J. CHEM. SOC. FARADAY TRANS.

, 1995, 91(16), 2671-2680 267 1

Auxetic Two-dimensional Polymer Networks

An Example of Tailoring Geometry for Specific Mechanical Properties

Ken E. Evans,* Andrew Alderson? and Frances R. Christian

School of Engineering, University of Exeter, North Park Road, Exeter, UK EX4 4QF

The Poisson's ratios and Young's moduli of 2D molecular networks having conventional and re-entrant honey-
comb forms have been modelled using molecular modelling. Three principle deformation mechanisms were
observed : bond hinging, flexure and stretching. Analytical models have also been developed that can be used to
describe each of these modes of deformation acting either independently or concurrently. A parametric fit of the
force constants in the concurrent analytical model calculations to the molecular model calculations yields good
agreement in the mechanical properties for all the structures studied. Specific trends in the force constants
required to fit the data are observed. Consequently, a force constants library can be compiled and has been
used to predict accurately the properties of more complex variants of the networks. This semi-analytical sub-unit
approach enables a more efficient use of computer-intensive molecular modelling programs.

1. Introduction quate approximation, as embodied in many commonly used

molecular mechanics programs, where atomic forces and
This paper considers one example of how the geometric
long-range interactions are modelled by spatial and angular-
arrangement of a polymer network may be tailored to
dependent, classical force constants. These issues have been
achieve specific macroscopic mechanical properties. The
thoroughly covered e l s e ~ h e r e . ' ~ * ' ~
example considered here is the development of auxetic molec-
In the next section we introduce and develop the molecular
ular structures ; polymer networks that exhibit a negative and analytical models used to describe the deformation of
Poisson's ratio. At the moment this work is in its early stages.
polymer networks having conventional and re-entrant honey-
Some research has been published on the design of novel
polymeric networks that exhibit this unusual effect '-' and (a 1
some work has been progressed on the synthesis of such net-
w o r k ~ .Applications
~,~ for such materials have also been iden-
tified.7-9 As yet, no molecular structure has been modelled,
synthesised and characterized, but we expect that this will not
be long in coming. The development of a molecular network
with a negative Poisson's ratio is an excellent example of
understanding form to obtain a particular function. The first
examples of structures (as opposed to materials) with auxetic
functionality (the ability to expand laterally when stretched)
are to be found in the analysis of the mechanical properties of
honeycombs.".' ' Here, one further attribute other than form
is required to demonstrate a function and that is the mode of
deformation. So, in the first analysis," the form, a hexagonal
honeycomb, was assumed to deform in a particular manner,
by flexure, leading to the mechanical property, a positive
Poisson's ratio. By modifying the form, using a re-entrant
honeycomb (see Fig. l), but maintaining the same mode of
deformation, a negative Poisson's ratio results.
Classical elasticity is not a scale-dependent formulation
and none of the simple analyses of structure that have been
used to predict negative Poisson's ratios contain any scaling
factors. One may therefore assume, naively, that geometric
structures and deformation mechanisms that apply at the
macrostructural level may also apply at the microstructural
and molecular level.
This assumption has been proved, experimentally, in the
microstructural case by the synthesis and testing of a novel
form of auxetic polyethylene.' 2*1 This paper considers the
further use of this approach at the molecular level. Of course,
at some length-scale classical mechanics will break down and
quantum mechanics take over. However, for the polymer net-
works considered here, classical mechanics is still an ade-

H
f Current address : British Nuclear Fuels plc, Company Research Fig. 1 Honeycomb cell geometry and co-ordinate system used in
Laboratory, B5 16, Springfields Works, Preston, Lancashire, UK PR4 the analytical models for (a) conventional and (b) reentrant molecu-
OXJ. lar networks
 2672 J. CHEM. SOC. FARADAY TRANS., 1995, VOL. 91

comb forms. The analytical models are based on the net-

works deforming by hinging, flexure and stretching of the
honeycomb cell walls, with each mode acting independently
of the others in the simplest (single-mode) models. A more
complex (multi-mode) analytical model where all three modes
act concurrently is then developed. Comparison between the
analytical and molecular model predictions of the mechanical
properties of the networks is then made and the results dis-
cussed accordingly.
2. Models
In this section we first describe the molecular model used to
predict the Poisson’s ratios and Young’s moduli of the molec-
ular network structures and then the analytical models. In all
cases expressions for the Poisson’s ratios and Young’s moduli
are derived.
2.1 Molecular Model
Examples of the earliest conventional and re-entrant molecu-
lar network sub-units designed’,’ are shown in Fig. 2(a) and
(b), respectively. The networks consist of branches of acety-
lene groups joined by benzene rings at the junctions. Each
benzene ring has three polyacetylene arms connected to it.
The connectivity of the polyacetylene arms to the benzene
rings determines the honeycomb geometry. When the arms
are connected to alternate benzene ring carbon atoms the
conventional honeycomb geometry is realised, whereas the
re-entrant structure is produced when the arms are connected
to three adjacent benzene ring carbon atoms. A naming con-
vention has been adopted:’ (n,m)-flexyne refers to the conven-
tional honeycomb structure, where n and M are the number
of acetylene links on the diagonal and vertical branches,
respectively. (n,m)-reflexyne refers, similarly, to the re-entrant
structure. Hence the sub-units shown in Fig. 2(a) and (b) are
E
(1,4)-flexyneand (1,4)-reflexyne,respectively.
The mechanical properties of these molecular networks
were determined using the POLYGRAF molecular modelling
programI6 (version 2.20) employing the DREIDING force
field,17 which is a well established package for predicting a
range of properties for polymeric materials. An infinite
system was approximated by periodically extending the
repeat unit in the x and y directions. The energy mini- Fig. 2 MolecuIar network sub-units: (a) (1,4)-flexyne, a positive
misation process (using the method of conjugate gradients) Poisson’s ratio molecular network; (b) (1,4)-reflexyne, a negative
described in ref. 2 was adopted in this work. For each struc- Poisson’s ratio molecular network
ture, minimisations were performed for the undeformed struc-
ture and for uniaxial applied loads of kO.5 GPa in each of
the x and y directions. Previous calculations on a wider range on conventional and re-entrant honeycomb networks. Hence,
of stress increments have shown that this is sufficient to give if we neglect the detailed molecular structure, it is possible to
accurate results.’ Values of Poisson’s ratio ( v i j ) and Young’s derive expressions for the elastic properies of these networks
modulus (Ei) were then evaluated from the stress and strain using conventional honeycomb theory.’, In this case the
data with network sub-units are represented by the conventional and
vij = -Ej/Ei re-entrant honeycomb cells shown in Fig. l(a) and (b), respec-
tively, and the deformation is due to flexure of the diagonal
Ei = b i / E i honeycomb arms. The repeat unit-cell lengths in the x and y
where bi is the stress applied in the i ( = x or y ) direction and directions are given by
and c j are the true strains calculated using
E~
x = 21 cos e (4)
ci = ln(I/Z,) (3) Y = 2(h + I sin 0) (5)
where I (= X or Y) and I, (= X, and Yo) are the deformed
where h and I are the lengths of the vertical and diagonal
and undeformed repeat unit-cell lengths in the i direction.
arms, respectively. 8 is the honeycomb angle which is positive
2.2 Single-mode Analytical Models for the conventional cell, and negative for the re-entrant cell,
see Fig. 1. The Poisson’s ratios and Young’s moduli associ-
In the derivations that follow we assume unit thickness in the ated with these networks have been developed elsewhere”
z-direction (perpendicular to the x-y plane). and are quoted directly here:
2.2.1Flexure Model cos e x
VXY = vy;’ = --
The molecular networks described in Section 2.1 are based sin 0 Y
J. CHEM. SOC. FARADAY TRANS., 1995, VOL. 91 2673

For an x directed load we have

(7)

and
F = F, sin 8/2 (23)
K , is the flexure force constant governing the flexure of the
ribs. For ribs of length I, thickness t, depth w and intrinsic where F, is the force applied in the x direction which, for unit
Young’s modulus E,, K , is found from standard beam thickness in the z direction, is given by
theory” to be F, = 0, Y (24)
K , = E, wt3/l (9) From eqn. (21)-(24) we have
From eqn. (6) we note that positive Poisson’s ratios are rea-
d0 = -(IY sin 8/2K,) dux (25)
lised owing to flexure when 8 is positive (conventional
honeycomb) whereas a negative value of 0 (re-entrant Given the possibility of non-linear elastic behaviour we use
honeycomb) yields negative Poisson’s ratios. the tangent Young’s modulus20
For an orthotropic material to have a symmetric stiffness
matrix and a positive definite strain energy for static equi- Ex = dadds, (26)
librium we require’ From eqn. (14), (16), (25) and (26) the Young’s modulus for
unit thickness in the z direction is
vxy Ey = v y x Ex
and
I v x y I d (EJE,)’” Similarly,
From eqn. (6)-(8) we have
sin 0 Y
vxyE, = vyxEx = K,/lz sin 8 cos 0 vyx = --
(12) cos e x
and and
Ivxy I = (EJEy)”2 (13)
Hence the flexure model satisfies the requirements of a sym-
metric stiffness matrix [eqn. (lo)] and a positive strain energy
Note that the hinging-model Poisson’s ratio expressions
Ceqn. (1 113. [eqn. (18) and (28)] are identical to the flexure model expres-
2.2.2 Hinging Model sions [eqn. (6)]. Furthermore, the hinging-model Young’s
We now consider the conventional and re-entrant honey- moduli [eqn. (27) and (2911 differ only by the nature of the
combs shown in Fig. 1 deforming by hinging of the honey- force constant from those for the flexure model [eqn. (7) and
comb cell walls, i.e. by varying the honeycomb angle 0. From (8), respectively]. Hence eqn. (10) and (11) will apply.
eqn. (4) and (5) we have 2.2.3 Stretching Model
dX = -21 sin 0 d8 (14) The final single-mode analytical model we consider is defor-
mation due to stretching of the arms of the honeycomb
dY = 21 cos 0 d8 (15) network shown in Fig. 1, i.e. by varying 1 and h. From eqn.
The increment of true strain in the x direction is defined by (4) and (5) we have

dE, = dX/X (16) axpi =2 cos e (30)

with a similar expression for the y direction. In the case of an aY/ah = 2 (3 1)
x-directed load the Poisson’s ratio is defined by
a Y / a l = 2 sin 8 (32)
vXy = -d~,,/d&, (17) The changes in the unit-cell lengths due to infinitesimal
which holds for non-linear as well as linear elastic behaviour. increments ds, and ds, in the lengths 1 and h, respectively are
Giving, from eqn. (4), (5), (16) and (17): then

cos 0 x d x = (ax/azyds, (33)

vxy = --
sin 8 Y dY = (aY/dl) ds, + (aY/ah)ds, (34)
To derive the Young’s moduli due to hinging we introduce Consider a load applied in the x direction. In this case the
a hinging force constant Kh defined in the usual manner by vertical ribs (length h) remain at constant length since there is
no resultant component of the applied force along the length
M = Kha (19) of these ribs, i.e.
where M is the moment applied to a diagonal honeycomb
ds, =0 (35)
arm and a is the angular displacement of the arm due to M.
M i s given by Therefore, from eqn. (30)-(35) we have
M = IF (20) dX =2 cos 8 ds, (36)
where F is the force applied perpendicular to the diagonal dY = 2 sin 8 ds, (37)
arm of length 1. Hence for an infinitesimal increment in and hence the Poisson’s ratio vXyis
applied force the change in angular displacement is
-sin 0 X
da =1 dF/Kh (21) VXY = --
cos e Y
2674 J. CHEM. SOC. FARADAY TRANS., 1995, VOL. 91

The stretching force constant is defined by For a y-directed load the force F,applied along the length of
the diagonal ribs is
F = K,s (39)
where F is the force applied along the length of an honey- F,= (X sin 8/2)o, (48)
comb arm of length s. K, can be related to the intrinsic and that along the length of the vertical ribs (Fh) is
material Young’s modulus and dimensions of the arm. For
example, consider the increment of extension ds of an arm of Fh = Xo, (49)
length s, thickness t, depth w and Young’s modulus E, due to Therefore, from eqn. (39), (48) and (49) we have
an infinitesimal increase d F in the force applied along the
length of the arm. From Hooke’s law ds, = (X sin 8/2K,) do, (50)
dsh = (X/K$ do, (51)
Hence, from eqn. (30)-(34), (50) and (51) the changes in the
unit-cell lengths due to an infinitesimal incremental change
Therefore, from eqn. (39) and (40) do, in a y-directed load are
K , = E,wt/s (41) dX = (X cos 8 sin 8/KJ do, (52)
For honeycombs consisting of homogeneous or uniform dY = [X(2 + sin’ 8)/K,] do,
material then s = I (and ds = ds,) in the case of the diagonal (53)
arms at angle 8. However, as will be shown, in the stretching and the Poisson’s ratio vyx is, therefore,
model the molecular structure is more important than it was cos 8 sin 8 Y
in the hinging and flexure models. It becomes necessary to Vyx = - (54)
consider which parts of the honeycomb network undergo (2 + sin’ e) X
stretching. Molecular model calculations (see Section 2.1) The increment of strain in the loading direction is in this case
were performed for (2,2)-flexyne for stresses of -2.0, 0 and
+2.5 GPa applied along the y direction. It was found that ca. ds, = [X(2 + sin’ 8)/K,y3 do, (55)
62.5% of the total extension of the vertical arms was due to and hence the Young’s modulus is
extension of the bonds connecting the acetylene arms to the
benzene rings. A further ca. 19% of the extension was due to E, =
K, y
distortion of the benzene rings themselves, with only ca. 19% (2 + sin’ 8) x
of the extension due to deformation of the acetylene From eqn. (38) and (54) we notice that the stretching model
branches. In other words, ca. 80% of the extension of the predicts negative Poisson’s ratios when 8 is positive, whereas
honeycomb arms was found to be related to the benzene ring positive Poisson’s ratios are realised for negative values of 8.
junctions (benzene ring and connecting bond). Hence, the In other words, the stretching model predicts the opposite
length s in eqn. (41) may not be directly related to either I or sign of Poisson’s ratio to the flexure and hinging models. We
h since the molecular structure introduces inhomogeneity also note from eqn. (38), (46),(54) and (56)
into the material forming the honeycomb networks on this
length-scale. As a first approximation then, we neglect acety- vXyE , = v,,E , = -K , sin 8/[cos 8(2 + sin’ 8)] (57)
lene branch elongation and consider all the extension in the and
stretching model to be due to the benzene ring and connect-
ing bond. Hence, since there are equal numbers of benzene (EJEy)’l’ = [(2 + sin’ 8)1/2X]/(cos 8Y) > I vXy I (58)
ring junctions for the vertical and diagonal arms (two) in all and hence the conditions for a symmetric stiffness matrix
cases, then s is the same in both arms and, therefore, from [eqn. (lo)] and a positive definite strain energy [eqn. (ll)] are
eqn. (41) the stretching force constant for both types of arm is satisfied by the stretching model.
equal.
In the case of an x-directed load, the force applied along 2.3 Concurrent Analytical Model
one of the diagonal honeycomb arms is given by
In a real molecular structure all three modes of deformation
may operate. Hence a multi-mode analytical model for the
networks where hinging, flexure and stretching act concur-
and we have
rently is developed.
s = s, (43) In the derivations that follow we once again assume unit-
thickness in the z direction.
where s, is the active length (i.e. benzene ring junction) of the The total incremental change dX in the unit-cell length X
diagonal arm. From eqn. (39), (42) and (43) we have due to an incremental change in the applied load for the con-
ds, = (Y cos 8/2K,) do, (44) current model is simply given by the sum of the incremental
changes due to each of the individual modes of deformation,
Hence the increment of strain in the x direction is, from eqn. i.e.
(W, (301, (33) and (44)
dX = dXh + dX, + dX, (59)
ds, = ( Y COS’ 8/K,X) do, (45)
where dX,, dX, and dX, are the incremental changes in X in
and from eqn. (26) and (45) the Young’s modulus in this case the hinging, stretching and flexure models, respectively. Simi-
is larly, the change in Y is given by
dY = dYh + dY, + dY, (60)
Consider the case of an x-directed load. We have for the
Now consider a load applied in the y direction. In this case
flexure model :’
the vertical ribs also extend, i.e. dX, = (YZ’ sin’ 8/K,) do, (61)
dSh # 0 (47) dY, = -(YZ2 sin 8 cos B/K,) do, (62)
J. CHEM. SOC. FARADAY TRANS., 1995, VOL. 91 2675

In the case of the hinging model we find from eqn. (14), (15) and, therefore
and (25)
dxh = (Y1’ Sin’ 8/Kh) da, (63)
dY, = -(Y1’ Sin 8 COS 8/Kh) da, (64)
For the stretching model eqn. (36), (37) and (44)yield 3. Results
dX, = ( Y COS’ 8 / K s )do, (65) 3.1 Molecular Model Data
dY, = (Y sin 8 cos O/KJ da, (66) Molecular model calculations were performed on a total of
12 structures with n = 1 or 2, 2 < rn d 8 and 8 = +3W. For
Substituting eqn. (61), (63) and (65) into (59) gives the total
each structure, energy minimisations were performed for the
change in X :
undeformed (zero stress) configuration and with uniaxial
dX = Y[(sin’ e/Khf) + (cos’ 8/K,)] do, (67) stresses of k0.5 GPa applied in each of the x and y direc-
tions. Previous calculations on a wider range of stress
where Khf is a function of the hinging and flexure force con-
increments have shown this to give accurate results.’ The
stants, defined by
Poisson’s ratio and Young’s modulus data were then calcu-
Khfl = 12(K, + K , ’) (68) lated following the procedure outlined in Section 2.1. In most
Similarly, from eqn. (60), (62), (64) and (66) we have cases different values of each elastic constant are realised for
tension and compression. This has been discussed elsewhere,’
dY = Y sin 6 cos 0(Ks-’ - Khfl) da, (69) where strain-dependent effects have been considered.
Eqn. (16) (and the de, equivalent), (17), (26), (68) and (69) give However, in this paper we are concerned only with the elastic
a Poisson’s ratio v,, of moduli of the undeformed state (although the molecular and
analytical models can be used to calculate strain-dependent
sin 8 cos O[(Ks/Khf)
- 11 -
X mechanical properties) and so the average of the values due
(70)
vxy =
[(K$KM)sin2 8 + 60s’ 01 Y to tension and compression will be compared to the analyti-
cal model calculations.
and a Young’s modulus E x of In Table 1 we present the average values of the elastic
moduli calculated from the molecular model for each struc-
ture, with an error estimated from the difference between
tension and compression. We also show the calculated unde-
Eqn. (70) illustrates that the sign of vxy is now dependent on formed unit-cell lengths for each structure. For reasons to
the geometry (0) and the force constants ratio ( K J K h f ) ,e.g. become apparent later, the data due to loading in the x direc-
negative and positive values of v,, are realised for 8 positive tion for the (2,rn)-reflexyne structures were not averaged (i.e.
with KJK,, < 1 and >1, respectively. That is, the sign and tension and compression in the x direction were considered
magnitude of the Poisson’s ratio is both geometry- and separately for these structures). The v,, E , us. vyx E x data,
mechanism-dependent . evaluated from the average elastic moduli in Table 1, are
Similarly, for a y-directed load we have plotted in Fig. 3. With the exception of the data point corre-
sponding to compressive loading in the x-direction for (2,5)-
reflexyne, the data are scattered about the equality line within
the calculated uncertainties. Hence the molecular model data
satisfy the requirement of a symmetric stiffness matrix [eqn.
(lo)]. Furthermore, the data are grouped into four clusters
with each cluster characterised by the value of n and 8 [i.e. (n,
Eqn. (70)-(73) yield 8) = (1, +30”), (1, -30”), (2, +30”) and (2, -30”)]. This will
Vxy Ey = VyxEx
be discussed later.
It is readily shown that the data in Table 1 also satisfy the
K , sin 8 cos 8[(:(K$Khf) - 11 condition of a positive definite strain energy for static equi-
-
-
(74)
+
[(K$Khf)sin’ 8 COS’ el librium [eqn. (1l)].
x [(K,/K&OS’ 8 + 2 + sin’ el
and 3.2 Comparison of Analytical Models with Molecular Model
I v,, I -= (EX/E,)”’ (75) The concurrent analytical model Poisson’s ratio expressions
can be rearranged to give the force constants ratio K J K ,
The expressions for the elastic moduli involve the unit-cell required to fit the analytical model to the molecular model
dimensions X and Y which are given by eqn. (4) and (9, data. From eqn. (70) and (72) we have
respectively, and are dependent on the honeycomb angle 8
and the cell-wall lengths h and 1. In order to relate X and Y Ks/Khf = [(vxy/tan exy/x) -k l1/C1 - ‘xy tan e(y/x)l (79)
to the molecular networks it is necessary to determine h and 1 and
in terms of the molecular dimensions. We define A and B to
be the acetylene and benzene junction ‘lengths’ defined in [(2 + sin’ 8)v,, + (Y/x)sin 8 cos 81
(80)
Fig. 2(a). A is simply the length of one acetylene link. B/2 is
KJKhf =
[(Y/X)sin e cos e - cosz ev,,]
the sum of the ‘radius’ of the benzene ring and :he length of For each structure four values of KJK,, , required to fit the
that part of the connecting bond not associated with an concurrent analytical model Poisson7s ratio data to the
acetylene link [i.e. where most of the stretching is found to molecular model data, were evaluated using eqn. (79) and
occur, see Section 2.2.3 and Fig. 2(a)]. Hence (SO), corresponding to tensile and compressive uniaxial
loading in each of the x and y directions. With the notable
l=B+nA (76) exception of tensile loading in the x direction for the (2,m)-
h=B+rnA (77) reflexyne structures, the values of K J K , were found to be
2676 J. CHEM. SOC. FARADAY TRANS., 1995, VOL. 91

TaMe 1 Average Poisson's ratios (v,, and vyx)and Young's moduli [ E x and E, (GPa)] calculated from the molecular model calculations.
Undeformed unit-cell lengths Xoand Yo(A) are also presented

structure vxY VYX EX EY XO YO

(12)-flexyne 0.46 [6] 0.696 16) 75 c41 120 [lo] 11.4709 24.6685
( 1,4)-flexyne 0.32 [3] 0.9 [l] 56.2 [4] 160 [lo] 11.4735 34.2045
(1,6)-flexyne 0.24 [3] 0.99 [4] 45 c13 220 [40] 11.4671 43.7892
(2,2)-flexyne 0.85 [2] 0.84 [S] 30 C5l 30 c31 15.6314 27.0193
(2,4)-flexyne 0.60[4] 1.11 [S] 23 c21 42 c41 15.5975 36.6377
(2,6)-flex yne 0.47 [2] 1.33 [4] 19 c41 55 c41 15.6375 46.1 108
( 1,4)-reflexyne -0.29 [8] -0.29 [2] 124 [63 110 [lo] 11.7017 21.3666
(1,5)-reflexyne -0.29 [S] -0.386 [2] 95 c11 116 [6] 11.7102 26.1698
( 1,6)-reflexyne -0.22 [S] -0.42 [3] 84 c41 140 [lo] 11.7151 30.9691
(2,5)-reflexyne -0.69" -0.53 [4] 57.4" 21 c41 15.8858 23.8009
- 1.04' 44.7'
(2,6)-reflexyne -0.72" -0.70 [3] 35.8" 31 c31 15.9271 28.6706
-0.88' 35.8*
(2,8)-reflexyne -0.50" -0.902 [2] 31.4" 40 ClOl 15.9098 38.1895
-0.Mb 31.5'

Numbers in square brackets are uncertainties in least significant figures. In the case of the (2,m)-reflexyne structures under loading in the x
direction actual values calculated due to compression' and tensionb are quoted (see text).

dependent on the number of acetylene links in the diagonal
branches (n)and on the sign of the honeycomb angle (O), i.e.
conventional or re-entrant honeycomb, but independent of
the number of acetylene links in the vertical branches (m).For
the (2,mkreflexyne structures the values of K J K , obtained
due to tensile loading in the x direction were found to be
significantly different to the values obtained from the other
loading conditions.
The mean values of KJK, for each particular combination
of n and 8 are given in Table 2. From the molecular model
calculations 8 was found to be ca. + 30 and cu. - 30" for the
conventional and re-entrant honeycomb geometries, respec-
tively, hence these are the values used in Table 2. With the
exception of the data for the (2,m)-reflexyne structures under
tensile loading in the x direction, the errors associated with
each KJK, value in Table 2 were taken to be the standard
deviation in the mean of the 12 values for each (n, 0) com-
bination since this dominates over the errors inherent in the
molecular modelling package. In the case of the data for the
(2,m)-reflexyne structures under tension in the x direction,
the uncertainty associated with KJK, was calculated by
adding the fractional errors in K , and K, (see later) in quad-
80
-60
-60
" I
rature since the smaller sample of K$K, values involved in
this case (three values instead of 12) make a statistical treat-
ment inappropriate.
In all structures K J K , > 1 in Table 2, indicating that
hinging and/or flexure dominate over stretching. Stretching is
most important for the (1,m)-reflexyne structures since they
have the lowest values of K J K , .
The concurrent analytical model Poisson's ratio calcu-
lations for the structures modelled in the molecular model
calculations are shown in Table 3. The calculations used the
library of mean K J K , values in Table 2 and the actual unit-
cell lengths determined for the undeformed structures in the
molecular model calculations (see Table 1). Also shown in
Table 3 are the hinging, flexure and stretching analytical
model Poisson's ratio calculations, employing the molecular
model unit-cell lengths. As noted in Section 2.2.2 the hinging
and flexure model Poisson's ratio expressions are identical
and so these models yield the same numerical values.
It is clear from Tables 1 and 3 that the stretching model
consistently predicts the incorrect sign of Poisson's ratio. The
hinging and flexure models predict the correct sign but con-
sistently overestimate the magnitude of the Poisson's ratio
(by as much as a factor of three in some cases), indicating
that some other mechanism (e.g. stretching) is necessary to
reproduce the molecular model data. The concurrent analyti-
cal model us. molecular model Poisson's ratios for loading in
the x and y directions are shown in Fig. 4(a) and (b), respec-
tively. The excellent scatter of the concurrent model calcu-
lations about the equality line in Fig. +a) and (b) confirms
that the deformation of these structures is principally due to
a combination of bond hinging, flexure and stretching. It also
indicates that the concurrent model and associated force con-
stants library may be used to predict the Poisson's ratios of
other (n,m)-flexyne and (n,m)-reflexyne structures, and other
related structures built from similar benzene ring and acety-
lene link subunits (see later).
The values of K , in the concurrent model can be found by
fitting the Young's moduli expressions [eqn. (71) and (7311 to
the molecular model Young's moduli in Table 1, using the
corresponding K J K M values from the fit to the Poisson's

-40
I

-20
I
0
I

20
I

40
I

60
I
80
ratio data. K, can then simply be evaluated from the values
vwEV IGPa
of K$K, and K, thus obtained. This was done for each
loading condition for all the structures considered in the
F'ig. 3 v,, E , 0s. vyxEx from the molecular model calculations. molecular model calculations. Following the same procedure
Uncertainties are calculated from the deviation of the elastic moduli
in tension and compression from the average value. 0, (1,m)-flexyne; used for the KJK, data, the mean values of K, and K , for
A, (2,m)-flexyne, 0, (2,mbreflexyne (tension in x direction); V,(2,m)- each (no)combination were evaluated and are also presented
reflexyne (compression in x direction) and 0 ,(1,m)-reflexyne. in Table 2.
J. CHEM. SOC. FARADAY TRANS., 1995, VOL. 91 2677

Table 2 Mean force constants and B / A [see text and Fig. %a)] data library for concurrent analytical model evaluated from parametric fits to
molecular model Poisson's ratio and Young's modulus data for (n,m)-flexyne and (n,m)-reflexyne

1 + 30 5.6 C1.21 358 [56] 65.1 C9.11 1.776 [0.017]

1 - 30 2.83 C0.57) 281 [57] 101 [20] 1.776 [0.017]
2 + 30 20.7 [6.8] 320 [llo] 15.6 C1.91 1.776 [0.017]
2" - 30 8.9 [1.9] 173 [58] 19.1 C4.81 1.776 [0.017]
2b - 30 37.7 C9.41 697 [41] 19.1 C4.8) 1.776 [0.017]

" Values for loads applied in y direction or compressive loads in x direction. Values for tensile load applied in x direction. Numbers in square
brackets are the estimated standard deviation in the mean value.

Table 3 Analytical model Poisson's ratio calculations for (n,m)-flexyne and (n,m)-reflexyne structures

concurrent stretching hingindflexure

structure VXY VYX VXY vYx VXY VYX

(1,2)-flexyne 0.43 [ 5 ] 0.66 [8] -0.27 -0.41 0.81 1.24

(1,4)-flexyne 0.31 [4] 0.9 [l] -0.19 -0.57 0.58 1.72
(1,6)-flexyne 0.24 [3] 1.2 [2] -0.15 -0.73 0.45 2.20
(2,2)-flexyne 0.83 [5] 0.83 [5] -0.33 -0.33 1.oo 1.oo
(2,4)-flexyne 0.61 [4] 1.13 [7] -0.25 -0.45 0.74 1.36
(2,6)-flexyne 0.49 [3] 1.42 [9] -0.20 -0.57 0.59 1.70
( 1,4)-reflexyne -0.30 [6] -0.33 [7] 0.32 0.35 -0.95 - 1.05
( 1,5)-reflexyne -0.24 [5] -0.41 [9] 0.26 0.43 -0.78 - 1.29
(1,6)-reflexyne -0.21 [4] -0.5 [l] 0.22 0.5 1 -0.66 - 1.53
(2,5)-reflexyne -0.77 [6]" -0.58 [5] 0.39 0.29 - 1.16 -0.87
-1.04 [31b
(2,6)-reflexy ne -0.64 [5]" -0.69 [6] 0.32 0.35 -0.96 - 1.04
-0.87 [2]'
(2,8)-reflexyne -0.48 [4]" -0.92 [8] 0.24 0.46 -0.72 - 1.39
-0.65 [21b

Concurrent model calculations employ K J K , library of Table 2. Molecular model Y / X values were employed in all analytical model calcu-
'
lations. " Compression in x direction; tension in x direction. Numbers in square brackets are the estimated uncertainty in the least significant
figures.

Considering, first, the data for (1, m)-flexyne, (1,m)-reflexyne rent model contains the essential features of the deformation
and (2,m)-flexyne we see that, within the calculated standard of these structures. (The Young's modulus expressions for the
deviation from the mean values, K, remains constant at single-mode analytical models cannot be fitted to the molecu-
K, x 320 N m-'. However, the value of K,, varies greatly lar model data using the same value of the appropriate force
between these three (no) combinations, leading to the trends constant for loading in both the x and y directions.) Further-
observed in the KJK, data. Hence, in these cases changing more, the agreement in the data indicates that more complex
the geometry primarily affects the hinging/flexure force coeffi- networks can be assembled and their properties predicted to
cient K, . This will be discussed later. reasonable accuracy using the concurrent model and the
In the case of the (2,m)-reflexyne data, K , showed similar associated force constants library (Table 2), see later.
variation about a mean value for all loading conditions, as The force constants trends also explain why the molecular
observed for the other structures. Hence in Table 2 we show
the mean K, value (with associated standard deviation) for Table 4 Concurrent analytical model Young's modulus calculations
all loading conditions. K , , however, showed a significant for (n,m)-flexyne and (n,m)-reflexyne, employing the K$KM and K ,
library of Table 2
increase when a tensile load was applied in the x direction
compared with the other loading conditions, for the (2,m)- structure EJGPa Ey/GPa
reflexyne networks. Hence two values of K, are given in
Table 2 for (no)= (2, -30"); these being the mean of the K, (1,2)-flexyne 77 c11 119 [2]
values for a tensile load in the x direction, and the mean of ( 1,4)-flexyne 56 c11 166 [3]
(1,6)-flexyne 43.6 [0.8] 212 [4]
the K, values for all other loading conditions. (2,2)-flexyne 31 c21 31 C2I
In Table 4 we present the concurrent analytical model (2,4)-flex yne 23 c11 42 c31
Young's modulus calculations for the networks considered in (2,6)-flex yne 18 c11 53 c31
this paper, using eqn. (71) and (73) employing the Y/Xratio (1,4)-reflexyne 110 [lo] 120 [lo]
from the molecular model calculations and KJK, and K, (1,5)-reflexyne 86 c91 140 [20]
values from Table 2. These are plotted against the molecular (1,6)-reflexyne 73 C81 170 [20]
(2,5)-reflexyne 39 171" 29 c51
model calculations (Table 1) in Fig. 5(a) and (b)for loading in 46 [8Ib
the x and y directions, respectively. Once again, excellent (2,6)-reflexyne 32 [6]" 35 C61
agreement is achieved in both cases. One might expect rea- 38 [7Ib
sonable agreement since the force constants used in the con- (2,8)-reflexyne 24 [4]" 47 C81
current model calculations were the mean of those derived 29 [5Ib
from a fit to the molecular model data for each structure.
Molecular model Y / X values were used in the calculations for all
However, the fact that good agreement is achieved in all the '
structures. " Compression in x direction; tension in x direction.
Poisson's ratio and Young's modulus data for uniaxial Numbers in square brackets are the estimated uncertainty in each
loading in both principal directions indicates that the concur- value.
 2678 J. CHEM. SOC. FARADAY TRANS., 1995, VOL. 91

1.2 and E, remain the same as for the simpler structures, i.e. eqn.
(70) and (71), respectively. However, the unit-cell lengths ratio
0.8 to be employed in these equations for the stabilised structures
is now
0.4 Y - (2[(B/A)+ rn]/[(B/A)+ n]
- + sin 01
-
n - (81)
;0.0
a
._
-
X cos e
a For the case of loading in the y direction it is easily shown
s2 -0.4 that
sin 8 cos 8[(K$KM)- 13 Y
(82)
-0.8
vyx = [ ( K , / K ~ ) c oes ~+ 4 + sin2 el
X
-1.2 E, = Ks Y
-
+ + sin2 81 x
[ ( K ~ / K ~ ) c8o s 4
~ (83)
- 1.6 where Y / X is again given by eqn. (81). When n = rn the
- 1.6 - 1 . 2 -0.8 -0.4 0.0 0.4 0.8 1.2
vxy( m oI ecu I ar m odeI I ing) expression for the unit-cell lengths ratio [eqn. (8 l)] becomes
trivial, depending only on 8. In order to predict the elastic
1.6 constants for the more complex networks when n is not equal
to rn, however, it is necessary to know the value of B / A to be
1.2 used in eqn. (81). This can be found by returning to the
simpler structures modelled earlier. For each structure mod-
0.8 elled in Table 1, B / A was calculated by substituting the unde-
-
n
formed unit-cell lengths, given in Table 1, from the molecular
.- 0.4 model calculations into eqn. (78). n, rn and 8 are known for
s
- each structure, enabling B / A to be determined. Unlike the
a
5 0.0 force constants, no discernible trend in B / A with n or 8 was
v
Si observed, the value remaining approximately constant for all
-0.4
structures. A mean value of B/A = 1.776 _+ 0.017 was found
for the simpler structures modelled in this paper, with the
calculated uncertainty being the estimated standard deviation
-0.8
in the mean value. This would be the value of B / A to use in
the calculations for the stabilised structures when n is not
-0.8 -0.4 0 . 0 0 . 4 0.8 1.2 1.6
equal to m. However, in this paper we will only consider the
vyx(molecular modelling) case of stabilised structures with n = rn, which leads to a can-
cellation of the B / A terms in eqn. (81).
Fig. 4 (a) Concurrent analytical model vxy us. molecular model vxy Table 5 contains the unit-cell lengths ratio, Poisson's ratio
for (n,m)-flexyne and (n,m)-reflexyne structures. (b) Concurrent ana- and Young's modulus calculations from the concurrent ana-
lytical model vyx us. molecular model vyx for (n,rn)-flexyne and (n,m)-
reflexyne. lytical model for some stabilised structures. No uncertainty in
Y / X is quoted since there is no dependence on B / A in these
model vxyE , us. vyx E x data are grouped into four sets accord- cases. The corresponding molecular model data are also
ing to their (n,8) combination (Fig. 3). From eqn. (74) we see shown in Table 5. It is important to note that the concurrent
~ )dependent on K,/K,, K, and 8 only,
that ~ , , E , ( = v , ~ E is analytical model calculations use the force constants estab-
and is independent of Y / X . Therefore, the concurrent analyti- lished from the earlier comparison between the concurrent
cal model predicts the value of v x y E y to be constant for a analytical model and molecular model calculations on the
given (n,O) combination, irrespective of m, as observed in the simpler (n,rn)-flexyne and (n,m)-reflexyne structures (Table 2).
molecular model. There has been no fitting of the concurrent model data to the
molecular model data for the stabilised structures in Table 5.
The values of Y / X from the molecular model are predicted
3.3 Model Predictions for More Complex Networks
to two decimal places by the concurrent analytical model.
Attempts to synthesise networks of the type so far discussed This level of agreement is easily acceptable when considering
indicate that one or more benzene rings need to be incorpor- the errors associated with the force constants to be employed
ated into the vertical branches in order for the networks to in the concurrent analytical model (Table 2). Excellent agree-
remain in a stable configuration.2'*22The force constants ment is found in the elastic moduli for each structure, con-
trends established in this paper indicate that the mechanical firming that the concurrent analytical model and its associ-
properties of these more complex networks can be modelled ated force constants library can be used to model more
to a reasonable degree of accuracy by the concurrent analyti- complex molecular networks accurately. This enables more
cal model. For example, we consider here the addition of an efficient and effective use of the computer intensive molecular
extra benzene ring, midway along the vertical arms [referred modelling package for the modelling of molecular networks
to here as stabilised (n,m)-flexyne and stabilised (n,m)- with specific mechanical properties.
reflexyne, where m is now the number of acetylene links In order to test the validity of using a fixed value of B / A to
between a junction benzene ring and the non-junction calculate the unit-cell lengths ratio from the concurrent ana-
benzene ring in the vertical branches]. lytical model when n # rn, the undeformed structure for sta-
Incorporating an extra benzene ring into the vertical poly- bilised (2,l)-reflexyne was determined using the molecular
acetylene branches introduces an extra component of stretch- model program. A value of Y / X = 1.112 was found for this
ing into the vertical branches. Following the procedure structure. Employing the mean value of B / A = 1.776
outlined in Section 2.3, the elastic constants of stabilised net- (established from the earlier work on the simple structures-
works can be derived. It is found that the expressions for vXy see above) in the concurrent analytical model yields
J. CHEM. SOC. FARADAY TRANS., 1995, VOL. 91 2679

their geometry and the behaviour of the sub-unit. The

network can be tailored by modifying the sub-unit, or by
altering the connectivity of the network. Most interestingly,
the force constants required to fit the concurrent analytical
model data to the molecular model data were found to be
directly related to the particular geometry of the network (see
Table 2). This enabled accurate predictions to be made of the
mechanical properties of the more complex stabilised struc-
tures using a concurrent analytical model, without recourse
L to a full molecular modelling computation. We now consider
why these force constants trends are observed.
So far we have used the analytical model Young's modulus
expressions for unit-thickness in the z direction. Hence one
possible source of deviation in the force constants is that the
value of the unit-cell length perpendicular to the x-y plane,
i.e. Z, may be different for different (n,@ combinations.
However, the molecular model calculations yield essentially
I I I I I I I
planar structures [slight deviation from planarity being
0 20 40 60 80 100 120 140
Edmolecular modelling)/GPa observed only for (2,5)-reflexyne], having similar values of 2
for all 12 simple structures, with a mean value of
Z = 3.572 & 0.015 A. Hence the observed force constants
trends cannot be attributed to changes in 2.
In the case of the (n,rn)-flexyne structures, increasing n from
m 1 to 2 reduces the mean value of the hingindflexure force
constant, K,, from 65.1 & 9.1 to 15.6 & 1.9 N m-' (see
Q)
Table 2), i.e. by a factor of 4.2 & 0.8. For (n,rn)-reflexyne K ,
160
decreases from 101 & 20 to 19.1 & 4.8 N m-l as n increases
from 1 to 2, i.e. a reduction in K, by a factor of 5.3 & 1.7,
0 which is consistent with the value obtained from the (n,rn)-
- flexyne data. From the definition of K,, [eqn. (68)] we see
m
m that K, is inversely proportional to the square of the length
v
lui 80 - of the diagonal honeycomb arm (i.e. K, proportional to l-').
However, for the molecular structures considered here, this
40 -
length should be modified to be the length of the diagonal
acetylene branches since flexure of the benzene rings is
unlikely to occur and hinging occurs about the benzene ring
0 40 80 120 160 200 240 280
Ej(molecular modelling)/GPa
(rather than hinging of the benzene ring itself). Hence doub-
ling the length of the diagonal acetylene branch by increasing
Fig. 5 (a) Concurrent analytical model E , us. molecular model E x n from 1 to 2 should yield a decrease in K, by a factor of 4
data for (n,rn)-flexyne and (n,rn)-reflexyne structures. (b) Concurrent (= 2*). In fact a factor of greater than 4 would be expected
analytical model E y us, molecular model E y data for (n,rn)-flexyne and since K , is also inversely proportional to the length of the
(n,rn)-reflexyne structures. diagonal acetylene branches [see eqn. (9) and (68)].
The effect of varying n on the value of K , appears to be
Y/X = 1.120 f 0.003. Hence the level of agreement in Y / X negligible, within the accuracy of the data of Table 2. K, =
for n # rn is of the same order as that for n = rn (see Table 5). 358 f 56 and 320 & 110 N m-' for (1,rn)-flexyne and (2,rn)-
flexyne, respectively, and hence K , remains constant as n
4. Discussion
increases from 1 to 2 within the calculated uncertainties. A
We have modelled 2D molecular networks based on conven- similar comparison cannot be made for the (n,rn)-reflexyne
tional and re-entrant honeycomb cells, using molecular mod- structures due to the anisotropy observed in K , for (2,rn)-
elling and analytical modelling techniques. The mechanical reflexyne under tensile and compressive loading in the x
properties of these molecular networks are determined by direction.

Table 5 Concurrent analytical model and molecular model undeformed unit-cell lengths ratio, Poisson's ratio and Young's modulus data for
stabilised (n,rn)-flexyne and (n,rn)-reflexyne structures

model structure y/x VXY vYx EJGPa E,/GPa

concurrent stabilised (1,l)-flexyne 2.8868 0.32 [4] 0.7 [l] 58 c11 122 [6]
analytical stabilised (1,l)-reflexyne 1.7321 -0.31 [7] -0.22 [ S ] 110 [lo] 80 [lo]
stabilised (2,2)-flexyne 2.8868 0.50 [3] 1.3 [l] 19 c11 47 c41
stabilised (2,2)-reflexyne 1.7321 -0.66 [6]" -0.54 [6] 34 [6]" 27 [GI
-0.90 [2]b 40 [7ib
molecular stabilised (1,l)-flexyne 2.8896 0.34 [l] 0.74 [l] 59 c21 123 [4]
stabilised (1,l)-reflexyne 1.7377 -0.36 [l] -0.24 [l] 120 [20] 90 [lo]
stabilised (2,2)-flexyne 2.8933 0.49 [4] 1.26 [S] 19 c31 49 c71
stabilised (2,2)-reflexyne 1.7335 - 0.85" -0.67 [S] 31" 27 P I
-0.82' 43'

" Compression in x direction; ' tension in x direction, Numbers in square brackets are estimated uncertainty in least significant figures.
2680
Considering now the effect of changing 8 from positive to
negative, we see that for n = 1 an increase in K, from
65.1 f 9.1 to 101 k 20 N m-l is observed for 0 = +30 and
-3o", respectively, which is an increase by a factor of
1.55 k 0.38. A corresponding decrease in K, (from K , = 358
56 to 281 & 57 N m-') by a factor of 0.78 f 0.20 is also
observed. In the case of the n = 2 structures K, increases
from 15.6 & 1.9 to 19.1 +_ 4.8 N m-' as 8 changes from +30
to -3O", i.e. K , increases by a factor of 1.22 f 0.34 which
agrees within the calculated uncertainties with the value
obtained above for the n = 1 structures. Again the anisotropy
in K, for the (2,m)-reflexyne structures renders a comparison
of the K, values meaningless in this case.
The increase in K, as 8 changes from +30 to -30" indi-
cates that hinging or flexure becomes more difficult to
achieve for the re-entrant structure. It is reasonable to
assume that the increase in K, is due to hinging since flexure
occurs along the acetylene arms whereas hinging occurs at
the benzene junction and it is the connectivity at the benzene
junction that is being investigated here. The reason for the
increase in K, is not clear. Naively, we may expect (n,m)-
flexyne structures to hinge more easily (and hence have a
lower K, value, as observed) than (n,m)-reflexyne structures
since there is an intermediate benzene ring carbon atom
about which to hinge in the conventional honeycomb
geometry [see Fig. 2(a) and (b)].
The slight decrease observed in K, as 8 changes from +30
to -30" for n = 1 structures may be due to the influence of
non-bonded interactions. It is known, for example, that the
value of the stretching force constant of the C-C bond is
reduced by a factor of ca. 5 for diamond owing to the pres-
ence of non-bonded interaction^.^^ It is clear from Fig. 2(a)
and (b) that the bonds connecting the polyacetylene branches
to the junction benzene rings are substantially more crowded
(leading to an increased number of non-bonded interactions)
for the (n,m)-reflexyne structures than for the (n,m)-flexyne
structures. These bonds have been found to make the largest
contribution to the stretching mode of deformation (see
earlier) and hence a reduction in K, would be expected for
(n,m)-reflexyne structures.
In the case of (2,m)-reflexyne, however, the value of K, was
found to be dependent on the loading condition (see Table 2).
This is probably due to the concurrent model not containing
all the deformation mechanisms in this case. It appears that
when n = 2 the benzene rings at the re-entrant junctions
become close enough to each other to interact in such a way
as to introduce other modes of deformation in the molecular
model calculations. This is consistent with the observation of
the slight deviation from planarity of the (2,5)-reflexynestruc-
ture.
5. Conclusions
The mechanical properties of 2D molecular networks, having
either positive or negative Poisson's ratios, have been calcu-
lated using molecular modelling and analytical modelling
techniques. The molecular networks are based on conven-
tional and re-entrant honeycomb sub-units and consist of
polyacetylene branches connected by benzene rings at the
junctions. An analytical model employing flexure, hinging
and stretching modes of deformation acting concurrently has
been found to produce excellent agreement with the molecu-
lar model Poisson's ratio and Young's modulus data when a
parametric fit of the force constants governing these deforma-
tion mechanisms is performed. The force constants required
to fit the concurrent analytical model to the molecular model
are found to follow certain trends, being dependent on the
number of acetylene links in the diagonal branches (n)and on
J. CHEM. SOC. FARADAY TRANS., 1995, VOL. 91
whether the honeycomb angle (0) is positive or negative. The
force constants trends enable a library of force constants to
be compiled for different (n,0)combinations, which can then
be used to predict the properties of other molecular networks
of the type considered here. The concurrent analytical model
can be extended to cover more complex variants of these net-
works. We developed the concurrent analytical model to
describe networks having an extra benzene ring in the verti-
cal branches of the sub-units. The mechanical properties of
these stabilised structures predicted from the concurrent ana-
lytical model and the associated force constants library
(established from the earlier work on the simpler structures)
were found to be in good agreement with the molecular
model calculations. In principle the properties of even more
complex networks (2D and 3D) can be modelled, as a first
approximation, using this concurrent analytical model
approach. This would enable more efficient use of molecular
modelling programs, providing first-order estimates of
properties having first obtained the properties of the appro-
priate sub-units. Hence, better selection of the network struc-
tures to be modelled using full molecular modelling
techniques should be possible.
The authors acknowledge the support of the Engineering and
Physical Sciences Research Council of the United Kingdom
and of Oxford Materials Ltd. K.E.E. currently holds an
EPSRC Advanced Fellowship.
References
1 K. E. Evans, M. A. Nkansah, I. J. Hutchinson and S. C. Rogers,
Nature (London), 1991,353, 124.
2 M. A. Nkansah, K. E. Evans and I. J. Hutchinson, Mod. Sim.
Mat. Sci. Eng., 1994,2, 337.
3 R. H. Baughman and D. S. Galvao, Nature (London), 1993, 365,
735.
4 K. W. Wojciechowski and A. C. Branka, Phys. Rev. A , 1989,40,
7222.
5 J . S. Moore and J. Zhang, Angew. Chem. Int. Ed. Engl., 1992,31,
922.
6 J . S. Moore, Nature (London), 1993,365,690.
7 K. E. Evans, Endeavour, 1991,15,170.
8 R. Lakes, A h . Muter., 1993,5,293.
9 A. W. Lipsett and A. I. Beltzer, J. Acoust. SOC.Am., 1988, 84,
2179.
10 L. J. Gibson and M. F. Ashby, in Cellular Solids: Structure and
Properties, Pergamon Press, Oxford, 1988, ch. 4.
11 K. E. Evans, Compos. Struct., 1990, 17,95.
12 K. L. Alderson and K. E. Evans, Polymer, 1992,33,4435.
13 K. E. Evans and K. L. Alderson, J. Muter. Sci. Lett., 1992, 11,
1721.
14 U. Burkert and N. L. Allinger, in Molecular Mechanics, Amer-
ican Chemical Society, Washington, 1982, ch. 1.
15 N. W. Ashcroft and N. D. Mermin, in Solid State Physics, Holt,
Rinehart and Winston, New York, 1976, p. 443.
16 POLYGRAF, Molecular Simulations Inc., 199 South Los
Robles Avenue, Suite 540, Pasadena, California 91 101, USA.
17 S. L. Mayo, B. D. Olefson and W. A. Goddard 111, J. Phys.
Chem., 1990,94,8897.
18 R. J . Roark and W. C. Young, in Formulas for Stress and Strain,
McGraw-Hill, New York, 5th edn., 1976.
19 B. M. Lempriere, Am. Inst. Aeronaut. Astronaut. J., 1968, 6,
2226.
20 J. E. Shigley, in Applied Mechanics of Materials, McGraw-Hill,
New York, 1976.
21 D. Leigh, UMIST, personal communication.
22 J. S. Moore, University of Illinois, personal communication.
23 M. O'Keeffe and B. G. Hyde, in Structure and Bonding in Crys-
tals, ed. M. O'Keeffe and A. Navrotsky, Academic Press, 1981,
vol. I, p, 247.
Paper 5/013251; Received 3rd March, 1995