996) 1123.
[23] S. Casassa, P. Ugliengo, C. Pisani, J. Chem. Phys. 106
(1997) 8030.
[24] R. Dovesi, M. Causa, R. Orlando, C. Roetti, V.R.
Saunders, J. Chem. Phys. 92 (1990) 7402.
[25] Notes on the starting geometry and the CRYSTAL input
decks can be found at the CRYSTAL home page: http://
www.ch.unito.it/ifm/teorica/supplement/index.html.
[26] H.B. Schlegel, Theor. Chim. Acta 66 (1984) 333.
[27] T.H. Fisher, J. Alml of, J. Phys. Chem. 96 (1992)
9768.
[28] R. Lindh, A. Bernhardsson, G. Karlstr  Malmq-
om, P.-A
vist, Chem. Phys. Lett. 241 (1995) 423.
[29] C. Peng, P.Y. Ayala, H.B. Schlegel, M.J. Frisch, J. Comp.
Chem. 17 (1996) 49.
[30] K.N. Kudin, G. Scuseria, H.B. Schlegel, J. Chem. Phys.
114 (2001) 2919.
[31] J. Andzelm, R.D. King-Smith, G. Fitzgerald, Chem. Phys.
Lett. 335 (2001) 321.