Professional Documents
Culture Documents
l k
2
@
l
u
l
2l k
2
l
02
@
i
@
l
u
l
@
i
@
j
u
j
ll
2
2
@
2
l
u
i
@
2
l
u
i
@
i
@
l
u
l
@
i
@
j
u
j
6
Two new coupling constants (in addition to the Lame parameters) now appear namely l
0
and l. Both
have units of length. From the energy density expression in Eq. (6), via appeal to the EulerLagrange equa-
tions, the Navier-like governing static equation can be derived as well as the response quantities (i.e. stres-
ses). The balance laws can then be written as
@
i
m
ij
e
jkl
r
a
kl
r
ij
r
a
ij
r
s
ij
d
ij
@
k
s
k
r
ji;j
0
7ac
Here r
s
is the symmetric part of the stress tensor (and thus coincides with the classical denition of stress
typically adopted). The remaining quantities, s and m, denote moment-stress like quantities respectively
characterizing resistance to strain gradients and rotational gradients. The physical stress tensor (r) is then
dened from these and yields (supercially) a balance law (7c) that appears supercially the same as in clas-
sical elasticity (Kleinert, 1989). e
ijk
here stands for the permutation symbol.
For materials that obey linear isotropic constitutive relations and subject to the usual symmetry and
invariance constraints, we can dene the following constitutive relations (Kleinert, 1989):
r
s
ij
2lu
i;j
kd
ij
u
l;l
s
i
2l kl
02
@
i
u
l;l
m
ij
4ll
2
@
i
x
j
8ac
For a comparison of the strain gradient model that we employ in the present work and others that are
prevalent in the literature, please see Appendix A.
We now tackle the inclusion problem. Noting that the transformation strain is only non-zero within the
inclusion domain, we can write the bulk-constitutive law for the inclusion-matrix as follows:
m
ij
4ll
2
@
i
x
j
x
j
H
e
ijk
r
a
jk
@
j
m
ij
r
ij
r
a
ij
2lu
i;j
u
i;j
H kd
ij
u
l;l
u
l;l
H d
ij
2l kl
02
r
2
u
l;l
u
l;l
H
9ac
X. Zhang, P. Sharma / International Journal of Solids and Structures 42 (2005) 38333851 3837
Here H is Heaviside equation, dened as Hx
1 x 2 X
0 x 62 X
_
. Using the equilibrium equations and the
constitutive relations, we can obtain the Navier-like governing equation for this problem i.e.
l@
2
u
i
l k@
i
@
l
u
l
2l kl
02
@
2
@
i
@
l
u
l
ll
2
@
4
u
i
@
2
@
i
@
l
u
l
@
k
P
T
ik
H 10
Here the symbol P
T
stands for the eigenstress corresponding to the eigenstrain, i.e. C
jlmn
e
mn
.
Clearly, identical to the classical case (Mura, 1987), the derivative of the eigenstrain acts as a body force.
Noting that the derivatives of the Heaviside function dened over the inclusion volume act as delta func-
tions across the inclusion surface, the displacement vector can be obtained using the Greens function (of
Eq. (12)) as
u
i
x
_
S
P
T
jl
G
ij
x x
0
dS
l
x
0
_
V
C
jlmn
e
mn
x
0
G
ij;l
x x
0
dV x
0
11
Here Gauss theorem has been used to convert the surface integral into a volume integral. Fortunately,
strain gradient Greens function has already been derived by Kleinert (1989) and is explicitly given by
G
ij
R
1
4plR
1 e
R
l
_ _
d
ij
1
4pl
@
i
@
j
R
2
l
2
1
R
1 e
R
l
_ _
_ _
1
4p2l k
@
i
@
j
R
2
l
02
1
R
1 e
R
l
0
_ _
_ _
12
where R = jx x
0
j.
Mere substitution of Eq. (14) into (13) results in:
u
i
x
1
4p
_
dS
k
x
0
1
lR
1 e
R
l
_ _
P
T
ik
P
T
jk
l
@
i
@
j
R
2
l
2
1
R
1 e
R
l
_ _
_ _
_
P
T
jk
2l k
@
i
@
j
R
2
l
02
1
R
1 e
R
l
0
_ _
_ _
_
1
l
/
;k
M
;k
P
T
ik
P
T
jk
l
@
i
@
j
w
;k
2
l
2
/
;k
M
;k
_ _
P
T
jk
2l k
@
i
@
j
w
;k
2
l
02
/
;k
M
0
;k
_ _
1
l
/
;k
M
;k
P
T
ik
P
T
jk
l
w
;ijk
2
l
2
/
;ijk
l
2
M
;ijk
_ _
P
T
jk
2l k
w
;ijk
2
l
02
/
;ijk
l
02
M
0
;ijk
_ _
13
Here we have made explicit the use of Gausss theorem to convert the surface integrals to volume inte-
grals and the displacement eld has been cast in terms of certain potentials dened below:
wx
1
4p
_
X
Rdx
0
/x
1
4p
_
X
dx
0
Mx; l
1
4p
_
X
e
R
l
R
dx
0
M
0
x; l
0
1
4p
_
X
e
R
l
0
R
dx
0
14
w(x) is the harmonic potential, /(x) is the biharmonic potential while M(x, l) and M
0
(x, l
0
) are the Yukawa
potentials with dierent coecients l and l
0
(i.e. the characteristic length scales in gradient elasticity theory).
3838 X. Zhang, P. Sharma / International Journal of Solids and Structures 42 (2005) 38333851
The rst two potentials are well known in classical potential theory (Kellogg, 1953) and the inclusion lit-
erature (see, Mura, 1987). The Yukawa potential is relatively less known and occurs in the study of
non-Newtonian gravitation (e.g. Gibbons and Whiting, 1981). Recently, it has been employed by Cheng
and He (1995, 1997) in their study of inclusions in micropolar elasticity.
For convenience we can write displacement in two parts: the classical one (already detailed by Eshelby
(1957) for ellipsoidal inclusions in terms of the harmonic and biharmonic potentials) and the second size-
dependent part which arises due to strain gradient eects. From Eq. (15), the strain gradient part can be
written as
u
GR
i
M
;k
l
d
ij
l
2
l
/
;ijk
M
;ijk
l
02
2l k
/
;ijk
M
0
;ijk
_ _
_ _
P
T
jk
15
The symmetric part of the strain, e
ij
1
2
@
j
u
i
@
i
u
j
, is then:
e
GR
il
1
2l
M
;kl
d
ij
M
;ki
d
lj
l
2
l
l
02
2l k
_ _
/
;ijkl
l
2
l
M
;ijkl
l
02
2l k
M
0
;ijkl
_ _
P
T
jk
16
Using Eshelbys convention, we can then dene the complete strain gradient Eshelby tensor to be:
S
ijkl
S
0
ijkl
S
GR
ijkl
S
0
ijkl
1
2
M
;lj
d
ik
M
;li
d
jk
M
;kj
d
il
M
;ki
d
jl
_
2l
l
2
l
l
02
2l k
_ _
/
;ijkl
l
2
l
M
;ijkl
l
02
2l k
M
0
;ijkl
_ _
k
l
M
;ij
k
l
2
l
l
02
2l k
_ _
/
;ijmm
l
2
l
M
;ijmm
l
02
2l k
M
0
;ijmm
_ _ _ _
d
kl
17
where S
0
is the classical Eshelbys tensor (known for various inclusion shapessee Mura, 1987) and S
GR
is
the gradient part.
Finally, the dilatation can be expressed as
tre
GR
il
1
l
M
;jk
l
2
l
l
02
2l k
_ _
/
;jkmm
l
2
l
M
;jkmm
l
02
2l k
M
0
;jkmm
_ _
P
T
jk
18
The present formulation was for arbitrary shaped inclusions and we specialize to the spherical shape in
the next section to obtain explicit expressions.
3. Spherical inclusions: closed form expressions
Assume a spherical inclusion of radius a embedded in an innite matrix (see Fig. 2).
Our results in the previous section were cast in terms of three potentials, the harmonic, biharmonic and
the so-called Yukawa potential. These can be written in closed form for the spherical shape (see, Mura,
1987; Kellogg, 1953; Cheng and He, 1995; Gibbons and Whiting, 1981):
wx
1
60
R
4
10a
2
R
2
15a
4
R 2 X
a
3
15
5R
a
2
R
_ _
R 62 X
_
_
19
X. Zhang, P. Sharma / International Journal of Solids and Structures 42 (2005) 38333851 3839
/x
1
6
R
2
3a
2
R 2 X
a
3
3R
2
R 62 X
_
_
20
Mx; k
k
2
k
2
k ae
a=k
shR=k
R
R 2 X
k
2
ach
a
k
ksh
a
k
_ _
e
R=k
R
R 62 X
_
_
21
Using, Eq. (19) and (21)(23), the strain gradient Eshelbys tensor for the spherical shape can be made
explicit. Although we have obtained the complete Eshelbys tensor for gradient elasticity and can thus be
employed for arbitrary eigenstrains Eq. (19), exceedingly simple expressions for the dilatation can be de-
rived; which is of frequent interest in various physical problems involving for example, thermal expansion,
lattice mismatch, phase transformations etc.
A dilatational eigenstrain implies a dilatational eigenstress P
T
ij
P
T
d
ij
. In this particular simple case, the
eigenstrain and eigenstress are related by P
T
3k 2le
T
. Some algebra and manipulations result in fol-
lowing relation for the dilatation:
tre
GR
P
T
2l k
l
0
ae
a
l
0
sh
R
l
0
R
R 2 X
P
T
2l k
ach
a
l
0
l
0
sh
a
l
0
_ _
e
R
l
0
R
R 62 X
_
_
22
Adding to it the dilatation from the well-known classical part, we obtain the total trace of the strain as
tre
e
T
3k 2l
k 2l
1 l
0
ae
a
l
0
1
l
0
sh
x
l
0
x
l
0
_ _
e
T
3k 2l
k 2l
ach
a
l
0
l
0
sh
a
l
0
_ _
1
l
0
e
x
l
0
x
l
0
_
_
23
The suitably normalized dilatational strain results are plotted in Fig. 3 as a function of position and var-
ious inclusion sizes. All the results in the present work are plotted parametrically in terms of the unknown
strain gradient constants. Please see Appendix B for more discussion on determination of the actual strain
gradient properties. The location x/a = 1 indicates the boundary of the spherical inclusion. The size eect of
Fig. 2. Spherical inclusion problem schematic.
3840 X. Zhang, P. Sharma / International Journal of Solids and Structures 42 (2005) 38333851
the strain gradient solution is clearly obvious. We note that, unlike the classical solution, the strain gradient
solution is inhomogeneous within the inclusion and asymptotically, our results converge to that of classical
elasticity (Eshelby, 1957) for large inclusion sizes. As well known, the classical result predicts zero dilatation
outside the inclusion for a dilatational eigenstrain. A further point to be noted is that our results (for the
dilatation) depend solely on l
0
and not on l. This is due to the fact that the latter is physically associated
with couple stresses or gradients of the rotation vector and (as well known), for isotropic materials, they
vanish for purely centrosymmetric loading. The parameter l
0
is associated with purely gradient eects of
Fig. 3. Strain dilatation as a function of position and inclusion size.
Fig. 4. Dilatational strain as a function of size for xed position (r = 0).
X. Zhang, P. Sharma / International Journal of Solids and Structures 42 (2005) 38333851 3841
the symmetric part of the strain and hence persists even in the highly symmetric dilatation case. In other
words, the conventional centrosymmetric isotropic micropolar elasticity theory would not dier from clas-
sical elasticity for the purely dilatational problem.
To emphasize on the size-dependency of our solution we also plot the dilatation as a function of size (for
a xed position, i.e. x = 0) in Fig. 4. We observe that while for large inclusion size, roughly >a = 7l
0
, the
strain gradient solution is indistinguishable from the classical one, the dilatation decreases signicantly be-
low this threshold.
4. The inhomogeneity problem in strain gradient elasiticty with couple stresses
The inhomogeneity problem is considerably more challenging and as alluded to earlier, unlike for the
inclusion problem, an exact general solution does not appear to be possible. Imagine that a far-eld strain,
e
1
ij
is applied to the inhomogeneity matrix system. We start of by writing the position dependent elastic
modulus in the following fashion:
C
ijkl
C
1
ijkl
DC
ijkl
Hx 24
Here, DC
ijkl
is the dierence between the elastic stiness tensors of the matrix and the inhomogeneity:
DC C
2
ijkl
C
1
ijkl
. Once again, from the equilibrium equations, $ r = 0, we obtain:
@
j
C
1
ijkl
u
k;l
_ _
@
j
HxDC
ijkl
u
k;l
_ _
0 25
This allows us to write (using approach outlined by Markov, 1979), the integral equations for the strain
in the inhomogeneity:
e
ij
x e
1
ij
x
_
X
Q
iklj
x x
0
DC
klmn
e
mn
x
0
d
3
x
0
Q
iklj
x
1
2
G
ik;lj
x G
jk;li
x
_ _
26
These integral equations, while not soluble exactly, may be tackled by a perturbation type approach. In
such a case, the rst approximation to the actual strain can be considered to be the average uniform
strain. We now proceed to evaluate this average strain from the point of view of later application to the
eective properties of composites.
We perform a volumetric averaging process over both sides of Eq. (28):
he
ij
xi e
1
ij
x
_
X
Q
iklj
x x
0
DC
klmn
e
mn
x
0
d
3
x
0
_ _
27
As a rst approximation, we assume that:
_
X
Q
iklj
x x
0
DC
klmn
e
mn
x
0
dx
0
_ _
_
X
Q
iklj
x x
0
DC
klmn
dx
0
_ _
e
mn
x
0
h i
It can be shown (see Markov, 1979, who uses this in the context of micropolar elasticity where also
strains are non-uniform) that this approximation is tantamount to adopting the rst term in a perturbation
series expansion involving the dierence in the moduli of the inhomogeneity-matrix. Implicitly or explicitly,
other researchers have also employed such an approximation where strain states are inhomogeneous e.g.
Nozaki and Taya (2001) in the context of polyhedral inhomogeneities and Sharma and Dasgupta (2002)
in the case of micropolar inhomogeneities.
3842 X. Zhang, P. Sharma / International Journal of Solids and Structures 42 (2005) 38333851
Thus we can relate/approximate the average elastic strain inside inhomogeneity to the ctitious eigen-
strain in terms of the average strain gradient Eshelby tensor: e
ij
_
S
ijkl
e
f
kl
_
$ S
ijkl
_
e
f
kl
_
.
The average modied Eshelby tensor in second strain gradient elasticity is:
S
ilmn
h i S
ilmn
h i
0
S
ilmn
h i
GR
hS
ilmn
i
0
3k
1
8l
1
15 k
1
2l
1
d
im
d
ln
3k
1
8l
1
15 k
1
2l
1
d
in
d
lm
3k
1
2l
1
15 k
1
2l
1
d
il
d
mn
hS
ilmn
i
GR
1
2l
1
M
;kl
d
ij
M
;ki
d
lj
_ _
l
2
l
1
M
;ijkl
l
02
2l
1
k
1
M
0
;ijkl
_ _
C
jkmn
_ _
1
6l
1
f d
l
d
kl
d
ij
d
ki
d
lj
_
1
15l
1
f d
l
1
15 2l
1
k
1
f d
0
l
_ _
_ _
H
ijkl
_ _
C
jkmn
1
5l
1
f d
l
2
15 2l
1
k
1
f d
0
l
_ _
_ _
C
ilmn
1
15l
1
f d
l
1
15 2l
1
k
1
f d
0
l
_ _
_ _
C
kkmn
d
il
_ _
28
Following Markov (1979) who encountered similar terms, we set
A
1
5l
1
f d
l
2
15 2l
1
k
1
f d
0
l
_ _
and B
1
15l
1
f d
l
1
15 2l
1
k
1
f d
0
l
_ _
The function f d
l
3
1d
l
d
3
l
e
d
l
d
l
chd
l
shd
l
with d
l
= a/l
Hence, more compactly,
hS
ilmn
i g
1
d
im
d
ln
g
1
d
in
d
lm
g
2
d
il
d
mn
29
where
g
1
l
1
A
3k
1
8l
1
15 k
1
2l
1
and g
2
k
1
A B 2l
1
3k
1
3k
1
2l
1
15 k
1
2l
1
Given our approximation of an average uniform eigenstrain in the inhomogeneity, we can now em-
ploy Eshelbys equivalent inclusion concept i.e. C
1
ijkl
e
kl
h i e
1
kl
_
e
f
kl
_ _ _
C
2
ijkl
e
kl
h i e
1
kl
_ _ _
. In other
words:
e
f
jk
H
1
d
ji
d
kl
H
1
d
jl
d
ki
H
2
d
jk
d
il
_ _
e
1
il
H
1
Dl
2 2g
1
Dl l
1
H
2
g
2
DlDK
k
1
DKK
1
Dk
2
2g
1
Dl l
1
2g
1
DK 3DKg
2
K
1
30
The equations developed above allow an approximate evaluation of the average strain within an inho-
mogeneity located in a strain gradient material that admits couple stresses. For more accurate assessment,
yet more terms (i.e. linear, quadratic etc.) in the series expansion must be considered.
5. Eective size-dependent non-local properties of composites
To obtain the size dependent eective properties of composites containing many inhomogeneities, we
employ the conventional formalism of concentration factors, see for example, Torquato (2001) and Markov
and Preziosi (2000). Concentration factor, A
ijkl
, connects the far-eld strain at innity to the average elastic
strain inside the inhomogeneity (and is sometimes called Wus tensor).
X. Zhang, P. Sharma / International Journal of Solids and Structures 42 (2005) 38333851 3843
The total average strain in the inhomogeneity is: hei
in
= hei + e
1
= hSe
f
i + e
1
$ hSihe
f
i + e
1
. Here, S, is
the complete strain gradient Eshelby tensor derived in this work. Implicit in our approximation (of the
inhomogeneity problem) is that terms higher order than linear in dierence moduli i.e.O(DC
2
) are neglected.
Employing the results of the previous section, we can then compactly write the concentration factor as
A
mnil
A
1
d
mi
d
nl
A
1
d
ml
d
ni
A
2
d
mn
d
il
A
1
l
1
2
1
2g
1
Dl l
1
A
2
k
1
DK K
1
Dk 2g
1
3g
2
2g
2
DlK
1
2 2g
1
Dl l
1
2g
1
DK 3g
2
DK k
1
31
Setting a = 2g
1
+ 3g
2
and b = 2g
1
, we write:
A
ijkl
K
1
K
1
aDK
I
0
ijkl
l
1
l
1
bDl
I
00
ijkl
I
0
ijkl
1
3
d
ij
d
kl
I
00
ijkl
1
2
d
ik
d
jl
d
il
d
jk
2
3
d
ij
d
kl
_ _
32
Here, K is bulk modulus. Eq. (32) appears identical to the classical results. The sole dierence is that
coecients a and b are redened:
a
3k
1
2l
1
3 k
1
2l
1
..
classical
2l
1
3k
1
A 3B
..
non-classical
b
23k
1
8l
1
15k
1
2l
1
..
classical
2l
1
A
..
non-classical
33
Here A and B are dened in the previous section. Having dened the concentration factors, we can em-
ploy any number of homogenization methods (e.g. self-consistent scheme, the eective eld or MoriTana-
ka method etc.). Since, for two phase materials with isotropic distribution of spherical particles, eective
eld or MoriTanaka method is reasonable and yet provides analytical expression, we adopt it in the fol-
lowing. In the eective eld method (see for example, Markov and Preziosi, 2000), the inuence of other
(nite concentration of) inhomogeneities is mimicked by an eective uniform eld. Once the concentration
factor is known (which we have already derived), the eective modulus within the eective eld or Mori
Tanaka scheme can be obtained via (Markov and Preziosi, 2000):
C
ijkl
C
1
ijkl
/
2
A
ijmn
DC
mnst
: /
1
I
stkl
/
2
A
stkl
1
34
C
ijkl
is the eective size-dependent stiness tensor while /
1
and /
2
are the volume fractions of the matrix
and inhomogeneities respectively and satisfy: /
1
+ /
2
= 1.I
ijkl
is the fourth order unit tensor which has
the form:
I
ijkl
1
2
d
ij
d
kl
d
ik
d
jl
Eq. (35) is exactly the expression of eective property under eective eld theory. In appearance it is
identical to the classical elasticity solution. The dierence of course lies in the redenition of the concentra-
tion factor (Eqs. (31)(33)).
3844 X. Zhang, P. Sharma / International Journal of Solids and Structures 42 (2005) 38333851
From Eq. (36) we can generate direct expressions for the eective bulk modulus K* and the eective
shear modulus l*:
K
K
1
K
1
/
2
DK
K
1
a/
1
DK
l
l
1
l
1
/
2
DK
l
1
b/
1
Dl
35a-b
Fig. 5. Eective bulk modulus with K
2
= 2K
1
and k
1
= l
1
.
Fig. 6. Eective bulk modulus with K
2
= 20K
1
and k
1
= l
1
.
X. Zhang, P. Sharma / International Journal of Solids and Structures 42 (2005) 38333851 3845
We rst present results for the eective bulk modulus in Fig. 5 for various volume fractions and as a
function of inclusion size. The following normalization is employed:
K K
=K
c
and x a=l where K
c
is
the eective bulk modulus predicted by classical elasticity. Fig. 5 is plotted for K
2
= 2K
1
k
1
= l
1
. As ex-
pected, for large inclusion sizes, our solution approaches the classical size-independent one.
In Fig. 5, the departure from classical solution is weak since the dierence in properties of the matrix and
inclusion is not very large. In Fig. 6 we plot results with the bulk moduli ratio of inhomogeneity to matrix
equal to 20. Signicant departure from classical results is now seen at small inclusion sizes.
Fig. 7. Eective shear modulus with l
2
= 2l
1
, k
1
= l
1
.
Fig. 8. Eective shear modulus with l
2
= 3l
1
, k
1
= l
1
, l
0
= 20l.
3846 X. Zhang, P. Sharma / International Journal of Solids and Structures 42 (2005) 38333851
Similar plots are shown for the eective shear modulus in Figs. 710. The following normalization is em-
ployed: l l
=l
c
and x a=l. Note that unlike in the case of bulk modulus, the eective shear modulus
depends both on l and l
0
.
6. Summary and conclusions
In summary, we have extended Eshelbys classical approach towards inclusions and inhomogeneities to
incorporate the size eect via the concept of strain gradient elasticity. The general formof modied strain gra-
dient Eshelbys tensor for arbitrary shaped inclusions was given in terms of three potentials. Explicit and exact
Fig. 9. Eective shear modulus with l
2
= 21l
1
, k
1
= l
1
, l
0
= 10l.
Fig. 10. Eective shear modulus with l
2
= 21l
1
, k
1
= l
1
, l
0
= 20l.
X. Zhang, P. Sharma / International Journal of Solids and Structures 42 (2005) 38333851 3847
solution was provided for special case of a spherical inclusion. An approximate solution to the inhomogeneity
problemwas also provided based on which and with an appeal to the eective eld theory, explicit expressions
for the eective size-dependent bulk and shear modulus of a composites material were derived.
We anticipate several applications and possible improvements of the present work. Clearly, the pres-
ent work can be employed to reanalyze several classical problems in the nanoregime e.g. phase trans-
formation at small scales, thermal mismatch problem for nanoinclusions, lattice mismatch problem in
quantum dots etc. While our solution to the inclusion problem is exact, the inhomogeneity problem
was solved with the assumption of a uniform strain approximation. This is tantamount to adopting
only the rst term in a perturbation series expansion in terms of the dierence in the elastic moduli of
the inclusion-matrix system. Further terms must be incorporated for more accurate results. Use of the
present work for constructing eective property solutions for non-linear gradient materials is also rele-
gated to future work.
Appendix A. Comparison of various strain gradient theories
The table below summarizes the various strain gradient theories and their dierences/similarities. This
compilation is by no means complete but serves to provide a benchmark for the more prevalent ones. Only
strain gradient theories are listed (for example, Eringens non-local integral formulation is not compared).
Model (Reference) Description Comparison to other works
Mindlins 2nd
gradient model
(1965)
W
1
2
ke
ii
e
jj
le
ij
e
ij
a
1
e
ijj
e
ikk
a
2
e
iik
e
kjj
a
3
e
iik
e
jjk
a
4
e
ijk
e
ijk
a
5
e
ijk
e
kji
b
1
e
iijj
e
kkll
b
2
e
ijkk
e
ijll
b
3
e
iijk
e
jkll
b
4
e
iijk
e
llkj
b
5
e
iijk
e
lljk
b
6
e
ijkl
e
ijkl
b
7
e
ijkl
e
jkli
c
1
e
ii
e
jjkk
c
2
e
ij
e
ijkk
c
3
e
ij
e
kkij
b
0
e
iijj
The most generalized strain gradient
model that includes also other higher
order stresses (beyond couple stresses).
This formulation which contains
up to 2nd gradients of strains has the
advantage to also incorporate surface
energies. Most other strain gradient
theories can be obtained by
appropriate simplication
of this work
e
ij
1
2
u
i;j
u
j;i
e
ijk
= u
i,jk
e
ijkl
= u
i,jkl
Kleinerts model
(1989used in
the present
work)
W x
l
2
@
i
u
j
l k
2
@
l
u
l
2l k
2
l
02
@
i
@
l
u
l
@
i
@
j
u
j
ll
2
2
@
2
l
u
i
@
2
l
u
i
@
i
@
l
u
l
@
i
@
j
u
j
1
4
e
ikl
u
l;kj
e
jkl
u
l;ki