ME340B – Elasticity of Microscopic Structures – Stanford University – Winter 2004

Lecture Note 2. Eshelby’s Inclusion I

Chris Weinberger and Wei Cai

c All rights reserved

January 12, 2004

Contents
1 Inclusion and eigenstain 2

2 Green’s function and Eshelby’s tensor Sijkl 3

3 Auxiliary tensor Dijkl 5

4 Ellipsoidal inclusion 7

5 Eshelby’s tensor in isotropic medium 10

1
1 Inclusion and eigenstain
Consider a homogeneous linear elastic solid with volume V and surface area S, with elastic
constant Cijkl , as shown in Fig. 1. Let a sub-volume V0 with surface area S0 undergo a uniform
permanent (inelastic) deformation, such as a martensitic phase transformation. The material
inside V0 is called an inclusion and the material outside is called the matrix. If we remove V0
from its surrounding matrix, it should assume a uniform strain e∗ij and will experience zero
stress. e∗ij is called the eigenstrain, meaning the strain under zero stress. Notice that both
the inclusion and the matrix have the same elastic constants. The eigenstress is defined as
σij∗ ≡ Cijkl e∗kl .
In reality, the inclusion is surrounded by the matrix. Therefore, it is not able to reach the
state of eigenstrain and zero stress. Instead, both the inclusion and the matrix will deform
and experience an elastic stress field. The Eshelby’s transformed inclusion problem is to
solve the stress, strain and displacement fields both in the inclusion and in the matrix.

V0
S0

V
S
Figure 1: A linear elastic solid with volume V and surface S. A subvolume V0 and surface S0
undergoes a permanent (inelastic) deformation. The material inside V0 is called an inclusion
and the material outside is called the matrix.

Figure 2: John Douglas Eshelby (1916-1981, United Kingdom).

2
2 Green’s function and Eshelby’s tensor Sijkl
Eshelby showed that the problem stated above can be solved elegantly by the superposition
principle of linear elasticity and using the Green’s function [1]. Eshelby used the following 4
steps of a “virtual” experiment to construct the desired solution.

Step 1. Remove the inclusion from the matrix.

V0

Apply no force to the inclusion, nor to the matrix. The strain, stress and displacement fields
in the matrix and the inclusion are,
matrix inclusion
eij = 0 eij = e∗ij
σij = 0 σij = 0
ui = 0 ui = e∗ij xj

Step 2. Apply surface traction to S0 in order to make the inclusion return to its original
shape

V0 T

∗
The elastic strain of the inclusion should exactly cancel the eigenstrain, i.e. eel
ij = −eij . The
strain, stress and displacement fields in the matrix and the inclusion are,
matrix inclusion
∗
eij = 0 eij = eel
ij + eij = 0
∗ ∗
σij = 0 σij = Cijkl eel
ij = −Cijkl eij = −σij
ui = 0 ui = 0
The traction force on S0 is Tj = σij ni = −σij∗ ni .

3
Step 3. Put the inclusion back to the matrix.

V0 T

The same force T is applied to the internal surface S0 . There is no change in the deformation
fields in either the inclusion or the matrix from step 2.

Step 4. Now remove the traction T. This returns us to the original inclusion problem as
shown in Fig. 1. The change from step 3 to step 4 is equivalent to applying a cancelling
body force F = −T to the internal surface S0 of the elastic body.

V0 F

Let uci (x) be the displacement field in response to body force Fj on S0 . uci (x) is called the
constrained displacement field. It can be easily expressed in terms of the Green’s function
∗
of the elastic body, (notice that Fj = −Tj = σjk nk )
Z Z
0 0 0 ∗
c
ui (x) = Fj (x )Gij (x, x )dS(x ) = σjk nk (x0 )Gij (x, x0 )dS(x0 ) (1)
S0 S0

The displacement gradient, strain, and stress of the constrained field are
Z
∗
c
ui,j (x) = σlk nk (x0 )Gil,j (x, x0 ) dS(x0 ) (2)
S0
Z
1 c 1
c
eij (x) = c
(u + ui,j ) = σ ∗ nk (x0 ) [Gil,j (x, x0 ) + Gjl,i (x, x0 )] dS(x0 ) (3)
2 i,j 2 S0 lk
σijc (x) = Cijkl eckl (x) (4)

In terms of the constrained field, the strain, stress and displacement fields in the matrix and
the inclusion are,

4
 matrix inclusion
eij = ecij eij = ecij
σij = σijc σij = σijc − σij∗ = Cijkl (eckl − e∗kl )
ui = uci ui = uci
To obtain explicit expressions for the stresses and strains everywhere, the constrained
field must be determined both inside and outside the inclusion. We can define a fourth order
tensor Sijkl that relates the constrained strain inside the inclusion to its eigenstrain,

ecij = Sijkl e∗kl (5)

Sijkl is often referred to as Eshelby’s tensor. Because it relates two symmetric strain tensors,
the Eshelby’s tensor satisfies minor symmetries,

Sijkl = Sjikl = Sijlk (6)

But it does not satisfy the major symmetry, i.e. Sijkl 6= Sklij . In the following sections, we
derive the explicit expressions of Eshelby’s tensor in an infinite elastic medium (V → ∞). In
principle, Eshelby’s tensor is a function of space, i.e. Sijkl (x). However, an amazing result
obtained by Eshelby is that,

For an ellipsoidal inclusion in a homogeneous infinite matrix, the

Eshelby tensor Sijkl is a constant tensor. Hence the stress-strain
fields inside the inclusion are uniform.

3 Auxiliary tensor Dijkl

For convenience, let us define another tensor Dijkl that relates the constrained displacement
gradients to the eigenstress inside the inclusion [3],
∗
uci,l (x) = −σkj Dijkl (x) (7)

Obviously, tensor Dijkl is related to Eshelby’s tensor,

Sijmn e∗mn = ecij (8)

1 c
= (u + ucj,i ) (9)
2 i,j
1 ∗ ∗
= − (σlk Diklj + σlk Djkli ) (10)
2
1 ∗
= − σlk (Diklj + Djkli ) (11)
2
1
= − Clkmn e∗mn (Diklj + Djkli ) (12)
2
Therefore,
1
Sijmn (x) = − Clkmn (Diklj (x) + Djkli (x)) (13)
2

5
 ∗
Rewrite Eq. (7) as uci,j (x) = −σkl Dilkj (x) and compare it with From Eq. (2), we obtain,
Z
Dilkj (x) = − nk (x0 )Gil,j (x − x0 ) dS(x0 ) (14)
S0

or equivalently,
Z
Dijkl (x) = − Gij,l (x − x0 )nk (x0 ) dS(x0 ) (15)
S0

Notice that we have used the fact that Gij (x, x0 ) = Gij (x − x0 ) for an infinite homogeneous
medium. Applying Gauss’s Theorem, we obtain
Z
∂
Dijkl (x) = − 0
Gij,l (x − x0 ) dV (x0 )
∂x
Z V0 k
∂
= Gij,l (x − x0 ) dV (x0 )
V0 ∂xk

Therefore,
Z
Dijkl (x) = Gij,kl (x − x0 ) dV (x0 ) (16)
V0

Recall that the Green’s function for an anisotropic medium is,

(zz)−1
Z
0 1 0 ij
Gij (x − x ) = exp [−ik · (x − x )] dk (17)
(2π)3 k2

where z = k/k. Substituting this into Eq. (16), we get

" #
−1
∂2 (zz)
Z Z
1 ij
Dijkl (x) = 3
exp [−ik · (x − x0 )] 2
dk dV (x0 )
V0 ∂x k ∂x l (2π) k
Z Z ∞" −1
#
1 (zz) ij
= dk(−ikk )(−ikl ) exp [−ik · (x − x0 )] dV (x0 )
(2π)3 V0 −∞ k2
Z Z
1
= − 3
exp [−ik · (x − x0 )] (zz)−1 0
ij zk zl dk dV (x ) (18)
(2π) V0
Because the integration over the inclusion volume V0 only depends on x0 but not on x, it is
helpful to rearrange integrals as,
Z Z
1 −1
Dijkl (x) = − dk exp(−ik · x)(zz)ij zk zl exp(ik · x0 ) dV (x0 )
(2π)3 V0
Z
1
= − dk exp(−ik · x)(zz)−1
ij zk zl Q(k) (19)
(2π)3
where
Z
Q(k) ≡ exp(ik · x0 ) dV (x0 ) (20)
V0

6
Therefore, for an infinite homogeneous medium, the auxiliary tensor Dijkl also satisfies minor
symmetries,
Dijkl = Djikl = Dijlk (21)
But it does not satisfy the major symmetry, i.e. Dijkl 6= Dklij (similar to Eshelby’s tensor
Sijkl ).

4 Ellipsoidal inclusion
Now let us restrict our attention to inclusions that are ellipsoidal in shape. The goal is to
prove that Dijkl (x) is a constant inside an ellipsoidal inclusion. The volume V0 occupied by
the inclusion can be expressed as,
 0 2  0  2  0 2
x y z
+ + ≤1 (22)
a b c
where a, b, c specify the size of the ellipsoid. Define new variables,
x0
X0 ≡ (23)
a
y0
Y0 ≡ (24)
b
z0
Z0 ≡ (25)
c
R ≡ X 0 e1 + Y 0 e2 + Z 0 e3 (26)
R ≡ |R| (27)
Then the integration over V0 becomes an integration over a unit sphere in the space of R,
Z Z
0
dV (x ) ⇒ abc dR (28)
V0 |R|≤1

Also define new variables in Fourier space,

λx ≡
akx (29)
λy ≡
bky (30)
λz ≡
ckz (31)
λ ≡
λ x e 1 + λy e 2 + λ z e 3 (32)
q
λ ≡ |λ| = a2 kx2 + b2 ky2 + c2 kz2 (33)
Therefore,
k · x0 = λ · R (34)
Z
Q(k) ≡ exp(ik · x0 ) dV (x0 )
V0
Z
= abc exp(iλ · R) dR (35)
|R|≤1

7
In polar coordinates,
Z 1 Z 2π Z π
Q(k) = abc R2 sin φ exp(iλR cos φ) dφ dθ dR (36)
0
Z 10 0
Z 1
2
= 2πabc dR R ds exp(iλRs) (37)
0 −1
Z 1  
2 2 sin(λR)
= 2πabc R dR (38)
0 λR
abc 1
Z
= 4π R sin λR dR (39)
λ 0
abc
= 4π 3 (sin λ − λ cos λ) (40)
λ
Substituting this result into Eq. (19), we have
Z ∞
1 4π
Dijkl (x) = − 3 dk exp(−ik · x)(zz)−1ij zk zl 3 abc(sin λ − λ cos λ)
8π −∞ λ
Z ∞
abc sin λ − λ cos λ
= − 2 (zz)−1
ij zk zl exp(−ik · x) dk (41)
2π −∞ λ3
Again we go to polar coordinates. Define new variables Φ, Θ, γ through,

kx =
k sin Φ cos Θ (42)
ky =
k sin Φ sin Θ (43)
kz =
k cos Φ (44)
γ ≡
(k · x)/k = x sin Φ cos Θ + y sin φ sin Θ + z cos Φ (45)
q
β ≡ λ/k = (a2 cos2 Θ + b2 sin2 Θ) sin2 Φ + c2 cos2 Φ (46)

Then
abc ∞ π 2π 2 sin λ − λ cos λ
Z Z Z
Dijkl (x) = − 2 k (zz)−1ij zk zl exp(−ikγ) sin Φ dΘ dΦ dk
2π 0 0 0 λ3
abc π 2π
Z Z
= − 2 (zz)−1
ij zk zl κ(γ) sin Φ dΘ dΦ (47)
2π 0 0
where
∞
sin λ − λ cos λ
Z
κ(γ) ≡ dk k 2 exp(−ikγ)
λ3
Z0 ∞
sin kβ − kβ cos kβ
= dk k 2 exp(−ikγ)
0 k3β 3
Z ∞  
1 sin kβ
= dk exp(−ikγ) − β cos kβ (48)
β3 0 k
Notice that the dependence of Dijkl on x is through γ = (k · x)/k in κ(γ). To evaluate κ(γ),
notice that the term in the square bracket is an even function of k. Because Dijkl is real,

8
κ(γ) must be real as well. Therefore, we can rewrite the integral as,
Z ∞  
1 sin kβ
κ(γ) = dk exp(−ikγ) − β cos kβ (49)
2β 3 −∞ k
Notice that
Z ∞
1 ∞
Z
dk exp(−ikγ) cos kβ = dk e−ikγ (eikβ + e−ikβ )
−∞ 2 −∞
1 ∞
Z
dk e−ik(γ−β) + e−ik(γ+β)
 
=
2 −∞
= π [δ(β − γ) + δ(β + γ)] (50)
Z ∞ Z ∞
d sin kβ
dk exp(−ikγ) = dk exp(−ikγ) cos kβ (51)
dβ k
Z−∞
∞
−∞
sin kβ
dk exp(−ikγ) = π [h(β − γ) + h(β + γ)] (52)
−∞ k
where
 1
 −2 if α < 0
h(α) = 0 if α = 0 (53)
 1
2
if α > 0
Therefore, if β ± γ > 0 then the κ(γ) reduces to
π
κ(γ) = [h(β − γ) + h(β + γ) − βδ(β − γ) − βδ(β + γ)]
2β 3
π
= (54)
2β 3
In other words, κ(γ) becomes a constant if β ± γ > 0. In this case, Dijkl (x) reduces to a
surface integral that is independent of x,

abc π 2π
Z Z
π
Dijkl (x) = − 2 (zz)−1
ij zk zl sin Φ dΘ dΦ (55)
2π 0 0 2β 3
We will now show that if x is within the ellipsoid, then β ± γ > 0. This will then prove that
Dijkl and Sijkl are constants within the ellipsoidal inclusion. To see why this is the case,
consider vector ρ such that,
x y z
ρ= e1 + e2 + e 3 (56)
a b c
If x lies within the ellipsoid, then
p
ρ ≡ |ρ| = (x/a)2 + (y/b)2 + (z/c)2 < 1 (57)

At the same time,

γ ≡ (k · x)/k = (λ · ρ)/k (58)
β ≡ λ/k (59)

9
Therefore,

|γ| = |λ · ρ|/k ≤ λρ/k < λ/k = β

β±γ > 0 (60)

Therefore, when x lies within the ellipsoid, the Dijkl tensor can be calculated by simply
performing a surface integral over a unit sphere,
abc π 2π
Z Z
sin Φ
Dijkl = − (zz)−1
ij zk zl dΘ dΦ (61)
4π 0 0 β3
When x lies outside the ellipsoid, Dijkl shall depend on x, and can be calculated directly
from the Green’s function,
Z
Dijkl (x) = − Gij,l (x − x0 )nk (x0 ) dS(x0 ) (62)
S

Once Dijkl is obtained, Eshelby’s tensor Sijkl can be found by Eq. (13).

5 Eshelby’s tensor in isotropic medium

The derivation of the Eshelby tensor in isotropic materials can be found in [1] (for isotropy)
and [2]. For isotropic medium, the Eshelby’s tensor can be expressed in terms of elliptic
integrals. Assuming that a > b > c and that the semi axis a aligns with the coordinate x
(and similarly b with y and c with z) then
3 1 − 2ν
S1111 = a2 I11 + I1
8π(1 − ν) 8π(1 − ν)
1 1 − 2ν
S1122 = b2 I12 + I1
8π(1 − ν) 8π(1 − ν)
1 1 − 2ν
S1133 = c2 I13 + I1
8π(1 − ν) 8π(1 − ν)
a2 + b 2 1 − 2ν
S1212 = I12 + (I1 + I2 )
16π(1 − ν) 16π(1 − ν)
S1112 = S1223 = S1232 = 0

The rest of the nonzero terms can be found by cyclic permutation of the above formulas.
Notice that we should also let a → b → c together with 1 → 2 → 3. The I terms are defined
in terms of standard elliptic integrals,
4πabc
I1 = [F (θ, k) − E(θ, k)]
(a2
− b2 )(a2
− c2 )1/2
 2
b(a − c2 )1/2

4πabc
I3 = − E(θ, k)
(b2 − c2 )(a2 − c2 )1/2 ac

10
where
r
a2 − c 2
θ = arcsin 2
r a
a2 − b 2
k=
a2 − c 2
and
I1 + I2 + I3 = 4π
4π
3I11 + I12 + I13 = 2
a
2 2 2
3a I11 + b I12 + c I13 = 3I1
I2 − I1
I12 = 2
a − b2
and the standard elliptic integrals are defined as
Z θ
dw
F (θ, k) = 2 2 1/2
(63)
0 (1 − k sin w)
Z θ
E(θ, k) = (1 − k 2 sin2 w)1/2 dw (64)
0
For a spherical inclusion (a = b = c), Eshelby’s tensor has the following compact expression,
5ν − 1 4 − 5ν
Sijkl = δij δkl + (δik δjl + δil δjk ) (65)
15(1 − ν) 15(1 − ν)
Notice that tensor itself does not depend on the radius of the sphere!
For an elliptic cylinder ( c → ∞ )
 2 
1 b + 2ab b
S1111 = + (1 − 2ν)
2(1 − ν) (a + b)2 a+b
 2 
1 a + 2ab a
S2222 = + (1 − 2ν)
2(1 − ν) (a + b)2 a+b
S3333 = 0
b2
 
1 b
S1122 = − (1 − 2ν)
2(1 − ν) (a + b)2 a+b
1 2νa
S2233 =
2(1 − ν) a + b
a2
 
1 a
S2211 = − (1 − 2ν)
2(1 − ν) (a + b)2 a+b
S3311 = S3322 = 0
 2
a + b2

1 (1 − 2ν)
S1212 = +
2(1 − ν) 2(a + b)2 2
1 2νb
S1133 =
2(1 − ν) a + b
a b
S2323 = S3131 =
2(a + b) 2(a + b)

11
 Another important geometry is the flat ellipsoid (a > b  c). The I integrals in this
limiting case reduce to
bc
I1 = 4π(F (k) − E(k)) 2
 a − b2 
c bc
I2 = 4π E(k) − (F (k) − E(k)) 2
b a − b2
 c 
I3 = 4π 1 − E(k)
 b 
c bc
I12 = 4π E(k) − 2(F (k) − E(k)) 2 /(a2 − b2 )
b a − b2
 
c bc
I23 = 4π 1 − 2E(k) + (F (k) − E(k)) 2 /b2
b a − b2
 
c bc
I31 = 4π 1 − E(k) − (F (k) − E(k)) 2 /a2
b a − b2
4π
I33 =
3c2
where E(k) and F (k) are complete elliptic integrals defined as
Z π
2 dw
F (k) = 2 2 1/2
(66)
0 (1 − k sin w)
Z π
2
E(k) = (1 − k 2 sin2 w)1/2 dw (67)
0

Eshelby’s tensor for various other shapes can be found in [2] and [4].

References
[1] J. D. Eshelby, Elastic Inclusions and Inhomogeneities, in Progress in Solid Mechanics,
2, ed. IN. Sneddon and R. Hill, (North-Holland, Amsterdam, 1961) pp. 89-140.

[2] T. Mura, Micromechanics of Defects in Solids, 2nd rev. ed., Kluwer Academic Publish-
ers, 1991.

[3] D. M. Barnett, ME340B Lecture Notes, Micromechanics of Solids, Stanford University,

2003-2004.

[4] S. Nemat-Nasser and M. Hori, Micromechanics: Overall Properties of Heterogeneous

Materials, 2nd rev. ed., Elsevier Science B.V., 1999

12