You are on page 1of 77

PLATE AND SHELL THEORY

Sren R. K. Nielsen, Jesper W. Strdahl and Lars Andersen

b)

a)
1

x3

x2
a
a

1
1

x1

Aalborg University
Department of Civil Engineering
November 2007

Contents

1 Preliminaries
1.1 Tensor Calculus . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
1.1.1 Vectors, curvilinear coordinates, covariant and contravariant bases
1.1.2 Tensors, dyads and polyads . . . . . . . . . . . . . . . . . . . . .
1.1.3 Gradient, covariant and contravariant derivatives . . . . . . . . .
1.1.4 Riemann-Christoffel tensor . . . . . . . . . . . . . . . . . . . . .
1.2 Differential Theory of Surfaces . . . . . . . . . . . . . . . . . . . . . . .
1.2.1 Differential geometry of surfaces, first fundamental form . . . . .
1.2.2 Principal curvatures, second fundamental form . . . . . . . . . .
1.2.3 Codazzis equations . . . . . . . . . . . . . . . . . . . . . . . .
1.2.4 Surface area elements . . . . . . . . . . . . . . . . . . . . . . .
1.3 Continuum Mechanics . . . . . . . . . . . . . . . . . . . . . . . . . . .
1.3.1 Three-dimensional continuum mechanics . . . . . . . . . . . . .
1.3.2 Fiber-reinforced composite materials . . . . . . . . . . . . . . .
1.4 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

.
.
.
.
.
.
.
.
.
.
.
.
.
.

7
7
7
12
19
23
28
28
35
44
47
49
49
59
67

2 Shell and Plate Theories


2.1 Plates and Shells . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.2 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

69
71
73

7 Index

74

8 Bibliography

76

.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.

Contents

Preface

This book has been written as a part of ......


Aalborg, November 2007
Sren R.K. Nielsen, Jesper W. Strdahl and Lars Andersen

Contents

C HAPTER 1
Preliminaries
Mathematical and mechanical preliminaries.
Sections 1.1 and 1.2: basic results of tensor calculus and differential theory of surfaces have
been outlined in below. Partly based on (Spain, 1965), (Malvern, 1969)
Section 1.3: Three-dimensional continuum mechanics. Composite materials. Based on (Reddy,
2004).

1.1 Tensor Calculus


1.1.1

Vectors, curvilinear coordinates, covariant and contravariant bases

Fig. 11 Spherical coordinate system.

Fig. 1-1 shows a Cartesian (x1 , x2 , x3 )-coordinate system, as well as a spherical (1 , 2 , 3 )coordinate system. 1 is the zenith angle, 2 is the azimuth angle, and 3 is the radial distance.
7

Chapter 1 Preliminaries

Notice that upper indices are use for the identification of the coordinates, which should not be
confused with power raising. Instead, this will be indicated by parentheses, so if x1 specifies
the first Cartesian coordinate, (x1 )2 indicates the corresponding coordinate raised to the second
power. With the restrictions 1 [0, ], 2 ]0, 2] and 3 0 a one-to-one correspondence
between the coordinates of the two systems exists except for points at the line x1 = x2 = 0.
These represent the singular points of the mapping. For regular points the relations become,
see e.g. (Zill and Cullen, 2005)

x1
sin 1 cos 2
2 3
1
2
x = sin sin
x3
3 cos 1

x3
1 cos p 1 2
(x ) + (x2 )2 + (x3 )2

2
2

x
=

tan1

x1
3

p
(x1 )2 + (x2 )2 + (x3 )2

(11)

We shall refer to l , l = 1, 2, 3, as curvilinear coordinates. Generally, the relation between the


Cartesian and curvilinear coordinates are given by relations of the type
xj = f j (l )

(12)

The Jacobian of the mapping (1-2) is defined as


f j
J = det
k


(13)

Points, where J = 0, represent singular points of the mapping. In any regular point, where
J 6= 0 the inverse mapping exists locally, given as
j = hj (xl )

(14)

For 1 =constant (1-3) defines a surface in space, defined by the parametric description
xj = f j (c1 , 2 , 3 )

(15)

Similar parametric description of surfaces arise, when 2 or 3 are kept constant, and the remaining two coordinates are varied independently. The indicated three surfaces intersect pairwise
along three curves sj , j = 1, 2, 3, at which two of the curvilinear coordinates are constant, e.g.
the intersection curve s1 is determined from the parametric description xj = f j (1 , c2 , c3 ). All
three surfaces intersect at the point P with the Cartesian coordinates xj = f j (c1 , c2 , c3 ). Locally, at this point an additional curvilinear (s1 , s2 , s3 )-coordinate system may be defined with
axes made up of the said intersection curves as shown on Fig. 1-1. We shall refer to these
coordinates as the arc length coordinates.

1.1 Tensor Calculus

The position vector x from the origin of the Cartesian coordinate system to the point P has the
vector representation
x = x1 i1 + x2 i2 + x3 i3 = xj ij

(16)

where ij , j = 1, 2, 3, signify the orthonormal base vectors of the Cartesian coordinate system. In the last statement the summation convention over dummy indices has been used. This
convention will extensively be used in what follows. The rules is that dummy latin indices involves summation over the range j = 1, 2, 3, whereas dummy greek indices merely involves
summation over the range = 1, 2. As an example aj bj = a1 b1 + a2 b2 + a3 b3 , whereas
a b = a1 b1 + a2 b2 . The summation convention is abolished, if the dummy indices are surrounded by parentheses, i.e. a(j) b(j) merely means the product of the jth components aj and bj .
Let dxj denote an infinitesimal increment of the jth coordinate, whereas the other coordinates
are kept constant. From (1-6) follows that this induces a change of the position vector given as
dx = i(j) dx(j) . Hence,

ij =

x
xj

(17)

A corresponding independent infinitisimal increment dj of the jth curvilinear coordinate induce a change of the position vector given as dx = g(j) d(j) , so

gj =

x
j

(18)

gj is tangential to the curve sj at the point P , see Fig. 1-1. In any regular point the vectors gj ,
j = 1, 2, 3, may be used as base vector for the arc length coordinate system at P . The indicated
vector base vectors is referred to as the covariant vector base, and gj are denoted the covariant
base vectors. Especially, if the motion is described in arc length coordinates, gj becomes equal
to the unit tangent vectors tj , i.e.

tj =

x
sj

(19)

Use of the chain rule of partial differentiation provides

gj =

x
x xk
=
= ckj ik
j
xk j

(110)

ij =

x
x k
=
= dkj gk
xj
k xj

(111)

where

10

Chapter 1 Preliminaries

f k
xk

= j =

j
k
hk

dkj = j =
j
x
x
ckj

(112)

ckj specify the Cartesian components of the covariant base vector gj . Similarly, dkj specifies
the components in the covariant base of the Cartesian base vector ij . Obviously, the following
relation prevails

ckl dlj

xk l
=
= jk
l
j
x

(113)

jk denotes Kroneckers delta in the applied notation, defined as

jk

0 ,
1 ,

j 6= k

(114)

j=k

Fig. 12 Covariant and contravariant base vectors.

Generally, the covariant base vectors gj are neither orthogonal nor normalized to the length
1, as is the case for the Cartesian base vectors ij . In order to perform similar operations on
components of vectors and tensorz as for an orthonormal vector base, a so-called contravariant
vector base or dual vector base is introduced. The corresponding contravariant base vectors gj
are defined from
gj gk = jk

(115)

where indicates a scalar product. (1-15) implies that g3 is orthogonal to g1 and g2 . Further,
the angle between g3 and g3 is acute in order that g3 g3 = +1, see Fig. 1-2. Generally, the
contravariant base vectors can be determined from the covariant base vectors by means of the
vector products

1.1 Tensor Calculus

11

g = g2 g3

1
2
g = g3 g1

3
g = g1 g2
V
1

gi =

1 ijk
e gj gk
V

(116)

where V denotes the volume of the parallelepiped spanned by the covariant base vectors gj , see
Fig. 1-2. This is given as
V = g1 (g2 g3 ) = g2 (g3 g1 ) = g3 (g1 g2 )

(117)

eijk is the permutation symbol defined as

ijk

1
= 1

,
,
,

(i, j, k) = (1, 2, 3), (2, 3, 1), (3, 1, 2)


(i, j, k) = (1, 3, 2), (3, 2, 1), (2, 1, 3)
else

(118)

The permutation symbol does not indicate the components of a 3rd order tensor in any coordinate system, and should merely be considered as an indexed sequence of numbers. For
this reason we shall not make distinction between upper and lover indices, so we may write
eijk = eijk . The permutation symbol and the Kronecker delta are related by the following
so-called e relation
j
i j
eijk eklm = li m
m
l

(119)

In the Cartesian coordinate system we have ij = ij , i.e. the dual vector basis is identical to
original.
A vector is envisioned as a geometric quantity in space (an arrow with a given length and
orientation). From this interpretation it follows that a vector is independent of any coordinate
system used for its specification. Actually, infinitely many coordinate systems can be used for
the representation (or decomposition) of one and the same vector. In the Cartesian, the covariant
and the contravariant bases a given vector v can be represented in the following ways
v = vj ij = v j gj = vj gj

(120)

where vj , v j and vj denotes the Cartesian vector components, the covariant vector components,
and the contravariant vector components of the vector v. Generally, Cartesian components of
vectors and tensors will be indicated by a bar. Use of (1-15), and scalar multiplication of the two
last relations in (1-20) with gk and gk , respectively, provides the following relations between
the covariant and contravariant vector components

12

Chapter 1 Preliminaries

vj = gjk v k
v j = g jk vk

(121)

where
gjk = gj gk = gkj

g jk = gj gk = g kj

(122)

The indicated symmetry property of the quantities gjk and g jk follows from the commutativity
of the involved scalar products. From (1-21) follows
vj = gjl v l = gjl g lk vk

gjl g lk = jk

(123)

By the use of (1-20) and (1-21) the following relations between the covariant and the contravariant base vectors may be derived
v j gj = vj gj = gjk v k gj = v j gjk gk
vj gj = v j gj = g jk vk gj = vj g jk gk

gj = gjk gk
gj = g jk gk

(124)

As seen gjk represent the components of gj in the contravariant vector base. Similarly, g jk
signify the components of gj in the covariant vector base. Use of (1-10) and (1-11) provides
the following relation between the Cartesian and the covariant vector components
vj ij = v j gj = v j ckj ik = cjk v k ij
v j gj = vj ij = vj dkj gk = djk vk gj

vj = cjk v k
v j = djk vk

(125)

Finally, using (1-21) the scalar product of two vectors u and v can be evaluated in the following
alternative ways
uv =

uj vj = uj vj = g jk uj vk
uj vj = uj v j = gjk uj v k

(126)

where it is noticed that uj = uj .

1.1.2

Tensors, dyads and polyads

A second order tensor T is defined as a linear mapping of a vector v onto a vector u by means
of a scalar product, i.e.
u = Tv

(127)

1.1 Tensor Calculus

13

Since the vectors u and v are coordinate independent quantities, the 2nd order tensor T as well
must be independent of any selected coordinate system chosen for the specification of the relation (1-27). Equations in solid mechanics are independent of the chosen coordinate system for
which reason these are basically formulated as tensor equations.
A dyad (or outer product or tensor product) of two vectors a and b is denoted as ab. The polyad
abc formed by three vectors a, b and c is denoted a triad, and the polyad abcd formed by
three vectors a, b, c and is denoted a tetrad.
For dyads the following associative rules apply

m (ab) = (ma) b = a (mb) = (ab) m

abc = (ab) c = a (bc)

(ma) (nb) = mn ab

(128)

where m and n are arbitrary constants. Further, the following distributive rules are valid
a (b + c) = ab + ac
(a + b) c = ab + bc

(129)

No commutative rule is valid for dyads formed by two vectors a and b. Hence, in general
ab 6= ba

(130)

If the outer product of two vectors entering a polyad is replaced by a scalar product of the same
vectors a polyad is obtained of an order less by two. This operation is known as contraction.
Contraction of a triad is possible in the following two ways
a bc = (a b) c

ab c = (b c) a

(131)

The scalar product of two dyads, the so-called double contraction, can be defined in two ways
ab cd = (a c)(b d)

ab cd = (a d)(b c)

(132)

As seen the symbol defines scalar multiplication between the two first and the two last
vectors in the two duads, whereas defines scalar multiplication between the first and the
last vector, and the last and the first vectors in the two dyads. Because of the commutativity of
the scalar product of two vectors follows
(a c)(b d) = (b d)(a c) = (c a)(d b) = (d b)(c a)

(a d)(b c) = (b c)(a d) = (c b)(d a) = (d a)(c b)

(133)

14

Chapter 1 Preliminaries

Use of (1-32) and (1-33) provides the following identities for double contraction of two dyads
ab cd = ba dc = cd ab = dc ba

ab cd = ba dc = cd ab = dc ba

(134)

From the second relation of (1-31) follows that the dyad ab/(b c) is mapping the vector
c onto the vector a via a scalar multiplication. From the definition (1-27) follows that this
quantity is a second order tensor. Now, it can be proved that any second order tensor can be
represented as a linear combination of nine dyads, formed as outer product of three arbitrary
linearly independent base vectors. Hence, we have following alternative representations of a
tensor T in the Cartesian, the covariant and the contravariant bases
T = T jk ij ik = T jk gj gk = Tjk gj gk

(135)

The dyads ij ik , gj gk and gj gk form so-called tensor bases. Obviously, (1-35) represents the
generalization to second order tensors of (1-20) for the decomposition of a vector in the corresponding vector bases. T jk , T jk and Tjk denotes the Cartesian components, the covariant
components, and the contravariant components of the second order tensor T. Using (1-10), (111), (1-13) and the associate rules (1-28) the following relations between the dyads and tensor
components related to the considered three tensor bases may be derived
ij ik = dlj dm
k gl gm
l m
gj gk = clj cm
k il im = gjl gkm g g

gj gk = g jl g km gl gm
T jk = cjl ckm T lm

T jk = djl dkm T lm

Tjk = gjl gkm T lm

T jk = g jl g kmTlm

(136)

(137)

As seen, clj cm
k and gjl gkm specify the Cartesian and contravariant tensor components of the dyad
gj gk , whereas g jl g km denotes the covariant tensor components of gj gk . (1-36) and (1-37) represent the equivalent to the relations (1-23) and (1-24) for base vectors and vector components.
In some outlines of tensor calculus the transformation rules in (1-37) between Cartesian and
curvilinear tensor components are used as a defining property of the tensorial character of a
doubled indexed quantity, (Spain, 1965), (Synge and Schild, 1966).
Alternatively, T may be decomposed after a tensor base with dyads of mixed covariant and
contravariant base vectors, i.e.
T = T jk gj gk = Tkj gk gj

(138)

T jk and Tkj represent the mixed covariant and contravariant tensor components. In general,
T jk 6= Tkj as a consequence of gj gk 6= gk gj , i.e. the relative horizontal position of the upper
and lower indices of the tensor components is important, and specify the sequence of covariant

1.1 Tensor Calculus

15

and contravariant base vector in the dyads of the tensor base. Tensor bases with dyads of mixed
Cartesian and curvilinear base vectors may also be introduced. However, in what follows only
the mixed curvilinear dyads in (1-38) will be considered. The following identities may be
derived from (1-35) and (1-38) by the use of (1-24)

jk
T = g lk T jl
T = T jk gj gk = T jl gj gl = T jl g lk gj gk

T jk = g jl T k
T = T jk gj gk = Tl k gl gk = Tl k g lj gj gk
l
(139)

T
=
g
T
T = Tjk gj gk = T lk gl gk = T lk glj gj gk

jk
jl
k

j k
l j
l
j k
Tjk = glk Tj l
T = Tjk g g = Tj g gl = Tj glk g g
Using (1-23) and (1-39) the following representations of the mixed components in terms of
covariant and contravariant tensor components may be derived

T kj = gjl T kl

k
lk
Tj = gjl T

(140)

T kj = g kl Tlj

k
kl
Tj = g Tjl

It follows from (1-40) that, if T jk = T kj then T kj = Tkj and Tjk = Tkj . A second order tensor
for which the covariant components fulfill the indicated symmetry property is denoted a symmetric tensor.
Let T jk denote the covariant components of a second order tensor T. The related so-called
transposed tensor TT is defined as the tensor with the covariant components T kj , i.e.
TT = T kj gj gk

(141)

For a symmetric tensor T jk = T kj for which reason TT = T.


The identity tensor is defined as the tensor with the covariant and contravariant components g jk
and gjk , i.e.
g = g jk gj gk = gjk gj gk = jk gj gk

(142)

The mixed components follows from (1-23) and (1-40)


g kj = gjl g kl = jk
gjk = gjl g lk = jk

(143)

Hence, the mixed components are equal to the Kroneckers delta, which explains the last statement for the mixed representation in (1-42). The name identity tensor stems from the fact that
g maps any vector v onto itself. Actually,

16

Chapter 1 Preliminaries

g v = g jk gj gk vl gl = g jk vl gj kl = g jk vk gj = v j gj = v

(144)

The length of a vector v is determined from


|v|2 = v v = v g v = g jk vj vk = gjk v j v k

(145)

Because of its relationship to the length of a vector the identity tensor is also designated the
metric tensor or the fundamental tensor.
The increment dx of the the position vector x as given by (1-6), due to independent increment
dxj and dj of the Cartesian or curvilinear coordinates becomes
dx = dxj ij = dj gj

(146)

Then, the length ds of the incremental position vector becomes, cf. (1-45)
ds2 = dxj dxj = gjk dj dk

(147)

The inverse tensor T1 of the tensor T is defined from the identity


T1 T = T T1 = g

(148)

Let Tjl1 denote the contravariant components of T1 , and T mk the covariant components of T.
Then,
l
gj gk = Tjl1 T lk gj gk
g = jk gj gk = T1 T = Tjl1 gj gl T mk gm gk = Tjl1 T mk m

Tjl1 T lk = jk

(149)

In matrix notation this means that the three dimensional matrix, which stores the components
Tjl1 , is the inverse of the matrix, which stores the components T lk . Similarly, the covariant
components (T 1 )jl of T1 and the contravariant components Tlk of T are stored in inverse
matrices. In contrast, the matrices which store the contravariant components Tjl1 and Tlk will
not be mutual inverse. From (1-15) follows that

k
gj T gk = gj T lm gl gm gk = T lm lj m
) = T jk

T jk = gj T gk
Similarly,

(150)

1.1 Tensor Calculus

17

Tjk = gj T gk

k
k
T = gj T g
j

T jk = gj T gk
Tjk = ij T ik

(151)

Next, the second Piola-Kirchhoff stress tensor S and its virtual work conjugated strain measure
the Green strain tensor E are considered. In the Cartesian, covariant, contravariant and mixed
tensor bases these are represented as follows
S = Sjk ij ik = S jk gj gk = Sjk gj gk = Sjk gj gk = S jk gj gk
E = E jk ij ik = E jk gj gk = Ejk gj gk = Ejk gj gk = E jk gj gk

(152)

Because the Piola-Kirchhoff stress tensor and the Green strain tensor are both symmetric tensors
their mixed tensor components fulfill Sjk = S jk and Ejk = E jk . For this reason we shall
simply use the notaions Sjk and Ejk for these quantities in what follows. The indicated Cartesian,
covariant, contravariant and mixed tensor components for the stress tensor are related as follows,
cf. (1-37), (1-39)
S jk = g jlg km Slm

S jk = g jlSlk

Sjk = gjl gkm S lm

Sjk = gjl Skl

Sjk = g kl Slj

Sjk = gjl S lk

S jk = djl dkm Slm

Sjk = cjl ckm S

(153)

lm

Completely analog relations prevail for the components of the Green tensor. For a physical
linear material a linear mapping of the Green tensor onto the Piola-Kirchhoff stress tensor via
a double contraction with the elasticity tensor C, i.e.
S = C E

(154)

C is a fourth order tensor, which can be specified in any tensor base with tetrads made up of
either the base vectors of the Cartesian, the covariant or the contravariant vector bases as follows

C = C jklm gj gk gl gm = Cj klm gj gk gl gm = C jk lm gj gk gl gm = C jkl m gj gk gl gm

jkl
m
lm j k
k m j
l
kl
j
m

= C m gj gk gl g = Cjk g g gl gm = Cj l g gk g gm = Cj m g gk gl g

j m
j l
jk
j
k l
k
m
l m
k l m
(155)
= C kl gj g g gm = C k m gj g gl g = C lm gj gk g g = C klm gj g g g

= Cj klm gj gk gl gm = Cjk lm gj gk gl gm = Cjkl m gj gk gl gm = Cjklm gj gk gl gm

jklm

= C
ij ik il im

18

Chapter 1 Preliminaries

Formally, the relations between the various mixed tensor components can be derived by raising
and lowering indices by means of the covariant components g jk and the contravariant components gjk of the identity tensor g, cf. (1-53). As an example
C jklm = glr gms C jkrs

(156)

Normally, the components of the elasticity tensor is specified in the Cartesian coordinate system. The relation between the Cartesian components C rstu and the covariant components C jklm
becomes, cf. (1-53)
rstu
C jklm = djr dks dlt dm
u C

(157)

The following relation between the covariant components of S and E may be derived, cf. (1-15),
(1-32)
S = S jk gj gk = C E = C jklm gj gk gl gm E rs gr gs =
C jklm E rs gj gk (gl gm gr gs ) = C jklm E rs gj gk (gl gr )(gm gs ) =
C jklm E rs rl sm gj gk = C jklm E lm gj gk

S jk = C jklm E lm

(158)

If the alternative definition of the double contraction in (1-32) is used, the following result is
obtained for the tensor representation of S
S = C E = C jklm E ml gj gk

(159)

Since, E lm = E ml , the indicated definitions provide the same result for the stress components.
Expressed in the mixed components of E and S the constitutive equations on components form
may be given in the following ways
Sjk = Cj k ml Elm = Cj klm Elm = C jk l m Elm = C jkml Elm

(160)

Clearly, the symmetry of the stress and strain tensor implies the following symmetry properties of the mixed elasticity tensor components C jklm = C kjlm = C jkml = C kjml and Cj kml =
Cj klm = C jk l m = C jkml .
Finally, the double contraction of S and E may be evaluated in the following alternative ways
S E = S E = S jk Ejk = Sjk E jk = Sjk Ekj

(161)

1.1 Tensor Calculus

1.1.3

19

Gradient, covariant and contravariant derivatives

Consider a scalar function a = a(xl ) = a(l ) of the Cartesian or curvilinear coordinates.


The gradient of a, denoted as a, is a vector with the following Cartesian and contravariant
representations
a =

a j
a
i
=
g
k
xj
j

(162)

Let k denote the increments of the curvilinear coordinated, and define the incremental vector
= k gk . The corresponding increment a of the scalar is determined by
a = a =

a j
a
g k gk = j j
j

(163)

The gradient of a vector function v = v(xl ) = v(l ), denoted as v, is a second order tensor,
which in complete analogy to (1-63) associates to any incremental vector = k gk a
v
j
corresponding increment v = v =
j of the vector v. The gradient tensor has
the following representations

v
(
v j ij )

vj

i
=
i
=
ij ik
k
k

xk
xk
xk

 j

v
(v j gj ) k
v
k
g =
g =
g + vj
v =
k
k
k j

vj j
v k (vj gj ) k

g =
g =
g + vj
k
k

gj
k
gj
k

gk

gk

(164)

At the derivation of the Cartesian representation it has been used that the base vector ij is
constant as a function of xl . In contrast, the curvilinear base vectors depend on the curvilinear
g
coordinates, which accounts for the second term within the parentheses in (1-64). Clearly, kj
j
and g
are vectors, which may hence be decomposed in the covariant and contravariant vector
k
bases as follows
gj
=
k

l
j k

gj
j
=
k

l k


j k

gl

gl

(165)

signify the covariant components of kj , and

j
l k

is the contravariant components of g


.
k

is denoted the Christoffel symbol. This is related to the co- and contravariant components
of the identity tensor as follows




1 lm gjm gkm gjk
l
= g
+
m
(166)
2
k
j

j k
j k

20

Chapter 1 Preliminaries

(1-65) and (1-66) have been proved in Box 1.1. From (1-66) follows that the Christoffel symbols
fulfill the symmetry condition

 

l
l
=
(167)
j k

k j

From (1-8) follows that


gj
2x
2x
gk
=
=
= j
k
j
k
k
j

(168)

Alternatively, the symmetry property (1-67) follows from (1-65) and (1-66).

Box 1.1: Proof of (1-65) and (1-66)


We consider the first relation in (1-65) as a definition of the Christoffel symbol, and prove
that this implies the second. From (1-15) and the first relation in (1-65) follows
 gm j
j

gj
j

=
g

g
=

g
+
g

=0
m
m
k m k
k
k





 

gj
l
l
j
j
j
j
l
gm =
gl g =
l =
=
m
=
k

m k
m k
m k
l k
 
j

gl gm = 0

l k

  
gj
j
+
gl gm = 0
k

l k

(169)

Since, (1-69) is valid for any of the three covariant base vectors the term within the
bracket must be equal to 0. This proves the validity of the second relation (1-65).
From (1-22) and the first relation (1-65) follows
gjm
(gj gm )
gj
gm
=
= k gm +
gj =
k
k

n
j k

gnm +

n
m k

gnj

(170)

1.1 Tensor Calculus

21

From (1-70) follows


gjm gkm gjk
+
m =
k
j













n
n
n
n
n
n
gnm +
gnj +
gnm +
gnk
gnk
gnj =
j k

n
j k

m k

k j

m j

j m

gnm

k m

(171)

where the symmetry property (1-67) has been used. Next, (1-66) follows from (1-71)
upon multiplication on both sides with g lm and use of (1-23).

Use of (1-65) in (1-66) provides the following representations of v


v = v j;k gj gk = vj;k gj gk

(172)

where

v j;k
vj;k

 
v j
j
vl
= k+

k l


vj
l
= k
vl

j k

(173)

Hence, v j;k specifies the mixed co-and contravariant components of v, and vj;k specifies the
contravariant components. For the partial derivative of the vector function v(l ) the following
representations may be derived by the use of (1-64)

(v j gj )
v j

=
gj +

v
k
k
=
j
k

(vj g ) = vj gj +
k
k

gj j
v = v j;k gj
k
gj
vj = vj;k gj
k

(174)

Hence, v j;k and vj;k may alternatively be interpreted as the covariant components and the conj
v
travariant components of the vector
k . For this reason v ;k is referred to as the covariant
derivative, and vj;k as the contravariate derivative of the components v j and vj , respective. In
the applied notation these derivatives will always be indicated by a semicolon.
Further, by the use of (1-65) the following results for the derivatives of the dyads entering the
covariant, mixed and contravariant tensor bases become

22

Chapter 1 Preliminaries

gj
gk

(gj gk =
gk + gj l =
l
l

gj k
gk

k
(g
g
)
=
g
+
g
j
j
l
l
l
gj
gk
j
(g gk ) =
gk + gj l
l
l

k
j k
gj k
j g
(g
g
)
=
g
+
g
l
l
l

gm gk +
gj gm

 



m
k
k
m

=
gm g
gj g

j l
m l


 

j
m
m

g gk +
gj gm
=

m l
k l






j
k
m k
j m

g g
gg
=
m
j l

m l

m
k l

(175)

m l

Finally, the covariant and covariant representations of the increment v of the vector v due to
the incremental vector = K gk of the curvilinear coordinates becomes
v = v = v j;k gj gk l gl = vj;k gj gk l gl
v = v j;k k gj = vj;k k gj

(176)

The gradient of a second order tensor function T = T(l ), denoted as T, is a third order
tensor with the following representations

 jk

(T jk gj gk ) l
T
gk

jk gj
jk

g =
gj gk + T
g + T gj l gl

l
l
l k

j
j
k
k

(T
g
g
)
T
g
g

j
j
j
j
k
k

gl =
gj gk + T k l gk + T k gj l gl

l
l

T
T = l gl =
!

k j
k

(T
g
g
)
T
g
g
k

k
j
j

gl =
gj gk + Tj k l gk + Tj k gj l gl

l
l




j k
j
k

(T
g
g
)
T
g
g

jk
jk
l
j
k
k
j

g =
g g + Tjk l g + Tjk g
gl

l
l
l





 jk
T
j
k

mk
jm

+T
gj gk gl
+T

m l
m l




!

T k
j
m

+ T mk
T jm
gj gk gl

m l
k l
=
 

!

T
m
k

Tmk
+ Tj m
gj gk gl

j
l
m
l


 
 

Tjk

m
m

Tkm
Tjm
gj gk gl
l
j l
k l

= T jk;l gj gk gl
= T jk;l gj gk gl
= Tj k ;l gj gk gl
= Tjk;l gj gk gl
(177)

1.1 Tensor Calculus

23

where
T jk;l
T jk;l
Tj

;l

Tjk;l





T jk
j
k

mk
jm

=
+T
+T

l
m l
m l






T k
j
m

m
j

T
+
T

m
k
l

m l
k l
 


Tj k

m
k
k
m

=
+ Tj
Tm

j l
m l

 
 

Tjk
m
m

T
=

km
jm
l

j l
k l

(178)

where (1-65) has been used in the last statement. Then, the partial derivative of the tensor
function T(m ) may be written in any of the following alternative second order tensor representations, cf. (1-74)
T
= T jk;l gj gk = T jk;l gj gk = Tjk;l gj gk
l

(179)

Partial differentiation of both sides of (1-44) provides


v
g
v
g
v
= k v+g k = k v+ k
k

g
v=0
k

(180)

Since v is arbitrary (1-80) implies that


g
=0
k

(181)

In turn this means that g = 0. Hence, the identity tensor is constant under covariant differentiation.

1.1.4

Riemann-Christoffel tensor

The gradient of a vector v with decomposition into Cartesian and curvilinear tensor bases have
been indicated by (1-64). The gradient of this second order tensor is given as, cf. (1-73), (1-74)






(
v j ij )
2 vj

i
i
=
i
i
=
ij ik il

k
l
k
l

xl xk
xl
xk
xk xl



 j 
 


v k

v
j
l
v m gj gk gl = v j;kl gj gk gl
+
(v) = l k g g = l
k

k
m





 

v
m

j
k
l

vm gj gk gl = vj;kl gj gk gl

l k g g = l
k
j k
(182)

24

Chapter 1 Preliminaries

where the following tensor components have been introduced




 m 
 m   j
 
2 vj
v
v
v
j
j
m
j
n
v ;kl = (v ;k );l = k l +
+

vm +
l
k
m

k m
l m
k l
n m
k l


 


j
j
n
m
v +
vm
(183)
l k m
l n
k m
j



 
 
 

2 vj
vm
vj
m
m vm
m
n
m
vj;kl = (vj;k );l = k l
vm

+

j k l
j l k
k l m
k l
n j


 


m
n
m
vm +
vm
(184)
l j k
j l
n k
From (1-83) and (1-84) follow that the indices k and l can be interchanged in the first five terms
on the right hand sides without changing the value of this part of the expressions, whereas
this is not the case for the last two terms. This implies that the sequence in which the covariant
differentiations is performed is significant, i.e. in general vj;kl 6= vj;lk . In order to investigate this
further the so-called Riemann-Christoffel tensor R is introduced, which is a fourth order tensor
with the mixed representation R = Rmjkl gm gj gk gl , where the mixed curvilinear components
are defined as
 
 

 
 


n
m
n
m
m
m
m
R jkl =

+ k
l
(185)
j l
j k
j l
n k
j k
n l
Obviously,
Rmjkl = Rmjlk

(186)

Further, the following relations are valid


)
Rmjkl + Rmklj + Rmljk = 0

(187)

Rjjkl = 0

The relations (1-87) follow by insertion of (1-85) and use of the symmetry property (1-67) of
the Christoffel symbol. From (1-84) and (1-85) follows that
Rmjkl vm = vj;kl vj;lk

(188)

The Cartesian components of R follow from (1-82)


2 j
2 j
jklm vm = v v = 0
R
xk xl xl xk
jklm = 0
R

(189)

Hence, it can be concluded that R = 0 in an Euclidean three-dimensional space, i.e. a space


spanned by a Cartesian vector basis. In turn this means that also the curvilinear components

1.1 Tensor Calculus

25

(1-85) must vanish in this space. A space, where everywhere R = 0 is called flat. Reversely,
a non-vanishing Riemann-Cristoffel tensor indicates a curved space. In a flat space Rmjkl = 0,
with the implication that vj;kl = vj;lk . The three-dimensional Eucledian space is flat, and any
plane in this space forms a flat two-dimensional subspace. In contrast, a curved surface in the
Eucledian space is not a flat subspace. An example of a curved four-dimensional space is the
time-space continuum used at the formulation of the general theory of relativity, where the indices correspondingly range over j = 1, 2, 3, 4.
1
Because of the identities (1-86), (1-87) it can be shown that only 12
N 2 (N 2 1) of the tensor components Rmjkl are independent and non-trivial, where N denotes the dimension of the
space, (Spain, 1965). Hence, for a two-dimensional space only one independent component
exists, which can be chosen as R1 212 . In the three-dimensional case six independent and nontrivial components exist, which are chosen as R1 212 , R1 213 , R1 223 , R1 313 , R1 323 and R2 323 .

Example 1.1: Covariant base vectors, identity tensor and Riemann-Christoffel tensor in
spherical coordinates
By the use of (1-10) the first equation (1-65) becomes
(
)
(cm
cm
l
j im )
j
=
im =
cm

l im
k
k
j k
)
(
cm
l
j
cm
=
l
k
j k

(190)

m
The Cartesian components cm
j of the covariant base vector gj is stored in the column matrix g j = [cj ]. Then,
(1-88) may be written in the following matrix form
(
)
(
)
(
)
(
)
g j
l
1
2
3
=
gl =
g1 +
g2 +
g3
(191)
k
j k
j k
j k
j k

The spherical coordinate system defined by (1-1) is considered. In this case the column matrices attain the form,
cf. (1-12)
3

cos 1 cos 2
sin 1 cos 2
sin 1 sin 2

(192)
g 1 = 3 cos 1 sin 2 , g 2 = 3 sin 1 cos 2 , g 3 = sin 1 sin 2
3
1
1
0
sin
cos

The covariant components of the identity tensor is given by (1-22) as gjk = gj gk = g Tj g k , where the last
statement is obtained by evaluating the scalar product in Cartesian coordinates. The covariant and contravariant
components of the identity tensor are conveniently stored in matrices. Using (1-92) these becomes

(3 )2

[ gjk ] = 0

0
2

3 2

( ) sin
0

1
(3 )2
 jk 

g
= 0

0
1
3
2
( ) sin2 1
0

(193)

The result for the contravariant components follows from (1-23). Next, (1-91) will be used to determine the
Christoffel symbols by direct specification of the decomposition on the right hand side. Using (1-92) the following
results are obtained

26

Chapter 1 Preliminaries

g 1
1

g 1
2

g 1
3

g 2
1

g 2
2

g 2
3

g 3
1

g 3
2

g 3
3

3 sin 1 cos 2

= 3 sin 1 sin 2 = 0 g 1 + 0 g 2 3 g 3
3

cos

3 cos 1 sin 2

cos 1 cos 2

= 3 cos 1 cos 2 = 0 g 1 +
g2 +0 g3
tan 1
0

cos 1 sin 2
sin 1

1
g1 +0 g2 +0g3
3

3 cos 1 sin 2

= 3 cos 1 cos 2 = 0 g 1 +
0

3 sin 1 cos 2

1
g2 +0 g3
tan 1

= 3 sin 1 sin 2 = sin(21 ) g 1 + 0 g 2 3 sin2 1 g 3


2
0

sin 1 sin 2

= sin 1 cos 2
0

sin

sin 1 sin 2

1
g2 +0g3
3

= 0 g1 +

sin 1 cos 2
0

0
= 0

1
g1 +0 g2 +0g3
3

cos 1 sin 2

1
g2 +0g3
3

= 0 g1 +

cos 1 cos 2

= 0 g1 + 0 g2 +0g3


1

= 0

1 1
2
= 0
1 1

 3

= 3
1 1

1

= 0

1 2
1
2
=
1 2

tan 1

 3 = 0
1 2

1
1

= 3

1 3

2
= 0
1 3

3 = 0
1 3

1

2 1 = 0

 2
1
=
2 1

tan
1
 3

= 0
2 1

1

2 1 2 = sin(21 )


2
2
=
0
2 2

3 = 3 sin2 1
2 2

1

= 0

2 3
1
2
= 3
2 3

 3 = 0
2 3

1
1

= 3

3
1


2
=
0

3 3 1

= 0
3 1

1

3 2 = 0


1
2
= 3
3 2

3 = 0
3 2

1

= 0

3 3
2
= 0
3 3

3 = 0
3 3
(194)

1.1 Tensor Calculus

27

The non-trivial components of the Riemann-Christoffel tensor follow from (1-85) and (1-94)

R1 212 =

2 2
n 1
2 1
n 2
2 2
2 1

1
1
1
1

2 1
1
3
1
1

sin(2 ) 0 + 0 0 sin 3 + 0 0 +

sin(2
)

cos(2
)

0
=
0

tan 2

(
)(
) (
)(
)
(
)
(
)

n
1
n
1
1
1

R 213 =

+ 1
3
=

2 3
n 1
2 1
n 3
2 3
2 1

1
1
1
1

00+ 3 0+0 3 0 3

0
+
0

0
=
0

tan

)(
) (
)(
)
(
)
(
)
(

n
1
n
1
1
1

+ 2
3
=
R 223 =

2 3
n 2
2 2
n 3
2 3
2 2

1 1
1
1

2 1
1
1
3

0 0 3 sin(2 ) + 0 0 + sin(2 ) 3 0 0 + sin 0 + 0 0 = 0

2
2

)(
) (
)(
)
)
)
(
(
(

n
1
n
1
1
1

R1 313 =

+ 1
3
=

3 3
n 1
3 1
n 3
3 3
3 1

1
1 1
1

00+00+0 3 3 3 0000+ 3 2 = 0


( )

(
)(
) (
)(
)
(
)
(
)

n
1
n
1
1
1
1

R 323 =

+ 2
3
=

3 3
n 2
3 2
n 3
3 3
3 2

1
1
1

0 0 0 sin(2 ) + 0 0 0 3 0 0 0 0 + 0 0 = 0

(
)(
) (
)(
)
(
)
(
)

n
2
n
2
2
2

R 323 =

+ 2
3
=

3 3
n 2
3 2
n 3
3 3
3 2

1
1
1 1
1

0
+
0

0
+
0

0
+
0
+
=
0

1
3
3
3
3
2
tan


( )
(

)(

)(

+ 1

(195)

As expected all components of the Riemann-Christoffel tensor vanish as a consequence of the flatness of the threedimensional Euclidean space.

28

Chapter 1 Preliminaries

1.2 Differential Theory of Surfaces


1.2.1

Differential geometry of surfaces, first fundamental form

Fig. 13 a) Parameter space. b) Surface space.

Let the spherical 3 coordinate be fixed at the value 3 = r. Then, the mapping (1-1) of the
spherical coordinates onto the Cartesian coordinates takes the form
1

x1
r
r sin 1 cos 2
2 2
1
2
x = r = r sin sin
x3
r3
r cos 1

(196)

(r 1 , r 2 , r 3 ) denotes the Cartesian components of the position vector r to a given point of the
surface. Obviously, (r 1 )2 + (r 2 )2 + (r 3 )2 = r 2 . Hence, with the zenith angle varied in the
interval 1 [0, ], and the azimuth angle varied in the interval 2 ]0, 2], (1-96) represents
the parametric description of a sphere with the radius r and the center at (r 1 , r 2 , r 3 ) = (0, 0, 0).
In what follows it is assumed that the parametric description of all considered surfaces is defined
by a constant value of the curvilinear coordinate 3 = c3 in the mapping (1-2). Then, a given
surface is given as, cf. (1-5)
r j = f j (1 , 2 , c3 ) = f j ( )

(197)

In the last statement of (1-97) the explicit dependence on the constant c3 is ignored, as will
also be the case in the following. Let the mapping (1-97) be defined within a domain in the
parameter space. For each point p determined by the parameters (1 , 2 ), a given point P
is defined on with Cartesian coordinates given by (1-97). Assume that a curve through p is
specified by the parametric description 1 (t), 2 (t) , where t is the free parameter. Then, this
curve is mapped onto a curve s = s(t) through P on as shown on Fig. 1-3. Especially, if the
curvilinear coordinate 1 is fixed, whereas 2 is varied independently a curve s1 through P is
defined on . Similarly, a curve s2 on the surface through P is obtained, if 1 is fixed and 2 is
varied. The positive direction of s1 and s2 are defined, so positive increments of correspond

1.2 Differential Theory of Surfaces

29

to positive increments of s . Then, these curves define a local two-dimensional arch length coordinate system (s1 , s2 ) through P . Assume that is a rectangular domain [a1 , a2 ][b1 , b2 ] with
sides parallel to the axes as shown on Fig. 1-3a. As an example this is the case for the mapping (1-96), where = [0, ]]0, 2]. In such cases the surface will be bounded by the arc
length coordinate curves given by the parameter descriptions r j = f j (a1 , 2 ), r j = f j (a1 , 2 ),
r j = f j (1 , b1 ) and r j = f j (1 , b2 ), see Fig. 1-3b.
Since d3 = 0, when merely 1 and 2 are varied in , the increment dr of the position vector r
to the point P on the surface is given as by the following Cartesian and curvilinear representations, cf. (1-6)
dr = dr 1 i1 + dr 2 i2 + dr 3 i3 = d1 a1 + d2 a2
dr = dr j ij = d a

(198)

where the covariant base vectors becomes, cf. (1-8)

a =

(199)

Fig. 14 Covariant and contravariant base vectors in the tangent plane.

a are tangential to the arch length curves at P , see Fig. 1-3b. a may then be interpreted as a
local two-dimensional covariant vector bases, which spans the tangent plane A at the point P ,
see Fig. 1-4. The indicated covariant vectors may be represented in the Cartesian vector basis
as, cf. (1-10)
a = cj ij
where the Cartesian components cj are given by (1-12).

(1100)

30

Chapter 1 Preliminaries

At the point P a unit normal vector a3 to the surface and the tangent plane is defined as
a3 =

a1 a2
1
= a1 a2
|a1 a2 |
A

(1101)

A denotes the area of the parallelogram spanned by a1 and a2 , and given as


A = |a1 a2 | = |a1 ||a2 | sin

(1102)

where denotes the angle between a1 and a2 , see Fig. 1-4. Using (1-16) with V = 1 A = A,
a contravariant basis at P may be defined by the relations

1
1 
1
2

a = a2 a3 =
a2 (a1 a2 ) =
|a2 | a1 (a1 a2 ) a2

A
(A)2
(A)2



1
1
1
2
2
(a1 a2 ) a1 =
|a1 | a2 (a1 a2 ) a1
a = a3 a1 =

A
(A)2
(A)2

a = a1 a2 = a3
A
1

(1103)

The last statements of the results for a1 and a2 follow from the vector identities a (b c) =
(a c) b (b a) c and (a b) c = (a c) b (b c) a. The basic relation between the
covariant and contravariant base vectors in the tangent plane becomes, cf. (1-15)
a a =

(1104)

(1-104) may alternatively be proved from the last statements in (1-103), using (1-102) and
a1 a2 = |a1 ||a2 | cos .
The unit normal vector a3 = a3 is orthogonal to both the covariant and the contravariant base
vectors in the tangent plane, so
a3 a = a3 a = 0

(1105)

Then, the covariant and contravariant components gjk and g jk of the three-dimensional identity
tensor can be stored in the following matrices

a11 a12 0

[ gjk ] = a21 a22 0


0
0 1

a11 a12 0
 jk 

g
= a21 a22 0
0
0 1

(1106)

A surface vector function v = v(1 , 2 ) is a vector field, which everywhere (i.e. for any
parameters (1 , 2 ) ) is tangential to the surface. Then, v may be represented by the following
Cartesian, covariant and contravariant representations

1.2 Differential Theory of Surfaces

v = vj ij = v a = v , a

31

(1107)

where vj , v and v denote the Cartesian, the covariant and the contravariant components of v.
v 3 = v3 = 0 for a surface vector, for which reason only two components enter the curvilinear
representations. The Cartesian and the covariant components are related as, cf. (1-25)
vj = cj v
v = dj vj

(1108)

Similarly, the covariant and contravariant components are related by the following relations
analog to (1-22)
v = a v
v = a v

(1109)

where
a = a a
a = a a

(1110)

As a consequence of the structure (1-106) of the covariant and contravariant components of the
identity tensor it follows that, cf. (1-23)
a a =

(1111)

where denotes the Kroneckers delta in two dimensions. Both a and a are surface vectors.
Then, these can be expanded in two-dimensional contravariant and covariant vector bases as
follows, cf. (1-24)
a = a a
a = a a

(1112)

In analogy to (1-27), a surface second order tensor is defined as a second order tensor, which
everywhere maps a surface vector v onto another surface vector u by means of a scalar product.
A second order surface tensor T admits the following Cartesian, covariant, contravariant and
mixed representations, cf. (1-35)
T = T jk ij ik = T a a = T a a = T a a = T a a

(1113)

32

Chapter 1 Preliminaries

T , T , T and T denotes the covariant, the contravariant and the mixed covafriant and
contravariant components of the surface tensor. These are related as, cf. (1-39), (1-40)
T = a a T

T = a a T

T = a T

T = a T

T = a T

T = a T

(1114)

The surface identity tensor a is defined as a second order tensor with the covariant components
a , the contravariant components a , and the mixed components , corresponding to the
following representations, cf. (1-42)
a = a a a = g a a = a a

(1115)

a will map any surface vector onto itself, i.e.


av=v

(1116)

(1-116) is proved in the same way as (1-44), using the representation (1-115). Let dr be a differential increment of the position vector r due to an increment d of the parameters. Obviously,
dr is a surface vector with the representation (1-98). The length ds of the incremental vector
becomes
ds2 = dr dr = dr a dr = d a a d = a d d

(1117)

Obviously, the tensor a determines the length of any surface vector, for which reason the alternative naming surface metric tensor is used. The quadratic form in the last statement of (1-117)
is called the first fundamental form of the surface.
The partial derivatives of the covariant and contravariant base vectors are unchanged given by
(1-65)
aj
=

l
j

aj
j
=

l


al
al

j
j

a +

a3

(1118)

a3

a3 = a3 is orthogonal to both a and a , so a3 a = 0 and a3 a = 0. Then, scalar


multiplication of the first equation in (1-118) with a3 and scalar multiplication of the second
equation with a3 provides




aj

a3 =

3
aj
a =

3
j

(1119)

1.2 Differential Theory of Surfaces

33

where it has been used that a3 a3 = 1. Because a3 = a3 , the left hand sides of (1-119) are
identical for j = 3. However, the right hand sides have opposite signs in this case. Both can then
only be truth if the right hand sides vanish, leading to the following result for the Christoffel
symbol


=0

(1-120) implies that


surface vector.

(1120)
a3

is orthogonal to the surface normal a3 = a3 , and hence must be a

Obviously, the increment vector = a with the covariant components is a surface


vector. The gradient v of a surface vector function v = v( ) is a surface second order tensor,
which to the increment vector associates a surface increment vector v = v =
v
of the vector v. The gradient tensor has the following curvilinear representations,

which merely follow by replacing the latin indices with greek indices in (1-72), (1-73)
v = v ; a a = v; a a

(1121)

where
v ;
v;



v

= +
v




v

=
v

(1122)

v ; and v; will be referred to as the surface contravariant derivative and surface covariant
derivative of the components v and v .
Correspondingly, the gradient T of a surface second order tensor function T = T( ) is a
surface third order tensor with the following curvilinear representations, cf. (1-77), (1-78)
T = T ; a a a = T ; a a a = T ; a a a = T; a a a

(1123)

where
T

T ;
T

T;





T

+T
+T
=






=
+
T







T

T
+
T






(1124)

34

Chapter 1 Preliminaries

The gradient of the gradient of a surface vector is a third order surface tensor with the following
representations, cf. (1-82), (1-83), (1-84)
(v) = v ; a a a = v; a a a

(1125)

where

 
 
 


2v
v
v
v

v +
=
= +
+













v +
v
(1126)


v ;

(v ; );










2 v
v
v
v

v
v; = (v; ); =

+












v +
v
(1127)



The surface Riemann-Christoffel tensor is denoted B = B(1 , 2 ) to distinguish it from the
equivalent tensor R in the three dimensional space. The mixed covariant and contravariant
representation becomes, cf. (1-85)
B = B a a a a



(1128)




(1129)

In analogy to (1-86) - (1-88) the tensor components B fulfill the following identities
B = B

B + B + B = 0
B = 0
B v = v; v;

(1130)

(1131)

Due to the identities (1-130) only one independent and non-trivial component exists, which is
taken as B 1212 .
From (1-85) and (1-129) follow that the components in the tangent plane R of the RiemannChristoffel tensor in the three-dimensional space is related to the corresponding components
B of the surface Riemann-Christoffel tensor as follows

1.2 Differential Theory of Surfaces

=B



35



(1132)

Since, the three-dimensional Eucledian space is flat, we have everywhere R = 0. Then, the
following simplified representation for tensor components B is obtained from (1-132)

1.2.2





(1133)

Principal curvatures, second fundamental form

Fig. 15 a) Unit tangent vector of a curve. b) Radius of curvature of a curve.

An arbitrary curve
s(t) on the surface is defined by the parametric description r = r(t) =

1
2
r (t), (t) , where t is a free parameter as explained subsequent to (1-97). P and Q denote
two neighboring points on the curve defined by the position vectors r and r + dr, corresponding
to the parameter values t and t + dt, respectively. The increment dr with the length ds is a
surface vector placed in the tangent plane at P , and specified by the covariant representation
dr = d a . Hence, the unit tangent vector to the curve at P is given as
t=

dr
d
=
a
ds
ds

(1134)

The unit tangent vector in Q deviates infinitesimally from t, and may be written as t + dt.
Given, that the length unchanged is equal to 1, the increment dt must fulfill
1 = (t + dt) (t + dt) = t t + 2 dt t + dt dt = 1 + 2 dt t
dt t = 0

(1135)

36

Chapter 1 Preliminaries

where the second order term dtdt is ignored. (1-135) shows that the increment dt is orthogonal
to t. Let n denote the unit normal vector n co-directional to dt. Then, the following identity
may be defined
dt
= 0 n
ds

(1136)

dt
(1-136) is referred to as Frenets formula. ds
and n are called the curvature vector and the
principal normal vector of the curve, respectively, and the proportionality factor 0 is referred
to as the curvature. Generally, this is always assumed to be positive, so (1-136) is defining the
orientation of n. In Fig. 1-5b lines orthogonal to s have been drawn through P and Q in the
plane spanned by t and n, which intersect each other at the curvature center O under the angle
d as shown on Fig. 1-5b. The position of O relative to s is defined by the orientation of n.
Then, the following geometric relation prevails, see Fig. 1-5b

|dt ds| = |t ds| d

|dt| = | t | d = 1 d =

ds
R

(1137)

where R = OP = OQ denotes the radius of curvature at P . Since, |n| = 1 it follows from


(1-136) and (1-137) that the curvature has the following geometric interpretation
0 =

1
R

(1138)

Fig. 16 Normal plane to unit tangent vector.

The unit tangent vector t is normal to a plane , which contains both to the unit normal vector
a3 = a3 and to the principal normal vector n, see Fig. 1-6. A third linear independent unit
vector ag in is given by the vector
ag = a3 t

(1139)

1.2 Differential Theory of Surfaces

37

We may then represent the curvature vector


follows

dt
ds

= 0 n as a linear combination of a3 and ag as

0 n = a3 + g ag

(1140)

The expansion components and are known as the normal curvature and the geodesic curvature of the curve s. Using that a3 ag = 0, and both are unit vectors, these may be expressed
in terms of the curvature 0 of the curve as follows

dt

= a3
= (a3 n) 0

ds

dt

g = a0
= (a0 n) 0
ds

(1141)

Using the chain rule of partial differentiation it follows from (1-134) that the curvature vector
dt
has the following representation
ds
dt
d2
a d d
=
a
+

ds
ds2
ds ds

(1142)

Since a3 = a3 is orthogonal to a , it follows from (1-119), (1-141) and (1-142) that


dt
a d d
3
= a3
=a
=
ds
ds ds

d d
d d
= b
ds ds
ds ds

(1143)

where the following indexed quantities have been introduced

a
= a3
=

= b

(1144)

The neighbouring points P and Q on the curve s are given by the position vectors r(s) and
r(s + ds) = r(s) + dr. By the use of (1-134) the increment dr is given by the Taylor expansion

dr = r(s + ds) r(s) =


dr
1 d2 r 2
1 dt 2
3
ds +
ds
+
O
ds
= t ds +
ds + O(ds3) (1145)
2
ds
2 ds
2 ds

The distance dz of Q from the tangent plane at P is given by the projection of dr on the surface
unit normal vector a3 , see Fig. 1-5. Hence,
dz = a3 dr =

1
dt 2 1
1
a3
ds = ds2 = b d d
2
ds
2
2

(1146)

where (1-143) and the orthogonality of a3 and t has been used. The larger the distance dz, the
larger is the curvature of the curve at P . In (1-146) this quantity is partly determined by the

38

Chapter 1 Preliminaries

components b and partly by the parameter increments d . As seen from (1-144) the components b are entirely determined from properties related to the surface, and independent of
the specific direction of the curve s. In contrast this direction is determined by the increments
d of the curvilinear coordinates. The quadratic form b d d entering (1-146) is called
the second fundamental form of the surface. In what follows b is considered the contravariant
components of a surface second order tensor b = b a a , which is called the curvature tensor
of the surface.
Using a = a a , it follows from (1-119), (1-144) that
a
a
(a a )
a
b = a3
=
a

a
+
a
a

=
a
a

= a
3




= a b = b
3

(1147)

where the orthogonality of a3 and a is used. Insertion of (1-144) and (1-147) into (1-133)
provides the following alternative representation of the components of the surface RiemannChristoffel tensor
B = b b + b b

B = b b + b b

(1148)

where the transformation rules (1-114) of covariant and contravariant tensor components have
been used in the final statement. (1-148) provides the following form for the selected independent tensor component B1212
 
B1212 = b21 b12 + b22 b11 = det b = b

(1149)

where b denotes the determinant formed by the contravariant components of the curvature tensor b. (1-149) is called Gausss equation. If B1212 = 0, all other contravariant components of
the tensor will vanish as well, and it may be concluded that B = 0. Then, a surface is flat, if
b = 0 everywhere. Noticed that if the determinant of the contravariant components vanish, the
determinant of the components of any other representation of b vanishes as well.
From (1-117) and (1-143) follows that the normal curvature of a curve on the surface with a
direction determined by the increment d is given as

b d d
a d d

(1150)

(1-150) may be interpreted as a Rayleigh quotient related to the following generalized eigenvalue problem, (Nielsen, 2006)

1.2 Differential Theory of Surfaces

39


b a t = 0

(1151)


b a t = 0

(1152)

t = dds represent the covariant components of a unit tangent vector t = t a to the surface,
which specifies the a direction for which (1-151) is fulfilled. With a = a a a and b =
b a a , (1-151) is seen to represent the curvilinear component relation of the following tensor
relation

(1-152) is fulfilled for two unit eigenvectors t1 and t2 related to the eigenvalues 1 and 2 of
(1-151). The directions specified on the surface by the eigenvectors t are denoted the principal curvature directions, and the related eigenvalues are called principal curvatures. The unit
eigenvectors fulfill the orthogonality conditions, (Nielsen, 2006)

t a t =

0 ,

t b t =

0 ,

1 ,

6=

(1153)

6=

(1154)

Let the principal curvatures be ordered, so 1 denotes the smallest and 2 the largest curvature.
As follows from the well-known bounds on the Rayleigh quotient the normal curvature of
an arbitrary curve passing through P will be bounded by the principal curvatures as, (Nielsen,
2006)
1 2

(1155)

Using the mixed representations a = a a and b = b a a the component form of (1-152)


attains the form

b t = 0

(1156)

Formally, (1-156) may be obtained from (1-151) by pre-multiplication of a and use of (1111), (1-114). Solutions t 6= 0 of the homogeneous system of linear equations (1-156) are
obtained, if the determinant of the coefficient matrix vanishes. This provides the following
characteristic equation for the determination of the principal curvatures




det b = 2 b11 + b22 + b11 b22 b21 b12 = 0

2 b +

b
=0
a

(1157)

40

Chapter 1 Preliminaries

 
 
In the last statement it has been used that det b = ab , where a = det a denotes the
determinant of the matrix formed by the contravariant components of the surface identity tensor.
This relation follows from the identities
= a b
 
 
 
det b = det a det b

 
b
det b =
a

(1158)

The solutions to (1-157) may be written as


1

H2 K

(1159)

1
1
H = (1 + 2 ) = b
2
2

(1160)

=H

where

K = 1 2 =

b
= b11 b22 b21 b12
a

(1161)

H is called the mean curvature of the surface, and K is the Gaussian curvature. As seen K
is equal to the determinant of the matrix formed by the mixed components b of the curvature
tensor, and 2H is equal to the trace of this matrix. It follows from that K = 0, if either 1 = 0
or 2 = 0. A surface, where the Gaussian curvature vanishes everywhere is developable, i.e. the
surface can be constructed by transforming a plane without distortion. Conical and cylindrical
syrfaces are developable, whereas a spherical surface in not. From (1-149) and (1-161) follows
B1212 = aK

(1162)

Since, the eigenvectors t are surface vectors, it follows from (1-116) and (1-153) that t1 at2 =
t1 t2 = 0. This means that the principal curvature directions through P are orthogonal to each
other. It is then possible to construct an (s1 , s2 ) arc-length coordinate system at each point on
the surface, which everywhere has the principal curvature eigenvectors t1 and t2 as tangents.
This so-called principal curvature coordinate system is specific in the sense that both [a ] and
[b ] become diagonal, i.e a12 = b12 = 0. This is analog to the formulation in modal coordinates
of the equations of motion of linear structural dynamics, where the modal mass and stiffness
matrices become diagonal, (Nielsen, 2006). In this specific coordinate system the principal
curvatures are given as

b()()
a()()

(1163)

1.2 Differential Theory of Surfaces

41

Next, various relations are indicated, which involve the partial derivative of the surface unit
normal vector a3 = a3 . Partial differentiation of the equation a3 a = 0 and use of (1-144)
provide
a
a3

a
+
a

=0



a3
a
3
a = a3 =
= b

(1164)

a3 is a unit vector, so a3 a3 = 1. Then, partial differentiation provides


a3

a3
=0

(1165)

a3
Hence,
is orthogonal to a3 , and accordingly can be decomposed in the covariant surface
vector base a as follows

a3
= l a

(1166)

Scalar multiplication of both sides with a and use of (1-110) and (1-164) provides the following
solution for the covariant components l

a3
= b = l a a = a l

l = a b = b

(1167)

Then, insertion into (1-166) provides the following alternative representations




a3
a3
3

= = b a = b a a = b a =
a

(1168)

where (1-112) and (1-144) have been used. Similarly, (1-118) and (1-144) provide the following
relation for the partial derivatives of the surface covariant base vectors
a
=

a + b a3

(1169)

(1-169) is called Gausss formula. Next, e and e are introduced as alternative two-dimensional
permutation symbols defined as

42

Chapter 1 Preliminaries

#
0 1

1 0
"
#

 
0 1

e
=

1
0
 
e =

"

(1170)

As seen [e ] = [e ]1 = [e ]T = [e ]. From (1-110) follows that


a1 = |a1 | = (a1 a1 )1/2 =
a2 = |a2 | = (a2 a2 )1/2 =
cos

a1 a2
|a1 ||a2 |

a11

a22

a12

a11 a22

sin =

a212
a11 a22

Then, (1-102) may be written as


A=

a11 a22 sin = a11 a22 (a12 )2 = a

(1171)

(1172)

where a = det[a ]. By the use of (1-170) and (1-172) the vector products in (1-103) can be
written in the following compact forms
a3 a =
a a =

a e a
a e a3

(1173)

Further, the following component identities may be derived

a = a e a e


a = e a e

e = a e a


e = a a e a

(1174)

The first relation in (1-174) follows from the following matrix identities
"
# "
#1
"
#
"
#"
#"
#T
a11 a12
a11 a12
a22 a12
0 1 a11 a12
0 1
= 21 22
=a
=a
a21 a22
1 0 a21 a22
a
a
a21
a11
1 0

(1175)

1.2 Differential Theory of Surfaces

43

where it has been used that det[a ] = 1/ det[a ] = a1 . The proof of the remaining relations in
(1-174), which is left an exercise, may be carried out by simple matrix operations on (1-175).
Next, a three-dimensional vector field v = v(1 , 2 ) is considered, defined on the parameter
domain . Hence, a vector is connected to each point of the surface . Since, v 3 = v3 6= 0, v is
not a surface vector. Physically, v may represent a force vector, a moment vector, a displacement
vector or a rotation vector related to a material point on the surface. In this case the following
representations prevail
v = v a + v 3 a3 = v a + v3 a3
Since,

v
3

= 0, the gradient becomes

v
a = v j ; aj a = vj; aj a

Similarly, (1-74) becomes


v =

v
j

3
= v ; aj = v ; a + v ; a3

3
=
v
a
=
v
a
+
v
a
j;
;
3;

where, cf. (1-73)


v ;
v 3;
v;
v3;

(1176)






v +
v3
= +






v
3
3

3
= +
v +
v


3





=
v
v3






v3

=
v
v3

3
3

(1177)

(1178)

(1179)

The right hand sides of (1-178) both consist of a surface vector and a vector in the normal
direction. The surface and normal components of these representations must pair-wise be equal
corresponding to
v ; a = v; a

v 3 ; a3 = v3; a3

v ; = a v;

v; = a v ;

v = v3
3;

(1180)

44

Chapter 1 Preliminaries

The final relations in (1-180) follow


from
the
initial vector
identities
using (1-112) and a3 = a3 .






Further, v3 = v 3 , v 3; = v3; , 3 3 = 3 3 = 0 and 3 = b , cf. (1-120), (1-144). Then,
withdrawal of the second equation in (1-179) from the fourth equation provides the following
relations for the components of the curvature tensor

v =

v = b v

a v = b v

= b a

b =

b =

1.2.3

(1181)

Codazzis equations

From (1-164), (1-168) and (1-169) follows




b

a3
a a3
2 a3
=








2 a3
2 a3

a + b a3 b a a = b
a



(1182)

Interchange of the indices and in (1-182) provides




2 a3
b

= b
a

(1183)

Withdrawal of (1-182) from (1-183) gives the relation






b
b

b
=
b

(1184)

(1-184) is trivially fulfilled for = . Further, the index combinations (, ) = (1, 2) and
(, ) = (2, 1) lead to the same relation. Hence, merely two non-trivial and independent relations exist, which are taken corresponding to the index combinations (, , ) = (1, 1, 2) and
(, , ) = (2, 2, 1), leading to the following relations

1.2 Differential Theory of Surfaces



b11

=
b1
2
1 2


b22

=
b2
1
2 1



b12

b2

1 1


b21

b1
2

2 2

45

(1185)

The relations (1-185) are called Codazzis equations. It can be shown that arbitrary component functions a = a (1 , 2 ) and b = b (1 , 2 ) of the curvilinear parameters 1 and
2 uniquely determine a surface (saved an arbitrary rigid body translation and rotation), which
has a d d and b d d as its first and second fundamental form, if and only if these
components fulfill the Codazzs equations (1-185) and the Gausss equation (1-165), (Spain,
1965).
In a principal curvature coordinate system we have a12 = a21 = 0 and b12 = b21 = 0. In this
case the Codazzis equations can be reduced to




a1
R1
a2
R2




1 a1
=
R2 2
1 a2
=
R1 1

(1186)

where a1 and a2 are given by (1-171), and R1 and R2 are the principal curvature radii . A proof
of (1-186) has been given in Box 1.2.

Box 1.2: Proof of (1-186)


Since b12 = b21 = 0 in the principal curvature coordinate system, (1-185) can be reduced
to




b11
1
2

= b11
b22

1 2
1 1
(1187)




b22

2
1

= b22
b11
1
2 1
2 2

46

Chapter 1 Preliminaries

Next, the Christoffel symbol in (1-187) are expressed in terms of the covariant and contravariant components of the identity tensor as given by the defining equation (1-66). Due
to the principal curvature coordinates and the special structure (1-106) these components
1
become gjk = g jk = 0 for j 6= k, g()() = a()() and g ()() = a()() = a()()
. Then





1 11 a11
1 1 a11
1

= a
+00 =

2 a11 2
1 2






a
1
1
a
1
2
11
11

22

= a
0+0
=

2
2
2 a22 2
1 1
(1188)




1 22 a22
1 1 a22

= a
+00 =

2
1
2 a22 1
2 1






1
a
1
1
a
1

22
22
11

= a
0+0
=

2
1
2 a11 1
2 2

Further, from (1-163) follows that b11 = (1/R1 )a11 and b22 = (1/R2 )a22 . Insertion of
(1-188) in (1-187) then provides

 



a11
1
1
1
a11

=
+

2 R1
2 R1 R2
2
(1189)
 



a22
1
1
1
a22

=
+

1 R2
2 R1 R2
1
Introduction of (a1 )2 = a11 and (a2 )2 = a22 provides the identities
 


 

a11

a1
a1 a1

a1

= 2 a1
=
+ a1 2

2
2

R1

R1
R1

R1

 


 

a22

a2
a2 a2

a2

= 1 a2
=
+ a2 1

1
1

R2

R2
R2

R2

a1
a11

=
2a
1

2
2

a22
a2

=
2a
2
1
1

(1190)

Insertion of the results (1-190) into (1-189) immediately provides (1-186).

Example 1.2: Identity tensor, curvature tensor, Riemann-Christoffel tensor and principal
curvatures in spherical coordinates
As seen from (1-93) the contravariant components of the surface identity tensor becomes

1.2 Differential Theory of Surfaces

"
r2
[a ] =
0

0
r2 sin2 1

47

(1191)

The contravariant components of the curvature tensor follow from (1-94) and (1-144)
"
#
r
0
[b ] =
0 r sin2 1

(1192)

Since both [a and [b ] are diagonal, it is concluded that the spherical coordinate system is a principal curvature
coordinate system. The principal curvatures and related principal curvature radii follow from (1-163)
1 = 2 =

1
r

R1 = R2 = r

(1193)

1 and 2 are negative, because unit surface normal vector is directed away from the curvature center, which in
this case is identical to the origin of the Cartesian coordinate system.
The selected independent component of the Riemann-Christoffel tensor becomes, cf. (1-149)
B1212 = r2 sin2 1

(1194)

Left and right hand sides of the Codazzi equations (1-186) vanish in this case. Hence, these equations are trivially
fulfilled.

1.2.4

Surface area elements

Fig. 17 Area elements in parallel surfaces.

Fig. 1.7 shows two surfaces and 0 , described by the same parameters 1 and 2 . The points
P0 and P denotes the mapping points for the same parameters, specified by the position vectors

48

Chapter 1 Preliminaries

r = r(1 , 2 ) and x = x(1 , 2 ), respectively. Further, the unit normal vectors related to the two
surfaces at the said points are assumed to be identical and equal to a3 = a3 . Then, the distance
between any such corresponding points P0 and P on the surfaces are everywhere constant and
equal to s3 , where s3 is measured from the surface 0 . On the surfaces arc length coordinate
coordinate systems (s1 , s2 ) and (S 1 , S 2) are introduced with origins P0 and P . The related
surface covariant base vectors are denoted a and g , respectively. Then, the position vectors
are related as follows
x = r + s3 a3

(1195)

dA0 denotes a differential area element of shape as a parallelogram with sides of the length ds1
and ds2 parallel to a1 and a2 , see Fig. 1.7. The length of the sides follows from (1-117)
ds1 =
ds2 =

a11 d1
a22 d2

(1196)

Let denote the angle between a1 and a2 . Then, cf. (1-102), (1-172)

dA0 = ds1 ds2 sin = a11 a22 sin d1 d2 =
p

a11 a22 (a12 )2 d1 d2 = a d1 d2

(1197)

The surface covariant base vectors g follows from (1-99), (1-168), (1-195)

x
r
a3
= + s3 = a s3 b a = s3 b a

Then, the contravariant components g of the surface identity tensor of becomes


g =

(1198)





g = g g = s3 b s3 b a a = s3 b s3 b a =




(1199)
a a s3 b a s3 b = a a s3 b a s3 b

Then, the differential area element dA on , corresponding to dA0 on 0 , becomes


dA =

g d1 d2

(1200)

where g = det[g ] follows from (1-199)


 



 1

2
3
3
3
g = det a det a s b det a s b =
det a s b
=
a
 2
 

2

2
det a 
1
3
3
det a det s b
=
det s b
=
a
a


2 2
b 2 2
a 1 b s3 + s3
= a 1 2H s3 + K s3
a

(1201)

1.3 Continuum Mechanics

49

where (1-157) and (1-158) have been used. Further, the mean curvature H and the Gaussian
curvature as given by (1-160) and (1-161) have been introduced. Then the ratio between the
differential area elements become

g
dA
= = 1 2Hs3 + K(s3 )2
dA0
a

(1202)

1.3 Continuum Mechanics


1.3.1

Three-dimensional continuum mechanics


V

V0

dX

dx
0 dV0

u(Xk , t)
dV

A0
x2

x(Xk , t)

i2
O
i3

i1

x1

Fig. 18 Referential and current configuration of a flexible body.

Fig. 1-8 shows a flexible body, i.e. a set of connected material points, which at the time t = 0
occupies the volume V0 with the surface A0 . The indicated configuration of the body is referred
to as the referential configuration. Material points (mass particles) are identified by their position vector X relative to the origin of a given frame of reference. Although any curvilinear
coordinate system may be used, vectors and tensors will in this and the next subsection exclusively be described in the Cartesian (x1 , x2 , x3 )-coordinate system with the origin O and the
base unit vectors (i1 , i2 , i3 ) shown on Fig. 1-8, with the exception that the dual Cartesian vector
base (i1 , i2 , i3 ) will be used as well. Because the unit base vectors in the indicated vector bases
are identical, i.e. ij = ij , covariant, contravariant and mixed Cartesian components will be
identical. Since no misapprehension with covariant and contravariant curvilinear coordinates is
possible, the bar atop the Cartesian components of vectors and tensors will be skipped throughout this subsection. The Cartesian components Xj of X = Xj ij are denoted the material or
Lagrangian coordinates of the particle. At the time t the body occupies a new volume V with
the surface A due to translation, rotation and mutual distortion of the particles. The indicated
configuration is referred to as the current configuration of the body. The position in the current configuration of a particle with the referential coordinates X j is defined by the position

50

Chapter 1 Preliminaries

vector x(Xk , t) = xj (Xk , t) ij . The current coordinates xj are known as the Eulerian cordinates of the particle. Traditionally, the governing equations of fluid mechanics are formulated
using Eulerian coordinates. This permits that large distortions in the continuum motion can be
handle with relative ease. By contrast formulations in Lagrangian coordinates is preferred for
the kinematical description in solid mechanics, which allows an easy tracking of the material
history. Its weakness lies in its inability to follow large distortions of the computational domain.
The displacement of the particle is defined as the difference between the position vector in the
current and referential configurations, see Fig. 1-8
u(Xk , t) = x(xk , t) X

(1203)

Consider a neighbor particle with the referential position vector X + dX. The indicated particle
is displaced to the position x(Xk + dXk , t) = x(Xk , t) + dx(Xk , t), where the incremental
vectors are related as
dx(Xk , t) = F(Xk , t) dX

(1204)

F = x is the so-called material deformation gradient, cf. (1-76). This second order tensor, as
well as the transposed tensor FT , the inverse tensor F1 and the transpose of the inverse tensor
FT have the following dual Cartesian representations, cf. (1-41), (1-47)
F
1

xj j k
= Fjk i i =
ii
Xk
Xj j k
1 j k
= Fjk
ii =
ii
xk
j k

,
,

xk j k
= Fkj i i =
ij
Xj
Xk j k
1 j k
= Fkj
ii =
ii
xj
j k

(1205)

The inverse material deformation gradient presumes the existence of the inverse mapping X =
X(xk , t). This requires that F is non-singular, i.e. that the Jacobian det[Fjk ] > 0. Let dV and
dV0 denote the volume of a particle in the current and the referential configuration. Further, let
and 0 be the mass density of the particle in these configurations. Since the mass of the particle
is unchanged we must have dV = 0 dV0 . The Jacobian specifies the quotient between the
differential volumes of the particle in the current and referential configurations corresponding
to

det[Fjk ] =

dV
0
=
dV0

(1206)

The distance between the neighboring particles in the referential and the current configuration
is given by the lengths ds and dS of the incremental position vectors dx and dX. Clearly, the
difference between these lengths is a measure of the local distortion of the material. Among
several strain measures suitable for description of large strains we shall adopt the so-called
Green strain tensor E, which is introduced in the following way


ds2 dS 2 = dx dx dX dX = dX FT F g dX = 2 dX E dX

(1207)

1.3 Continuum Mechanics

51

where (1-44) and (1-204) have been used. g denotes the identity tensor, which has the following Cartesian representation g = jk ij ik = jk ij ik = jk ij ik . jk = jk = jk specify the
Kroneckers delta (1-14) in various tensor bases. Hence, the Green strain tensor is related to the
material deformation gradient as follows

E=


1 T
F Fg
2

(1208)

The dual Cartesian components of E may be expressed in terms of the components of the
displacement field by (1-203), (1-205) and (1-208)
 1

1
Flj ij il Fmk im ik jk ij ik =
Flj Fmk lm jk ij ik =
2
2




1
1 xl xl
Flj Flk jk ij ik =
jk ij ik =
2
2 Xj Xk




1 
ul 
ul 
1 uj
uk
ul ul
j k
+
+
ij ik

lj +
lk +
jk i i =
2
Xj
Xk
2 Xk Xj Xj Xk


1 uj
uk
ul ul
Ejk =
+
+
(1209)
2 Xk Xj Xj Xk
E = Ejk ij ik =

For deformation gradients fulfilling | Xjk | 1 the quadratic term in (1-209) may be ignored.
The corresponding Green small strain tensor e has the components

ejk

1
=
2

uj
uk
+
Xk Xj

(1210)

The Eulerian small strain tensor represents the corresponding strain tensor, where the deformation gradient is taken with the respect to the Eulerian coordinates. This tensor has the dual
Cartesian components

jk

1
=
2

uj
uk
+
xk
xj

(1211)

The Cauchy equations of motion of a mass particle is formulated in Eulerian coordinates. In


tensor notation these read, (Malvern, 1969)
a = + b

(1212)

a(xk , t) denotes the acceleration vector of the particle, and b(xk , t) is the external load per
unit mass. (xk , t) represents the Cauchy stress tensor, which is a symmetric tensor, = T .
The divergence operation specifies a vector with the components x l ( jk ij ik ) il =
jk
ii
xl j k

il =

jk
i,
xk j

cf. (1-77). Then, the dual Cartesian component form of (1-212) becomes

52

Chapter 1 Preliminaries

A1

dA

A2

Fig. 19 Definition of boundary conditions.

aj =

jk
+ bj
xk

(1213)

The surface traction (xk , t) at the current surface A is related to Cauchys stress tensor (xk , t)
by Cauchys boundary condition, (Malvern, 1969)
(xk , t) = n

(1214)

where n = n(xk , t) signifies the outward directed unit vector to the surface A. A is divided
into two disjoint sub-surfaces A1 and A2 , at which either mechanical or kinematical boundary
conditions are prescribed, see Fig. 1-9. Then, (1-212) is solved with the following boundary
conditions
(xk , t) = (xk , t)
(xk , t)
u(xk , t) = u

,
,

xk A1
xk A2

(1215)

(xk , t) is the pre (xk , t) denotes the prescribed surface traction on the sub-surface A1 , and u
scribed displacement on the sub-surface A2 .
As a basis for any finite element method the equations of dynamic equilibrium and related
boundary conditions are reformulated in a weak form. This is a variational principle for some
functional of the unknown fields of the problem. At stationarity the varied functional determines the discretized unknown fields, so these fulfil the equilibrium equation and the boundary
conditions at best or in average throughout V and A, when the fields are varied within
a predefined functional sub-space. For so-called displacement based finite element formulations, where the only unknown is the displacement field, the relevant variational principle is the
principle of virtual displacements, which may be formulated as
Z

dV =


u b a dV +

with the component form

A1

u dA

(1216)

1.3 Continuum Mechanics

jk

jk dV =

53

uj b a dV +

uj j dA

(1217)

A1

u(xk ) = uj (xk ) ij denotes an arbitrary variational displacement field with the restrictions that
u(xk ) = 0 on A2 . Then, (1-217) can be derived by scalar multiplication of both sides (1-213)
with uj (xk ), followed by an integration over V . Next, the identity is obtained by the use of the
divergence theorem, the mechanical boundary condition on A1 , and the kinematical constrain
that allowable variations fulfill uj (xk ) = 0 on A2 . Save that certain continuity requirements
need to fulfilled to make the indicated mathematical operations meaningful, the variational field
u(xk ) may else be prescribed with an arbitrary magnitude. Especially, the variations do not
necessarily have any relation to the actual displacement field.
= jk ij ik in (1-216) denotes the Eulerian small strain tensor for the variational displacement field u(xk ), i.e. the components are given as, cf. (1-211)

jk

1
=
2

uj uk
+
xk
xj

(1218)

Because uj (xk ) may be of finite magnitude, this is not identical to the variation of the components of the Eulerian small strain tensor of the actual displacement field as given by (1-211).
This is so because xj depends on uj , for which reason the Eulerian gradient x k will be varied
as well. Actually

uj
xk

uj Xl
Xl xk

uj Xl
uj
=
+

Xl xk
Xl

Xl
xk

uj
uj
+
Flk1
xk
Xl

(1219)

where it has been used that the Lagrangian coordinates Xk are not subjected to any variation, i.e. the variation concerns the displacement of a specific mass particle, for which reason
u
u
u
u
( Xjk ) = Xkj . Only for small displacement variations will ( xkj ) = xkj .
Since, and are symmetric tensors, it follows that = , cf. (1-59), (1-61).
Hence, the same internal virtual work is obtained in (1-216) independent of which definition of
the double contraction of the involved tensors is used.
Finally, the acceleration vector a enters the volume integral on the right hand side of (1-216) via
the combination b a. Formally, b a may be interpreted as an equivalent quasi-static load
per unit mass, which takes the dynamic effects into consideration via the the so-called inertial
load a dV on the particle. This is the essence of the dAlemberts principle. Then, the
right hand side of (1-216) represents the external virtual work We performed by the equivalent
load per unit mass b a throughout V , and the surface traction on A1 . The left hand side
represents the internal virtual work Wi performed by the internal forces (stress components)
on the particles. Then, the principle of virtual displacements postulates that the internal and
external virtual works are equal, i.e.

54

Chapter 1 Preliminaries

Wi = We

(1220)

The idea is to formulate finite element theories based on the Lagrangian formulation, for which
reason (1-216) at first is reformulated into the following equivalent form based on quantities
defined in the referential configuration
Z

V0

E S dV0 =

V0


u 0 b a dV0 +

u 0 dA0

A01

(1221)

with the component form


Z

jk

Ejk S dV0 =

V0

V0

uj 0 b a dV0 +

A01

uj 0j dA0

(1222)

(b a) dV = (b a) 0 dV0 specify the equivalent static load on a given particle, so the first
integrals on the right hand side of (1-216) and (1-221) denote the summation of these loads
from all particles within the structure. Hence, these integrals are equal.
The surface A0 in the referential configuration is divided into the sub-surfaces A01 and A02 ,
which are mapped onto the A1 and A2 in the current configuration. Correspondingly, a surface
area element dA0 on A01 is mapped onto the surface area element dA on A1 . 0 entering the
second integral on the right hand side of (1-221) denotes a fictitious surface traction defined
in the referential configuration, which is related to the actual surface traction in the current
configuration as

0 dA0 = dA

(1223)

Then, the second integrals on the right hand side of (1-216) and (1-221) represent the summation of identical loads from corresponding area elements in the two configuration. It follows
that also these integrals are equal. 0 will be referred to as a pseudo surface traction, since
the referential configuration is actually free of any stresses, save for a possible known residual
stress configuration with no relation to the indicated external loads and boundary conditions.
As mentioned, denotes the Eulerian small strain tensor of the variational displacement field
u(xk ), and must not be confused with the variation of the Eulerian small strain tensor of the
actual displacement field. For this reason it could be anticipated that E entering the integral on
the left hand side of (1-221) in the same way represents the Green strain tensor of the variational
displacement field. This is not so. Indeed, E denotes the variation of the finite Green strain
tensor due to the variation the displacement field, and depends on both u and u. In order to
derive the relation between E and the variation of the material gradient tensor is at first
determined. Use of (1-205) provides with xj = (Xj + uj ) = uj

1.3 Continuum Mechanics

F =

xj
Xm

ij im =

55

xj j m uj xk j m
uj xl j kl m
ii =
ii =
i ( i ) =
Xm
xk Xm
xk Xm

uj j k xl l m uj j k
uj xl j k l m
i (i i i ) =
ii
ii =
i i F
xk Xm
xk
Xm
xk

(1224)

Then, the variation of the Green strain tensor follows from (1-208) and (1-222)

E =


1  uk uj  j k
1 T
F F + FT F g = FT
+
i i F = FT F (1225)
2
2 xj
xk

where it has been used that g = g u = 0, cf. (1-81). The related component form
becomes, cf. (1-205)
Ejk = E jk =

xl lm xm

Xj
Xk

(1226)

The tensor and component form of the inverse relation read


= FT E F1

jk = jk =

(1227)

Xl
Xm
E lm
xj
xk

(1228)

The tensor S entering the integral on the left hand side of (1-221) is another pseudo-stress
quantity denoted the second Piola-Kirchhoff stress tensor. This is the virtual work conjugated
stress measure to E to insure that the left hand sides of (1-216) and (1-221) become identical.
This necessitates that jk jk dV = Elm S lm dV0 . Then, use of (1-206) and (1-226) provides
the following relation between the components of the Cauchy stress tensor and the components
of the second Piola-Kirchhoff stress tensor
xj
xk lm
jk
S dV0
Xl
Xm
dV0 xj xk lm
xj lm xk
=
S =
S
dV Xl Xm
0 Xl
Xm

jk jk dV = Elm S lm dV0 =
jk = jk

(1229)

The related tensor form reads


=

F S FT
0

The tensor and component form of the inverse relation read

(1230)

56

Chapter 1 Preliminaries

S=

0 1
F FT

S jk = Sjk =

(1231)

0 Xj lm Xk

xl
xm

(1232)

For hyperelastic materials the existence of an elastic potential W (E) is postulated, depending
entirely on the Green strain tensor, from which the components of the second Piola-Kirchhoff
tensor may be obtained from the gradient relation
S jk =

W (E)
Ejk

(1233)

For a St.Venant-Kirchhoff type of material the following linear relation between S and E is
postulated
S = S0 + C E

(1234)

S0 signifies a known residual stress configuration in the referential configuration, which has no
dependence on the external loads, the surface traction on A01 , or on the prescribed displacements on A02 . The fourth order tensor C is known as the elasticity tensor. In the first place this
has 34 = 81 independent components C jklm. However, because S = ST and E = ET are both
symmetric tensors, the following symmetry relations prevail C jklm = C kjlm = C jkml , which
reduces the number of independent components to 66 = 36. Additionally, if the linear material
is hyperelastic, it follows from (1-233) and (1-234) that
C jklm =

2 W (E)
2 W (E)
=
= C lmjk
Ejk Elm
Elm Ejk

(1235)

The symmetry condition (1-235) further reduces the number of independent elastic constants to
21. In this case the constitutive relation between the 6 non-trivial stress components (due to the
symmetry of S), and the 6 non-trivial components of E can be given in the following matrix
form
S = S0 + C E

S 11
22
S
33
S

S=
S 12

13
S
S 23

(1236)

S011
22
S0
33
S0

S0 =
S 12
0
13
S0
S023


E11

E22

E33

E=
E
12

E13
E23

(1237)

1.3 Continuum Mechanics

C 1111
1122
C
1133
C
C=
C 1112

1113
C
C 1123

C 1122
C 2222
C 2233
C 2212
C 2213
C 2223

57

C 1133
C 2233
C 3333
C 3312
C 3313
C 3323

C 1112
C 2212
C 3312
C 1212
C 1213
C 1223

C 1113
C 2213
C 3313
C 1213
C 1313
C 1323

C 1123

C 2223

C 3323

C 1223

1323
C

1313
C

(1238)

For an orthotropic material the number of material parameters further reduces to 9, corresponding to the constitutive matrix

C 1111 C 1122 C 1133


0
0
0
1122

C
C 2222 C 2233
0
0
0
1133

C
C 2233 C 3333
0
0
0

C=

0
0
0
2G
0
0
12

0
0
0
0
2G
0

13
0
0
0
0
0
2G23

(1239)

G12 , G13 and G23 denotes the shear moduli of the material. Finally, in case of an isotropic
material merely two independent parameters remains. In this case the constitutive matrix may
be written, (Malvern, 1969)

C=

+ 2

0 0 0

+ 2

0 0 0

+ 2 0 0 0

0
0
0
2 0 0

0
0
0
0 2 0
0
0
0
0 0 2

(1240)

where and are known as the Lame parameters. these may be related to Youngs modulus E,
and the Poissons ratio as follows, (Malvern, 1969)

E
(1 + v)(1 2)

=G=

E
2(1 + )

(1241)

where G is the shear modulus. It should be noticed that although the relation (1-236) is physical
linear, the relation becomes geometrical non-linear, when the Green tensor is expressed in the
displacement components.
The use of the variational formulation (1-221) as a basis for a finite element formulation requires that the variational field u(Xk ) and displacement field u(Xk , t) need to be interpolated

58

Chapter 1 Preliminaries

between nodal values by means of suitable selected shape function. The shape function for the
variational field must vanish, and the shape function for the displacement field must comply
with the kinematical boundary conditions on A2 .

Fig. 110 Shell like structure in referential configuration.

In the present context the body V0 in the referential state is either a plate or a shell. Shell
structures have a surface like appearance, since the thickness is considerable smaller than the
two other dimensions. The geometry of a shell structure is determined by a parameterized
surface and a thickness function. In this case the surface A0 is conveniently divided into three
sub-surfaces 0 u , 0 l and 0 . 0 u and 0 l form the upper and lover surfaces of the body,
whereas 0 makes up the remaining cylindrical part of the boundary. 0 is further sub-divided
into the subsets 0 1 and 0 2 . Kinematical boundary conditions are only prescribed on 0 2 .
Hence, 0 2 is identical to the sub-surface A0 2 on Fig. 1-9, whereas A0 1 consists of 0 u , 0 l
and 0 1 . Further, the so-called midsurface 0 has been shown in fig. 1-10, which is an artificial
plane in the shell defined as the midst between 0 u , 0 l . The thickness t is the distance between
the upper and lower surface in the direction of the normal of the midsurface.

Fig. 111 Plate and shell surfaces in referential and current state.

The referential position of a material point within the shell is specified by the position vector
X(1 , 2 , 3 ), indicating a vectorial parametric description in terms of the curvilinear coordi-

1.3 Continuum Mechanics

59

nates 1 , 2 , 3 . The parametrization in the 3 -direction is centralized in a way that the midsurface is obtained for 3 = 0, i.e. the position vector to a particle or material point P0 of the
midsurface is given as
R(1 , 2 ) = X(1 , 2 , 0)

(1242)

The thickness t = t(1 , 2 ) related to a point of the midsurface may vary over the surface.
For a given value = 3 a surface is defined in the referential state by the position vector
X(1 , 2 , ) to a particle P . Especially, the upper and lower surfaces are described by the
parametric descriptions
)
Ru (1 , 2 ) = X 1 , 2 , t(1 , 2 /2)

Rl (1 , 2 ) = X 1 , 2 , t(1 , 2 /2)

(1243)

At the material points P0 and P local arc-length coordinate systems have been defined in both
the referential and the current states. The related covariant and contravariant base vectors in the
referential state are denoted Aj and Aj , and Gj and Gj , respectively. The corresponding base
vectors in the current state are denoted aj and aj , and gj and gj . The surface covariant base
vectors become, cf. (1-99)
1

A = A ( , )

G = G (1 , 2 , ) =
1

a = a ( , )

g = g (1 , 2 , ) =

(1244)

The displacement vector of the particles P0 and P from the referential state to the current state
are given as, see Fig. 1-11
v(1 , 2 ) = r(1 , 2 ) R(1 , 2 )

u(1 , 2 ) = x(1 , 2 ) X(1 , 2 )

(1245)

The idea in shell theory is to relate all vectors and tensors related to a particle P in the current
state to the covariant or contravariant base vectors Aj and Aj of the related particle on the
midsurface in the referential state. This will be the subject of section 2.

1.3.2

Fiber-reinforced composite materials

Composite materials consist of two or more ingredients with specific mechanical properties.
Typically, a composite contains a reinforcement material (fibres, steel bars), which are embedded into a matrix material (epoxy, cement). The strength and stiffness of the reinforcement

60

Chapter 1 Preliminaries

Fig. 112 Transfer of axial stress in fiber to matrix material.

material are much larger than those of the matrix material, as well as those of the combined
material. The reinforcement material is primarily carrying axial stresses, which are transferred
to the matrix material as shear stresses at the interface between the two materials. The basic
mechanism can be explained with reference to Fig. 1-12, which shows two cross-sections separated by a distance la of a circular fiber with the radius r embedded in a surrounding matrix
material. At the left section a tensile force F = f Af has been applied to the fiber, where f is
the axial stress, and Af is the area of the fiber. Over the length la , which is denoted the anchor
length, the axial stress drops from f to zero as a shear stress f develops at the interface between the materials. From static equilibrium follows that f = 2r df /dx. At the right section
both f and f are zero, and a tensile axial stress m has developed in the matrix material in
static equilibrium with F . Since, m is distributed to a significantly larger area than Af , this
stress is correspondingly smaller than the fiber stress at the left cross-section.

Fig. 113 Buckling of fiber in axial compression.

During axial compression the fiber may tend to buckle. Whether this happens depends on the
capability of the matrix material in providing lateral elastic support to the fiber as illustrated by
the Winckler foundation model shown on Fig. 1-13.
A lamina or ply represents the fundamental building block in a laminate. The fiber reinforcement may consist of discontinuous or continuous fibers. These may be unidirectional, bidirectional or randomly distributed, or consist of woven fiber nets as shown on Fig. 1-14. A
unidirectional fiber arrangement only provides extra strength and stiffness in the direction of

1.3 Continuum Mechanics

61

Fig. 114 Fiber arrangements. a) Discontinuous randomly distributed fibres. b) Unidirectional fibers. c) Bidirectional fibers. d) Woven fiber net. (Reddy, 2004).

the alignment. The strength and stiffness of continuously unidirectional fibers are larger than
that of a corresponding discontinuous fiber arrangement.
= 0

x1

> 0
= 90
x2

< 0

x3

Fig. 115 Laminate built up of unidirectional laminae with different fiber orientations. (Reddy, 2004).

A laminate consists of several stacked laminae to achieve the desired stiffness, strength and
thickness. In Fig. 1-15 is illustrated how unidirectional fiber reinforced laminae may be stacked
to achieved this. The layers are usually bonded together with the same matrix material as that
of the adjacent laminae. Mismatch of the material properties between the layers may produce
significant shear stresses at the interfaces, which in turn may cause delamination. Similarly,
mismatch of the fiber and matrix materials may cause debonding of the fibers.
The idea in what follows is to model the lamina as an equivalent smeared out homogeneous
orthotropic material, with averaged mechanical properties determined by the constituent materials. In order to derive the material law a so-called material (x 1 , x 2 , x 3 )-coordinate system
is introduced, with the (x 1 , x 2 )-plane placed in the mid-plane of the lamina, and the x 1 - and

62

Chapter 1 Preliminaries

x2

x3

x1
Fig. 116 Orthotropic lamina. Definition of material coordinate system. (Reddy, 2004).

x 2 -axes in the direction of the fibers, see Fig. 1-16.


The Piola-Kirchhoff stress components S 33 and S 3 in the said coordinate system vanish at the
upper and lower surface of the lamina. Then, continuity suggests that these stress components
can be ignored throughout the lamina. From (1-239) follows that the shear strain components

E3
of the Green tensor vanish as well. However, the normal strain components E33
is non-zero
as seen from the third equation in (1-239)). As an example the isotropic case provides with
(1-240)

S 33 = 0 = (E11
+ E22
) + ( + 2) E33

E33
=

(E11
+ E22
)=
(E + E22
)
+ 2
1 11

(1246)

where E33
in general is non-vanishing. The axial strain in the x 1 -direction is given as E11
=

S11
/E1 12
S22
/E2 , where E1 and E2 denote the Youngs moduli for normal stresses in the x 1
and x 2 -directions, and 12
denotes the Poissons ratio for the contraction in the x 1 -direction
due to an axial strain in the x 2 -direction. Similarly, the axial strain in the x2 -direction is given

as E22
= S22
/E2 21
S11
/E1 , where 21
denotes the Poissons ratio for the contraction in the

x2 -direction due to an axial strain in the x1 -direction. For plates and shells shear stresses S 3
are present in bending deformation, and cannot be disregarded. For this reason these will also
be included in the constitutive relation for the laminae. Then, the material law for the retained
second Piola-Kirchhoff stress and Green strain components can be presented by the following
flexibility relation on matrix form

E = D S
where

(1247)

1.3 Continuum Mechanics


E11
E
22

E = 12


13

63

S 11
S 22

S = S 12
13
S

(1248)

S 23

23

1
E1


21
E

D = 0

12
E2
1
E2

1
G12

1
G13

1
G23

(1249)

As seen in (1-248) the angle strains 12


= 2E12
, 13
= 2E13
, 23
= 2E23
have been introduced
as shear strain measures. The symmetry of the flexibility matrix requires that the Poisson ratios
fulfill

12
21
=
E2
E1

(1250)

Hence, at most six independent material parameters enter the constitutive law (1-248), (1-249).
The inverse stiffness relation reads
S = C E

(1251)

E1
1
12 21

21 E2

C = 1 12 21

12
E1

1 12
21
E2

1 12
21

G12

G13

G23

(1252)

In finite element formulations of plate and shell theories the kinematical constraint E33
= 0 is
often presumed, with the consequence that the element thickness is preserved during deformation. This constraint can only be fulfilled if a normal stress S 33 6= 0 is applied to prevent the

64

Chapter 1 Preliminaries

Poisson contraction as given by (1-245), thus stiffening the configuration. The resulting stiffening effect has been termed thickness locking or Poisson thickness locking, which is manifesting
in an overestimation of the bending stiffness. Despite the obvious contradiction this problem is
usually eliminated in finite element analysis by applying the plane stress constitutive equations
(1-251), (1-252).
x2
x1
x2

x3 , x3

x1

Fig. 117 Orthotropic lamina. Material and global coordinate systems. (Reddy, 2004).

Next, the constitutive relations are formulated in a referential (x1 , x2 , x3 )-coordinate system
with the (x1 , x2 )-plane coinciding with the (x 1 , x 2 )-plane, and with the x3 -axis co-directional
to the x 3 -axis. Further, the x 1 -axis is rotated the angle from the x1 -axis in the positive x3 direction, see Fig. 1-17. The base unit vectors ij of the material coordinate system are related
to the base unit vectors ij of the referential coordinate system as
ij = ljk ik

(1253)

The relation for the corresponding dual base unit vectors is written
i j = mjk ik

(1254)

ljk and mjk specify the referential coordinates of the base unit vectors ij and i j .With j indicating
a row index, and k a column index, these may be stored in the matrix

cos sin 0

[ljk ] = [mjk ] = sin


cos 0
0

(1255)

From the tensor representations S = S jk ij ik = S jk ij ik it follows from (1-253) that the referential and material components of the second Piola-Kirchhoff stress tensor are related as
k
S jk = llj lm
S lm

Evaluation of (1-256) provides the following explicit relations

(1256)

1.3 Continuum Mechanics

65

S 11 = cos2 S 11 + sin2 S 22 2 sin cos S 12


S 22 = sin2 S 11 + cos2 S 22 + 2 sin cos S 12

S 12 = sin cos S 11 sin cos S 22 + (cos2 sin2 ) S 12


S

13

= cos S

13

sin S

23

S 23 = sin S 13 + cos S 23
(2-257) may be written in the following matrix form

(1257)

S = T S () S

(1258)

where S and S are column matrices storing the non-trivial stress components as indicated in
(1-248), and the transformation matrix T S () is given as

T S () =

cos2
sin2
2 sin cos
0
0
sin2
cos2
2 sin cos
0
0

2
2
sin cos sin cos cos sin
0
0

0
0
0
cos sin
0
0
0
sin
cos

(1259)

j k
Similarly, the dual tensor representations E = Ejk ij ik = Ejk
i i provides the following
analog relation between the referential and material components of the Green tensor

Ejk = mlj mm
k Elm

(1260)

Explicit evaluation of (1-260) provides the following relation between the non-trivial strain
measures on matrix form
E = T E () E

(1261)

where E and E are column matrices storing the non-trivial stress components a

T E () =

cos2
sin2
sin cos
0
0
sin2
cos2
sin cos
0
0

2
2
2 sin cos 2 sin cos cos sin
0
0

0
0
0
cos sin
0
0
0
sin
cos

It can easily be shown that T S () and T E () are related as

(1262)

66

Chapter 1 Preliminaries

T 1
() = T TS ()
E

(1263)

Then,
() E = T S () C T TS () E
S = T S () S = T S () C E = T S () C T 1
E

(1264)

From (1-264) is concluded that the components of the elasticity tensor in the two coordinate
systems are related by the following relation on matrix form
C() = T S () C T TS ()

(1265)

C() has the structure

C 1111 C 1122 C 1112


0
0
C 1122 C 2222 C 2212
0
0

1112

2212
1212
C() = C
C
C
0
0

0
0
0
C 1313 C 1323
0
0
0
C 1323 C 2323

(1266)

Although not indicated, the components in (1-266) can easily be explicitly calculated from (1252) and (1-259).

1.4 Exercises

67

1.4 Exercises
1.1 Given the following vectors a, b, c and d. Prove the following vector identities
(a.) a (b c) = (a c) b (a b) c
(b.) (a b) (c d) = (a c)(b d) (a d)(b c)
(c.) (a b) c = (b c) a = (c a) b
1.2 Prove the e relation (1-19).
1.3 Given the vectors a, b, c and d with the Cartesian components




1
3
1
 j
 j
 j
 j  3

a
= 2 ,
b = 2 ,
c = 2 ,
d = 2
3
1
3
3
Calculate the dyadic scalar products ab cd and ab cd.

1.4 Prove the relations (1-37) and (1-40).


1.5 Given the Cartesian components C rstu of the elasticity tensor C. Calculate the mixed
curvilinear components Cj kl m .
1.6 Prove (1-81) by applying (1-77) to the mixed representation g = jk gj gk of the identity
tensor.
b)

a)
1

x3

x2
a
a

1
1

x1

Fig. 118 a) Parameter domain. b) Shell surface.

1.7 The mid-surface of a shell structure is defined by the parametric description


x1
a 1

2
x = a 2
x3
2f 1 2

where the parameter domain has been indicated on Fig. 1-18a. a and f are parameters
defining the length and the height of the shell structure as shown on Fig. 1-18b.

68

Chapter 1 Preliminaries

(a.) Calculate the covariant and contravariant base vectors gj and gj in terms of the Cartesian base vectors.
(b.) Calculate the contravariant components of the surface identity tensor and the curvature tensor, and determine principal curvature, the mean curvature, the Gaussian
curvature and the principal curvature directions.
(c.) Formulate Codazzis equations for the surface in terms of the selected curvilinear
coordinates.
1.8 Prove the last three relations in (1-74).
1.9 Derive (1-217) from (1-213).

C HAPTER 2
Shell and Plate Theories
The history of plate theories started with Jacob II Bernoulli, who formulated a grid model of
overlapping beams, where the torsional mode was neglected. The differential equations of a
plate were derived independently by Lagrange, Poisson and Navier. However, their theories
all had flaws with respect to the elastic constants and the boundary conditions. These problems were finally solved by Kirchhoff, (Kirchhoff, 1850), for which reason he is considered the
founder of modern plate theory. It took almost a century before E. Reissner, (Reissner, 1944)
and (Reissner, 1945), and Mindlin (Mindlin, 1951), extended the model to include transverse
shear deformations, in much the same way as Timoshenko included shear deformation in the
Bernoulli-Euler beam theory.
Dealing with the eigenvibrations of a bell, Euler in 1764 developed the first mathematical model
to shell theory by analyzing the axisymmetric bell structure as independent ring beams, ignoring
the torsional and meridional bending effects. Again, a century passed before a consistent theory
was derived by Love, (Love, 1888). It is based on Kirchhoffs method, and thus are referred
to as the Kirchhoff-Love model. The inclusion of transverse shear deformations, although proposed by Naghdi, (Naghdi, 1972), is today mostly referred to as the Reisnner-Mindlin model,
in recognition of their analog contributions to plate theory.
The list of names contributing to plate and shell theory in the twentieth century is very long.
Hardly any other area of structural mechanics has produced so many publications, which is
partly an indication of the complexity of the involved mechanics, and partly of the practical
relevance of the subject. While the initial interest in plate and shell theory grew from the need
of solutions to problems in civil and architectural engineering, the focus has gradually changed
to a wider scale of applications in mechanical engineering, aerospace and vehicle industry and
biomechanics, which has been the driving force for further developments. Subjects as large
displacements (e.g. v. Karman plate theory), finite rotations, buckling, composites, anisotropic
materials, material nonlinearities and higher order plate and shell theories have been investigated. Central in these developments has been the concurrent development of numerical methods based on the finite element methods. In turn the numerical modeling has also changed the
standard mechanical model for plate and shell elements from a Kirchhoff-Love to a ReissnerMindlin kinematical formulation, motivated by the reduced compatibility requirements of the
element shape functions.

69

70

Chapter 2 Shell and Plate Theories

When applying a finite element method to plates and shell it is important to realize that errors
on the numerical predictions stem from two sources, namely modeling errors intrinsic to the
applied plate or shell theory, and discretization errors caused by the numerical analysis of this
theory by a finite-degrees-of-freedom model.
Modeling errors are caused by certain approximations applied to the three-dimensional continuum in terms of specific kinematic constraint on the motion of the mass particles, resulting
in Love-Kirchhoff, Reissner-Mindlin or even higher order theories. Additionally, decisions on
the inclusion or exclusion of material and geometrical non-linearities, of anisotropy and nonhomogeneity will influence the qualities of the theory. While the discretization errors can be
removed by mesh refinement, the modeling errors always persist.
When developing and applying a finite element formulation the aim is to achieve a certain accuracy of the numerical prediction relative to the theoretical model, using a minimum number
of degrees of freedom. This rises a number of questions, which have to be answered in advance. Should one start from a genuine shell theory, or should the plate and shell elements be
derived as degenerated continuum elements. Should the two-dimensional elements be derived
from a variational formulation of the a genuine plate or shell theory, or should the dimensional
reduction be achieved via degenerated three-dimensional continuum elements. Further, which
simplifications in the formulation are admissible, and in which applications.
The aim of the following outline is to provide an overview of the mathematical formulation of
the theory for thin-walled based on the finite element formulation. It is not the intention to go
in into depth with these theories, which consequently will be given in a somewhat descriptive
manner. However, we do need to define some fundamental concepts precisely.
Since, non-linear plates turn into shells after the first load increment has been applied the focus
is from the start on shells.
The formulations aim at the largest possible generality. Both thin and thick shells are covered. The shell structure may be layered (sandwich or laminar structures), anisotropic or nonhomogeneous.
Only static problems will be covered. Further, the non-linear behavior is assumed to be of
geometric nature. The extension to dynamic problems or material non-linearities are possible without fundamental difficulties. In the first case a mass and damping matrix need to be
included. In the last case the incremental elastic stiffness matrix need to be modified.

2.1 Plates and Shells

71

2.1 Plates and Shells

Fig. 21 Bending and membrane deformation modes of axis-symmetric shells.

The structural behavior of the shell is characterized by two principal different modes of deformations, namely membrane and bending deformations. Fig. 2-1 shows some illustrative
examples for axis-symmetric shells, which have been chosen merely because the actions are
easily illustrated in this case. The bending deformation typically causes an ovalization of the
shell. Correspondingly, the cross-section is exposed to a linearly varying normal stress over the
section with the zero line placed at the midsurface. In contrast, membrane dominated deformations induce normal stresses in the section, which are approximately constant over the thickness.
Obviously, the membrane state is the desirable mode of deformation. If this can be established,
the shell may carry very large external loads compared to its self-weight.
If the external load varies significantly, or the supports are inappropriately designed, bending
deformations may arise. These deformations and related stresses are considerably larger than
those caused by the same loads in case of membrane action. if a membrane state

72

Chapter 2 Shell and Plate Theories

Fig. 22 Strain energy fractions in bending for clamped hemisphere shell

Fig. 23 xxxxx

2.2 Exercises

2.2 Exercises
2.1 Given the following vectors a, b, c and d. Prove the following vector identities
(a.) a (b c) = (a c) b (a b) c
(b.) (a b) (c d) = (a c)(b d) (a d)(b c)
(c.) (a b) c = (b c) a = (c a) b

73

Index

principal curvature direction, 39

double contraction, 13, 17, 53


dual vector base, 10
dyad, 13

acceleration vector, 51
anchor length, 60
angle strain, 63
arc length coordinates, 8, 9, 48
azimuth angle, 7, 28

e-delta relation, 11, 67


elasticity tensor, 17, 18, 56
Eulerian coordinates, 50
Eulerian small strain tensor, 51, 53
external load per unit mass, 51
external virtual work, 53

base unit vectors, 9, 49, 64


Cartesian tensor components, 14
Cartesian vector components, 11
Cauchy stress tensor, 51
Cauchys boundary condition, 52
characteristic equation, 39
Christoffel symbol, 19
Codazzis equations, 45
composite material, 59
contraction, 13
contravariant base vector, 10, 12
contravariant tensor components, 14
contravariant vector base, 10
contravariant vector components, 11
covariant base vector, 9, 12, 29
covariant derivative, 21
covariant tensor components, 14
covariant vector base, 9
covariant vector components, 11
current configuration, 49
curvature, 36
curvature center, 36
curvature tensor, 38
curvature vector, 36
curvilinear coordinates, 8, 59

first fundamental form, 32


flat space, 25
fourth order tensor, 17
Frenets formula, 36
fundamental tensor, 16
Gausss equation, 38, 45
Gausss formula, 41
Gaussian curvature, 40, 49
geodesic curvature, 37
gradient of scalar, 19
gradient of tensor, 22
gradient of vector, 19, 43
Green small strain tensor, 51
Green strain tensor, 17, 50
hyperelastic material, 56
identity tensor, 15, 18, 23, 30
inertial load, 53
internal virtual work, 53
inverse tensor, 16, 50
isotropic material, 57
Jacobian, 8, 50

dAlemberts principle, 53
discretization error, 70
displacement, 50

kinematical boundary condition, 52


Kroneckers delta, 10, 11, 15, 31, 51

74

INDEX

Lagrangian coordinates, 49
Lame parameters, 57
lamina, 60
laminate, 60
material coordinate system, 61
material coordinates, 49
material deformation gradient, 50
matrix material, 59
mean curvature, 40, 49
mechanical boundary condition, 52
metric tensor, 16
midsurface, 58
mixed covariant and contravariant component,
14
modeling error, 70

75

surface covariant derivative, 33


surface identity tensor, 32, 46
surface metric tensor, 32
surface Riemann-Christoffel tensor, 34, 38
surface second order tensor, 31, 33
surface traction, 52
surface vector, 30
symmetric tensor, 15

normal curvature, 37, 39

tensor base, 14
tensor product, 13
tetrad, 13
thickness, 58
thickness locking, 64
thickness Poisson locking, 64
third order tensor, 22, 33
transposed tensor, 15, 50
triad, 13

orthotropic material, 57
outer product, 13

unit normal vector, 30, 41


unit tangent vector, 9, 35

permutation symbol, 11
physical linear material, 17
ply, 60
Poissons ratio, 57, 62
polyad, 13
principal curvature, 39, 47
principal curvature coordinate system, 40, 45,
47
principal normal vector, 36
principle of virtual displacement, 52, 53
pseudo surface traction, 54

vector, 19

radial distance, 7
radius of curvature, 35, 36, 45, 47
Rayleigh quotient, 38
referential configuration, 49
regular point, 8
reinforcement material, 59
Riemann-Christoffel tensor, 23, 24, 27, 34, 47
second fundamental form, 38
second order tensor, 12, 19
second Piola-Kirchhoff stress tensor, 17, 55
shear modulus, 57
singular point, 8
spherical coordinate system, 7, 25
summation convention, 9
surface contravariant derivative, 33

Winckler foundation, 60
Youngs modulus, 57, 62
zenith angle, 7, 28

76

INDEX

Bibliography

Kirchhoff, G. (1850). Uber das gleichgewicht und die bewegung einer elastischen scheibe. J. Reine
Angew. Math., 40:5158.
Love, A. (1888). On the small vibrations and deformations of thin elastic shells. Philos. Trans. R. Soc.,
179:491.
Malvern, L. (1969). Introduction to the Mechanics of a Continous Medium. Printice-Hall.
Mindlin, R. (1951). Influence of rotatory inertia and shear flexural motions of isotropic elastic plates. J.
Appl. Mech., 18:3138.
Naghdi, P. (1972). The Theory of Shells. In: Handbuch der Physik, Vol. VI/2, Flugge, S. (ed.). Springer
Verlag.
Nielsen, S. (2006). Structural Dynamics, Vol. 1. Linear Structural Dynamics, 4th Ed. Aalborg Tekniske
Universitetsforlag.
Reddy, J. (2004). Mechanics of Laminated Composite Plates and Shells. Theory and Analysis. CRC
Press.
Reissner, E. (1944). On the theory of bending of elastic plates. J. Math. Phys., 23:184191.
Reissner, E. (1945). The effect of transverse shear deformation on the bending of elastic plates. J. Appl.
Mech., 12:6976.
Spain, B. (1965). Tensor Calculus. Oliver and Boyd.
Synge, J. and Schild, A. (1966). Tensor Calculus. University of Toronto Press.
Zill, D. and Cullen, M. (2005). Differential Equations with Boundary-Value Problems, 6th Ed. BrooksCole.

77

You might also like