Theory of Elasticity

Liping Liu
December 11, 2009
Contents
1 Tensor Analysis 2
1.1 A review of linear algebra . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.2 The divergence theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
1.3 Curvilinear coordinate systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
2 Kinematics 12
2.1 Geometric interpretation of the deformation gradient . . . . . . . . . . . . . . . . . . 12
2.2 Small strain and linearization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
2.3 Transformations of a deformation gradient . . . . . . . . . . . . . . . . . . . . . . . . 14
2.4 Compatibility conditions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
3 Concepts of Stress 20
3.1 Cauchy stress and balance laws . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
3.2 Implications of the rst and second laws of thermodynamics . . . . . . . . . . . . . 21
4 Constitutive Laws 23
4.1 Hookes law . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
4.2 Frame indierence and material symmetry . . . . . . . . . . . . . . . . . . . . . . . . 24
4.3 Approximation of the stored energy functions . . . . . . . . . . . . . . . . . . . . . . 25
5 Formulations of Linear (Linearized) Elasticity Problems 26
5.1 PDE formulation, Variational formulation, Weak formulation . . . . . . . . . . . . . 26
5.2 Uniqueness of the solution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
5.3 Plane strain problem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
5.4 Antiplane shear problem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
5.5 Plane stress problem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
6 St Venants Problems and Semi-inverse Method 34
6.1 Simple extension . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
6.2 Pure bending . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
6.3 Torsion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
6.4 Bending . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
1
7 Important Problems and their Solutions 40
7.1 The fundamental solution Kelvins solution . . . . . . . . . . . . . . . . . . . . 40
7.2 Pressured vessel . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
7.3 Half-space problem Boussinesqs solution . . . . . . . . . . . . . . . . . . . . . . 46
7.4 Contact problem Hertzs solution . . . . . . . . . . . . . . . . . . . . . . . . . . 51
7.5 Inclusion problem Eshelbys solution . . . . . . . . . . . . . . . . . . . . . . . . 57
8 Introduction to Micromechanics 62
8.1 Cracks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
8.2 Composite materials . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
8.3 Dislocations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
8.4 Plasticity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
9 A Hierarchy of Plate Theories 62
9.1 Kirchho-Love plate theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
9.2 von Karman plate theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
9.3 Mindlin plate theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
10 Waves in Solid Structures 62
10.1 Bulk waves . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
10.2 Surface waves . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
10.3 Interfacial waves . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
1 Tensor Analysis
1.1 A review of linear algebra
Vector Space Let IR be the scalar eld of real numbers. We consider only real vector spaces. Let
V
n
be a set. V
n
is a vector space (also called a linear space) if it is equipped with two operations:
scalar product IR V
n
V
n
,
vector addition V
n
V
n
V
n
,
and it is closed under these two operations. That is, V
n
is a vector space if , IR & a, b V
n
,
a +b V
n
.
The vector space V
n
is n-dimensional if we can nd a basis e
1
, , e
n
V
n
such that for any
a V
n
, we have a unique decomposition
a =
n

i=1
a
i
e
i
,
where a
i
IR (i = 1, n) are the components of vector a under the basis e
1
, , e
n
.
Tensor Space Let V
n
(V
m
) be n-dimensional (m-dimensional) vector space. A mapping A : V
n

V
m
is a tensor if A is linear. That is, , IR & a, b V
n
,
A(a +b) = A(a) +A(b). (1)
2
Let Lin(V
n
, V
m
) be the collection of all linear mappings (i.e., tensors) with domain V
n
and range
V
m
. For any IR and any A
1
, A
2
Lin(V
n
, V
m
), dene two operations
scalar product (A
1
)(a) = A
1
(a) a V
n
,
vector addition (A
1
+A
2
)(a) = A
1
(a) +A
2
(a) a V
n
.
(3pts) 1. Show that for any , IR and any A
1
, A
2
Lin(V
n
, V
m
), A
1
+ A
2
is a linear
mapping (from V
n
to V
m
).
The above claim implies that the set Lin(V
n
, V
m
) is also a vector space.
Inner Product We equip a n-dimensional vector space V
n
with a mapping V
n
V
n
IR, called
inner product such that for any , IR and any a, b, c V
n
, the inner product is
1. Positive-denite: a a 0; a a = 0 i a = 0,
2. Linear: a (b +c) = a b +a c,
3. Symmetric: a b = b a.
Geometric interpretations:
Length of a vector: [a[ =

a a,
Angle between two vectors: cos() =
ab
|a||b|
.
Euclidean Space IR
n
For a n-dimensional vector space V
n
equipped with an inner product, we
can nd an orthonormal basis e
i
: i = 1, , n such that for all i, j = 1, , n,
e
i
e
j
=
ij
=
_
1 if i = j,
0 if i ,= j.
With respect to this basis, for any vector a V
n
, we nd its components (a
1
, , a
n
) (or coordinates
if a is a point in space)
a =
n

i=1
a
i
e
i
, a
i
= a e
i
IR i = 1, , n.
We may further identify the space V
n
with the familiar Euclidean space IR
n
. However, one shall
keep in mind, IR
n
, as a vector space equipped with an inner product, is more than a collection of
arrays of real numbers. One should not think of a vector in IR
n
as an array of real numbers unless
we specify a basis or a frame.
3
Tensor Product For given vectors a IR
n
and b IR
m
, dene
b a : V
n
V
m
(b a)(c) = (a c)b c IR
n
.
(3pts) 2. Show that for any a IR
n
and b IR
m
, the mapping b a (from V
n
to V
m
) is linear.
(6pts) 3. Let e
i
: i = 1, n be an orthonormal basis of IR
n
and e
p
: p = 1, m be an
orthonormal basis of IR
m
. Show that
e
p
e
i
: i = 1, , n, p = 1, m Lin(IR
n
, IR
m
)
forms a basis of the linear space Lin(IR
n
, IR
m
).
Subspace of IR
n
, Orthogonal Subspace A subset M IR
n
is a subspace if , IR & a, b
M,
a +b M.
Let M

= b : b a = 0 a M.
(Bonus problem 1) (1pts) Show that M

is a subspace of IR
n
.
Projection Theorem Let M be a subspace of IR
n
. For any x IR
n
, we have
x = y +z where y M, z M

.
4
The vector y, z are uniquely determined by x.
(Bonus problem 2) (5pts) Prove the above theorem
Proof:
Transpose of a Tensor Let A Lin(IR
n
, IR
m
), e
i
: i = 1, n be an orthonormal basis of IR
n
and e
p
: p = 1, m be an orthonormal basis of IR
m
. Then A admits the following decomposition
A =

p,i
A
pi
e
p
e
i
where A
pi
= e
p
A(e
i
) i = 1, , n, p = 1, , m.
Dene
A
T
: IR
m
IR
n
,
A
T
=

p,i
A
pi
e
i
e
p
Lin(IR
m
, IR
n
).
(3pts) 4. Show that for any a V
n
and b V
m
,
b A(a) = a A
T
(b).
Symmetric and Skew-symmetric Tensor Let A Lin(IR
n
, IR
n
). A is symmetric if A = A
T
;
A is skew-symmetric if A
T
= A.
Let e
i
: i = 1, n, e
p
: p = 1, n be two orthonormal bases of IR
n
. We have shown
A =

p,i
A
pi
e
p
e
i
where A
pi
= e
p
A(e
i
) p, i = 1, , n.
If ( e
1
, , e
n
) = (e
1
, , e
n
), show
1. if A = A
T
, then A
pi
= A
ip
for all p, i = 1, , n;
2. if A = A
T
, then A
pi
= A
ip
;
5
3. For any A Lin(IR
n
, IR
n
), we have a unique decomposition A = E +W, where E = E
T
and
W = W
T
.
(3pts) 5. A = A
T
i for any a, b IR
n
,
b A(a) = a A(b).
(3pts) 6. If A = A
T
and a A(a) = 0 for any a IR
n
, then A = 0.
(1pt) 7. Conceive a nonzero 3 3 matrix [A
pi
] such that
a Aa =

p,i=1,2,3
a
p
A
pi
a
i
= 0 a IR
3
.
Product of tensors Let A Lin(IR
n
, IR
m
), B Lin(IR
m
, IR
k
). Then
BA : IR
n
IR
k
,
BA(a) = B(A(a)).
Orthogonal Tensor Let Q Lin(IR
n
, IR
n
). Q is orthogonal if Qa Qb = a b for all a, b IR
n
.
Remark Orthogonal tensor preserve the length of a vector and the angle between two vectors.
1. [a[ = [Qa[.
2. a b = Qa Qb.
Remark Q is orthogonal i Q
T
Q = QQ
T
= I.
Trace of a tensor Let A Lin(IR
n
, IR
n
) and A =

p,i
A
pi
e
p
e
i
. Then Tr(A) =

n
p=1
A
pp
.
Determinant of a tensor Let A Lin(IR
n
, IR
n
) and A =

p,i
A
pi
e
p
e
i
. Then det A = det[A
pi
].
(6pts) 8. Prove that Tr, det : Lin(IR
n
, IR
n
) IR is independent of the choice of basis.
6
Rigid Rotation Tensor An orthogonal tensor R Lin(IR
n
, IR
n
) is a rigid rotation if det R = +1.
Representation theorem: For any A Lin(IR
n
, IR), there is an a IR
n
such that
A(b) = a b b IR
n
.
A =

i
A
1i
e
1
e
i
, e
1
= 1 and a =
n

i=1
A
1i
e
i
.
Cross product in IR
3
For a, b IR
3
,
a b =

e
1
e
2
e
3
a
1
a
2
a
3
b
1
b
2
b
3

= W(b),
where W =

p,i
W
pi
e
p
e
i
,
[W
p,i
] =
_
_
0 a
3
a
2
a
3
0 a
1
a
2
a
1
0
_
_
.
11. Verify the following properties of cross products:
1. (2pt) b a = a b, a (a b) = 0, b (a b) = 0.
2. (2pt) (a b) c = (b c) a = (c a) b.
3. (Bonus problem 3) (2pt) Geometric interpretation: show that [ab[ =area of the parallelogram
formed by a and b; [c (a b)[= volume of the parallelepiped formed by a, b, c.
7
1.2 The divergence theorem
Tensor Fields Let IR
n
be an open domain.
: IR is a scalar eld;
V : IR
m
is a vector eld;
T : Lin(IR
m
, IR
k
) is a tensor eld.
Dierentiation Let be a scalar eld on IR
n
. For any a IR
n
,
D(x) : IR
n
IR,
(D(x))(a) = lim
0
(x +a) (x)

.
Denition: is dierentiable on if D(x) Lin(IR
n
, IR) for all x
(x +a) = (x) +(D(x))(a) +o()
= (x) +(x) a +o() x , a IR
n
.
and
(x) =
n

i=1

,i
e
i
,
,i
= e
i
= lim
0
(x +e
i
) (x)

=
(x
1
e
1
+ +x
n
e
n
)
x
i
.
Denition: V : IR
m
is dierentiable on if every component is dierentiable
V (x) =

p
V
p
(x) e
p
.
Denition:
V (x) = DV (x) : IR
n
IR
m
(V (x))(a) =
m

p=1
e
p
V
p
(x) a
Thus,
V (x) =

p,i
V
p,i
e
p
e
i
.
Divergence: If m = n, div(V ) = Tr(V )
div(V ) =

i
V
i,i
e
i
e
i
=

i
V
i,i
.
Further, if T : Lin(IR
n
, IR
m
) and
T(x) =

p,i
T
pi
e
p
e
i
.
8
Then
div(T) : IR
m
IR
div(T)(a) =

p,i
T
pi,i
a e
p
.
One may identify div(T) with a vector eld IR
m
(instead of Lin(IR
m
, IR)). With an abuse of
notation, we write
div(T) =

p,i
T
pi,i
e
p
.
Field of class C
0
, C
1
, C
2
, , C

Identities: Assume , V, U, T : IR, IR

n
, IR
n
, Lin(IR
n
, IR
m
), W : IR
m
are smooth elds
on .
1. (V ) = V () +V ;
2. div(V ) = () V +divV ; V = divV
3. (V U) = (V )
T
U + (U)
T
V
4. div(V U) = V div(U) + (V )U
5. div(T
T
W) = T W +W divT
6. div(T) = divT +T
Proof: Tacitly, an orthonormal basis e
1
, , e
n
IR
n
and an orthonormal basis e
1
, , e
m

IR
m
are chosen and xed. Notation: Einstein summation, i.e., summation over double index is
understood. For example, to show 5, we have
div(T
T
W) = (T
pi
W
p
)
,i
= T
pi,i
W
p
+T
pi
W
p,i
= W divT +T W
(8pts) 9. Show the above equations 1-6.
9
Divergence Theorem
Let be a smooth simply connected domain in IR
n
, V : IR
m
is a smooth vector eld on
. Then we have
_

V nda =
_

V dv, (2)
where n : IR
n
is the outward unit normal on the boundary . If m = n, take the trace of
Eq. (2), we have
_

V nda =
_

divV dv. (3)

For a smooth tensor eld T : Lin(IR
n
; IR
m
), we have
_

Tnda =
_

divTdv. (4)
10
curl operator Let V : IR
3
.
curlV = V =

e
1
e
2
e
3

1

2

3
V
1
V
2
V
3

=
3

i=1
c
ijk
V
k,j
e
i
,
where
c
ijk
=
_

_
1 if (ijk) = (123) = (231) = (312),
1 if (ijk) = (132) = (213) = (321),
0 otherwise.
(6pt) 12. Let IR
3
be a domain in IR
3
. Assume that V : IR
3
, : IR are smooth
elds. Show the following identities
1. = 0.
2. div( V ) = 0.
3. If a, b, c IR
3
, a (b c) = (a c)b (a b)c
4. ( V ) = ( V ) V .
(hints:) c
ijk
is antisymmetric and called Levi-Civita symbol in literature. In particular, one notices
that c
ijk
= 0 if any two of the indices i, j, k take the same value and
c
ijk
= c
ikj
, c
ijk
= c
jik
, etc.
To prove the above identities, e.g., the rst one, we notice that in Einstein notation,
= c
ijk

,kj
= c
ijk

,jk
( why? ) = c
ikj

,kj
( why? ) = c
ijk

,kj
( why? ) = .
Therefore, = 0. The following identity will be useful in prove the third and fourth one,
which you may use in your proof.
c
ijk
c
ipq
=
jp

kq

jq

kp
,
where
ij
= 1 if i = j and = 0 otherwise.
1.3 Curvilinear coordinate systems
11
HW2, Problem 13-25
Due on Monday, November 30th, 2009
2 Kinematics
Consider a homogeneous deformation y : IR
3
with y = F IR
33
and det(F) > 0. Note
that the deformation gradient F is independent of the position x.
2.1 Geometric interpretation of the deformation gradient
1. The length of a material line. Let p
0
, p
1
be two points in , L = [p
1
p
0
[ the length of the
line p
0
p
1
in the reference conguration. After deformation, the deformed line is between Fp
0
and Fp
1
, and the length of the deformed line is
L

= [F(p
1
p
0
)[ = [(p
1
p
0
) F
T
F(p
1
p
0
)]
1/2
, (5)
where C = F
T
F is called the Cauchy-Green strain tensor.
2. The area of a material surface. Let p
0
, p
1
, p
2
be three points in , A =
1
2
[(p
2
p
0
) (p
1

p
0
)[ , = 0 the area of the triangle p
0
p
1
p
2
. After deformation, the deformed triangle have
Fp
0
, Fp
1
, Fp
2
as vertices and the area of the deformed triangle is
A

= [F(p
2
p
0
) F(p
1
p
0
)[ = det(F)[q F
1
F
T
q]
1/2
. (6)
where q = (p
2
p
0
) (p
1
p
0
).
To show the identity (6), we consider three vectors a, b, c IR
3
. An use of (8) yields
Fa (Fb Fc) = det(F)a (b c) a F
T
(Fb Fc) = det(F)a (b c).
Since a is arbitrary, we conclude
F
T
(Fb Fc) = det(F)(b c) Fb Fc = det(F)F
T
(b c). (7)
3. The volume of a material volume. Let p
0
, p
1
, p
2
, p
3
be four points in , V =
1
2
[(p
3
p
0
)
[(p
2
p
0
) (p
1
p
0
)][ , = 0 the volume of the tetrahedron with p
0
p
1
p
2
p
3
as vertexes. After
deformation, the deformed tetrahedron have Fp
0
, Fp
1
, Fp
2
, Fp
3
as vertexes and the volume
of the deformed tetrahedron is
V

=
1
2
[(Fp
3
Fp
0
) [(F(p
1
p
0
) F(p
2
p
0
)][ = det(F)V. (8)
12
2.2 Small strain and linearization
Assume the deformation gradient is small: F = I+H with << 1 and I being the identity matrix.
Let
E =
1
2
(F +F
T
) I =
1
2
(H+H
T
).
(9pt) 13. Let p
1
, p
2
, p
3
IR
3
be three points in . Without loss of generality we choose
p
0
= 0 be the origin. Show the following identities (9), (10) and (12). (Hints: Taylor expansion
with respect to the small number )
1. (The normal strain along direction p
1
= p
1
/[p
1
[ or stretch.) Assume [p
1
[ ,= 0, show
lim
0
1

[(p
1
F
T
Fp
1
)
1/2
[p
1
[] =
1
2
p
1
(H+H
T
)p
1
/[p
1
[. (9)
That is, to the leading order (c.f., (5)),
L

L
L
p
1
E p
1
.
2. (The change of volume.) Show that
lim
0
1

(det(F) 1) = Tr(H). (10)

That is, to the leading order (c.f., (8)),
V

V
V
Tr(E).
3. (The change of area.) First, we have
F
1
= (I +H)
1
I H. (11)
Assume q = p
2
p
1
,= 0. Show (Hints: use (11), (10) and (9).)
lim
0
1

[[Fp
2
Fp
1
[ [p
2
p
1
[] = Tr(H)[q[
1
2
q (H+H
T
)q/[q[. (12)
13
That is, to the leading order (c.f., (6)),
A

A
A
Tr(E) q E q where q = q/[q[.
2.3 Transformations of a deformation gradient
Consider a homogeneous deformation with y = Fx and F IR
33
. For the reference congu-
ration, we choose the orthonormal basis e
1
, e
2
, e
3
; for the current conguration, we choose the
orthornormal basis e
1
, e
2
, e
3
. In these bases, we have
y = y
p
e
p
, x = x
i
e
i
.
Let F
pi
=
y
p
x
i
be the numerical matrix of the deformation gradient. Then the tensor
F = y = F
pi
e
p
e
i
.
1. Passive viewpoint. We have a change of the bases:
e
1
, e
2
, e
3
e

1
, e

2
, e

3
, e
1
, e
2
, e
3
e

1
, e

2
, e

3
.
We assume the new bases are orthonormal as well. In terms of the old bases, the new bases
can be expressed as
e

i
= H
ij
e
j
, e

p
= Q
pq
e
q
.
(3pt) 14. Show that
H
ij
= e

i
e
j
, H
ik
H
jk
=
ij
or HH
T
= I;
Q
pq
= e

p
e
q
, Q
pr
Q
qr
=
pq
or QQ
T
= I.
14
In these new bases, the same tensor F admits the following decompositions:
F = F

pi
e

p
e

i
= F

pi
Q
pq
e
q
H
ij
e
j
= F

pi
Q
pq
H
ij
e
q
e
j
= F
pi
e
p
e
i
. (13)
The equality in the second line exactly means that tensor F remains unchanged in the course
of changing bases. Since the representation of a tensor in a basis is unique, from (13),
QQ
T
= HH
T
= I it folows
F

pi
Q
pq
H
ij
= F
qj
or F

pi
= Q
pq
F
qk
H
ik
. (14)
2. Active viewpoint. Assume H, Q So(3) be two rigid rotation matrix. Here we consider the
following composition of mappings:
x

= Hx x = H
T
x

rotate the reference conguration by H,

y

= Qy y = Q
T
y

rotate the current conguration by Q.

Then with respect to the old deformation x y(x), we consider the new deformation
given by

(x

) := Qy(x) = Qy(H
T
x

).
If since the old deformation is homogeneous, y = Fx, we have
F

=
x
y

(x

) = QFH
T
or F

= F

pi
e
p
e
i
= Q
pq
F
qk
H
ik
e
p
e
i
, (15)
which implies
F

pi
= Q
pq
F
qk
H
ik
. (16)
Note that the relation between the new matrix F

pi
and the old matrix F
pi
of the deformation
gradients are exactly the same for the two viewpoints, see (14) and (16).
(2pt) 15. Consider a vector V = V
1
e
1
+V 2e
2
in 2D and a rotation matrix
Q =
_
cos() sin()
sin() cos()
_
.
Draw two diagrams that show the dierences between the passive viewpoint and the active
viewpoint. (Hints: In one diagram, plot V, its components in the bases e
1
, e
2
, Qe
1
, Qe
2
;
in the other, plot V and its rotation QV, and their components in the bases e
1
, e
2
.)
15
2.4 Compatibility conditions
Let be a simply-connected open domain in IR
3
,
1
,
2
two disjoint open set, and an interface
such that =
1

2
.
1. Compatibility conditions for a vector eld V : IR
3
to be a gradient of a scalar potential.
(a) Assume V : IR
3
is dierentiable.
A necessary condition (curl-free condition) If V = , then
V = = c
ijk

,jk
= 0,
That is,
V = 0 on . (17)
A sucient condition. It turns out that equation (17) is also sucient. To see this,
let us dene
(x, ) =
_
(x
0
,x

)
V
i
(x)e
i
(dx

j
e
j
) =
_
(x
0
,x)
V
i
(x

)dx

i
.
The above integral in general depends on the integration path . However, if equa-
tion (17) is satised, by the Stokes theorem we have
_

c
V
i
(x

)dx

i
=
_

Vda = 0.
Assume there is a second path

between x
0
and x
1
. By the above equation we have
_

V
i
(x

)dx

i
=
_

c
V
i
(x

)dx

i
= 0, i.e., (x, ) = (x,

) =: (x).
Further, we verify that, indeed,
= V on .
Henceforth, we have proved that the curl-free condition (17) is a necessary and
sucient condition for a dierentiable vector eld V : IR
3
to be a gradient of
a scalar potential : IR.
Remark: In higher dimension, the necessary and sucient condition is that the
antisymmetrization of V vanishes on
1
2
(V(V)
T
) = 0.
16
(b) Assume V : IR
3
is dierentiable on
1

2
but discontinuous across .
A necessary and sucient condition
V = 0 on
1

2
, [[V]] t = 0 or [[V]] = c(x)n on , (18)
where c : IR is a scalar function on , n is the unit normal on and t is any
unit vector parallel to the surface, i.e., t n = 0.
(2pt) 16. Assume the vector eld V = , where : IR is continuous. Show that if
V is discontinuous across , then
[[V]] t = 0 or [[V]] = c(x)n on .
(Hint: Use equation (17) and the Stokes theorem.)
2. Compatibility conditions for a tensor eld F : IR
3
to be a gradient of a vector eld, i.e.,
a deformation gradient.
(a) Assume F : IR
33
is dierentiable.
A necessary condition. If F = y, we have
c
ijk
F
pj,k
= c
ijk
y
p,jk
= 0 i = 1, 2, 3&p = 1, 2, 3 on . (19)
A sucient condition. First, we may write F = F
pi
e
p
e
i
and dene
V
p
= F
T
( e
p
) = F
pi
e
i
: IR
3
.
From (19), we have
V
p
= c
ijk
(V
p
e
j
)
,k
= c
ijk
F
pi,k
= 0.
Applying (17) to each of the vector eld V
p
for p = 1, 2, 3, we see that (19) is also
sucient and
y
p
(x) =
_
(x
0
,x)
V
p
(x

) dx

j
e
j
=
_
(x
0
,x)
F
pi
(x

)dx

i
.
(b) Assume F : IR
33
is dierentiable on
1

2
but discontinuous across .
A necessary and sucient condition
F
T
( e
p
) = 0 or (19) on
1

2
, [[F]]t = 0 or [[F]] = a n on . (20)
3. Compatibility conditions for a symmetric tensor eld E : IR
33
sym
to be the symmetrized
gradient of a vector eld
E =
1
2
(u + (u)
T
), u : IR
3
. (21)
17
(a) Assume E : IR
33
sym
is of C
2
E =
_
_

xx

xy

xz

yx

yy

yz

zx

zy

zz
_
_
=
_
_
u
x,x
1
2
(u
x,y
+u
y,x
)
1
2
(u
x,z
+u
z,x
)
1
2
(u
x,y
+u
y,x
) u
y,y
1
2
(u
y,z
+u
z,y
)
1
2
(u
x,z
+u
z,x
)
1
2
(u
y,z
+u
z,y
) u
z,z
_
_
.
Note that E = E
T
.
A necessary condition:

xx
yz
=

x

2
u
x
yz
=

x
[

zy
x
+

zx
y


xy
x
] (22)

xy
xy
=
1
2
(

3
u
x
xy
2
+

3
u
y
yx
2
)
=
1
2
[

xx
y
2
+

2

yy
x
2
]
Similarly, we have ( (2pt) 17. Complete the right hand side of the following equa-
tions):

yy
xz
=

zz
xy
=

xz
xz
= (23)

yz
yz
=
A sucient condition: it can be shown that the above conditions are also sucient
for the existence of a displacement such that equation (21) admits a solution u for a
given strain eld that satises the compatibility equations (22)-(23). This was rst
shown by Volterra (1907).
(b) Assume E : IR
33
is of C
2
on
1

2
but discontinuous across .
A necessary and sucient condition for the existence of a continuous displacement
u : IR
3
such that (21) is satised is
E satises the compatibility equations (22)-(23) on
1

2
, and
[[E]] =
1
2
(a n +n a) on , (24)
where a : IR
3
is a vector eld on , and the jump condition is called Hadamards
jump conditions.
18
4. Assume V : IR
3
is dierentiable on
1

2
but discontinuous across . Let b : IR
is continuous and bounded. By
divV = b on , (25)
we mean the following
divV = b on
1

2
(26)
and
[[V]] n = 0 on . (27)
Note that V is discontinuous across , V is not dierentiable and so divV is not literally
well-dened on .
5. Assume : IR
33
is of C
1
on
1

2
but discontinuous across . Let b : IR
3
be
continuous and bounded. By
div = b on , (28)
we mean the following
div = b on
1

2
(29)
and
[[]]n = 0 on . (30)
To see the motivation behind the denition, e.g., (25)-(27), we assume the divergence theorem
is valid even though V and is discontinuous across . That is, we have for any subdomain
D ,
_
D
divV =
_
D
b =
_
D
V n.
Choosing D to be a domain as shown in the gure, we obtain (27).
19
3 Concepts of Stress
3.1 Cauchy stress and balance laws
Cauchy hypothesis: On any interface in the material body, there exists contact force between
the two parts separated by the interface. This contact force on an innitesimal area da can be
expressed as
(x, n)da,
where x is the position of the area element da and n is the unit normal on da, see the following
gure.
Cauchy theorem:
Theorem 1 The contact force can be expressed as
(x, n)da = (x)nda,
where : IR
33
is a tensor eld.
The meaning of this theorem is that the dependence of (x, n) on the unit normal n is linear.
Balance laws and their implications:
First, let us assume the following theorem.
Theorem 2 (Localization theorem) Let IR
n
be an open bounded domain. Assume f is a
continuous function.
1. If
_

fg = 0 continuous functions g : IR, or (31)

2. if
_
D
f = 0 subdomains D , (32)
then
f = 0.
Proof: We prove it by contradiction. Assume f ,= 0 at x
0
. By continuity of f, we know f > 0
on a neighborhood U of x
0
. Then equation (31) is violated for the choice of g which is a continuous
positive function and vanishes outside U or the subdomain D = U.
20
Let be the material body. Assume the body is subject to a body force b : IR
3
and
traction t : IR
3
and is in equilibrium. Consider a part of the body D with D = .
Then the balance of linear momentum implies
_
D
nda +
_
D
b = 0 =
_
D
(div +b) = 0; (33)
the balance of angular momentum implies
_
D
x (n)da +
_
D
x b = 0 =
_
D
[x (div +b) +c
ijk

kj
] = 0. (34)
From (33), (34), and part 2) of Theorem 2, we arrive at the equilibrium equation
div +b = 0 and =
T
on . (35)
Thus, the Cauchy stress is symmetric.
3.2 Implications of the rst and second laws of thermodynamics
.
We begin from the following hypothesis
H1. Assume the homogeneous material body has a deformation y : IR
3
. Let F = y be
the deformation gradient. At a constant temperature, the internal energy of the body can be
expressed as
U(y) =
_

W(y). (36)
Assume that the body is subject to a body force b : IR
3
, a traction t :
N
IR
3
, and
y = x on
D
, where
N
and
D
=
N
. In equilibrium, the deformation of the body is
y : IR
3
. Now let us imagine the deformation is perturbed innitesimally with new deformation
y

= y + z with z = 0 on
D
. By the rst law of thermodynamics, within the leading order we
have
U(y

) U(y) =
_

b z +
_

N
t z (37)
_

[z
W
F
(y) b z]
_

N
t z = 0.
Note that an immediate implication of the above formula is that the body is not moving, i.e., in
equilibrium, since we have not included the kinetic energy in (37). Dene S : IR
33
IR
33
as the
rst order derivative of the scalar function W : IR
33
IR
S =
W
F
or S
pi
=
W
F
pi
. (38)
For the deformation gradient y : IR
33
, we evaluate the function S and obtain the Piola-
Kirchho stress on point x as
S(y(x)).
21
Further, by an use of the divergence theorem, equation (37) can be rewritten as
_

z [divS(y) b] +
_

N
[S(y)n t] z = 0. (39)
If we a priori assume that [divS(y) b] and [S(y)n t] are continuous functions on and

N
, respectively, by the part 1) of the localization theorem 2, we conclude the equilibrium equation
and boundary conditions
_

_
div(S(y)) = b on ,
S(y)n = t on
N
,
y = x on
D
.
(40)
(6pt) 18. By index notation, carry out the derivation of (39) in details.
22
The second law of thermodynamics implies that when the body is in equilibrium, the Gibbs
free energy
G(y

) :=
_

[W(y

) b y

]
_

N
y

t. (41)
is minimized among all admissible deformation satisfying y

= x on
D
.
(6pt) 19. Assume that y : IR
3
is a minimizer of the energy functional G(y

):
G(y) = min
y

G(y

). (42)
Consider a small perturbation of y, i.e., y

= y +z with z = 0 on
D
. Show that the minimizer y
necessarily satises (37), and hence (40) if the relevant elds are a priori assumed to be continuous.
The PDE (40) is called the Euler-Lagrange equation of the variational principle (42).
Pialo-Kirchho stress vs Cauchy stress:
(2pt) 20. Explain in words the dierence between Pialo-Kirchhoff stress and Cauchy
stress. Draw a diagram if necessary.
4 Constitutive Laws
4.1 Hookes law
To complete our theory of elasticity, we need an additional equation that describes the properties
of the material. Such a equation or a law is called the constitutive law (of the material). The
classic Hookes law states that the stress depends on the strain linearly
= CE, (43)
23
where is the (Cauchy) stress tensor, E is the symmetrized strain, and C : IR
33
sym
IR
33
sym
is called
stiness tensor.
A constitutive law describes the behaviors of a material. At the continuum level, a constitutive
law cannot be derived from the laws of physics. Instead, we should think of the constitutive law is
obtained by experimental measurements. Nevertheless, the symmetries of the underlying material
and the physical space and the laws of physics place nontrivial restrictions on the forms of the
stiness tensor C. Though these restrictions are realized long time ago, a systematic derivation of
these restrictions mature after the development of the general framework of continuum mechanics.
The modern viewpoint on the Hookes law (43) is that it is the rst-order approximation of the
actual behavior of the material, and hence valid only for small strain. For materials, e.g., rubber
that remain to be elastic for large strain, the Hookes law fails completely.
4.2 Frame indierence and material symmetry
The modern viewpoint of the constitutive laws, instead of specifying the stress-strain relation,
species the stored energy function W : IR
33
IR, see (36). Consider a material body with a
homogenous deformation y = Fx. From (36), we see that the stored energy in the material is
U = W(F)[[,
where [[ is the volume of . Now let us consider a new deformation y

= Qy(x) with Q So(3),

i.e., the new conguration of the material body is an rigid rotation of y(). The principle of
relativity (Galilean invariance) tells us the stored energy in the material shall be the same as
before, i.e., (c.f. (15))
W(F) = W(QF) F IR
33
& Q So(3). (44)
The above equation is referred as the principle of material frame indierence, which is in fact valid
in a much broader context.
Material symmetry. Without loss of generality we may assume is a perfect sphere. Before
applying the deformation y : y(), we transform the reference conguration to

= R.
If the new reference conguration

is exactly the same as , then we will have

W(F) = W(FR
T
) F IR
33
& R (, (45)
where ( is called the symmetry group of the material. If the material is a crystal with a Bravais
lattice
L :=
3

i=1

i
a
i
:
i
are integers and a
i
are lattice vectors,
then ( is the point group L, which is the collection of orthogonal matrices R such that
L = RL.
Multiplying the above equation by R
T
, we see that if R (, then R
T
(.
Note that the frame-indierence (44) is material-independent, but equation (45) depends on the
material through the group (.
A material is isotropic if ( So(3).
(1pt) 21. Read the article http://en.wikipedia.org/wiki/Crystal

system and write below the

names of seven crystal systems.
24
4.3 Approximation of the stored energy functions
Clearly, the stored energy function W : IR
33
IR depends on the reference conguration. We
now make a choice of a reference conguration. Assume the body is free of body force and traction.
From the variational principle (42), we see that the deformation gradient shall take the matrix that
minimizes the energy function
W(F

) = min
FIR
33
,det(F)>0
W(F).
We assume the minimizer exists. If a minimizer does not exist, our solid wont be stable by the
second law of thermodynamics. If the minimizer F

,= I (I is the identity matrix), we redene our

energy function as W(F) = W(F

F). So, without loss of generality, we assume F

= I, i.e., y = x
is the minimizing deformation when the material is free of body force and surface traction. Upon
a Taylor expansion of the function W at a neighborhood of the point I, we have
W(I +H) = W(I) +H
W
F
(I) +
2
1
2
H CH+o(
2
), (46)
where
C =

2
W
FF
(I) or C
piqj
=

2
W
F
pi
F
qj
(I). (47)
22. Answer the following questions:
(i) (2pt) Explain why
W
F
(I) = 0 and C is positive semi-denite, i.e., H CH 0 H IR
33
.
(ii) (3pt) Read the article http://en.wikipedia.org/wiki/Matrix

exponential and show that for

any skew-symmetric matrix W, Q = exp(W) is a rigid rotation, i.e.,
QQ
T
= I and det(Q) = 1.
(iii) (2pt) Using the frame-indierence equation (44), show that the stiness tensor C dened
by (47) satises (Hint: consider rigid rotation Q = exp(W) and (46))
CW = 0 W
T
= W or C
piqj
= C
pijq
.
25
(iv) (2pt) Let ( be the symmetry group of the material. By (45), (44) and (46), show that
H CH = (RHR
T
) C(RHR
T
) H IR
33
& R (( So(3)). (48)
(Hint: rst show that W(I +H) = W(R(I +H)R
T
), and then use (46))
(v) (2pt) Assume the 180
0
rigid rotation around z-axis
R =
_
_
1 0 0
0 1 0
0 0 1
_
_
(.
By (48), show that
C
1113
= C
2213
= C
1213
= C
1123
= C
2223
= C
1223
= 0.
(vi) (2pt) If ( = So(3), from similar calculations as in the last problem, one can show that the
stiness tensor must take the following form
C
piqj
= (
pq

ij
+
pj

iq
) +
pi

qj
, (49)
where and are called Lam e constants. Show that the stored energy function
W(F) = W(I) +
1
2
(F I) C(F I), (50)
with C given by (49) violates the frame-indierence requirement (44).
Is there a contradiction? Why? In fact, there is no quadratic energy function as in (50) can
fulll the frame-indierence requirement (44). This is why you may see statements such as
there exists no linear material in nature or terminology such as linearized elasticity.
5 Formulations of Linear (Linearized) Elasticity Problems
5.1 PDE formulation, Variational formulation, Weak formulation
Assume y = x + u with u being small. Neglect the higher order terms in the energy function
expansion (46), we can write the stored energy function as in (50), where the stiness tensor is
positive semi-denite and satises
C
piqj
= C
qjpi
= C
pijq
.
26
Based on stability requirement, we shall further assume that for some c > 0,
E CE c[E[
2
E
T
= E.
Collecting results from previous chapters, in particular, by (38), (40), (50) we have the PDE
formulation of the linear elasticity problem
_

_
div[Cu] = b on ,
(Cu)n = t on
N
,
u = 0 on
D
.
(51)
By (38), (39), (50) we have the weak or integral formulation of the linear elasticity problem that
for any z = 0 on
D
,
_

z [div(Cu) b] +
_

N
[(Cu)n t] z = 0. (52)
By (42), (39), (50) we have the variational formulation of the linear elasticity problem that u is the
minimizer among all admissible v satisfying v = 0 on
D
G
l
(u) = min
v
_
G
l
(v) :=
_

[
1
2
v Cv b v]
_

N
t u
_
. (53)
The equivalence between the PDE formulation (51) and the weak formulation (52) follows from
part 2) of the localization Theorem 2 if all relevant quantities are a priori assumed to be continuous.
(5pt) 23. Show that if u is a solution (minimizer) of the variation problem (53), then u
necessarily satises that for any z C
1
with z = 0 on
D
,
_

[z Cu z b]
_

N
t z = 0,
which, by the divergence theorem, implies (52), and hence (51).
5.2 Uniqueness of the solution
(6pt) 24. Show that if u and u

are both minimizers of the variation problem (53), then u and

u

necessarily satises
_

[(u u

) + [(u u

)]
T
[
2
= 0,
27
which implies
1
2
(u u

) + [(u u

)]
T
= 0 on . (54)
(3pt) 25. By (54), show that there is a skew-symmetric matrix W IR
33
and a constant
vector c IR
3
such that
u u

= Wx +c on .
Well-posedness of a problem means the following three properties
Existence. The existence theorem is usually the most important question in a theory. However,
a proof of the existence theorem is usually dicult. We will not address the existence problem
in this course.
Uniqueness. Note that, though the displacement may not be unique in linear elasticity, the
strain and stress elds are unique.
Stability.
28
Homework 3, Problem 26-32, due on Tuesday, December 15, 2009
5.3 Plane strain problem
In a plane strain problem, we consider an innite cylindrical body = D (, +), where
D IR
2
is a two dimensional region on xy-plane. Consider the elasticity problem
_

_
div[Cu] = b
x
(x, y)e
x
+b
y
(x, y)e
y
on ,
(Cu)n = t
x
(x, y)e
x
+t
y
(x, y)e
y
on

N
(, +),
u = u
0
x
(x, y)e
x
+u
0
y
(x, y)e
y
on

D
(, +),
(55)
where b
x
, b
y
, t
x
, t
y
, u
0
x
, u
0
y
are given data and independent of z, C is independent of x, y, z, i.e., the
material is homogeneous,

N
,

D
are a mutual disjoint subdivision of D, see the following gure.
(2pt) 26 (i). Show that if u(x, y, z) satises (55), then for any z
0
IR,
u

(x, y, z) = u(x, y, z +z
0
) (56)
satises (55) as well.
From the uniqueness theorem, by (56) we conclude that u

(x, y, z) = u(x, y, z +z
0
) = u(x, y, z)
(the arbitrary translation is ignored). Therefore, any solution to (55) is in fact independent of z.
Thus,

z
u = 0. (57)
In another word,
u =
_
_
u
x,x
u
x,y
0
u
y,x
u
y,y
0
u
z,x
u
z,y
0
_
_
.
Let , ,

1, 2 or x, y, i.e., the in-plane indices. By (57), the stress can be written as

(summation over double indices , )
=
_
_

xx

xy

xz

yx

yy

yz

zx

zy

zz
_
_
=
_
_
C
11
u
,
+C
113
u
3,
C
12
u
,
+C
123
u
3,
C
13
u
,
+C
133
u
3,
C
21
u
,
+C
213
u
3,
C
22
u
,
+C
223
u
3,
C
23
u
,
+C
233
u
3,
C
31
u
,
+C
313
u
3,
C
32
u
,
+C
323
u
3,
C
33
u
,
+C
333
u
3,
_
_
,
29
and the equations in (55) are
_

_
C

+C
3
u
3,

= b

on D,
C
3

+C
33
u
3,

= 0 on D,
C

,
n

+C
3
u
3,
n

= t

on

N
,
C
3

,
n

+C
33
u
3,
n

= 0 on

N
,
u

= u
0

(x, y), u
3
= 0 on

D
.
(58)
For plane strain, we seek a solution satisfying
u
3
(x, y) = 0 on D. (59)
Using (55), equation (58) can be written as (the in-plane components)
_

_
C

= b

on D,
C

,
n

= t

on

N
,
u

= u
0

(x, y) on

D
,
(60)
plus the out-of-plane equations
_

_
C
3

= 0 on D,
C
3

,
n

= 0 on

N
,
u
3
= 0 on

D
.
(61)
Note that (61) is not automatically satised for all materials. However, if the stiness tensor
satises
C
3

= 0

, , 1, 2, (62)
then equation (61) is trivial.
The equation (60) is the plane strain problem. For given data b
x
, b
y
, t
x
, t
y
, u
0
x
, u
0
y
and domain
D, it can be shown it is a well-posed problem, i.e., there exists a solution, the strain is unique and
the solution is stable with respect to given data.
5.4 Antiplane shear problem
The antiplane shear problem is similar to the plane strain problem. The dierence is that the given
body force b, traction t, and displacement u
0
have only e
z
component. More precisely, we pose the
antiplane shear problem as
_

_
div[Cu] = b
z
(x, y)e
z
on ,
(Cu)n = t
z
(x, y)e
z
on

N
(, +),
u = u
0
z
(x, y)e
z
on

D
(, +).
(63)
By similar arguments as before, we conclude that u is independent of z. Further, we seek a solution
satisfying
u
x
(x, y) = u
y
(x, y) = 0 (x, y) D. (64)
30
By the above equations, the antiplane shear problem (63) can be written as (c.f. (58))
_

_
C
3
u
3,

= 0 on D,
C
33
u
3,

= b
z
on D,
C
3
u
3,
n

= 0 on

N
,
C
33
u
3,
n

= t
z
on

N
,
u
3
= u
0
z
(x, y) on

D
.
(65)
That is,
_

_
C
33
u
3,

= b
z
on D,
C
33
u
3,
n

= t
z
on

N
,
u
3
= u
0
z
(x, y) on

D
.
(66)
plus the in-plane components
_
C
3
u
3,

= 0 on D,
C
3
u
3,
n

= 0 on

N
.
(67)
Note again that equation (67) is not always trivial. If
C
3
= 0 , ,

1, 2, (68)
then indeed equation (67) is automatically satised regardless what values is u
3
(x, y). In this case,
the equation (66) is the governing equation for antiplane shear.
(6pt) 26 (ii). Consider an isotropic stiness tensor C with Lam e constants , .
(i) Is equation (62) equivalent to (68)? If yes, please explain why. Does the isotropic stiness
tensor C satises (62)?
(ii) Write down the equations in (60) in terms of , (instead of C
,

).
(iii) Write down the equations in (66) in terms of , (instead of C
,

).
31
5.5 Plane stress problem
In a plane stress problem, we consider a thin body = D(h/2, h/2), where h << 1 and D IR
2
is a two dimensional region on xy-plane. Consider the elasticity problem
_

_
div[Cu] = b
x
(x, y)e
x
+b
y
(x, y)e
y
on ,
(Cu)n = t
x
(x, y)e
x
+t
y
(x, y)e
y
on

N
(h/2, h/2),
(Cu)e
z
= 0 on x : z = h/2 x : z = h/2,
u = u
0
x
(x, y)e
x
+u
0
y
(x, y)e
y
on

D
(h/2, h/2),
(69)
where b
x
, b
y
, t
x
, t
y
, u
0
x
, u
0
y
are given data and independent of z, C is independent of x, y, z, i.e., the
material is homogeneous,

N
,

D
are a mutual disjoint subdivision of D.
Based on the boundary condition (Cu)e
z
= 0 on x : z = h/2 x : z = h/2 and the fact
that the body is a thin (h << 1), we seek a solution with stress eld of the following form
=
_
_

xx
(x, y)
xy
(x, y) 0

yx
(x, y)
yy
(x, y) 0
0 0 0
_
_
. (70)
Note that the stress eld is assumed to be independent of z. Let S = C
1
is the compliance tensor.
Then in terms of the compliance tensor, the strain can be written as

= S

,
3
= S
3

,
33
= S
33

, (71)
which is independent of z. Therefore, the displacement u is also independent of z. The rst of (71)
denes a linear mapping S

: IR
22
sym
IR
22
sym
with components given by S

. It can be shown S

is invertible and we denote by C

= S
1
. Therefore, if u

= (u
x
, u
y
),

=
_
u
x,x
u
x,y
u
y,x
u
y,y
_
, (72)
then the in-plane components of (69) can be written as
_

_
C

= b

on D,
C

,
n

= t

on

N
,
u

= u
0

(x, y) on

D
(h/2, h/2).
(73)
(6pt) 27. Consider an isotropic stiness tensor C with Lam e constants , .
32
(i) Calculate C

= (S

)
1
. That is, for a given 22 symmetric matrix

, what are C

?
(Hint: see the handout I gave in class)
(ii) Write down the equations in (73) in terms of , (instead of C

).
(iii) Compared with plane strain problem, list at least four dierences between plane strain problem
and plane stress problem (Hint: geometric feature of the bodies, out-of-plane component of
strain and stress, boundary conditions, the modulus, etc)
33
6 St Venants Problems and Semi-inverse Method
St Venants Principle: The elastic eld produced by a self-balanced force system on a local region
on the body is also local. More precisely, consider the Dirichlet problem or the Neumann problem
_
div[Cu] = b on ,
u = 0 on ,
or
_

_
div[Cu] = 0 on ,
u = 0 on
D
,
[Cu]n = t on
N
.
(74)
Let us denote the support of b or t by D
D = suppb := x : [b(x)[ ,= 0 or = x : [t(x)[ ,= 0.
If D is small and
_
D
b = 0 and
_
D
x b = 0 or
_

N
t = 0 and
_

N
x t = 0
then u(x) is small for points x that is far away from D.
St Venants Problem: Consider a long prismatic bar D(0, L), where D is the two dimensional
cross-section. Denote by the two end surfaces by S
0
and S
L
, and the side surface by S
s
:
S
0
= x : z = 0, S
L
= x : z = L, S
s
= D (0, L),
see the following gure.
Consider the problem
_

_
div[Cu] = 0 on ,
[Cu]n = 0 on S
s
,
[Cu]n = t on S
L
,
u = 0 on S
0
.
(75)
From the general theory of linear elasticity, we know the above problem admits a unique solution.
By St Venants principle, we can instead consider a relaxed problem
_

_
div[Cu] = 0 on ,
[Cu]n = 0 on S
s
,
_
S
L
[Cu]n =
_
S
L
t = R,
_
S
L
x [Cu]n =
_
S
L
x t = M.
(76)
34
Clearly, the solution of (75) satises (76). In general equation (76) admits innitely many solutions.
However, since the scale of the cross-section D is small compared with L, by the St Venants
principle, we conclude that any solution of (76) is a good representation of the true solution of the
original problem (75). In practice, we of course shall focus on the solution of (76) that can be made
as simple as possible.
Classication of St Venants Problems: For St Venants problems, what is given is the domain
= D (0, L), and the total force R IR
3
and torque M IR
3
applied on one end of the bar,
say, S
L
. According to the directions of vector R and M, we classify the St Venants problems as
follows.
(i) Simple extension: R | e
z
, M = 0.
(ii) Pure bending: R = 0, M e
z
.
(iii) Torsion: R = 0, M | e
z
.
(iv) Bending: R e
z
, M = 0.
To solve the St Venants problems (76), we often focus on the stress eld : IR
33
sym
. In
terms of stress eld (instead of the displacement), the St Venants problems (76) can be rewritten
as
_

_
div = 0 on ,
n = 0 on S
s
,
_
S
L
n =
_
S
L
t = R,
_
S
L
x n =
_
S
L
x t = M.
(77)
plus the compatibility equation: the strain eld
_
_

xx

xy

xz

yx

yy

yz

zx

zy

zz
_
_
satises (22)-(23), where
_
_

xx

xy

xz

yx

yy

yz

zx

zy

zz
_
_
= C
1
. (78)
If C is homogeneous and isotropic with Youngs modulus E and Poissons ration , then
_
_

xx

xy

xz

yx

yy

yz

zx

zy

zz
_
_
=
1
E
_
_

xx
(
yy
+
zz
) (1 +)
xy
(1 +)
xz
(1 +)
yx
(1 +)
yy
(
xx
+
zz
) (1 +)
yz
(1 +)
zx
(1 +)
zy

zz
(
yy
+
xx
)
_
_
. (79)
Thus, in terms of the stress eld, the compatibility equations (22)-(23) can be written as

2
yz
(
xx
(
yy
+
zz
)) = (1 +)

x
[

zy
x
+

zx
y


xy
x
]
(1 +)

xy
xy
=
1
2
[

2
y
2
(
xx
(
yy
+
zz
)) +

2
x
2
(
yy
(
xx
+
zz
))] (80)
.
.
.
.
.
.
.
.
.
In particular, if there is no body force, i.e., div = 0 on , then the above equation can be rewritten
as

2
+
1
1 +
Tr() = 0 on . (81)
This equation is called Michell stress compatibility equation.
35
6.1 Simple extension
R = R
z
e
z
and M = 0. By the last of (77), we make a guess on the stress eld : IR
33
sym
=
_
_

xx

xy

xz

yx

yy

yz

zx

zy

zz
_
_
=
_
_
0 0 0
0 0 0
0 0
R
z
|D|
_
_
, (82)
where [D[ denotes the area of the cross-section D. Note that is constant throughout the domain
D. Further, the equilibrium equation (the rst of (77)) and the compatibility equation (78) is
automatically satised. Thus, we conclude that (82) is indeed a solution of the St Venants problem
(77) and (78) with R = R
z
e
z
and M = 0 for general anisotropic medium.
To nd strain and displacement when the medium is isotropic, by (82) and (79) we have
_
_

xx

xy

xz

yx

yy

yz

zx

zy

zz
_
_
=
1
E
_

R
z
|D|
0 0
0
R
z
|D|
0
0 0
R
z
|D|
_

_
. (83)
Thus, a solution of displacement (u
x
, u
y
, u
z
) to
1
2
(u + (u)
T
) = E is
_

_
u
x
=
R
z
E|D|
x
u
y
=
R
z
E|D|
y
u
z
=
R
z
E|D|
z
. (84)
6.2 Pure bending
R = 0 and M = M
x
e
x
+M
y
e
y
. By the last of (77), we make a guess on the stress eld : IR
33
sym
=
_
_

xx

xy

xz

yx

yy

yz

zx

zy

zz
_
_
=
_
_
0 0 0
0 0 0
0 0 A
x
x +A
y
y
_
_
, (85)
where A
x
, A
y
are constants to be determined by the third of (77). Note that is linear throughout
the domain D. Thus, the equilibrium equation (the rst of (77)) and the compatibility equation (78)
is automatically satised. Thus, we conclude that (82) is indeed a solution of the St Venants
problem (77) and (78) with R = 0 and M = M
x
e
x
+M
y
e
y
for general anisotropic medium if
_
R =
_
S
L
(A
x
x +A
y
y)e
z
= 0,
M =
_
S
L
x (A
x
x +A
y
y)e
z
= M
x
e
x
+M
y
e
y
.
(86)
Choose the centroid of S
L
as our origin, then the rst of (86) is automatically satised. Choose
e
x
, e
y
such that the momentum of inertia is diagonized
_
S
L
_
x
2
xy
yx y
2
_
= diag[I
xx
, I
yy
].
By the second of (86), we obtain
A
x
=
M
y
I
xx
, A
y
=
M
x
I
yy
. (87)
36
To nd strain and displacement when the medium is isotropic, by (85) and (79) we have
_
_

xx

xy

xz

yx

yy

yz

zx

zy

zz
_
_
=
_

_
(x
M
y
EI
xx
+y
M
x
EI
yy
) 0 0
0 (x
M
y
EI
xx
+y
M
x
EI
yy
) 0
0 0 x
M
y
EI
xx
+y
M
x
EI
yy
_

_
. (88)
Thus, a solution of displacement (u
x
, u
y
, u
z
) to
1
2
(u + (u)
T
) = E is
_

_
u
x
=
M
y
2EI
xx
(z
2
(x
2
y
2
)) xy
M
x
EI
yy
,
u
y
=
M
x
2EI
yy
(z
2
+(x
2
y
2
)) +xy
M
y
EI
xx
,
u
z
= xz
M
y
EI
xx
+yz
M
x
EI
yy
.
(89)
6.3 Torsion
R = 0 and M = M
z
e
z
. We assume the material is isotropic. By physical intuition, we make a guess
on the stress eld : IR
33
sym
=
_
_

xx

xy

xz

yx

yy

yz

zx

zy

zz
_
_
=
_
_
0 0
xz
0 0
yz

xz

yz
0
_
_
, (90)
where
xz
,
yz
: D IR, depending on (x, y) but independent of z, are shear stress to be determined.
By (77) we have
_

xz
x
+

yz
y
= 0 on D,

xz
n
x
+
yz
n
y
= 0 on D,
_
S
L
(
xz
e
x
+
yz
e
y
) = 0,
_
S
L
x (
xz
e
x
+
yz
e
y
) = M
z
e
z
.
(91)
Note that
_
S
L
x (
xz
e
x
+
yz
e
y
) =
_
S
L
z
yz
e
x
+z
xz
e
y
+ (x
yz
y
xz
)e
z
.
Further, by the compatibility equation (81), we have

yz
=
2

xz
= 0 on . (92)
By the rst of (91), we have [
yz
,
xz
, 0] = 0, i.e., the vector eld [
yz
,
xz
] is curl-free.
Therefore, the following scalar potential
F(x, y) =
_
(x,y)
(x
0
,y
0
)
[
yz
,
xz
] dl.
is well-dened and

xz
=

y
F(x, y),
yz
=

x
F(x, y) (x, y) D. (93)
Plugging the above equation to the stress compatibility equation (92) we obtain

2
F = 0,

x

2
F = 0 on D.
37
We thus conclude

2
= 2, F = on D, (94)
where is the shear modulus, is a constant to be determined, and is called Prandl stress
function. Let (n
x
, n
y
) be the unit normal on D at a point p D and (t
x
, t
y
) be the unit tangential
unit vector along D at the same point p. Clearly,
[t
x
, t
y
] [n
x
, n
y
] = 0 = (t
x
, t
y
) = (n
y
, n
x
). (95)
Further, from the second of (91), (93) and (95) we have
[t
x
, t
y
] F = t
x
F
x
+t
y
F
x
= (n
y

yz
+n
x

xz
) = 0 on D.
Therefore, we conclude that for some c IR,
F = c on D. (96)
To see the consequence of the last of (91), we rst verify the total force
_
S
L

xz
=
_
S
L
F
y
=
_
D
[n
x
, n
y
] [0, 1]F = c
_
D
n
y
= c
_
S
L
div[0, 1] = 0.
Similarly, we have
_
S
L

yz
= 0. Second, we calculate the total torque in e
z
direction
_
S
L
(x
yz
y
xz
) =
_
S
L
(x

x
+y

x
) =
_
S
L
(
x
x
+
y
x
2)
= [
_
S
L
(n
x
x +n
y
y)
_
S
L
2] = 2 = M
z
, (97)
where
=
_
S
L

1
2
_
S
L
(n
x
x +n
y
y) (98)
is called torsion stiness.
Note that torsion stiness is an geometric property of the cross-section D, i.e., S
L
= D. To
calculate the torsion stiness, from (94) and (96) we notice that the stress function satises that
for some constant c

IR,
_
= 2 on D,
= c

on D.
(99)
28. Consider the torsion of a long prismatic bar = D (0, L), i.e., the problem (75) with
R = 0 and M = M
z
e
z
. Assume the material is isotropic with Lam e constants , . As discussed
above, the torsion stiness is determined by (99) and (98).
(i) (2pt) Show that the torsion stiness is independent of the constant c

.
(ii) (2pt) Assume D = (x, y) : x
2
+y
2
< R
2
is the circle centered at the origin and with radius
R. Solve (99) for c

= 0 and calculate the torsion stiness .

38
(iii) (4pt) By (93) and (94), calculate the stress eld
xz
,
yz
, the strain eld, and the displacement
(Assume u = 0 if z = 0). (Note that the elds are functions dened on ).
(iv) (2pt) If we express the displacement u
x
, u
y
in the following form
_
u
x
u
y
_
= W
_
zx
zy
_
,
where W IR
22
. Is it true W
T
= W? Sketch a gure that illustrates the geometric
meaning of the constant in (94). What is the relation between and M
z
, ?
6.4 Bending
39
7 Important Problems and their Solutions
7.1 The fundamental solution Kelvins solution
Formulation of the problem. Consider an innite homogeneous body with an elastic stiness
tensor C in IR
n
. At the origin, there is a concentrated body force b(x) = b
0
(x), where (x) is the
Dirac function that satises
_
IR
n
f(x)(x) = f(0) f C

0
(IR
n
).
We are interested in nding the elastic eld, i.e., solving the following problem for u : IR
n
IR
n
,
_
div[Cu] = b
0
(x) on IR
n
,
[u[ 0 as [x[ .
(100)
Solution. We will solve this problem by the Fourier method. First let us recall the denition of
Fourier transformation and the inversion theorem. Let f : IR
n
IR be a generalized function
(Precisely, f needs to be a tempered distribution. We notice that the Fourier transformation
and the inversion theorem are valid for almost all functions we encountered in elasticity, including
the Dirac function and its derivatives). Then the Fourier transformation of f, denoted by

f : IR
n
IR, is given by

f(k) =
_
IR
n
f(x) exp(ik x)dx. (101)
Further, the Fourier inversion theorem holds
f(x) =
1
(2)
n
_
IR
n

f(k) exp(ik x)dk. (102)

29 (i). (2pt) Let u(k) be the Fourier transformation of the solution of (100). Show that u(k)
satises
u(k) = N(k)b
0
, (103)
where N(k) IR
nn
sym
is the inverse of the matrix C
piqj
k
i
k
j
.
Equation (103) is the k-space equation corresponding to (100). In eect, the Fourier transfor-
mation converts a partial dierential equation, e.g., (100) into an algebraic equation, e.g. (103).
This is possible because our problem is homogeneous in the sense that the material properties C is
independent of the position x and the domain is special, i.e., IR
n
.
An application of the inversion theorem to (103) yields
u(x) =
1
(2)
n
_
IR
n
N(k)b
0
exp(ik x)dk. (104)
Further, if C is an isotropic stiness tensor with Lam e constants , (see (49)), we nd
N(k) =

[k[
2
I

[k[
4
k k. (105)
40
29 (ii). (2pt) Calculate the constants , in terms of , .
Plugging (105) into (104), we arrive at
u(x) = (x)b
0
[(x)]b
0
, (106)
where
(x) =
1
(2)
n
_
IR
n
1
[k[
2
exp(ik x)dk,
(x) =
1
(2)
n
_
IR
n
1
[k[
4
exp(ik x)dk.
29 (iii). (2pt) Show that the above functions and satisfy
(x) = (0), (x) = (0). (107)
By symmetry, we seek solutions to (107) that can be written as
= (r), = (r),
where r = (x
2
1
+ +x
2
n
)
1/2
. In this case, we have
(r) = (r)
,x
i
x
i
= (

(r)r
,x
i
)
,x
i
=

(r) +
n 1
r

(r), (108)
where ( )

=
d
dr
( ).
29 (iv). (2pt) Show in detail the identity (r) = (

(r)r
,x
i
)
,x
i
=

(r) +
n1
r

(r).
Therefore, by (107) we have

(r) +
n 1
r

(r) = 0 r > 0. (109)

From the theory of ordinary dierential equation, we know that a solution to (109) can be written
as
(r) =
_
C
2
log(r) +C
0
if n = 2,
C
n
1
r
n2
+C
0
if n 3,
(110)
41
where C
i
are constants determined by the boundary conditions. In particular, the constant C
0
is
immaterial since it does not aect the strain and stress eld, see (106). It is the usual convention
that we choose it to be zero. Further, to nd the constants C
i
(i = 2, ), we integrate the rst of
(107) over a unit ball centered at the origin B
1
and obtain
1 =
_
B
1
(0) =
_
B
1
(r) =
_
B
1
e
r
(r) =
_
B
1

(r)
29 (v). (2pt) Plug (110) into the above identity and nd out what is the constants C
2
and
C
3
. (Bonus problem: 1pt, nd C
n
for n 4)
We now calculate the other potential function (r) in (107). From (107), (108), and (110), we
see that
(r) =

(r) +
n 1
r

(r) = (r) =
_
C
2
log(r) if n = 2,
C
n
1
r
n2
if n 3.
(111)
Again, from the theory of ordinary dierential equation, we nd
(r) =
_

_
D
2
(r
2
1) log(r) if n = 2,
D
n
r
4n
if n 3, n ,= 4,
D
4
log(r) if n = 4.
(112)
29 (vi). (2pt) Verify that the above dened (r) indeed satises (111) for appropriately
chosen constants D
i
IR and nd the relation between the constants D
i
and C
i
for i = 2, 3.
(Bonus problem: 1pt, nd D
i
for i 4)
In summary, from (106), (110) and (112), we conclude that the solution to (100) is
u(x) = G
(n)
(x)b
0
,
G
(n)
(x) =
_

_
C
2
log(r)I D
2
[(1 2 log(r))I 2e
r
e
r
] if n = 2,
r
2n
C
n
I D
n
[(4 n)(2 n)e
r
e
r
+ (4 n)I] if n 3, n ,= 4,
r
2n
C
n
I D
n
[2e
r
e
r
+I] if n = 4,
(113)
where e
r
= x/[x[, and we have used the identity
(r) = (

(r)

(r)
r
)e
r
e
r
+

(r)
r
I.
The above solution (113) is called the Kelvins solution in three dimensions (n = 3). It is also
referred to as the Greens function or the fundamental solution.
Direct calculations verify that for any b [C
0
(IR
n
)]
n
, the solution to
_
div[Cu] = b(x) on IR
n
,
[u[ 0 as [x[
(114)
42
is given by
u(x) =
_
IR
n
G
(n)
(x x

)b(x

)dx

,
which is called the Greens formula.
43
7.2 Pressured vessel
Formulation of the problem. Consider a homogeneous body with an elastic stiness tensor C
on an annulus region = x : r
a
< [x[ < r
b
subject to a hydrostatic pressure p
a
from inside and
pressure p
b
from outside. We are interested in nding the elastic eld, i.e., solving the following
problem for u : IR
n
,
_

_
div[Cu] = 0 on ,
(Cu)e
r
= p
a
e
r
if r = r
a
,
(Cu)e
r
= p
b
e
r
if r = r
b
.
(115)
Solution. For simplicity, we assume the stiness C is isotropic. First, we show that if u is a
solution to (115), then
u

(x

) = Qu(x), x

= Qx (116)
is also a solution to (115), where Q So(n) is any rigid rotation. To see this, by the chain rule we
have
u
q,x
j
= Q
sq
u

s,x
j
= Q
sq
u

s,x

k
x

k,x
j
= Q
sq
Q
kj
u

s,x

k
C
piqj
u
q,x
j
x
i
= (C
piqj
Q
sq
Q
kj
u

s,x

k
)
,x

l
x

l,x
i
= C
piqj
Q
sq
Q
kj
Q
li
u

s,x

k
x

l
.
Since u satises the rst of (115), by the second of the above equation we obtain
C
piqj
Q
sq
Q
kj
Q
li
Q
rp
u

s,x

k
x

l
= 0 i.e., div
x
[C

x
u

(x

)] = 0 x

Q = , (117)
where
(C

)
piqj
= (C)
p

j
Q
pp
Q
qq
Q
ii
Q
jj
.
Since C is an isotropic tensor, by (48) we see that C

= C, and hence equation (117) is identical to

the rst of (115). Similarly, we can verify u

satisfy the second and third of (115) as well. We thus

conclude that u

is also a solution of (115) if u is so. From the uniqueness theorem, we infer that
u(x) = Qu(Qx) x & Q So(n) = u(x) = u
r
(r)e
r
.
Therefore, the strain
u
p,i
= (
u
r
r
x
p
)
,x
i
= (
u
r
r
)

x
i
x
p
r
+
u
r
r

pi
, (118)
44
the stress

pi
= C
piqj
u
q,j
= 2[(
u
r
r
)

x
i
x
p
r
+
u
r
r

pi
] +[(
u
r
r
)

r +n
u
r
r
]
pi
, (119)
and
div[Cu] = C
piqj
u
q,ji
= 2[(
u
r
r
)

x
i
x
p
r
]
,i
+ 2(
u
r
r
)
,p
+[(
u
r
r
)

r +n
u
r
r
]
,p
=
x
p
r
_
2[(
u
r
r
)

r + (n + 1)(
u
r
r
)

] +[(
u
r
r
)

r + (n + 1)(
u
r
r
)

]
_
(120)
30 (i). (2pt) Start from (119). Show (120) in details.
Thus, the rst of (115) implies
(
u
r
r
)

r + (n + 1)(
u
r
r
)

= 0.
From the theory of ordinary dierential equation, we conclude that
u
r
r
=
r
n
n
+C
1
r +C
0
,
where the constants C
0
, C
1
are determined by the boundary conditions in (115).
30 (ii). (3pt) Find the constants C
1
, C
0
in terms of p
a
, p
b
, r
a
, r
b
.
30 (iii). (3pt) Let n = 2 and e

= [x
2
, x
1
]/r. Calculate the stress tensor in the frame
e
r
, e

, i.e., calculate the following matrix

=
_

rr

r

_
=
_
e
r
e
r
e

e
r
e
r
e

_
,
where the stress is given by (119) in the rectangular frame e
1
, e
2
.
30 (iv). (2pt) Assume n = 2, r
b
= cr
a
(c > 1), p
b
= 0 but p
a
,= 0. What is

/p
a
at
r = a? Sketch the curve

(c) below.
45
7.3 Half-space problem Boussinesqs solution
Formulation of the problem. Consider a homogeneous body with elastic stiness tensor C in
the half space H = x IR
n
: z < 0. Our problem is to solve for u : H IR
n
(n = 3) such that
_

_
div(Cu) = 0 on H,
(Cu)e
z
= t(x, y) on H,
[u[ 0 as z .
(121)
Solution. Let k

= (k
x
, k
y
), x

= (x, y) IR
n1
be the in-plane wave vector and position vector,
respectively. Upon Fourier transformation on variable (x, y), the rst of (121) can be written as
C
pq
( u
q
k

) +iC
pq3
k

d u
q
dz
+iC
p3q
k

d u
q
dz
+C
p3q3
d
2
u
q
dz
2
= 0, (122)
where , 1, 2 are in-plane indices,
u(k
x
, k
y
, z) =
_
IR
n1
u(x, y, z) exp(ik

)dxdy. (123)
Similarly, the second of (121) can be written as
iC
p3q
u
q
k

+C
p3q3
u
q
dz
=

t
p
(k
x
, k
y
) =
_
IR
n1
t
p
(x, y) exp(ik

)dxdy. (124)
Denote by
T
pq
= C
p3q3
, R
pq
= iC
pq3
k

(i.e., R
qp
= iC
p3q
k

), Q
pq
= C
pq
k

. (125)
The equation (122) can be written as
T
pq
d
2
u
q
dz
2
+ (R
pq
+R
qp
)
d u
q
dz
+Q
pq
u
q
= 0. (126)
Let
= T
d u
dz
+R
T
u, =
_
u

_
. (127)
Note that
p
= R
qp
u
q
+T
pq
d u
q
dz
= iC
p3q
u
q
k

+C
p3q3
u
q
dz
is the Fourier transformation of the traction
(Cu)e
z
. By (126) and (127) we arrive at
d
dz
= H, H =
_
T
1
R
T
T
1
RT
1
R
T
Q RT
1
_
. (128)
46
Thus, in general a solution to (122) can be expressed as
u = [I, 0] exp(zH)
0
, (129)
where the constant vector
0
is determined by the boundary conditions at z and z = 0.
Example 7.1 For an isotropic material with the Lam e constants , (c.f. (49)), the matrices Q,
R, T in Eq. (125) can be identied as
T =
_
_
0 0
0 0
0 0 2 +
_
_
, R =
_
_
0 0 ik
x
0 0 ik
y
ik
x
ik
y
0
_
_
,
Q =
_
_
[k

[
2
+ ( +)k
2
x
( +)k
x
k
y
0
( +)k
x
k
y
[k

[
2
+ ( +)k
2
y
0
0 0 [k

[
2
_
_
. (130)
From Eqs. (128), direct calculations reveal that
H(k
x
, k
y
) =
_

_
0 0 ik
x
1

0 0
0 0 ik
y
0
1

0
k
x

i(+2)
k
y

i(+2)
0 0 0
1
2+
4(+)k
2
x
2+
+k
2
y
(2+3)k
y
k
x
2+
0 0 0
k
x

i(+2)
(2+3)k
y
k
x
2+
4(+)k
2
y
2+
+k
2
x
0 0 0
k
y

i(+2)
0 0 0 ik
x
ik
y
0
_

_
. (131)
The Jordan normal form of H(k
x
, k
y
) is (k
z
= [k

[)
H(k
x
, k
y
) = P
_

_
k
z
0 0 0 0 0
0 k
z
1 0 0 0
0 0 k
z
0 0 0
0 0 0 k
z
0 0
0 0 0 0 k
z
1
0 0 0 0 0 k
z
_

_
P
1
,
where the column vectors in P IR
66
are the generalized eigenvectors. Note that the eigenvalues
of H(k
x
, k
y
) are [k

[ and the above decomposition is unique. By (129) we conclude that

u = (a +zb) exp(k
z
z), (132)
and by (127)
= [Tb + (R
T
+k
z
T)a +z(R
T
+k
z
T)b] exp(k
z
z) (133)
Plugging (132) and (133) back to (128) or (126) we nd that for all z < 0,
z[k
2
z
T+k
z
(R+R
T
) +Q]b + [k
2
z
T+k
z
(R+R
T
) +Q]a + (2k
z
T+R+R
T
)b = 0,
which implies
_
[k
2
z
T+k
z
(R+R
T
) +Q]b = 0
[k
2
z
Tk
z
(R+R
T
) +Q]a + (2k
z
T+R+R
T
)b = 0
(134)
47
Direct calculations show that the rank of k
2
z
T+k
z
(R+R
T
) +Q is two (because k
z
is an eigenvalue
of H(k
x
, k
y
), k
2
z
T+k
z
(R+R
T
) +Q is necessarily singular) and equation (134) is equivalent to
_

_
b
1
k
x
b
2
k
y
+ib
3
k
z
= 0
b
1
k
y
b
2
k
x
= 0
k
x
a
1
+k
y
a
2
ik
z
a
3
i(3 4)b
3
= 0
(135)
Further, the boundary condition (124) implies
Tb + (R
T
+k
z
T)a =

t. (136)
Equations (135)-(136) have six unknowns and six equations. The coecient matrix is non-singular
and hence they admit a unique solution for any given

t. In particular, if

t = [0, 0,

t
z
], we nd that
_

_
a
1
=
k
x
(12)

t
z
2ik
2
z
,
a
2
=
k
y
(12)

t
z
2ik
2
z
,
a
3
=
(1)

t
z
k
z
,
_

_
b
1
=
k
x

t
z
2ik
z
,
b
2
=
k
y

t
z
2ik
z
,
b
3
=

t
z
2
.
(137)
Further, if t(x) = P
2
(0)e
z
, then

t(k
x
, k
y
) = Pe
z
(k
x
, k
y
) IR
2
.
Thus, by (132) and (137) we obtain
u(k
x
, k
y
, z) = P
_

_
k
x
(12)
2ik
2
z
k
y
(12)
2ik
2
z
(1)
k
z
_

_
exp(k
z
z) P
_

_
k
x
2ik
z
k
y
2ik
z

1
2
_

_
z exp(k
z
z). (138)
Let
(x, y, z) =
1
(2)
2
_
IR
2
1
k
2
x
+k
2
y
exp(ik
x
x +ik
y
y + (k
2
x
+k
2
y
)
1/2
z)dk
x
dk
y
. (139)
Then, by (138) we arrive at
u(x, y, z) =
1
(2)
2
_
IR
2
u(k
x
, k
y
, z) exp(ik
x
x +ik
y
y)dk
x
dk
y
,
which, by (156), can be identied as
_
_
u
x
u
y
u
z
_
_
= P
_

_
(12)
2

,x
(12)
2

,y
(1)

,z
_

_
P
_

_
1
2
z
,zx
1
2
z
,zy

1
2
z
,zz
_

_
. (140)
To evaluate , let r = (x
2
+y
2
)
2
, cos = (k
x
x +k
y
y)/rk
z
, and the integral formula that for C ,
_

0
exp(t)dt =
1

,
_

0
1
t
exp(t)dt = log .
48
Thus,
(x, y, z) =
1
(2)
2
_

0
_
2
0
1
k
z
exp(ik
z
r cos +k
z
z)ddk
z
=
1
2(2)
2
_
2
0
log(z +ir cos )d
=
1
2(2)
2
_
2
0
log(z
2
+r
2
cos
2
)d
and hence
_
_

,x

,y

,z
_
_
=
1
2
_

x(R+z)
rR

y(R+z)
rR
1
R
_

_
,
where R = (z
2
+r
2
)
1/2
. Finally, by (140) we conclude that for any (x, y, z) H,
_
_
u
x
u
y
u
z
_
_
=
P
4
_

_
xz
R
3
+ (1 2)
x
R(Rz)
yz
R
3
+ (1 2)
y
R(Rz)

z
2
R
3
(1 2)
1
R
_

_
. (141)
From the above equation, if there is a distributed surface load p(x, y) in e
z
direction on x :
z = 0, then the displacement is given by the Greens formula
_
_
u
x
u
y
u
z
_
_
=
1
4
_

_
z
_
p(x

,y

)(xx

)
R
3
+ (1 2)
_
p(x

,y

)(xx

)
R

(R

z)
z
_
p(x

,y

)(yy

)
R
3
+ (1 2)
_
p(x

,y

)(yy

)
R

(R

z)
z
2
_
p(x

,y

)
R
3
(1 2)
_
p(x

,y

)
R

_
, (142)
where
R

= [(x x

)
2
+ (y y

)
2
+z
2
]
1/2
.
49
31. Consider a homogeneous half space H = x IR
3
: z < 0. For given k > 0, let
u : H IR
3
be given by
u(x) = (a +zb) exp(kz) exp(ikx), (143)
where a, b IR
3
are independent of position x. Consider the equilibrium equation
div(Cu) = 0 on H, (144)
where C is an isotropic stiness tensor, i.e., the above equation is equivalent to
u + ( +)( u) = 0 on H. (145)
(i) (3pt) Plug (143) into (145). Find the conditions on the vectors a and b such that (145) is
satised.
(ii) (3pt) Calculate the traction on the plane x : z = 0 in terms of a, b, i.e., the quantity
t(x, y, z = 0) = (Cu)e
z

z=0
(iii) (4pt) Assume the traction
(Cu)e
z

z=0
=

t
0
exp(ikx),
where

t
0
= [t
0
1
, 0, t
0
3
] IR
3
is given. Solve for vectors a, b in terms of

t
0
, k, , .
50
7.4 Contact problem Hertzs solution
Formulation of the problem. Consider a homogeneous body with an elastic stiness tensor C
on a half-sphere = x : x
2
+ y
2
+ (z + R
0
)
2
< R
2
0
, z > R
0
. Assume the base of the body is
hold still, i.e., u = 0 on
D
= x : z = R
0
and is initially in contact with a rigid plate
at the point x = 0. Assume that there is no friction between the rigid plate and . Now we push
down the rigid plate by a small amount h in z-direction. We are interested in the elastic eld,
i.e., solving for u : IR
3
,
_

_
div(Cu) = 0 on ,
(Cu)n = 0 on
C
,
u = 0 on
D
,
(x +u(x)) e
z
h x ,
(Cu)n = p(x, y)e
z
on
C
,
(146)
where the contact area

C
= (x, y, z) : z +u
z
(x, y, z) = h. (147)
Solution. Let

C
= (x, y) : (x, y, z)
C
be the contact area in the current conguration
(projected to the xy-plane). Then for a point (x, y, z)
D
, we have
z
1
2

=
1
R
0
(x
2
+y
2
). (148)
Since h << 1, the contact area is small compared with the curvature of at the contact point.
Therefore, near the contact point, the solution to (146) is given by the Boussinesqs solution (142).
In particular,
u
z
(x, y, z = 0) =
1 2
4
_

C
p(x

, y

)
r

dx

dy

, (149)
where
r

= [(x x

)
2
+ (y y

)
2
]
1/2
.
From (147), (148) and (149), we have

1
2R
0
(x
2
+y
2
)
1 2
4
_

C
p(x

, y

)
r

dx

dy

= h (x, y)

C
. (150)
51
The above equation is peculiar in the sense that we need to determined the unknown function
p(x

, y

) and the unknown domain

C
simultaneously. There is a general theory called the variational
inequalities which addresses such kind of free-boundary problems, see e.g. Freedman (1982).
Below, by our knowledge of potential theory and elliptic integrals, we solve such an equation which
includes (150) as a special case

1
x
2
+
2
y
2
+
1
2
_

f(x

, y

)
r

dx

dy

= h (x, y) , (151)
where
1
,
2
> 0 are two constants, f : IR and IR
2
are to be determined.
First, let us recall a theorem.
Theorem 3 Let = x :

n
i=1
x
2
i
a
2
i
= 1 be an ellipsoid in IR
n
. Consider a Poissons equation
(x) =

(x) on IR
n
,
where IR is a constant and

= 1 on but vanishes otherwise. Then the solution

(x) =
_

(x x

)dx

(152)
satises
(x) = D

1
2
x Q

x x , (153)
where is the Greens function given by (110), the constant D

> 0 and the symmetric matrix Q

are given by (158) and (161), respectively.

Proof: By the Fourier method, we have that for any x IR
n
,
(x) =

(2)
n
_
IR
n
1
[k[
2

(k) exp(ik x)dk,

(x) =

(2)
n
_
IR
n
k k
[k[
2

(k) exp(ik x)dk, (154)

where

(k) =
_

exp(ik x)dx.
Let
k = [k[,

k = k/k, A = diag[a
1
, , a
n
],
B
n
:= A
1
= A
1
x : x , S
n1
= B
n
. (155)
Note that B
n
is the n-dimensional unit ball centered at the origin, and S
n1
the spherical surface
of the unit ball B
n
, which is (locally) homeomorphic with (n 1)-dimensional space.
Direct integration yields
(x) =

(2)
n
_
S
n1
_

0
k
n1
k
2
_

exp(ik

k (x x
1
))dx
1
dkd

k
=

(2)
n
_
S
n1
_

0
k
n3
_

exp(ik(A

k) (A
1
x A
1
x
1
))dx
1
dkd

k
=

(2)
n
_
S
n1
_

0
1
[A

k[
n3
([A

k[k)
n3
_

exp(ik[A

k[
(A

k)
[A

k[
(A
1
x A
1
x
1
))dx
1
dkd

k
=
_
S
n1
det(A)
[A

k[
n2
g
B
n
(
A

k
[A

k[
, A
1
x)d

k (156)
52
where
g
B
n
(

k, x) =
1
(2)
n
_

0
k
n3
_
B
n
exp(ik

k (x x
1
))dx
1
dk.
In particular, if x = 0, we have that for any R So(n),
g
B
n
(

k, 0) =
1
(2)
n
_

0
k
n3
_
B
n
exp(ik

k x
1
)dx
1
dk
=
1
(2)
n
_

0
k
n3
_
B
n
exp(ik(R

k)
T
Rx
1
)dx
1
dk = g
B
n
(R

k, 0) =:
1
. (157)
Thus, g
B
n
(

k, 0) is in fact independent of

k and
D

= det(A)
1
_
S
n1
1
[A

k[
n2
d

k. (158)
Similarly, we have
(x) = det(A)
_
S
n1

k

k
[A

k[
n
g

B
n
(
A

k
[A

k[
, A
1
x)d

k, (159)
where
g

B
n
(

k, x) =
1
(2)
n
_

0
k
n1
_
B
n
exp(ik

k (x x
1
))dx
1
dk.
32 (i). (4pt) Show in details the above equation (159) (as in (156)).
It is a fundamental property of spherical domain that
g

B
n
(

k, x
1
) = g

B
n
(

k, x
2
) =:
2
x
1
, x
2
B
n
. (160)
That is, the function x g

B
n
(

k, x) is a constant on B
n
. Further, choosing x = 0, by a similar
argument as in (157), we can show the constant
2
is independent of

k as well. Thus, we identify
Q

= det(A)
2
_
S
n1

k

k
[A

k[
n
d

k, (161)
To nd
1
,
2
, we consider the case is the unit ball B
n
and so A = I. Immediately, we have
D
B
n
=
1
_
S
n1
1
[

k[
n2
d

k =
1
= D
0
/[S
n1
[,
Q
B
n
=
1
n
I,
1
n
2
=
_
S
n1

k
2
1
d

k = =
_
S
n1

k
2
n
d

k =
2
= 1/[S
n1
[, (162)
53
where [S
n1
[ denotes the area of the spherical surface S
n1
, and D
B
n
, Q
B
n
are determined by

B
n
(x) =
B
n
,
B
n
(x) = D
B
n

1
2
x Q
B
n
x on B
n
. (163)
32 (ii). (3pt) Assume n = 3. Calculate the value of D
B
n
dened in (163).
Note that for any continuous function f : S
n1
IR,
_
S
n1
f(

k
1
, ,

k
n
)d

k =
_
B
n1
f(

k
1
, ,

k
n1
, (1

k
2
1

k
2
n1
)
1/2
)
(1

k
2
1

k
2
n1
)
1/2
d

k
1
d

k
n1
+
_
B
n1
f(

k
1
, ,

k
n1
, (1

k
2
1

k
2
n1
)
1/2
)
(1

k
2
1

k
2
n1
)
1/2
d

k
1
d

k
n1
. (164)
We now continue our solution to (151). From (152)-(153), in three dimensional space (n = 3)
we have the identity

4
_

1
[(x x

)
2
+ (y y

)
2
+ (z z

)
2
]
1/2
dx

dy

dz

= D

1
2
x Q

x. (165)
Now send a
3
0 while keep a
3
= 1. Then the domain approaches to a at elliptic area
= (x, y) : x
2
1
/a
2
1
+y
2
/a
2
2
= 1 on the xy-plane, and we denote by
D

= lim
a
3
=1/0
D

= a
1
a
2

1
lim
a
3
0
_
S
n1
1
(a
2
1

k
2
1
+a
2
2

k
2
2
+a
2
3

k
2
3
)
(n2)/2
d

k
= 2a
1
a
2

1
_
B
n1
1
(a
2
1

k
2
1
+a
2
2

k
2
2
)
(n2)/2
(1

k
2
1

k
2
2
)
1/2
d

k
1
d

k
2
, (166)
where the last equality follows from (164). Similarly, we have
Q

= lim
a
3
=1/0
Q

= diag[Q
1

, Q
2

, Q
3

], (167)
where
Q
i

= a
1
a
2

2
lim
a
3
0
_
S
n1

k
2
i
(a
2
1

k
2
1
+a
2
2

k
2
2
+a
2
3

k
2
3
)
n/2
d

k
= 2a
1
a
2

2
_
B
n1

k
2
i
(a
2
1

k
2
1
+a
2
2

k
2
2
)
n/2
(1

k
2
1

k
2
2
)
1/2
d

k
1
d

k
2
, (168)
54
where

k
2
3
= 1

k
2
1

k
2
2
. Further, in this limit the l.h.s. of (165) can be written as
lim
a
3
=1/0

4
_

1
[(x x

)
2
+ (y y

)
2
+ (z z

)
2
]
1/2
dx

dy

dz

= lim
a
3
=1/0

4
_

_
a
3
z(x

,y

)
a
3
z(x

,y

)
1
[(x x

)
2
+ (y y

)
2
+ (z z

)
2
]
1/2
dz

dx

dy

=
1
2
_

z(x

, y

)
[(x x

)
2
+ (y y

)
2
+z
2
]
1/2
dx

dy

(169)
where
z(x, y) = (1 x
2
/a
2
1
y
2
/a
2
2
)
1/2
.
32 (iii). (1pt) Explain in words why the last equality in (169) holds.
Therefore, taking into account (165)-(169), choosing z = 0 we arrive at
1
2
_

z(x

, y

)
[(x x

)
2
+ (y y

)
2
]
1/2
dx

dy

= D

1
2
(Q
1

x
2
+Q
2

y
2
) (x, y) . (170)
Comparing (170) with (151), we see that if for some l IR,
h = lD

,
1
= lQ
1

,
2
= lQ
2

, (171)
then equation (151) is satised with

C
= , f(x

, y

) = lz(x

, y

).
We remark that equation (171) determines a
1
, a
2
and the constant l uniquely.
In particular, if a
1
= a
2
= a
0
, by (168) we see that (By (162)
2
= 1/4)
Q
1

= Q
2

=
1
2a
0
_
B
2

k
2
1
(

k
2
1
+

k
2
2
)
3/2
(1

k
2
1

k
2
2
)
1/2
d

k
1
d

k
2
,
and
Q
1

+Q
2

=
1
2a
0
_
B
2
1
(

k
2
1
+

k
2
2
)
1/2
(1

k
2
1

k
2
2
)
1/2
d

k
1
d

k
2
=
1
a
0
_
1
0
dt

1 t
2
=

2a
0
,
where the last integral is found by using Mathematica (which you should learn how to use). Similarly,
by (166), we nd (By (162)
1
= 1/8)
D

=
a
0
4
_
B
n1
1
(

k
2
1
+

k
2
2
)
1/2
(1

k
2
1

k
2
2
)
1/2
d

k
1
d

k
2
=
a
0
4
.
By (170), we obtain the formula that for = (x, y) : x
2
+ y
2
< a
2
0
and z(x, y) = (1 x
2
/a
2
0

y
2
/a
2
0
)
1/2
,
1
2
_

z(x

, y

)
[(x x

)
2
+ (y y

)
2
]
1/2
dx

dy

=
a
0
4


8a
0
(x
2
+y
2
) (x, y) . (172)
55
32 (iv). (6pt) Using the above formula (172) and the procedure outlined in (171), solve (150),
i.e., nd the unknown

C
and p(x

, y

). Further, let F =
_

C
p(x

, y

)dx

dy

. What is the functional

relation between F and h? This relation, i.e., F = F(h), is called the Hertz contact force law.
56
7.5 Inclusion problem Eshelbys solution
Formulation of the problem. Let := x :

n
i=1
x
2
i
/a
2
i
= 1 be an ellipsoidal inclusion.
Consider a homogeneous body with elastic stiness tensor C
0
in IR
n
. We are interested in solving
for u : IR
n
IR
n
,
_
div(C
0
u +P
0

) = 0 on IR
n
,
[u[ 0 as [x[ ,
(173)
where P
0
IR
nn
is called eigenstress. This problem is called the homogeneous Eshelbys inclusion
problem.
Solution. By Fourier transformation, we have
u(x) =
1
(2)
n
_
IR
n

k cofD(

k)
T
P
0

k
det(D(

k))
_

exp(ik (x x
1
))dx
1
dk
= det(A)
_
S
n1

k cofD(

k)
T
P
0

k
det(D(

k))[A

k[
n
g

B
n
(
A

k
[A

k[
, A
1
x)d

k
where
[D(k)]
pq
= [C
0
]
piqj
k
i
k
j
, [D(k)]
1
= cofD(k)
T
/ det(D(k)).
By (160), we nd
u(x) = det(A)
2
_
S
n1

k cofD(

k)
T
P
0

k
det(D(

k))[A

k[
n
d

k x .
That is, inside the ellipsoidal inclusion , the induced strain is uniform and can be written as
u = RP
0
on , (174)
where
[R]
piqj
= det(A)
2
_
S
n1
[cofD(

k)
T
]
pq
(

k)
i
(

k)
j
det(D(

k))[A

k[
n
d

k (175)
is called the Eshelbys tensor. The uniformity of the induced strain is now called the Eshelbys
uniformity property, which is remarkable since we solve a PDE but obtain a simple quadratic
solution.
57
In particular, if C
0
is isotropic with Lam e constants , , by direct calculations we nd the
above Eshelbys tensor can be written as
R =

2

[S
1

+
2 +
S
2
],
where
[S
1
]
piqj
= det(A)
_
S
n1

pq

k
i

k
j
[A

k[
n
, [S
2
]
piqj
= det(A)
_
S
n1

k
p

k
q

k
i

k
j
[A

k[
n
.
Note that the tensors S
1
and S
2
depends only on the geometric properties of ellipsoid.
A critical observation made by Eshelby in his famous 1957 paper is that the solution the
inhomogeneous inclusion problem
_
div(C(x)u +P

) = 0 on IR
n
,
[u[ 0 as [x[ ,
C(x) =
_
C
1
if x ,
C
0
if x

c
(176)
if the eigenstress for the homogeneous problem P
0
is appropriately chosen. To see this, let us
formally rewrite equations (173)and (176) in a less concise form as
_
_
_
div[C
0
u] = 0 in

c
,
div[C
0
u] = 0 in ,
[[C
0
u +P
0

]]n = 0 on ,
(177)
and
_
_
_
div[C
0
u] = 0 in

c
,
div[C
1
u] = 0 in ,
[[C(x)u +P

)]]n = 0 on ,
(178)
respectively, where [[ ]] denotes the jump across the . Clearly, a solution to (177) satises the rst
two of (178) automatically since on , the rst of (177) coincides with the rst of (178), and the
uniformity of (174) guarantees the second of (178). Finally, to verify the last of (178), we rewrite
the last of (177) as
[C
0
u
+
C
0
u

P
0
]n = 0 on , (179)
and the last of (178) as
[C
0
u
+
C
1
u

]n = 0 on , (180)
where u
+
(u

) is the boundary value approached from the outside (inside) of . From (179) and
(174), we have
[C
0
u
+
]n = [C
0
u

+P
0
]n = [C
0
RP
0
+P
0
]n on .
Plugging the above equation into (180), we verify that equation (180) is satised as well if
[(C
1
C
0
)R+II]P
0
= P

, (181)
where II : IR
nn
IR
nn
is the identity mapping, i.e., for any P
0
IR
nn
, IIP
0
= P
0
. From the
denition of R (see (175)), we can show that the the above linear algebraic equation admits a unique
solution P
0
IR
nn
sym
for any given P

sym
. The above solutions are referred to as the Eshelbys
solutions.
33. Important physical quantities can be explicitly computed for the inhomogeneous inclusion
problem (176).
58
(i) (2pt) Calculate the elastic energy in terms of the Eshelbys tensor R, which is given by
c(P

) =
1
2
_
IR
n
u C(x)u.
( Hints: use the divergence theorem
_
IR
n
v =
_
IR
n
v div.
In particular, you may choose = C(x)u +

. )
Now let us consider a physical situation: let V be a nite but large elastic body containing
the ellipsoidal inhomogeneity . Assume the body is subjected to a uniform applied stress
P
a
IR
nn
sym
, i.e., the stress satises
n = P
a
n = t on V,
and the material properties are given by
C(x) =
_
C
1
if x ,
C
0
if x

c
= V

see the following gure.

Figure 1: An elastic body with an ellipsoidal inhomogeneity
(ii) (2pt) What are the governing equation and boundary conditions for the displacement?
59
(iii) (2pt) The displacement u may be written as
u = u
0
+u
1
, u
0
= [C
1
0
P
a
]x.
What is the governing equation and boundary conditions for u
1
?
(iv) (4pt) In particular, we notice that u
1
satises
(C
0
u
1
)n = 0 on V.
Since V is much larger than , to solve the governing equation for u
1
, we may replace V by
IR
n
and require the strain and stress approach to zero at innity. Calculate the strain, i.e.,
u
1
, on in terms of the applied stress P
a
and the Eshelbys tensor. (Hint: identify the
equivalent eigenstress P
0
for the homogeneous problem such that
div[C
0
u
1
+P
0

] = 0.
That is, nd the relation between P
a
and P
0
.)
(v) (3pt) Now let us calculate the total free energy of the system which can be dened as
T(C
1
, R; P
a
) =
1
2
_
V
u C(x)u
_
V
u P
a
n
=
1
2
_
V
_
u [C
0
+ (C
1
C
0
)

]u u P
a
_
=
1
2
_
V
_
u C
0
u +u (C
1
C
0
)

]u u P
a
_
=
1
2
_
V
_
P
a
C
1
0
P
a
+

u (C
1
C
0
)u u P
a
_
= (182)
Complete the rest of the calculations, i.e., obtain the functional relation T = T(C
1
, R; P
a
).
(Hint:
_
V
u P
a
=
_
V
(C
1
0
P
a
+u
1
) P
a
= P
a
C
1
0
P
a
+P
a

_
V
u
1
= P
a
C
1
0
P
a
. Think
for a while about the last equality.)
60
(vi) (20pt, bonus problem) In 2D, assume = (x, y), x
2
/a
2
+ y
2
/b
2
= 1, C
0
(C
1
) is isotropic
with Youngs modulus and Poissons ratio E
0
,
0
= 0.3 (E
0
,
1
= 0.3), and P
a
= diag[0, t].
Rewrite the above energy function as
T = T(E
0
, E
1
, a, b, t)
and dene
T(E
0
, E
1
, a, b, t) = T(E
0
, E
1
, a, b, t) +
1
2
P
a
C
1
0
P
a
=
1
2
_

u (C
1
C
0
)u.
Plot the following curves using Mathematica or Matlab (submit your Matlab or Mathematica
codes if you choose to work on this problem):
Curve1 : y
1
= y
1
(b) =T(E
0
= 1, E
1
= 10, a = 1, b, t = 1),
Curve2 : y
2
= y
2
(b) =T(E
0
= 1, E
1
= 10, a = 1, b, t = 1),
Curve3 : y
3
= y
3
(b) =T(E
0
= 1, E
1
= 0.1, a = 1, b, t = 1),
Curve4 : y
4
= y
4
(b) =T(E
0
= 1, E
1
= 0.1, a = 1, b, t = 1),
Curve5 : y
5
= y
5
(E
1
) =T(E
0
= 1, E
1
, a = 1, b = 0.1, t = 1),
Curve6 : y
6
= y
6
(E
1
) =T(E
0
= 1, E
1
, a = 1, b = 0.1, t = 1),
Curve7 : y
7
= y
7
(E
1
) =T(E
0
= 1, E
1
, a = 1, b = 10, t = 1),
Curve8 : y
8
= y
8
(E
1
) =T(E
0
= 1, E
1
, a = 1, b = 10, t = 1).
61
8 Introduction to Micromechanics
8.1 Cracks
8.2 Composite materials
8.3 Dislocations
8.4 Plasticity
9 A Hierarchy of Plate Theories
9.1 Kirchho-Love plate theory
9.2 von Karman plate theory
9.3 Mindlin plate theory
10 Waves in Solid Structures
10.1 Bulk waves
10.2 Surface waves
10.3 Interfacial waves
References
62