UNIVERSITY OF CALIFORNIA AT BERKELEY

DEPARTMENT OF MECHANICAL ENGINEERING

ME 185

P. M. NAGHDI’S NOTES ON CONTINUUM MECHANICS

1994
 UNIVERSITY OF CALIFORNIA
Department of Mechanical Engineering

ME185

Aiiiiiiiiiiiiiiiiiiiiiii
i List of General References

1. i
iiiiiiiiiiiiii
Cartesian Tensors, by Sir Harold Jeffreys, Cambridge University
Press, 1952 [lst ed., 1931, 7th impression 1969].

2. Schaum’s Outline Series - Theory and Problems of Continuum

Mechanics, by G. E. Mase, McGraw-Hill Book Co., 1970.

3. iiiiiiiii, by E. A. Fox, Harper & Row, 1967.

Chapters 1-7 of Mechanics

4. i
iiiiiiiiiiiiiiiiii
Continuum Mechanics, by Y. C. Fung, Prentice-Hall Inc., 1969.

5. i
iiiiiiiiiiiiiiiiii
Continuum Mechanics, by P. G. Hodge, Jr., McGraw-Hill, 1970.

6. iiiiiiiiiiiiiiiiiii
Mechanics of Continua, by A. C. Eringen, John Wiley & Sons,
Inc., 1967.
7. i
iiiiiiiiiiiiiiiiii
Continuum Mechanics, by P. Chadwick, G. Allen & Unwin
Ltd., London, 1976.

8. i
iiiiiiiiiiiiiiiiii
Continuum Mechanics, by A.J.M. Spencer, Longman Group
Ltd., London, 1980.
9. i
iiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiii
An Introduction to the Theory of Elasticity, by R. J. Atkin and
N. Fox, Longman Inc., New York, 1980.
 Table of Contents

I. Kinematics .......................................................................................................................... 1
1. Bodies, configurations, motions, mass, and mass density. ..................................... 1
2. Deformation gradient. Measure of strain. Rotation and stretch. ............................ 13
3. Further developments of kinematical results. ......................................................... 20
4. Further developments of kinematical results. ......................................................... 20
5. Velocity gradient. Rate of deformation. Vorticity. ................................................. 21
6. Superposed rigid body motions. ............................................................................. 25
7. Infinitesimal deformation and infinitesimal strain measures. ................................. 35
8. The transport theorem. ............................................................................................ 41
II. Conservation Laws and Some Related Results ................................................................. 44
1. Conservation of mass. .............................................................................................. 45
Appendix to Sec. 1 of Part II. .................................................................................. 47
2. Forces and couples. Euler’s laws. .......................................................................... 49
3. Further consideration of the stress vector. Existence of the stress tensor and
its
relationship to the stress vector. ............................................................................. 51
4. Derivation of spatial (or Eulerian) form of the equations of motion. ..................... 58
5. Derivation of the equations of motion in referential (or Lagrangian form). .......... 61
6. Invariance under superposed rigid body motions. .................................................. 64
7. The principle of balance of energy. ........................................................................ 71
III. Examples of Constitutive Equations and Applications. Inviscid and Viscous Fluid
and Nonlinear and Linear Elasticity. .......................................................................... 74
1. Inviscid fluid. .......................................................................................................... 75
2. Viscous fluid. .......................................................................................................... 79
3. Elastic solids: Nonlinear constitutive equations. .................................................... 83
4. Elastic solids: Linear constitutive equations. ......................................................... 86
5. Some steady flows of viscous fluids. ...................................................................... 91
6. Some unsteady flow problems. ............................................................................... 96
7. Further constitutive results in linear elasticity and illustrative examples. .............. 98
8. Saint-Venant torsion of a circular cylinder. ............................................................ 102
Appendix to Part III (Secs. 5-8). .................................................................................. 104
Supplements to the Main Text ................................................................................................ 110
Appendix L: Some results from linear algebra ...................................................................... 150
 ME185

Part I: Kinematics
1. Bodies, configurations, motions, mass, and mass density.

A body is a set of particles. Sometimes in the literature a body B is referred to as a mani-

fold of particles. Particles of a body B will be designated as X (see Fig. 1.1) and may be

identified by any convenient system of labels, such as for example a set of colors. In mechanics

the body is assumed to be smooth and, by assumption, can be put into correspondence with a

domain of Euclidean space. Thus, by assumption, a particle X can be put into a one-to-one

correspondence with the triples of real numbers X1,X2,X3 in a region of Euclidean 3-space E3.

The mapping from the body manifold onto the domain of E3 is assumed to be one-to-one, inver-
tible, and differentiable as many times as desired; and for most purposes two or three times

suffice.

Each part (or subset) S of the body B at each instant of time is assumed to be endowed with

a positive measure M(S), i.e., a real number > 0, called the mass of the part S and the whole body

is assumed to be endowed with a nonnegative measure M(B) called the mass of the body. We

return to a further consideration of mass later in this section.

Bodies are seen only in their configurations, i.e., the regions of E3 they occupy at each
instant of time t (−∞ < t < +∞). These configurations should not be confused with the bodies

themselves.

Consider a configuration of the body B at time t in which configuration B occupies a region

R in E3 bounded by a closed surface ∂R (Fig. 1.2). Let x be the position vector of the place
occupied by a typical particle X at time t. Since the body can be mapped smoothly onto a

domain of E3, we write

h
x = χ(X,t) . (1.1)
h
In (1.1), X refers to the particle, t is the time, x [the value of the function χ] is the place occupied
h
by the particle X at time t and the mapping function χ is assumed to be differentiable as many

times as desired both with respect to X and t. Also, for each t, (1.1) is assumed to be invertible

so that
 -2-

h
X = χ−1 (x,t) , (1.2)

where the symbol χ−1 designates the inverse mapping. The mapping (1.1) is called a motion of
the body B. The description of motion in (1.1) is similar to that in particle mechanics and is

traditionally referred to as the material description.

During a motion of the body B, a particle X (by occupying successive points in E3) moves
through space and describes a path C ; the equation of this path is parametrically represented by

(1.1). The rate of change of place with time as X traverses C is called the velocity and is tangent

to the curve C. Similarly, the second rate of change of place with time as X traverses C is the

acceleration. Thus, in the material description of the motion, the velocity v and acceleration a

are defined by

h h
. ∂χ(X,t) .. ∂2χ(X,t)
v = x = hhhhhhh , a = x = h
hhhhhhh . (1.3)
∂t ∂t2
h
In the above formulae, the particle velocity v is the partial derivative of the function χ with

respect to t holding X fixed; and, similarly, the particle acceleration a is the second partial
h
derivative of χ with respect to t holding X fixed. Also, the superposed dot, which designates the

material time differentiation, can be utilized in conjunction with any function f and signifies dif-

ferentiation with respect to t holding X fixed.

Reference configuration. Often it is convenient to select one particular configuration and

refer everything concerning the body and its motion to this configuration. The reference

configuration need not necessarily be a configuration that is actually occupied by the body in any

of its motions. In particular, the reference configuration need not be the initial configuration of

the body.

Let κR be a reference configuration of B in which X is the position vector of the place occu-

pied by the particle X (Fig. 1.3). Then, the mapping from X to the place X in the configuration

κR may be written as
 -3-

X = κR (X) (1.4)

which specifies the position vector X occupied by the particle X in the reference configuration

κR. The mapping (1.4) is assumed to be differentiable as many times as desired and invertible.

The inverse mapping is

−1
X = κ (X) . (1.5)
R

Then, the mapping (1.1) at time t from the place x to the particle X may be expressed in terms of

(1.5), i.e.,

h −1 h
x = χ [κ (X),t] = χκR (X,t) . (1.6)
R

h
The second of (1.6), which involves the function χκR (with a subscript κR, emphasizes that the

motion described in this manner represents a sequence of mappings of the reference

configuration κR. In the future, however, for simplicity’s sake we omit the subscript κR and

write (1.6)2 and its inverse as

x = χ(X,t) , X = χ−1(x,t) (1.7)

with the understanding that X in the argument (1.6) is the position vector of X in the reference
h
configuration κR. Returning to (1.6)2, we observe that for each κR a different function χκR

results; and the choice of reference configuration, similarly to the choice of coordinates, is arbi-

trary and is introduced for convenience.

A necessary and sufficient condition for the invertibility of the mappings of (1.7)1,2 is that

the determinant J of the transformation from X to x be nonzero, i.e.,

∂χ ∂χi
J = det( hhh ) = det( h hhh ) ≠ 0 . (1.8)
∂X ∂XA
 -4-

Description of the motion. There are several methods of describing the motion of a body.

Here, we describe three but note that, because of our smoothness assumptions, they are all

equivalent.

It was noted earlier that the description utilized in (1.1) is the material description. In such

a description one deals with abstract particles X which together with time t are the independent

variables.

A description such as (1.7) in which the position X of X at some time, e.g., t = 0, is used as

a label for the particle X is called the referential description. In the referential description,

which is also known as Lagrangian, the independent variables are X and t. For some purposes,

it is convenient to display and to utilize (1.7) in terms of its rectangular Cartesian coordinates.

Thus, let EK be constant orthonormal basis vectors associated with the reference configuration

and similarly denote by ek constant orthonormal basis vectors associated with the present

configuration at time t. Then, the positions X and x referred to EK and ek, respectively, are

X = XK EK , x = xk ek (1.9)

where XK are the rectangular Cartesian coordinates of the position X and similarly xk are the

rectangular Cartesian coordinates of the position x. Referred to the basis ek, the component

form of (1.7)1 can be displayed as

xk = χk(XK,t) , (1.10)

where without loss in generality (since EK are constant orthonormal basis) we have also replaced

the argument X by its components XK. In the future we often display the various formulae both

in their "coordinate-free" forms, as well as their component forms.

The particle velocity and acceleration in a material description of motion are given, respec-

tively, by (1.3)1,2. The corresponding expressions for velocity and acceleration in the referential

description (1.7) will have the same forms as (1.3)1,2 but with the argument X replaced by X:
 -5-

. ∂χ .. ∂2 χ
v = x = hhh (X,t) , a = x = hhhh (X,t) . (1.11)
∂t ∂t2

So far we have introduced two descriptions of motion. The third is called the spatial

description in which attention is centered in the present configuration, i.e., the region of space

occupied by the body at the present time t. In the spatial description, which is also known as

Eulerian, the independent variables are the place x and the time t. To elaborate further, consider

an entity f defined by

f = f̂(X,t) . (1.12)

Since (1.1) is invertible in the form (1.2), the function f̂(X,t) can be expressed as a different

function of x,t, i.e.,

h
f̂(X,t) = f̂(χ−1(x,t),t) = f̃(x,t) . (1.13)

Moreover, the function f̃ is unique. It is clear that with the use of the transformation of the form

(1.13), the velocity and acceleration in (1.3)1,2 can be expressed in terms of different functions of

x,t:

v = v̂(X,t) = ṽ(x,t) = ṽk ek , a = â(X,t) = ã(x,t) = ãk ek . (1.14)

In the above spatial description, the spatial form f̃(x,t) on the right-hand side of (1.13) was

obtained from a representation (1.12) in which f̂ is a function of particle X and t. Analogous

results hold if we start with f = f̂(X,t) and then use (1.7)2 to obtain

f = f̂(X,t) = f̃(x,t) . (1.15)

 -6-

Material derivative. Consider again a function such as f̂(X,t) in (1.12). The partial
.
derivative of this function with respect to t holding the variable X fixed, denoted by f, is called

the material derivative of f. But f, with the use of (1.13)2, can also be expressed as a different

function of x,t. Thus, using the chain rule for differentiation, we have

h
. ∂f̂ ∂f̃ ∂χ ∂f̃
f = hhh c = hhh . hhh + hhh , (1.16)
∂t ∂x ∂t ∂t
cX

where on the left-hand side of the above we have temporarily emphasized that the variable X is

held fixed when calculating the partial derivative of f̂ with respect to t. With the use of the first

of (1.3)1, we may rewrite (1.16)2 as

. ∂f̂(X,t) ∂f̃ ∂f̃ ∂f̃ ∂f̃

f = h hhhhh c = hhh + hhh . ṽ = hhh + ṽk hhhh (1.17)
∂t ∂t ∂x ∂t ∂xk
cX

which is the material derivative of f̃(x,t). The results of the type (1.15) and (1.17) are applicable

to any scalar-valued, vector-valued or tensor-valued field. In particular, consider the referential

form of the velocity vector v which by (1.11)1 and transformation of the type (1.15) is

v = v̂(X,t) = ṽ(x,t) . (1.18)

Then, the acceleration or the material derivative of v where these are regarded as functions of x,t

are

. ∂ṽ ∂ṽ
a = v = hhh + hhh ṽ = ã(x,t) . (1.19)
∂t ∂x

Material curve, material surface and material volume. Let a body B with material

points (or particles) X in a fixed reference configuration κR occupy a region R R in E3 bounded

by a closed surface ∂R R. Any subset SR of B in E3 will be designated by PR (⊆ R R) bounded

by a closed boundary surface ∂PR. Because of the nature of the smoothness assumption
 -7-

imposed on the mapping (1.7)1 from the reference configuration κR to the current configuration κ

of B at time t, all curves, surfaces and volumes in κR are carried by the motion χ into curves, sur-

faces and volumes in the configuration κ. Thus, the region R R with the closed boundary surface

∂R R is mapped into a region R with corresponding closed boundary surface ∂R in κ. More-

over, the parts PR (⊆ R R) bounded by ∂PR is mapped into a part P (⊆ R) bounded by a closed

surface ∂P. The part P is occupied by the same material points (or particles) as those which

occupied PR and such a region or volume is called a material volume.

A surface in the reference configuration κR is defined by equations of the form

F(X) = 0 or X = X(U1,U2) , (1.20)

where U1,U2 are parametric variables on the surface. With the use of (1.7)2, the surface (1.20)

can be mapped into a corresponding surface

f(x,t) = F(χ−1(x,t)) = 0 or x(U1,U2,t) = χ[X(U1,U2),t] (1.21)

in the configuration κ at time t. The surface (1.21) is called a material surface, since its material

points (or particles) are the same as those of the surface (1.20).

A curve in the reference configuration may be regarded as the intersection of two surfaces

of the form

F(X) = 0 , G(X) = 0 , (1.22)

and is defined by

X = X(U) , (1.23)

where U is a parameter and can be identified with the arc length. The intersection of the surfaces

(1.22) can be mapped into a curve in the configuration κ which is the intersection of the surfaces

f(x,t) = F(χ−1(x,t)) = 0 , g(x,t) = G(χ−1(x,t)) = 0 (1.24)

 -8-

or

x(U,t) = χ[X(U),t] . (1.25)

The curve (1.25) is called a material curve, since its material points (particles) are the same as

those of the curve (1.23).

Lagrange’s criterion for a material surface. A surface f(x,t) = 0 is said to be material if

and only if the material derivative of f vanishes, i.e.,

. ∂f ∂f
f = hhh + hhh . v = 0 . (1.26)
∂t ∂x

First we write f(x,t) = F(X,t) with the use of (1.7)1 and suppose that f(x,t) is material.
. .
Then, F(X,t) must be independent of t and hence F = 0 and this implies that f = 0.
. .
Next, suppose that f = 0. Then, it follows that F = 0 and this implies that F must be

independent of t and a function of X only so that F(X) = 0 and after using (1.7)2 we have the

result that F(χ−1(x,t)) = f(x,t) = 0 is material.

In a similar manner, we may establish that the intersection of surfaces f(x,t) = 0 and

g(x,t) = 0 is a material curve if and only if

. ∂f ∂f . ∂g ∂g
f = hhh + hhh . v = 0 , g = hhh + hhh . v = 0 . (1.27)
∂t ∂x ∂t ∂x

Mass density. Before closing this section, consider again a configuration κ of the body B

at time t in which B occupies a region of E3 R bounded by a closed surface ∂R and recall that a
part (or a subset) of B will be denoted by S. The part S in a neighborhood of the place x in κ

occupies a region of space P (⊆ R) bounded by a closed surface ∂P. We introduce explicitly the

mass M(S) of the part and the mass M(B) of the whole body at time t. These measures are non-

negative and may be functions of t. Assuming that the measure M(S) is absolutely continuous,
 -9-

then the limit

M(St)
ρ = lim hhhhh (1.28)
v→0 v

exists, where v = v(P) is the volume of the region of space P. The scalar field ρ = ρ(x,t) is

called the mass density of the body in the configuration at time t. The mass density is nonnega-

tive and is a function of both x and t. The mass of the part S of the body and the mass of the

whole body, both at time t, can be expressed in terms of the mass density ρ:

M(St) = ∫ ρ dv , M(Bt) = ∫ ρ dv , (1.29)

P R
where the subscript t attached to S and B emphasizes that the left-hand sides of (1.29)1,2 refer to

the masses of S and B in the configuration κ at time t and dv is an element of volume in this

configuration, the regions of integration being over P and R in E3. Similarly, we define the mass
density ρ in a reference configuration and write
R

M(SR) = ∫ ρR dVR , M(BR) = ∫ ρR dVR , (1.30)

PR RR
where dVR denotes an element of volume in the reference configuration and PR,R R (PR ⊆ R R)

are the regions of space occupied by the part and the whole body in the reference configuration.

If the reference configuration is identified with the initial configuration at time to, then instead of

(1.30) we write

M(So) = ∫ ρo dV , M(Bo) = ∫ ρo dV , (1.31)

Po Ro
where ρo and dV are, respectively, the mass density and the volume element in the initial

configuration while Po and R o (Po ⊆ R o) are the regions of space occupied, respectively, by the

part and the whole body in the initial configuation.

It should be emphasized that, at this stage in our development, the mass of a part of the

body and that of the whole body depend on the particular configuration occupied by the body.
 - 10 -

2. Deformation gradient. Measures of strain. Rotation and stretch.

Henceforth we identify the particle (material point) X by its position vector X in a fixed

reference configuration, which (for convenience) we take to be coincident with the initial

configuration κo. The deformation gradient (also called displacement gradient) F relative to the

reference position is defined by

∂χ(X,t)
F = F(X,t) = hhhhhhh , J = det F ≠ 0 . (2.1)
∂X

The components of F are designated by FiA or sometimes written as xi,A with a comma designat-

ing partial differentiation. They are given by

∂χi(XB,t)
FiA = hhhhhhhh (2.2)
∂XA

and may be regarded as a 3×3 square matrix.

At a fixed time t, line elements dx are related to line elements dX in κo by

dx = F dX or dxi = xi,A dXA . (2.3)

The deformation gradient F in (2.3) describes the local deformation of a particle (material point)

whose position vector is X in κo. In other words, (2.3) is a linear transformation of a small

neighborhood of X from the reference configuration κo into the current configuration κ at time t.

Let the magnitudes of dX and dx be denoted by dS and ds, respectively; and let M and m

be the unit vectors in the directions of dX and dx, respectively. Then,

dX = MdS , M.M = 1 ,
(2.4)
dx = m ds , m.m = 1 ,

or in component forms
 - 11 -

dXA = MA dS , MA MA = 1 ,
(2.5)
dxi = mi ds , mi mi = 1 .

Measures of strain.

In general, the material line element dX undergoes both stretch and rotation. The ratio ds/dS ,

denoted by λ, is called the stretch of the line element, i.e.,

ds
λ = hhh . (2.6)
dS

It then follows from (2.3)-(2.5) that

m ds = dx = F dX = F M dS ,
(2.7)
or mi ds = dxi = xi,A dXA = xi,A MA dS .

Next, using (2.6), from (2.7) we have

λm = F M or λmi = xi,A MA . (2.8)

By taking the scalar product of each side of (2.8) with itself we obtain

λ2 = F M . F M = M . C M , (2.9)

where in writing (2.9)3 we have also defined the tensor C by

C = FT F . (2.10)

The component forms of the equations (2.9)-(2.10) are:

λ2 = CAB MA MB , (2.9a)

and
 - 12 -

CAB = xi,A xi,B . (2.10a)

The symmetric tensor C (with components CAB) is called the right Cauchy-Green tensor.

Clearly, for a material line element which coincides with dX in the reference configuration

κo, the value of the stretch λ in the direction M can be calculated from (2.9) once C is known.

Alternatively, λ can be calculated in terms of the deformed line element dx in the current

configuration κ. To see this, since F is invertible, we rewrite (2.3) as

F−1 dx = dX or XA,i dxi = dXA , (2.11)

where the inverse function F−1 using (1.7)2 is defined by

−1
−1 ∂χ
∂χ
F−1 = hhhhh
A
or XA,i = hhhhh . (2.12)
∂x ∂xi

Then, from (2.4), (2.5), (2.11) and (2.12) we have

M dS = dX = F−1 dx = F−1 m ds ,
(2.13)
or MA dS = dXA = XA,i dxi = XA,i mi ds ,

which with the use of (2.6) can be written in the form

λ−1 M = F−1 m or λ−1 MA = XA,i mi . (2.14)

The result (2.14) can also be obtained immediately by inverting (2.8). By taking the scalar pro-

duct of each side of (2.14) with itself we obtain

λ−2 = F−1 m . F−1 m = m . (F−1)T F−1 m = m . B−1 m , (2.15)

where in writing (2.15)3 we have also defined the tensor B by

 - 13 -

B = F FT . (2.16)

The component forms of equations (2.15)-(2.16) are:

λ−2 = cij mi mi , (2.15a)

−1
cij = b = XA,i XA,j , bij = xi,A xj,A , (2.16a)
ij

where the notation cij designates the components of the inverse of the tensor B. The symmetric

tensor B whose components are given in (2.16a) is called the left Cauchy-Green tensor. The ten-

sors C and B represent, respectively, the referential (or Lagrangian) and the spatial (or Eulerian)

descriptions of strain. When the motion is rigid, both tensors become identity tensors, i.e.,

C = B = I, where I is the unit tensor.

For some purposes, it is convenient to employ relative measures of strain such that these

measures vanish when the motion is rigid. Recalling (2.9) and (2.15) we may write

λ2 − 1 = M . (C − I)M = (CAB − δAB)MA MB

(2.17)
−1
1 − λ−2 = m . (I − B−1) m = (δij − b ) mi mj
ij

or equivalently,

λ2 − 1 = 2M . E M = 2EAB MA MB ,
(2.18)
h
−2
1 − λ = 2m . E m = 2eij mi mj ,

where

1 1
E = hh (C − I) or EAB = hh (CAB − δAB) , (2.19)
2 2

h 1 1
E = hh ( I − B−1) or eij = hh (δij − cij) . (2.20)
2 2
 - 14 -

The relative measure E defined by (2.19) is known as the relative Lagrangian strain, and the
h
corresponding relative spatial measure of strain E is defined by (2.20).

Rotation and stretch.

The deformation gradient F in (2.3) describes the local deformation of a material line element at

X from the reference configuration κo to the current configuration κ; it involves, in general, both

rotation and stretch. Now, since by (1.8) the deformation gradient F is nonsingular, using the

polar decomposition theorem it may be expressed in the polar forms

F = R U = V R or xi,A = RiB UBA = Vij RjA , (2.21)

where U and V are positive definite symmetric tensors called, respectively, the right and left

stretch tensors and R is the proper orthogonal tensor satisfying

R RT = RT R = I det R = 1 , (2.22)

or in component form

RiA RiB = δAB or RiA RjA = δij . (2.22a)

The effect of (2.21) is to replace the linear transformation (2.3) by two linear transformations:
Either by

dX′ = U dX and dx = R dX′ or dXB′ = UBA dXA and dxi = RiB dXB′ , (2.23)

or by

dx′ = R dX and dx = V dx′ or dxj′ = RjA dXA and dxi = Vij dxj′ . (2.24)

Associated with each of the stretch tensors U, V are three positive principal values (eigen-

values) and three orthogonal principal directions (eigenvectors). These will be discussed later.
 - 15 -

Here, it should be emphasized that, in general, F, R, U, V are all functions of X and t and they

can vary from one material point to another during motion. In the remainder of this section, we

provide physical interpretations for the decompositions (2.23) and (2.24).

(i) Stretch followed by pure rotation. In this case, let the deformation of dX into dX′

involve only pure stretch and the deformation from dX′ into dx be one of pure rotation. By

(2.21)1 and (2.23), as well as (2.9)-(2.10), we have

dx . dx = R dX′ . R dX′ = dX′ . dX′ , (2.25a)

or

dxi dxi = RiA RiB dXA′ dXB′ = δAB dXA′ dXB′ = dXA′ dXA′ , (2.25b)

and

C = FT F = (R U)T (R U) = U2 , (2.26)

or

CAB = xi,A xi,B = RiK UKA RiL ULB = δKL UKA ULB = UKA UKB , (2.27a)

λ2 = UKA UKB MA MB = CAB MA MB . (2.27b)

According to (2.25), the length of the line element dX′ is the same as the length of dx so that all

the stretching is represented by U in (2.26). However, line elements are generally also rotated

by the action of U. The part of the deformation described by R is a pure rotation.

(ii) Pure rotation followed by stretch. In this case, let the deformation of dX into dx′ be

one of pure rotation and the deformation from dx′ to dx involve only stretch and rotation. By

(2.21) and (2.24), as well as (2.15)-(2.16), we have

dx′ . dx′ = R dX . R dX = RT R dX . dX = dX . dX , (2.28)

 - 16 -

or

dxi′ dxi′ = RiA RiB dXA dXB = δAB dXA dXB = dXA dXA (2.29)

and

B = F FT = (V R) (V R)T = V2 , (2.30a)

1
hhh = m . B−1 m , (2.30b)
λ2

or

bij = xi,A xj,A = Vir RrA Vjs RsA = Vir Vjr (2.31a)

λ−2 = (B−1)ij mi mj . (2.31b)

According to (2.28), the length of the line element dx′ is the same as the length of dX so that the

deformation described by R is one of pure rotation.

Keeping in mind the decomposition (2.21), it is evident from (2.26) and (2.30a) that

U = (FT F) ⁄2 and V = (F FT) ⁄2 involve the square root operation which often is difficult to exe-
1 1

cute and this is the main reason for the choice of U2 = C and V2 = B as the strain tensors.

Moreover, since U2 is symmetric positive definite, it possesses a unique symmetric positive

definite square root; a parallel remark applies to the square root of V2. In this regard the expres-

sions for the square root of the stretch λ and its inverse, in the forms (2.9) and (2.15), are espe-

cially noteworthy.
 - 17 -

Secs. 3 & 4. Further developments of kinematical results.

 - 18 -

5. Velocity gradient. Rate of deformation. Vorticity.

Consider the spatial (Eulerian) form of the particle velocity in the form given by (1.14),

v = ṽ(x,t) and define the spatial velocity gradient tensor by

L = grad ṽ(x,t) (5.1)

or

∂vi
h∂v
hh = v e × e , vi,j = hhh , (5.2)
i,j i b j
∂x ∂xj

where vi,j are the components of L referred to the orthonormal basis ei and in recording (5.2) we

have used the same symbol for the function v and its value. The tensor L can be uniquely

decomposed into its symmetric part D and its skew-symmetric part W (see Appendix L, Sect. 8)

as follows:

L = D + W , (5.3)
where

1 1
D = hh (L + LT) = dij ei b× ej , W = hh (L − LT) = wij ei b× ej . (5.4)
2 2

The corresponding component forms of (5.3) and (5.4) are:

vi,j = dij + wij , (5.3a)

where

1 1
dij = hh (vi,j + vj,i) = dji , wij = hh (vi,j − vj,i) = − wji . (5.4a)
2 2

The tensor D with components dij is called the rate of deformation tensor and W with com-

ponents wij is called the vorticity or the spin tensor.

There is a one-to-one correspondence between a skew tensor and an axial vector (see

Appendix L, Sect. 12). The axial vector w which corresponds to the skew vorticity tensor W in

(5.3) or (5.3a) is defined by

1
Wz = w×z or w = hh curl v (5.5)
2
 - 19 -

for every vector z. Referred to the basis ei, w reads as

w = wi ei (5.6)

and the corresponding component forms of (5.5)1,2 are:

1 1
wij = − εijk wk = εjik wk , wi = hh εijk wkj = hh εijk vk,j , (5.5a)
2 2

where εijk stand for the components of the permutation symbol. Motion in which w = 0 is
1
called irrotational motion. It is usual in some books to identify curl v (instead of hh curl v in
2
(5.5)2) as vorticity and denote it by w = curl v.

The foregoing results have been obtained starting from (5.1), where v = v(x,t). Alterna-

tively, we may begin by recalling the referential (Lagrangian) form of the particle velocity in the

form given by (1.14)1, where v = v(X,t). Then, the material derivative of the deformation gra-

dient (2.1) is
.
F = LF , (5.7)

or in component form
.
hhh .
xi,A = xi,A = vi,A = vi,k xk,A . (5.7a)

Additional useful results may be obtained from (5.7) or (5.7a). Thus, the material derivative of
the Jacobian of transformation (1.8) is

. .
hhhh
J = det F = J tr (L F F−1) = J tr L

= J tr D = J vk,k = J dkk (5.8)

or
.
J = J XA,i vi,k xk,A = J δik vi,k = J vk,k . (5.8a)

A motion which is volume-preserving, i.e., a motion corresponding to which the volume occu-
pied by any material region remains unchanged, is called isochoric. For an isochoric motion,
 - 20 -

.
J = 1 , J = 0 => tr D = dkk = div v = vk,k = 0 . (5.9)

Next, consider the material derivative of the tensor C defined by (2.10):

. .
hhhh . .
C = FTF = FT F + FT F = FT( LT + L)F = 2FT D F , (5.10)

or in component form

. .
hhhhhh .
hhh .
hhh
CAB = xi,A xi,B = xi,A xi,B + xi,A xi,B

= vi,k xk,A xi,B + vi,k xk,B xi,A

= (vi,k + vk,i)xk,A xi,B = 2dik xk,A xi,B (5.10a)

In order to discuss a physical interpretation of the tensors D and W (or equivalently the

axial vector w), we take the material derivative of (2.8) with respect to t and make use of (5.7) to

obtain

.
h hh . . .
hhhh
λm = λm + λm = FM = LFM = λLm , (5.11)

or in component form

.
hhhh . . .
hhhhhh
λ mi = λ mi + λ mi = xi,A MA = vi,k xk,A MA = λ vi,k mk . (5.11a)

The scalar product of (5.11) with the unit vector m, after also using the results m . m = 1,
.
m . m = 0, yields

. .
λ hhhh
hh = l n λ = L m . m = (D + W) . m b× m = D . m b× m , (5.12)
λ

or
. .
λ hhhh
hh = l n λ = vi,k mi mk = (dik + wik)mi mk = dik mi mk . (5.12a)
λ
 - 21 -

It follows from (5.12) that the rate of deformation D determines the material time derivative of

the logarithmic stretch for a material line element having the direction m in the current

cofiguration κ. Further, using (5.12), the result (5.11)2 may be rewritten as

. .
. λ λ
m = L m − hh m = W m + [D − ( hh ) I]m , (5.13)
λ λ
or from (5.11a)
. .
. λ λ
mi = vi,j mj − ( hh )mi = wij mj + [dij − ( hh )δij ] mj . (5.13a)
λ λ

Now, let m be a principal direction of the eigenvectors associated with the rate of deformation

tensor D. Then,

Dm = γm or dij mj = γ mi (5.14)

where the associated scalar eigenvalue γ is

.
λ
γ = D m . m = D . m b× m = dij mi mj = hh (by (5.12a)) (5.15)
λ

where we have also used (5.12). It follows from (5.14) and (5.15) that
. .
λ λ
D m = hh m or dij mj = hh mi . (5.16)
λ λ

Moreover, from combination of (5.13) and (5.16) we deduce that

. .
m = Wm = w×m or mi = wij mj = εikj wk mj . (5.17)

Thus, the material time derivative of a unit vector m (which represents the direction of a

material line in κ) in a principal direction of D is determined by (5.17). In other words, the axial

vector w (or the vorticity) is the angular velocity of the line element which is parallel to a princi-

pal direction.

(For further results, see the Supplement to Part I, Section 5.)

 - 22 -

6. Superposed rigid body motions.

A rigid motion is one that preserves relative distance. In this section, we examine the effect

of such a motion on the various kinematic measures introduced in preceding sections. More

specifically, we consider here motions which differ from a prescribed motion, such as (1.7)1,

only by superposed rigid body motions of the whole body, i.e., motions which in addition to a

prescribed motion involve purely rigid motions of the body. For later reference, we recall here

the motion defined in Eq. (1.7)1, namely

x = χ (X, t) .

Consider a material point (or particle) of the body, which in the present configuration at

time t occupies the place x as specified by (1.7)1. Suppose that under a superposed rigid body

motion, the particle which is at x at time t moves to a place x+ at time t+ = t + a, a being a con-
stant. In what follows for all quantities associated with the superposed motion we use the same

symbols as those associated with the motion (1.7)1 to which we also attach a plus "+" sign.

Thus, we introduce a vector function χ+ and write

x+ = χ̂+ (X, t+) = χ+ (X, t) . (6.1)

It is clear that the difference between the two functions χ̂+ and χ+ is due to the presence of the
constant a in the argument of the former.

Similarly, consider another material point of the body which in the present configuration at

time t occupies the place y specified by

y = χ (Y, t) . (6.2)

Suppose that under the same superposed rigid body motion, the particle which is at y at time t

moves to a place y+ at time t+. Then, corresponding to (6.1), we write

y+ = χ̂+ (Y, t+) = χ+ (Y, t) (6.3)

Recalling the inverse relationship X = χ−1(x, t) and the analogous result for Y = χ−1(y, t), the

function χ+ on the right-hand sides (6.1) and (6.3) may be expressed as different functions of x, t
 - 23 -

and y, t, respectively, i.e.,

h h
x+ = χ+(X, t) = χ+(x, t) , y+ = χ+(Y, t) = χ+(y, t) . (6.4)

During the superposed rigid body motions of the whole body, the magnitude of the relative dis-
h h
placement χ+(x, t) - χ+(y, t) must remain unaltered for all pairs of material points X, Y in the

body and for all t in some finite time interval [t1, t2]. Hence,
h h h h
[χ+(x, t) − χ+(y, t)] . [χ+(x, t) − χ+(y, t)] = (x − y) . (x − y) (6.5)

for all x, y in the region R occupied by the body at time t.

Since x, y are independent, we may differentiate (6.5) successively with respect to x and y
h
to obtain the following differential equation for χ+:
h T h
R ∂χ+(x, t) H R ∂χ+(y, t) H
J hhhhhhhh J J hhhhhhhh J = I , (6.6)
Q ∂x P Q ∂y P
h
where x, y are any points in R and t is any time in the interval [t1, t2]. Since the tensor ∂χ+(y,

t)/∂y is invertible, (6.6) can be written in the alternative form

h T h −1
R ∂χ+(x, t) H R ∂χ+(y, t) H
J hhhhhhhh J = J hhhhhhhh J = QT(t) (6.7)
Q ∂x P Q ∂y P

for all x, y ⊆ R and t in [t1, t2]. Thus, each side of the equation (6.7) is a tensor function of t, say

QT(t), and we may set

h
∂χ+(x, t)
hhhhhhhh = Q(t) for x ∈ R . (6.8)
∂x
h
∂χ+(y, t)
Therefore we also have hhhhhhhh = Q(t). >From (6.6) or (6.7), we conclude that
∂y

QTQ = I , det Q = ± 1 , (6.9)

and hence Q is an orthogonal tensor. The motion under consideration must include the particu-
h
lar case χ+(x, t) = x and for this particular case Q = I and det Q = +1. Since the motions are con-

tinuous, we must iiiiii have

always
 - 24 -

det Q = 1 (6.10)

and therefore Q(t) is a proper orthogonal tensor.

The differential equation (6.8) may be integrated to yield

h
χ+(x, t) = a(t) + Q(t)x , (6.11)
h
where a(t) is a vector function of time. The last result is a general solution of (6.5) for χ+(x, t).

For later convenience, we put

a(t) = c+(t+) − Q(t)c(t) , (6.12)

where c+, c are vector functions of t+ and t, respectively. With the use of (6.12), the (6.11) can

be expressed in the form

x+ = c+ + Q(x − c) , (6.13)

where

QT Q = Q QT = I , det Q = 1 . (6.14)

The transformation (6.13) is a rigid transformation since it is distance preserving, i.e.,

| x+ − y+ | 2 = (x+ − y+) . (x+ − y+) = Q(x − y) . Q(x − y)

= (x − y) . [QTQ(x − y)]
= (x − y) . I(x − y)
= (x − y) . (x − y) = | x − y | 2 . (6.15)

The transformation (6.13) also preserves the angle between any two nonzero vectors x − y and

x − z since

x+ − y+
cosθ+ = hhhhhhh
| x+z+ |
Q(x − y)
= hhhhhhhh . Q(x
|x − y|
(x − y) . [QTQ(x − z)]
= hhhhhhhhhhhhhhhhhhh
|x − y| |x − z|
(x − y) (x − z)
= hhhhhhh . hhhhhhh = cosθ , (6.16)
|x − y| |x − z|
 - 25 -

where z is any place in the configuration occupied by the body at time t which moves to z+ at

time t+ as a consequence of the rigid body motion. Since (6.13) is both length and angle preserv-

ing, it follows that the element of length, element of area and element of volume all remain unal-

tered under superposed rigid body motions.

The component forms of the various results between (6.5) and (6.10) proceed as follows.

First, the component form of (6.5) is

h h h h
[χi+(xj, t) − χi+(yj, t)][χi+(xj, t) − χi+(yj, t)] (6.5a)

To record the results (6.6) in component form, we differentiate (6.5a) successively with respect

to xm and yn and obtain

h
∂χi+(xj t)
hhhhhhhh , t)] = δ (x − y )
∂xmj im i i

and
h h
∂χi+(xj, t) ∂χi(yj, t)
h hhhhhhh hhhhhhhh (6.6a)
∂xm ∂yn

∂yn
Next, multiply (6.6a) by h hhhhhhh
h and use the chain rule for differentiation, i.e.,
+
∂χk (yj, t)
h
∂χi+(yj, t) n ∂y
h hhhhhhh h hhhhhhh
h
∂yn
∂χk+(yj, t)

to deduce
h
∂χk+(xj, t) m ∂y
h hhhhhhh = h hhhhhhh . (6.7a)
h
∂xm
∂χk+(yj, t)

Since the left-hand side of (6.7a) is independent of yj and the right-hand side is independent of

xj, each side must be a function of time only and we have

h
∂χk+(xj, t)
h hhhhhhh = Q (t) . (6.8a)
∂xm km
 - 26 -

Integration of the differential equation (6.8a) yields

h
xk+ = χk+(xj, t) = ak(t) (6.11a)

Since xj is any point, we must also conclude from (6.11a) that

h
yk+ = χk(yj, t) = ak(t) + Qkm(t)ym

and hence
h
∂χk+(yj, t)
h hhhhhhh = Q (t) .
∂yn kn

Substitution of the last result and (6.8a) into (6.6a) results in the component for of

Qkm(t)Qkn(t) = δmn . (6.9a)

Also, for the convenience of the reader, we also record here the component forms of most of the

equations given above. Thus, for example, the component form of (6.1) is

xi+ = χ̂i+(XA, t+) = χi+(XA,t) . (6.1a)

Similarly, we identify the component form of (6.2), (6.3), etc., as

yi = χi(YA, t) , (6.2a)

yi+ = χ̂i+(YA, t+) = χi+(YA, t) , (6.3a)

h h
xi+ = χi+(XA, t) = χi+(xj, t) , yi+ = χi+(YA, t) = χi+(yj, t) , (6.4a)

h h h h
[χi+(xj, t) − χi+(yj, t)][χi+(xj, t) − χi+(yj, t)] = (xi − yi)(xi − yi) , (6.5a)

h h
∂χi+(xj, t) ∂χi+(yj, t)
h hhhhhhh h hhhhhhh = δ
mn , (6.6a)
∂xm ∂yn

h
∂χi+(xj, t)
h hhhhhhh = Q (t) , (6.8a)
∂xk ik

Qim(t)Qin(t) = δmn , (6.9a)

 - 27 -

h
χi+(xj, t) = ai(t) + Qik(t)xk , (6.11a)

ai(t) = ci+(t+) − Qik(t)ck(t) , (6.12a)

xi+ = ci+ + Qij(xj − cj) , (6.13a)

Qim Qin = δmn = Qmi Qni , (6.14a)

(xi+ − yi+)(xi+ − yi+)k)

= δjk(xj − yj)(xk − yk) = (xj − yj)(xj − yj) , (6.15a)

| x+ − y+ | | x+ − z+ | cosθ+i+)
= Qim(xm − ym)Qin(xn − zn)
= δmn(xm − ym)(xn − zn)
= (xm − ym)(xm − zm)
= | x − y | | x − z | cosθ . (6.17a)

Recall
h
x+ = χ̂+(X, t+) = χ+(X, t) = χ+(x, t) ,
or
h
x+ = χ+(x, t+) = a(t) + Q(t)x . (6.11)

Then,

. + hdx
+
+ ∂χ̂+(X, t+)
hhh
v =x = + = hhhhhhhhh
dt ∂t+

∂χ+(X, t)dt
= hhhhhhhh hhh
∂t dt+

∂χ+(X, t)
= hhhhhhhh (6.18)
∂t

Therefore

. d
v+ = x+ = hhh (a(t) + Q(t)x)
dt+
 - 28 -

d dt
= h h (a(t) + Q(t)x) hhh
dt dt+

. . .
= a(t) + Q(t)x + Q(t)x

. .
= a(t) + Q(t)x + Q(t)v . (6.19)

Let
.
Ω (t) = Q(t)QT(t) . (6.20)

Then
. .
Ω Q = Q QT Q = Q . (6.21)

But, from (6.9) QT Q = I, so that

. .
QT(t)Q(t) + QT(t)Q(t) = O ,

[Ω (t)Q(t)]TQ(t) + QT(t)Ω(t)Q(t) = O ,

QT(t)ΩT(t)Q(t) + QT(t)Ω(t)Q(t) = O ,

ΩT(t) + Ω(t) = O

or

ΩT(t) = −Ω(t) . (6.22)

Since Ω (t) is skew-symmetric, it posseses an axial vector

1
ωi = − hh εijk Ωjk , (6.23)
2

or

Ωjk = −εjki ωi .

Alternatively, since Ω is skew-symmetric, there exists a vector-valued function ω such that for

any vector V,
 - 29 -

ΩV=ω×V . (6.24)

Returing to (6.19), we may write

.
v+ = a + Ω Q x + Q v . (6.25)
 - 30 -

iiiiiiiii rigid body motions. In order to relate our results to

So far we have been discussing superposed

the more familiar equations of rigid body dynamics, let us now consider a rigid motion defined

by

x=X , (6.26)

x+ = a(t) + Q(t)x= a(t) + Q(t)X . (6.27)

The equation (6.27)1 may be solved for x in the form

x = QT(t)(x+ − a(t)) . (6.28)

Then
. .
v+ = a(t) + Q(t)x

.
= a(t) + Ω Q(t)[QT(t)(x+

.
= a(t) + Ω (t)[x+ − a(t)]

.
= a(t) + ω (t) × [x+ − a(t)] . (6.29)

Thus ω (t) is recognized to be the angular velocity of the body.

The component form of the foregoing equations are listed below:

xi+ = ai(t) + Qij(t)xj , (6.11a)

. . . .
vi+ = xi+ = ai(t) + Qij(t)xj + Qij(t)xj , (6.19a)

.
Ωik(t) = Qim(t)Qkm(t) , (6.20a)

. .
ΩikQkl = QimQkmQkl = Qil , (6.21a)

Recalling

. .
Qil Qjl + Qil Qjl = 0 ,
 - 31 -

we obtain

Ωij(t) = −Ωji(t) , (6.22a)

ΩimVm = −εimp ωp Vm = εipm ωp Vm , (6.24a)

.
vi+ = ai + Ωim Qmn xn + Qim vm , (6.25a)

xi+ = ai(t) + Qim xm , (6.27a)

Qij xi+ = Qij ai + Qij Qim xm

Qij(xi+ − ai) = δjm xm = xj , (6.28a)

.
vi+ = ai + Ωij(xj+ − aj)

.
= ai + εijk ωj(xk+ − ak) . (6.29a)

We return now to the more general considerations of superposed rigid body motions.

>From (6.25) and (6.11), it follows that

.
v+ = a + Ω Q x + Q v

.
= a + Ω Q[QT(x+ − a)] + Q v

.
= a + Ω (x+ − a) + Q v , (6.30)

or in component form
.
vi+ = ai + Ωij[xj+ − aj] + Qijvj . (6.30a)

Then,
 - 32 -

∂v +
i ∂x +
j j ∂v ∂x
n
hhhh = Ωij hhhh + Qij hhhh hhhh
+
∂xm +
∂xm ∂xn ∂xm
+

∂vj ∂
= Ωij δjm + Qij hhhh hhhh [Q (x + − al(t))]
∂xn ∂xm+ ln l

∂v
j ∂x +
l
= Ωim + Qij hhhh Qln hhhh
∂xn +
∂xm

∂vj
= Ωim + Qij hhhh Qln δlm
∂xn

∂vj
= Ωim + Qij hhhh Qmn
∂xn

∂vj
= Ωim + Qij Qmn hhhh , (6.31a)
∂xn

or in direct notation

+
hhh = Ω + Q h∂v
h∂v hh QT . (6.31)
∂x+ ∂x

It follows from (6.31) and the results of section 5 that

D+ = Q D QT ,
W+ = Q W Q T + Ω , (6.32)

or, in indicial notation,

dij+ = Qim Qjn dmn ,

wij+ = Qim Qjn wmn + Ωij . (6.32a)

In particular, suppose that the body in its superposed motion is at time t passing through the

configuration κ with angular velocity Ω, so that

.
Q=I , Q=ΩQ=Ω . (6.33)
 - 33 -

Then (6.32)

D+ = D , W+ = W + Ω . (6.34)
 - 34 -

7. Infinitesimal deformation and infinitesimal strain measures.

We recall from section 1 that a motion of the body is defined by

x = χ(X, t) or xi = χi(XA, t) . (7.1)

The deformation function χ in (7.1) may be written in terms of the relative displacement u (see

section 4), i.e.,

χ = X + u(X, t) or χi = δiA XA + ui(XA, t) . (7.2)

We further recall the relative strain measures

1 1
E = hh (C − I) or EAB = hh (CAB − δAB) , (7.3)
2 2

and

1 1
e = hh (I − c) or eij = hh (δij − cij) , (7.4)
2 2

where

C = FT F or CAB = FiA FiB (7.5)

and
−1 −1
c = (F−1)TF−1 = (F FT)−1 = B−1 or cij = FAi FAj . (7.6)

Also, the relative displacement gradient is defined by

H = F − I or ui,A = FiA − δiA . (7.7)

In order to obtain infinitesimal kinematical results from those of the finite theory discussed

iiiiiiii by a nonnegative function

in sections 1-5, we introduce a measure of smallness

ε = ε(t) = max{ sup | u,K | } (K = 1, 2, 3) , (7.8)

K X∈ Ro

where sup stands for the supremum (or the least upper bound) of a nonempty bounded set of real

numbers. If f(u,K) is any scalar-, vector-, or tensor-valued function of u,K = {u,1, u,2, u,3}

defined in a neighborhood of u,K = 0 (K = 1, 2, 3) and satisfying the condition that there exists a
 - 35 -

nonnegative real constant C such that | f(u,K) | < Cεn as ε → 0, then we write f = 0(εn) as ε → 0.

Thus, in particular, the components of u,K referred to either ei or EA are of 0(ε) as ε → 0, i.e.,

ui,A = 0(ε) and uB,A = 0(ε) as ε → 0 . (7.9)

The components EAB of the relative strain measure E in (7.3) can also be expressed in

terms of relative displacement gradients (7.9)2 in the form

1
EAB = hh (uA,B + uB,A + uM,AuM,B) . (7.10)
2

Clearly, if terms of 0(ε2) as ε → 0 can be neglected, we can write (7.10) approximately as

1
EAB = hh (uA,B + uB,A) = 0(ε) as ε → 0 . (7.11)
2

Next, consider the expression

∂uA
FiA(δAj − hhhh δBj) (7.12)
∂XB

∂ui
and substitute for FiA = δiA + h hhh from (7.7)2. Thus,
∂XA

∂ui ∂uA
Expression (7.12) = (δiA + h hhh )(δAj − hhhh δBj)
∂XA ∂XB

∂ui ∂ui
= δij + h hhh δAj − hhhh δBj + 0(ε2)
∂XA ∂XB

= δij + 0(ε2)

= δij , (7.13)

where in writing (7.13)4, terms of 0(ε2) as ε → 0 have been neglected. Hence, to the order ε2 as

ε → 0, we can identify the coefficient of FiA in (7.12) as the inverse of FiA, i.e.,

−1 ∂XA
FAi = hhhhh . (7.14)
∂xiBi
 - 36 -

With the use of (7.14) and the chain rule of differentiation, it can be readily verified that

∂uA A∂u B ∂XB ∂u

hhhh = hhhh hhhh hhhh δCi)
∂xi ∂XB ∂xi ∂XC

∂uA
= hhhh δBi + 0(ε2) as ε → 0 (7.15)
∂XB

and similarly

∂u ∂u ∂X ∂u
hhhi = h hhh
i h hhh
A hhhh
A
δBj)
∂xj ∂XA ∂xj ∂XB

∂ui
= h hhh δAj + 0(ε2) as ε → 0 . (7.16)
∂XA

It follows from (7.15) and (7.16) that to 0(ε2), it is immaterial whether the partial derivatives of
∂u1 ∂u1
the displacement field u is taken with respect to xi or XA so that hhhh = hhhh to 0(ε2) as ε → 0.
∂x1 ∂X1

Hence in dealing with infinitesimal kinematics, it is not necessary to distinguish between

Eulerian and Lagrangian form of kinematical quantities.

>From (7.6)2 and (7.14), we have

−1 −1 ∂uA ∂uA
cij = FAi FAj = (δAi − hhhh δBi)(δAj − hhhh δCj)
∂XB ∂XC

∂uA ∂uA
= δij − ( hhhh δBiδAj + hhhh δCj δAi) + 0(ε2) as ε → 0 . (7.17)
∂XB ∂XC

Hence, if terms of 0(ε2) as ε → 0 can be neglected, then

1
eij = hh (δij − cij)
2

1 ∂uA ∂uB
= hh ( hhhh + h hhh )δiAδjBε → 0 , (7.18)
2 ∂XB ∂XA

so that any differences between the two strain measures disappear upon linearization.
 - 37 -

In view of the remark made following (7.15), to the order of ε2 we may display the com-

ponents of the relative displacement gradients either as uA,B or ui,j. For example, if we make the

latter choice, then we may write

ui,j = eij + ωij = 0(ε) ,

eij = 0(ε) , ωij = 0(ε) as ε → 0 , (7.19)

where eij and ωij are defined by

1 1
eij = hh (ui,j + uj,i) , ωij = hh (ui,j − uj,i) . (7.20)
2 2

We observe here that ui,i = eii = 0(ε) is called the cubical dilatation and that ωii = 0.

We consider now the linearized version of the Cauchy-Green measure CAB, the stretch UAB

and the rotation RiA. The Cauchy-Green measure CAB defined by (7.5)2 is related to EAB by

CAB = δAB + 2EAB. But, when the approximation (7.11) is adopted, then CAB can be written as

CAB = UAC UCB = δAB + 2EAB

= δAB + 0(ε) as ε → 0 (7.21)

and from an examination of

(δAC + EAC)(δCB + ECB) = δAB + 2EAB + 0(ε2) as ε → 0 (7.22)

to the order ε2 we can identify UAB as

UAB = δAB + EAB = δAB + 0(ε) as ε → 0 . (7.23)

Also, by considering an expression of the form

−1
UAB(δBCneglected,then.EQI(7.24)UAB = δAB − EAB = δAB + 0(ε) as ε → 0 .

We now turn to the rotation tensor RiA. Recalling the polar decomposition theorem, the

components of the rotation tensor R can be written as

−1
RiA = FiB UBA . (7.25)
 - 38 -

It then follows from (7.7), (7.24) and (7.11) that

1
RiA = (δiB + ui,B)[δBA − hh (uA,B + uB,A)]
2

1
= δiA + hh [ui,A − uA,BδiB]
2

1 ∂ui ∂uA
= δiA + hh ( h hhh − hhhh )
2 ∂XA ∂xi

1 ∂ui ∂uA
= δiA + hh ( hhhh − hhhh ) , (7.26)
2 ∂xA ∂xi

where terms of 0(ε2) as ε → 0 have been neglected and where use has been made of the result
(7.15). To the order of approximation considered, we may also write (7.26) as

R = I + Ω or Rij = δij + ωij , (7.27)

where

1
Ω = −ΩT = hh (H − HT)j,i) , (7.28)
2

which is consistent with (7.20)2.

In the remainder of this section, we employ the linearization procedure discussed above and

obtain the linearized version of such expressions as

J = det F = det(xi,A) (7.29)

and its inverse. For this purpose, we first recall that J can be represented as

1
J = hh εijk εLMN FiL FjM FkN . (7.30)
6

Substituting the components of the displacement gradient in terms of ui,A given by (7.7)2 in

(7.30) we get
 - 39 -

1
J = hh εijk εLMN(δiL + ui,L)(δjM + uj,M)(δk,N + ukN)
6

1
= hh εijk εLMN[δiL δjM δkN + 3δiL δjM uk,N + 0(ε2)]
6

= 1 + δkN uk,N = 1 + uk,k = 1 + ekk as ε → 0 , (7.31)

1
where in obtaining (7.31)3 use has been made of the identities hh = εijk εijk and εijk εijm = 2δkm
6
and where terms of 0(ε2) as ε → 0 have been neglected. The inverse of (7.31) is given by

J−1 = [1 + eii + . . . ]−1

= 1 − e + (e )2 + . . .
ii ii
= 1 − eii as ε → 0 (7.32)

and again in writing (7.32) terms of 0(ε2) as ε → 0 have been neglected.

(For a discussion of the interpretation of the infinitesimal strain measures, see Supplement

to Part I, Section 7.)

 - 40 -

8. The transport theorem.

Let S be an arbitrary part (or subset) of the body B and suppose that S occupies a region Po,

with a closed boundary surface ∂Po, in a fixed reference configuration. Similarly let P, with

closed boundary surface ∂P, be the region occupied by S in the configuration at time t.

Let φ be any scalar-valued or tensor-valued field with the following representations:

φ = φ̃(x, t) = φ̃(χ(X, t), t) (8.1)

and consider the volume integral

I = ∫ φ̃(x, t)dv = ∫ φ̂(X, t)J dV , (8.2)

P Po

where we have used dv = J dV, J = det F > 0. Often we shall encounter an expression of the type
(8.2) and we need to calculate its time derivative dI/dt. Thus, we write

hdI
hh = hdh φ̂(X, t)J dV
dt ∫
dt
Po

d
= ∫ hdth (φ̂(X, t)J)dV
Po

∂φ̂ .
= ∫ [ hhh (X, t)J + φ̂(X, t)J]dV
∂t
Po

∂φ̂
= ∫ [ h∂t
hh (X, t)J + Jv φ̂(X, t)]dV
i,i
Po

∂φ̂(X, t)
= ∫ [ hhhhhhh
∂t
+ φ̂(X, t)vi,i]JdV
Po

.
= ∫ [φ + φ̂(X, t)div v]JdV
Po

.
= ∫[φ + φ̃(x, t)div (8.3)
P
 - 41 -

.
where in the fourth of (8.3) we have also used J = J vi,i and where

. ∂φ̂(X, t) ∂φ̃(x, t) ∂φ̃(x, t)

φ = hhhhhhh = hhhhhhh + hhhhhhh vi . (8.4)
∂t ∂t ∂xi

It then follows that the time derivative of (8.2)1 is given by

.
hdh
dt ∫ φ̃(x, t)dv = ∫ [φ + φ̃(x, t)div v]dv . (8.5)
P P

>From the result (8.5) follows the various expressions given below:

hdh ∂φ̃(x, t) ∂φ̃(x, t)

dt ∫ φ̃(x, t)dv = ∫ [ hhhhhhh
∂t
+ hhhhhhh vi
∂xi
P P

∂φ̃(x, t) ∂(φ̃(x, t)vi)

= ∫ [ hhhhhhh + hhhhhhhhhh ]dv
∂t ∂xi
P

∂φ̃(x, t)
= ∫ hhhhhhh dv + ∫ φ̃(x, t)vinida
∂t
P ∂P

∂φ̃(x, t)
= ∫ hhhhhhh dv + ∫ φ̃(x, t)v . n da , (8.6)
∂t
P ∂P
h
iiii spatial region P in the
where the divergence theorem has been used. Next, apply (8.6)4 to a fixed

configuration at time t and write

hdh φ̃(x, t)dv = ∂φ̃(x, t)

∫
dt h ∫h hhhhhhh
∂t
dv + ∫ φ̃(x, t)v . n da
h
P P ∂P

∂
= h h ∫ φ̃(x, t)dv + ∫h φ̃(x,t)v . n da . (8.7)
∂t h
P ∂P

The form (8.7)2 is another statement of the transport theorem. It is used when it is convenient to
h
focus attention on a fixed region of space P at time t and consider the motion of the body over

this region. This form (8.7)2 can also be deduced directly from (8.6)4 without the introduction of
h
the fixed spatial region P by the following line of argument: Since in the calculation of the first
 - 42 -

integral in (8.6)4 the variable x (and hence the spatial region P) is fixed, the operator ∂/∂t com-
h
mutes with the volume integral in (8.6)4 over a region P and the form (8.7)2 with P replaced by P

follows from (8.6)4. The commuting of ∂/∂t with the volume integral in (8.6)4 may be contrasted

with a similar calculation involving the volume integral in (8.3)2. In the latter integral, since X

(and hence the material region Po, is fixed, the operator d/dt commutes with the volume integral

in (8.3)2 over the region Po.

 - 41 -

Part II: Conservation Laws and Some Related Results

 - 42 -

1. Conservation of mass.

We recall from section 1 of Part I that the mass of any part P of the body, i.e., M is a con-

tinuous function of its volume and that there exists a scalar mass density ρ such that

M(St) = ∫ ρ dv , P ⊆ R , (1)
P

where ρ = ρ(x, t) depends on the particular configuration occupying the region of space R and dv

is the element of volume in the present configuration. A statement of the conservation of mass

for any material part of the body in the present configuration is as follows:

hdh
dt ∫ ρ dv = 0 . (2)
P
.
By the transport theorem, (2) can be written as ∫ [ρ + ρvk,k]dv = 0. The last result must hold for
P

ii arbitrary parts P; hence, by the argument outlined previously, assuming that ρ is continuously
all

differentiable we have
.
ρ + ρvk,k = 0 , (3)

. ∂ρ ∂ρ
where a superposed dot denotes material derivative, i.e., ρ = hhh + vk hhhh . The result (3) can
∂t ∂xk

also be expressed as

∂ρ
hhh ∂ρ
+ (ρvk),k = 0 or hhh + div(ρv) = 0 . (4)
∂t ∂t

Equation (3) or (4) represents the local equation for conservation of mass. It is also referred to

iiiiii form of the "continuity equation."

as spatial

Another form of the principle of conservation of mass may be stated as:

∫ ρ dv = ∫ ρo dV , (5)
P Po

where Po is the material part in the initial reference configuration, dV is the element of volume in

the reference configuration and ρo denotes the mass density in the reference configuration.

Using dv = J dV, J = det F = det xi,A, from (5) we get

 - 43 -

ρ = J−1 ρo or ρ = (det F)−1 ρo , ρo = Jρ , (6)

which is the iiiiiii form of continuity equation.

material
 - 44 -

Appendix to Sec. 1 of Part II

Theorem: If φ(x, t) is continuous in R and

∫ φ dv = 0 (1)
P

for every part P ⊆ R, then the necessary and sufficient condition for the validity of (1) is that

φ = 0 in R . (2)

Proof: Let us first recall the definition of continuity.

Definition: A function φ(x, t) is said to be continuous in a region R if for every x ∈ R and every

ε > 0 there exists a δ > 0 such that | φ(x, t) − φ(ox, t) | < ε whenever | x − ox | < δ.

Sufficiency: If φ = 0 in R, then (1) is trivially satisfied.

Necessity: (Proof by contradiction). Suppose that there is a point ox ∈ P (⊆ R) at which

φo = φ(ox, t)>0. Then, by the continuity of φ, there exists a δ > 0 such that

φo
| φ(x, t) − φo | < hhh for all | x − ox | < δ . (3)
2

Now let Pδ be the region | x − ox | < δ and Vδ be the volume of this region so that

Vδ = ∫ dv > 0 . (4)
Pδ

>From (3) and the fact that φo > 0 we have

φo
φ > hhh in Pδ . (5)
2

Now let P be a part which contains Pδ, then

∫ φ dv > ∫ 1⁄2 φo dv = 1⁄2 φo Vδ > 0 . (6)

Pδ Pδ

Since the result (6) contradicts (1), it follows that there can exist no point ox such that φ(ox, t) >
 - 45 -

0. Similarly, we realize that if φo = φ(ox, t) < 0, then by the continuity of φ there exists a δ > 0

such that

φo
| φ(x, t) − φo | < − hhh for all | x − ox | < δ . (7)
2

Thus we can conclude that

φo
φ(x, t) < hhh for all x ∈ Pδ . (8)
2

Hence

∫ φ dv < ∫ 1⁄2 φo dv = 1⁄2 φo Vδ < 0 . (9)

Pδ Pδ

Since (9) contradicts (1), again it follows that no point ox exists for which φ(ox, t) < 0. Combin-

ing this result with that above we conclude that (2) is also a necessary condition for the validity

of (1).
 - 46 -

2. Forces and couples. Euler’s laws.

We admit two types of forces, namely the body force per unit mass b and the surface force

(or contact force) per unit area) as follows:

b = b(X,t) = body force per unit mass,

(1)
t = t(X,t ; n) = contact force per unit area.

It is important to note that the surface force t depends on the orientation of the surface area (with

outward unit normal n) upon which it acts. Similar to the definitions (1) above, we could also

admit body couple per unit mass c and surface couple per unit area m, but these are ruled out in

the construction of classical continuum mechanics.

We also note here the following:

.
momentum per unit mass = v = x
(2)
.
moment of momentum per unit mass = x × v = x × x

We consider now the basic balance laws (also called conservation laws) for momentum and

moment of momentum (also called angular momentum) in the context of purely mechanical

theory. These balance laws, which are known as Euler’s laws, may be stated (in words) as fol-

lows:

IRate of change of momentum M IAll external forces acting M

K NK N ,
for any part of the body on the part
L OL O

(3)
IRate of change of moment of momentum M IMoment of all external forces M
K NK N .
for any part of the body acting on the part
L OL O

The above laws can be translated into mathematical forms both with respect to the current
 - 47 -

configuration (Eulerian form) or with respect to the reference configuration (Lagrangian form).

In this section we limit the discussion to the Eulerian (or spatial) form of the balance laws (3).

Thus, remembering the definitions (2)1,2, the momentum for any part P of the body occupying a

region P with boundary surface ∂P and the moment of momentum for any part P are:

∫vdm , ∫x × vdm , (4)

P P

where dm = ρdv is the element of mass. Also, the total force acting on P and the moment of total

force acting on P are:

∫bdm + ∫ tda , ∫(x × b)dm + ∫ (x × t)da . (5)

P ∂P P ∂P

Keeping the expressions (4) and (5) in mind, the spatial (or Eulerian) form of the balance

laws (3) are:

hdh vdm = bdm +

dt ∫ ∫ ∫ tda , (6)
P P ∂P

hdh x × vdm = x × bdm +

dt ∫ ∫ ∫ x × tda . (7)
P P ∂P

In the next few sections of Part II we exploit the implications of (6) - (7) and also derive their

local forms.
 - 48 -

3. Further consideration of the stress vector. Existence of the stress tensor and its rela-
tionship to the stress vector.

Consider an arbitrary part of the material region of the body B which occupies a part P in

the present configuration at time t. Let P be divided into two regions P1, P2 separated by a sur-

face σ (see Fig. 3.1). Further, let ∂P1, ∂P2 refer to the boundaries of P1, P2, respectively; and let

∂P′, ∂P′′ be the portions of the boundaries of P1, P2 such that

∂P′ = ∂P1 ∩ ∂P ,
∂P′′ = ∂P2 ∩ ∂P . (1)

Thus, a summary of the above description is as follows:

P = P1 ∪ P2 , ∂P = ∂P′ ∪ ∂P′′ ,
∂P1 = ∂P′ ∪ σ , ∂P2 = ∂P′′ ∪ σ . (2)

Now recall the linear momentum principle, i.e.,

hdh .
dt ∫ ρ x dv = ∫ ρ b dv + ∫ t(n)da (3)
P P ∂P

or with the use of dm = ρ dv in the form:

hdh .
dt ∫ x dm = ∫ b dm + ∫ t(n) da (4)
P P ∂P

which holds for an arbitrary material region P ⊆ R. Application of (4) separately to the parts P1,

P2 and again to P1 ∪ P2 = P yields

hdh .
dt ∫ x dm − ∫ b dm − ∫ t(n)da = 0 , (5)
P1 P1 ∂P1

hdh .
dt ∫ x dm − ∫ b dm − ∫ t(n)da = 0 (6)
P2 P2 ∂P2

and

hdh .
dt ∫ x dm − ∫ b dm − ∫ t(n)da = 0 . (7)
P1∪P2 P1∪P2 ∂P′∪∂P′′
 - 49 -

The stress vector t(n) in (5) acting over the boundary ∂P1 results from contact forces exerted

ii the material on one side of the boundary (exterior to P1) ii

by on the material of the other side.

Similar remarks hold for the stress vector in (6) and in (7). We emphasize that the stress vector

in (5) over ∂P′ ∪ σ represents the contact force exerted on P1 across the surface, etc. The

appropriate normals associated with t(n) over the surface σ are equal and opposite in sign. To

elaborate, let n be the outward unit normal at a point on σ when σ is a portion of ∂P1. Then, the

outward unit normal at the same point on σ when σ is a portion of ∂P2 is −n.

>From combination of (5) and (6) and after subtraction from (7), we obtain the following

equation:

∫ [t(n) + t(−n)]da = 0 (8)

σ

over the arbitrary surface σ. Assuming that the stress vector is a continuous function of position
and n, it follows that

t(n) = −t(−n) or t(x , t ; n) = − t(x , t ; −n) . (9)

According to the result (9), the stress vectors acting on opposite sides of the same surface at

a given point are equal in magnitude and opposite in direction. The result (9) is known as
Cauchy’s lemma.

Again we suppose that a body B is mapped into the present configuration κ, at time t, which

occupies the region R. Consider some interior particle Xo of B having position vector ox in R.

Construct at ox a tetrahedron T, lying entirely within R, and in such a way that the side i is per-

pendicular to the ei-direction (see Fig. 3.2) and inclined plane - with outward unit normal on -

falls in the octant where x1 , x2, x3 are all positive. We refer to the side i of T by Si and to the

inclined plane by S, respectively. Let h denote the height of the tetrahedron, i.e., the perpendicu-

lar distance from ox to the inclined face S, and let S denote the area of this face. Then, the areas

of the three orthogonal faces Si are

Si = S on . ei = S oni (10)
 - 50 -

and the volume of the tetrahedron is

1
Vtet = hh h S . (11)
3

Recalling that dm = ρdv, by virtue of the transport theorem and the conservation of mass

(see Sec. 1 of part II), the principle of linear momentum (4) reduces to
..
∫ x dm = ∫ b dm + ∫ t(n)da . (12)
P P ∂P

Next, apply (12) to the material region T and obtain

..
∫ x dm = ∫ b dm + ∫ t(n)da . (13)
T T ∂T

Observing that the surface integral in (13) represents contributions from all four boundary planes
of T, we have

3
∫ t(n)da = Σ ∫ t(−e )da + ∫ t( n)da .
i o
(14)
i=1 S
∂T i S

Now according to Cauchy’s lemma in (9)

t(−ei) = − t(ei) (15)

and then with the use of (14) and (15), the linear momentum equation (13) can be rewritten in

the form
..
∫ (x − b)dm = ∫ t( n)da − ∫ o
t(ei)da . (16)
3
T S
Σ
i=1
Si

Now recall that dm = ρdv and that ρ is already assumed to be bounded (see Sec. 1 of Part I).
..
Further, we assume that the fields x and b are bounded. Then, since (by a theorem of analysis)

| ∫ f dv | ≤ ∫ | f | dv

where | f | denotes the absolute value of f, we obtain the following estimate for the integral on

the left-hand side of (16):

 - 51 -

.. ..
| ∫ ρ(x − b)dv | ≤ ∫ | ρ(x − b) | dv
T T

= ∫ K dv
T

Sh
= K* ∫ dv = K* hhh
3
T

where use has been made of the mean value theorem for integrals, we have set
..
K(x ,t) = | ρ(x − b) | and K* stands for some specific interior value of K in T. Hence, we may

conclude that there exists a fixed, positive real number K* such that

.. Sh
c ∫ ρ(x − b)dv c ≤ K* hhh . (17)
3
T

Next, assume that the stress vector field is continuous in both x and n. Then, by the mean value

theorem for integrals, the two surface integrals in (16) yield:

∫ t(ei)da = ti*Si = ti* S oni (18)

3

Σ
i=1
Si
and

∫ t( n)da = t( n) S ,
*
o o
(19)
S

where in writing (18)2 use has been made of (10) and where ti* and t* stand for some specific

interior values (at the point ox) of the stress vectors on the respective faces Si and on the plane S

of T. Each stress vector ti* acts on that side of coordinate plane i which is associated with the

outward unit normal ei. It follows from (16), (17), (18) and (19) that (since K* is fixed)

h1h K*Sh ≥ | ..
3 ∫ ρ(x(on)da − ∫ t(ei)da |
3
T
Σ
i=1
Si

= | t(*on) S − ti* S oni |

= S | t(*on) − ti* oni | .

 - 52 -

Therefore,

1
| t(*on) − ti* oni | ≤ hh K*h . (20)
3

Consider now a sequence of tetrahedra T1, T2, ..., of diminishing heights h1 > h2 > ..., each

member of which is similar to T with three mutually orthogonal faces having outward unit nor-

mals −ei and an inclined plane with outward unit normal on. Next, apply (20) to each member of

the sequence and, in the limit as h→ 0, obtain

| t(*on) − ti* oni | ≤ 0 , (21)

where the stress vectors in (21) are evaluated at the point ox which is the common vertex of the

family of tetrahedra. It follows from (21) that

t(*on) = ti* oni . (22)

Since (22) must hold at any point ox and corresponding to any direction on, without ambiguity,

we suppress henceforth the designation star, replace ox by x, on by n and write

t(n) = tini . (23)

We now define tki by

tki = ti . ek or ti = tki ek . (24)

Let tk denote the components of the stress vector t(n), i.e., tk = t(n) . ek. Then, from (23) and (24)

we have

tk = t . ek = tini . ek = tkini , (25)

where in (25)1 we have written for simplicity t in place of t(n). Thus, if t = t(x , t ; n) is the stress

vector at the point x acting on a surface whose outward unit normal is n, then it is clear that

tki = tki(x , t) are defined at x and are independent of n. The relation (25) establishes the

iiiiiiii of the set of quantities tki. It remains to show that tki are Cartesian components of a
existence

second order tensor. To this end, consider the transformation of two sets of Cartesian coordi-
 - 53 -

nates and recall

xk′ = ajkxj , ek′ = ajkej , nk′ = ajknj , etc. , (26)

where

aik = ei . ek′ . (27)

From

t = tkek = tk′ek′ = tkiniek = tki′ni′ek′ (28)

and with the use of ni′ = arinr , ek′ = ask es where ari are components of an orthogonal tensor, we

obtain (tsr − ariasktki′)nres . But es are linearly independent. Therefore,

(tsr − ariasktki′)nr = 0 , (29)

which holds for all n and the quantity in parentheses is independent of n. Hence,

tsr = askaritki′ , (30)

which establishes the tensor character of tki. The second order tensor whose Cartesian com-

ponents are defined by (24) is called the Cauchy stress tensor.

With reference to a rectangular Cartesian system of spatial coordinates, consider a paral-

lelepiped shown in Fig. 3.3 and recall the formula

ti = tkiek (31)

In (31), ti is the stress vector acting on the face whose outward unit normal is ei. For example,

on the front face of the parallelepiped in Fig. 3.3, the outward unit normal coincides with e1; and

thus, from (31) with i = 1, the stress vector on this face is given by

t1 = tk1ek = t11e1 + t21e2 + t31e3 .

Keeping the above in mind, an examination of the subscripts of the stress tensor tki in (31) easily

reveals that (i) the second index "i" refers to the stress vector ti on the face whose outward unit

normal is ei, while (ii) the first index "k" refers to the component of ti in the coordinate
 - 54 -

directions. Utilizing this convention, which stems from the definition of (24), the tractions (i.e.,

the forces per unit area) are sketched in Fig. 3.3.

 - 55 -

4. Derivation of spatial (or Eulerian) form of the equations of motion.

We derive in this section the spatial (or Eulerian) form of the equations of motion from the

integral balance laws (6) and (7) of section 2. We begin with the balance of linear momentum

(6) in section 2 of Part II, which for the present purpose can be written as (recall that dm = ρdv):

hdh ρvdv = ρbdv +

dt ∫ ∫ ∫ t(n)da . (1)
P P ∂P

By the transport theorem discussed in Part I, the left-hand side of (1) is given by

.
hhhh
LHS of (1) = ∫[(ρv) + ρvdivv] dv
P
. .
= ∫[ρv + ρv + ρvdivv] dv
P
. .
= ∫{ρv + v[ρ + ρdivv] dv (2)
P

= ∫ρadv ,
P

where in arriving at the results, (2)4, we have also used the local conservation of mass given by

(3) of section 1. Next, we consider the surface integral in (1), substitute the relation (23) of sec-
tion 3, namely t = tini, and use the divergence theorem obtain

∫ tda = ∫ tinida = ∫ti,i dv , (3)

∂P ∂P P

where a comma following a subscript indicates partial differentiation.

After substitution of the results (2) and (3) into (1), the resulting equation can be put in the

form

∫[ti,i + ρb − ρa] dv = 0 , (4)

P
 - 56 -

which must hold for all arbitrary material volumes. Then, since the integrand is a continuous

function, by the usual procedure we arrive at the local form

ti,i + ρb = ρa . (5)

Next, we turn to the balance of moment of momentum (7) in section 2 of Part II, which can

be rewritten as

hdh (x × ρv) dv = (x × ρb) dv +

dt∫ ∫ ∫ x × t(n) da (6)
P P ∂P

Again by the transport theorem, the left-hand side of (6) can be shown to yield

LHS of (6) = ∫x × ρa dv . (7)

P

Similarly, the surface integral in (6) after substitution from (23) of section 3 and the use of the

divergence theorem gives

∫ x × t(n) da = ∫ (x × tini) da = ∫(x × ti),i dv

∂P ∂P P
(8)
= ∫[(x,i × ti) + x × ti,i] dv
P

Introduction of (7) and (8) into (6) results in

∫[(x,i × ti) + x × (ti,i + ρb − ρa)] dv = 0 ,

P

or

∫x,i × ti dv = 0 , (9)
P
 - 57 -

where in obtaining the last result we have also used (5). Since (9) must hold for all arbitrary

material volumes P and since the integrand is continuous, by the usual argument we conclude

that

x,i × ti = 0 ,

or equivalently

e i × ti = 0 . (10)

The two results (5) and (10) represent the consequences of the balance of linear momentum

and moment of momentum. Although the foregoing derivation in vectorial form leading to

equations of motion (5) and the restriction (10) is simple and attractive, it conceals some of the

features of the stress tensor introduced earlier in section 3. Thus, we now recall the relation (24)

for ti = tkiek and substitute this into (5) and (10) to obtain the alternative forms of the equations

of motion and the restriction on ti. With the use of the identities

ti,i = (tkiek),i = tki,iek ,

(11)
ei × ti = ei × tkiek = εikjtkiej ,

the component forms of (6) and (10) result in

tki,i + ρbk = ρak , (12)

tki = tik . (13)

Thus, it is easily seen that the vector equation (5) is equivalent to the three scalar equations of

motion (12). Similarly, the restriction (10) on the three vectors ti implies the symmetry of the

stress tensor tki. This last observation reveals the fact that the three scalar equations of motion

involve only six components of the stress instead of nine.

 - 58 -

5. Derivation of the equations of motion in referential (or Lagrangian) form.

In previous developments, the stress vector t (and therefore the stress tensor tij) acting on P

are measured per unit area of surfaces in the present configuration at time t. In view of the

transformation x = χ (X , t) all surfaces in R can be mapped into corresponding surfaces in Ro

occupied by the body in the reference configuration. For some purposes, it is more convenient to

measure the stress vector and the stress tensor acting on P per unit area of surfaces in Po.

Corresponding to an arbitrary part P with boundary ∂P in the present configuration, we have a

part Po with boundary ∂Po in the reference configuration (see Fig. 1.1). We denote the outward

unit normal to ∂Po by N, where

N = NAeA . (1)

We denote the stress vector acting on ∂P, but measured per unit area of the surface ∂Po in the

reference configuration by p.

In terms of quantities measured in the reference configuration, the momentum principles

(i.e., linear and moment of momentum) are:

∫ ρo a dV = ∫ ρo b dV + ∫ p dA , (2)
Po Po ∂Po

and

∫ (x × ρo a)dV = ∫ (x × ρo b)dV + ∫ x × p dA (3)

Po Po ∂Po

where p = p(X , t ; N) depends on position, time and the unit normal N to ∂Po. By a procedure

similar to that used previously (Part II: Sec. 3), we can prove that

p = NApA , pA = piAei , (4)

p = piA NA ei , (5)

p=PN. (5a)

The vectors pA (A = 1,2,3) represent stress vectors acting across surfaces at a point P in the
 - 59 -

present configuration κ which were originally coordinate planes XA = const. through the

corresponding point Po in the reference configuration and measured per unit area of these planes

(Fig. 5.1). Also, the components piA (i.e., of the stress tensor P) represent surface forces acting

in the present configuration, but measured per unit area of an XA-plane in the reference

configuration and resolved parallel to the ei-directions. Introducing (5) into (2), by a procedure

similar to that used previously (Part II, Sec. 4), we obtain

piA,A + ρobi = ρoai or pA,A + ρob = ρoa , (6)

Then, from (3), (5), and after using (6), we also obtain

piAxj,A = pjAxi,A . (7)

Introducing the notation sAB through

piB = xi,AsAB , (8)

it follows from (7) that

sAB = sBA . (9)

The last two are called, respectively, the symmetric and the nonsymmetric Piola-Kirchhoff

stress.

Also, it can be verified that

JT = P FT = F S FT or Jtik = xi,ApkA = xk,ApiA = xi,Axk,BsAB , (10)

which relates the Cauchy stress T to the stress tensors P and S. To verify the truth of (10), we

may start with

df = t da = p dA (11)

which represents an element of contact force acting on the current configuration of the body B in

terms of both the spatial and referential description of the stress vector. Using the derived rela-

tion between the area elements da and dA, namely n da = J F−T N dA, we have
 - 60 -

df = t da = T n da
(12)
= p dA = P N dA .

Hence (J T F−T − P)N dA = 0 for all N and we may conclude that J T F−T − P = 0 or

J T = P FT . (13)

Recalling (8) or P = F S, we are led to the results (10)1.

The coordinate-free forms of the various results between (6)-(10) may be displayed as fol-
lows:
.
Div P + ρo b = ρo v , (6a)

P FT = F PT , (7a)

P=FS, (8a)

S = ST . (9a)
 - 61 -

6. Invariance under superposed rigid body motions.

It was established previously that a particle X of a body, which in the reference

configuration is at X and at time t in κ occupies the place x = χ(X, t), under superposed rigid
h
motions of the whole body at a different time t+ = t + a occupies in κ+ the place x+ = χ+(X, t+)
specified by

x+ = a + Q x or xi+ = ai + Qijxj . (1)

In (1), a is a vector-valued function of t, Q is a proper orthogonal tensor-valued function of t and

a is a constant. The vector a can be interpreted as a rigid body translation and Q as a rotation

tensor.

We have already studied how various kinematical quantities transform under superposed

rigid body motions and shall presently extend these considerations to the dynamical quantities

which appear in the equations of motion, namely

.
tij,j + ρbi = ρvi , tij = tji . (2)

However, we first need to establish some further kinematical results:

(a) Further kinematical results:

Recall that an element of area dA of a material surface having outward unit normal vector

N in the reference configuration of a body is deformed at time t into the element of area da

whose outward unit normal vector is n. Then,

∂XK
dak = J h hhh dAK , (3)
∂xk

where

dak = da nk , dAK = dA NK . (4)

h
Under the motion χ+ the element of area dA is deformed into da+ and N is deformed into n+ so
that

∂XK
dak+ = J+ h hhh dAK , (5)
∂xk+
 - 62 -

and

dak+ = da+nk+ . (6)

It follows from (1)2 that

∂x +
i ∂x
j
hhhh = Qij hhhh = Qijδjk = Qik (7)
∂xk ∂xk

and consequently

+
∂xi+
FiA = h hhh j (8)
∂XA

i.e.,

F+ = Q F . (9)

Therefore,

J+ = det F+ = det(Q F)

= (det Q)(det F)

= +1 det F (10)

=J .

Also, by the law of conservation of mass,

ρo = ρJ = ρ+J+ , (11)

which together with (10)5 implies

ρ+ = ρ . (12)

Returning again to (1)2, we see that

Qik xi+ = Qik ai + QikQijxj

= Qikai + δkjxj
(13)
= Qikai + xk ,
 - 63 -

xk = Qik(xi+ − ai) or x = QT(x+ − a) . (14)

Therefore

∂x k
hhhh = Qik , (15)
∂xi+

and consequently

∂X K ∂X K j∂x ∂X
K
h hhh = h hhh hhhh = h hhh Qkj (16)
∂xk+ ∂xj ∂xk+ ∂xj

which, in direct notation, reads

(F+)−1 = F−1QT (17)

and could also have been deduced by taking the inverse of both sides of (9).

Substituting the results (10)5 and (16)2 into (5) yields

∂XK
dak+ = J h hhh QkjdAK
∂xj

= Qkjdaj , (18)

where (3) has been used in completing the last step. The combination of (18)2 with (4)1 and (6)

gives

da+nk+ = Qkjda nj . (19)

Now square both sides of (19) and remember that n and n+ are unit vectors and Q is a proper
orthogonal tensor so that

(da+)2nk+ nk+ = (da+)2

= Qkj nj Qkl nl (da)2

= δjl nj nl (da)2

= nj nj (da)2 = (da)2 . (20)

 - 64 -

Since area is always a positive number, it follws from (20)5 that

da+ = da (21)

and hence from (19) that

nk+ = Qkj nj or n+ = Q n . (22)

It is worthwhile noticing that we have been able to deduce the behavior of F, F−1, J, ρ, da
and n under superposed rigid body motions without making any additional physical assumptions

in (a).

(b) The stress vector and the stress tensor.

It is clear from the developments of section 2 that not all kinematical quantities transform

according to formulae of the type

u+ = Q(t)u , U+ = Q(t)U QT(t) (23)

under superposed rigid body motions, where u and U in (23) stand for a vector and a second

order tensor field, respectively. We investigate now the relationships between† t and t+ and

between T and T+. For convenience, we recall the formulae for ti and tij, i.e.,

t = t(X, t; n) , ti = tijnj , (24)

t+ = t+(X, t+; n+) , ti+ (25)

The mechanical fields which enter the linear and moment of momentum principles are the

stress vector and the body force per unit mass. We first consider the former, which in the motion

χ(X, t) is a vector field defined by (24)1. Consider now a second motion which differs from the

first only by superposed rigid body motions specified in the form (1). The second motion

imparts a change in the orientation of the body, so that the outward unit normal vector n to the

same material surface (in the configuration κ) becomes the unit normal vector n+ (in the
configuration κ+). The stress vector in the second motion takes the form indicated by (25)1 and
hhhhhhhhhhhhhhh
† Our development follows Naghdi (1972, pp. 484-486). See also Green and Naghdi (1979).
 - 65 -

we have already seen that the outward unit normal vector n+ transforms according to (22). The

geometrical properties of the transformation (1) were discussed in section 2. Keeping these in

mind and recalling that t is linear in n, it is reasonable to expect for the stress vector t+(X, t; n+)

(i) to have the same magnitude as t(X, t; n) and (ii) to have the same relative orientation to n+ as

t has relative to n. These remarks lead us to introduce the following assumption:

t+ = Q t , ti+ = Qijtj . (26)

We observe that (26) implies that

| t+ | 2 = t+ . t+ = ti+ ti+ = Qij Qik tj tk = δjk tj tk = tj tj = | t | 2 (27)

and, recalling (22), we also have

t+ . n+ = ti+ ni+ = Qij Qik tj nk = δjk tj nk = tj nj = t . n . (28)

Now

t+ . n+ = | t+ | | n+ | cosθ+ = | t+ | cosθ+ (29)

and

t . n = | t | | n | cosθ = | t | cosθ ,

which together with (27)7 and (28)5 lead to

cosθ+ = cosθ . (30)

The results (27) and (30) verify the motivating remarks made prior to (26) to the effect that (i)

the magnitude of t is the same as the magnitude of t+ and (ii) the magnitude of the angle between

t and the outward unit normal n is the same as the magnitude of the angle between t+ and n+.

We now turn attention to the stress tensor. >From the results (24)2, (25)2 and (22) we have

ti+ = tij+ nj+ = tij+ Qjk nk . (31)

But, by assumption (26), we also have

ti+ = Qij tj = Qij tjk nk . (32)

 - 66 -

Combination of (31) and (32) yields

(tij+ Qjk − Qij tjk)nk = 0 . (33)

The last result must hold for all nk and the coefficient of nk is independent of nk. Therefore

tij+ Qjk − Qij tjk = 0 ,

tij+ Qjk Qmk − Qmk Qij tjk = 0 ,

+
tim = Qij Qmk tjk ,

and hence

tij+ = Qim Qjn tmn (34)

or, in direct notation,

T+ = Q T QT . (35)

A scalar, vector or tensor quantity which under superposed rigid body motions transforms as

iiiiiiii. Of course, not all physical quantities are

(12), (22) or (35), respectively, is called objective

objective. For example

. . .
v+ = a + Q x + Q x ,
(36)
+
W =QWQ +Ω T

are clearly not objective.

(c) Body forces and accelerations.

h
The equations of motion for the two motions χ and χ+ of the body are, respectively, given
by

∂tij .
hhh = ρ(vi − bi) ,
∂xj
(37)
∂tij+ .
hhhh = ρ+(vi+ − bi+) .
∂xj+
 - 67 -

But

∂t +
ij ∂ k∂x
hhhh = hhhh (Qim Qjn tmn) hhhh
∂xj+ ∂xk ∂xj+
∂tmn
= Qim Qjn h hhh Qjk
∂xk
∂tmn
= Qim δkn h hhh
∂xk
∂tmn
= Qim h hhh , (38)
∂xn

where (15) has been employed in (38)2. It follows from (38)4 and (37) that
. .
ρ+(vi+ − bi+) = Qim ρ(vm − bm) , (39)

which with the aid of (12) becomes

. .
vi+ − bi+ = Qim (vm − bm) (40)

or
. .
v+ − b+ = Q(v − b) .

References
..
1. Naghdi, P.M. 1972 The theory of shells and plates. In S. Flugge’s Handbuch der Physik,
Vol. VIa/2 (edited by C. Truesdell), Springer-Verlag, 425-640.

2. Green, A.E. & Naghdi, P.M. 1979 A note on invariance under superposed rigid body
motions. J. Elasticity 9, 1-8.
 - 68 -

7. The principle of balance of energy.

We begin by assuming the existence of a scalar potential function per unit mass ε = ε(x,t),

called the specific internal energy. The internal energy for each part P in the present

configuration is defined by the volume integral

∫ ρ ε dv . (1)
P

We introduce a scalar field r = r(x,t) per unit mass per unit time, called the specific heat supply

(or heat absorption), as well as the heat flux across a surface ∂P (with the outward unit normal n)

by the scalar h = h(x,t;n) per unit area per unit time. The integrals

H = ∫ ρ r dv − ∫ h da , (2)
P ∂P

where ∂P is the boundary of P, defines the heat per unit time entering the part P in the present
configuration. The first term on the right-hand side of (2) represents the heat transmitted into P

by radiation and the second term the heat entering P by conduction. We also recall that the

kinetic energy for each part P in the present configuration is defined as

1
∫ h2h ρ v . v dv . (3)
P

We record now the rate of work by the body and contact forces, i.e., R = Rb + Rc, where

Rb = ∫ ρ b . v dv , Rc = ∫ t . v da . (4)
P ∂P

In (4), the scalar b . v which occurs in the volume integral represents the rate of work per unit

mass by the body forces. Similarly, t . v is a scalar representing rate of work per unit area by the

iii of
contact force t. Each of the integrals in (4) is a rate ii work
i iii term and thus has the dimension of

"work per unit time".

With the foregoing background, the law of balance of energy may be stated as follows: The

rate of increase of internal energy plus kinetic energy is equal to the rate of work by body force

and contact force plus energies due to heat per unit time entering the body. Thus, we write
 - 69 -

hdh 1
dt ∫ ρ[ε + h2h v . v] dv = ∫ ρr dv + ∫ ρb . v dv
P P P

+ ∫ t . v da − ∫ h da . (5)
∂P ∂P

We observe that the negative sign in front of the last integral in (5), as well as in (2), is in accord

with the convention that heat h = h(x,t;n) is assumed to flow into the surface in the direction

opposite to that of the outward unit normal to ∂P.

By application of (5) to an elementary tetrahedron, it can be shown that h(n) = −h(−n) or

h(x,t;n) = −h(x,t;−n), together with

h = qini = q . n , q = qiei . (6)

The last surface integral in (5) can then be transformed into a volume integral as follows (using

the divergence theorem):

∫ h da = ∫ qini da = ∫ qi,i dv . (7)

∂P ∂P P

Similarly, using t = tkiniek and the divergence theorem

∫ t . v da = ∫ (ti . v),i dv
∂P ∂P

= ∫ [tki,ivk + tkivk,i] dv . (8)

P

By the transport theorem, the left-hand side of (5) can be expressed

. . .
hdh 1 1
dt ∫ ρ[ε + h2h v . v] dv = ∫ {ρ(ε + v . v) + (ε + h2h v . v)(ρ + ρvm,m)} dv
P P

. .
= ∫ ρ(ε + v . v) , (9)
P

where in obtaining the last result conservation of mass has been used.
 - 70 -

Now, substitute (7), (8) and (9) into (5) and rearrange the terms to obtain
. .
0 = ∫ {ρ r − qi,i − ρ ε + tkivk,i + vk[tki,i + ρ(bk − vk)]} dv ,
P

and after using also the equations of motion we get

.
∫ {ρ r − qi,i − ρ ε + tkivk,i} dv = 0 , (10)
P

which holds for all arbitrary parts P. Hence, it follows that the local energy equation is
.
ρ r − qi,i − ρ ε + P = 0 , (11)

where the stress power P is defined by

P = tkivk,i = tkidki , (12)

and where we have made use of the fact that vk,i = dki + wki and tki is symmetric.
 - 71 -

Part III: Examples of Constitutive Equations and Applications

Inviscid and viscous fluid and nonlinear and linear elasticity

 - 72 -

Examples of Constitutive Equations

Generally materials are classified as solids or fluids. However, we also encounter materials

which do not fall into either of these categories and which possess properties common to both

solids and fluids. A strict definition of these classifications is not attempted here, but we mention

here some of the main characteristic features of solids and contrast these with corresponding

features in the case of fluids. One of the main characteristic features of a solid is that it has a

reference state (or a reference configuration); and, in general, the deformed state (or

configuration) is not too far from the reference state. Moreover, regardless of the amount of

deformation, to some extent a solid remembers its reference state. A fluid, on the other hand,

does not possess a reference state (or configuration) and does not necessarily remain near such a

state in any motion.

1. Inviscid fluid

In hydrodynamics, an inviscid or an ideal fluid (including an ideal gas) is conceived of as a

medium possessing the following properties: (i) it cannot sustain shearing motion and (ii) the

stress vector (or the force intensity) acting on any surface in the fluid is always directed along

the normal to the surface, i.e., t is parallel to n. However, it is desirable to begin from a more

general viewpoint. Thus, we assume that an inviscid fluid is characterized by a constitutive

equation of the form

T = T̂(ρ) or tij = t̂ij(ρ) . (1.1)

In (1.1), the response function T̂ is a single-valued function of its argument and ρ is the mass

density of the fluid. Morever, it is instructive to consider an even more general assumption and

suppose the constitutive response function to depend also on the velocity vector†v. This is
because at this stage it is not clear why the explicit dependence on v (or even x) should be ruled

hhhhhhhhhhhhhhh
† We rule out inclusion of the velocity gradient which gives rise to a shearing motion. The latter is not compati-
ble with the properties of an ideal fluid stipulated in the opening paragraph of section 1.
 - 73 -

out. With this background, we begin by examining the more general constitutive assumption:

T = T̂(ρ, v) or tij = t̂ij(ρ, vk) . (1.2)

It is convenient to recall here (from Parts I & II) that under a superposed rigid body motion

the position x moves to x+ in accordance with

x+(τ) = a(τ) + Q(τ)x , (τ ≤ t)

(1.3)
+
v (τ) = a′(τ) + Q(τ)v(τ) + Q′(τ)x ,

where Q′ = Ω Q, Ω is the rigid body angular velocity, the temporary notation Q′(τ) is defined by
d
Q′(τ) = hhh Q(τ) with τ being real. Consider now a special case of (1.3) for which at time τ = t,
dτ
.
a′(t) = a(t) is a constant vector and Ω is a constant Ωo, such that at time t
. .
a(t) = c , Q(t) = Ωo Q(t) ,

where c is a constant vector and Ωo is a constant rigid body angular velocity tensor. We note for

later reference that when the above special motion represents a rigid body translational velocity,

we obtain
.
Q(t) = I , Q(t) = 0 ,
(1.4)
+
v =c+v .

Also, when the special motion involves only a constant rigid body angular velocity, we have
.
c = 0 , Q(t) = I , Q(t) = Ωo ,
(1.5)
+ +
v = v + Ωox , L = L + Ωo

We now examine (1.2)1 under a constant rigid body translational velocity of the form (1.4).

Since the constitutive equation (1.2)1 must hold for all motions, including a s.r.b.m., and since

on physical grounds the response function T̂ (but not its arguments) must remain unaltered under

such superposed rigid body motions, we have

 - 74 -

T+ = T̂(ρ+, v+) = T̂(ρ+, v + c) . (1.6)

Now under the special motion (1.4), the stress tensor transforms as

T+ = T . (1.7)

Substituting (1.2)1 and (1.6)2 with (1.7) we get

T̂(ρ, v) = T̂(ρ, v + c) , (1.8)

which must hold for all constant values c. It follows that (1.8) can be satisfied only if the depen-

dence of T̂ on v is suppressed and we are left with the original assumption (1.1). It should be

evident that the same type of conclusion can be reached if we had initially included in (1.2) also

dependence on the current position x.

Once more we examine the constitutive equation (1.1) under a general s.r.b.m., and recall

that under such motions

T+ = Q T QT .

Introducing the assumption (1.1) into the above invariance requirement, we obtain

T̂(ρ+) = Q T̂ (ρ) QT , (1.9)

which must hold for all proper orthogonal tensors Q. Since the condition (1.9) is also unaltered

if Q is replaced by − Q, it follows that (1.9) must hold for all orthogonal Q and not just the

proper orthogonal ones. Hence, T̂(ρ) is an isotropic tensor and can only be a scalar multiple of

the identity tensor I (see the Supplement), i.e.,

T = T̂(ρ) = −p(ρ)I or tij = t̂ij = −p(ρ)δij . (1.10)

In (1.10), p is a scalar function called the pressure and the minus sign is introduced by conven-

tion, in order to conform to the traditional form in which the stress tensor T = −pI for an inviscid

fluid is displayed in the fluid dynamics literature.

Substitution of the result (1.10) into the relation t = T n (see Part II) yields

t = −p(ρ) n or tk = −p(ρ)nk , (1.11)

 - 75 -

which shows that the stress vector acting on a surface in an inviscid fluid is always directed

along the normal to the surface. In most books on fluid dynamics, (1.11)1 is taken as a defining

property of an inviscid fluid, as was also indicated in the opending paragraph of this section. If

(1.11) instead of (1.1) is taken as a starting point, then with the use of t = T n we have

T n = −p(ρ)n or (T + pI)n = 0 which must hold for all outward unit normal n, and we may con-

clude that T = −pI in agreement with the earlier result (1.10) for the stress tensor in an inviscid

fluid.

We are now in a position to obtain the differential equation of motion for the determination

of the velocity field v in an inviscid fluid. Thus, after substituting (1.10) into the equations of

motion (see Part II), we arrive at

. .
−grad p + ρb = ρv or p,i + ρbi = ρvi . (1.12)

For a specified body force b, the vector equation (1.12)1 [or (1.12)2], together with the equation

for mass conservation represent four scalar equations for the determination of the four unknown

(p, v). The simple structure of (1.12) at first sight may be somewhat deceptive: these equations

are difficult to solve by analytical procedures (or even numerical methods). This is because of
.
the nonlinearity due to the convective terms in v.
 - 76 -

2. Viscous fluid.

Although our main objective here is the development of linear constitutive equations for a

(Newtonian) viscous fluid, it is enlightening to start with a more general constitutive assumption.

For a viscous fluid we assume that the stress tensor T depends on the present value of the mass

density ρ, the velocity vector v and the velocity gradient tensor L. Thus, we write
h h
T = T(ρ, v, L) or tij = tij(ρ, vk, vk,l) , (2.1)
h
where T is a single-valued function of its arguments and satisfies any other continuity or dif-

ferentiability conditions that may be required in subsequent analysis. Alternatively, for conveni-

ence and without loss in generality, we recall that L = D + W and rewrite (2.1) as

T = T̃(ρ, v, D, W) or tij = t̃ij(ρ, vk, dkl, wkl) . (2.2)

The constitutive assumption (2.2)1 [or (2.2)2] must hold for all admissible motions, includ-

ing such superposed rigid body motions as those specified by (1.4) and (1.5). First, by a pro-

cedure parallel to that discussed in the previous section (see the development between Eqs.

(1.6)-(1.8) of section 1) we may suppress the dependence of the response function in (2.2)1 on v

and arrive at

T = T̂(ρ, D, W) or tij = t̂ij(ρ, dkl, wkl) . (2.3)

Next, from the invariance requirement for T, namely T+ = Q T QT, we obtain the restriction

T̂(ρ+, D+, W+) = Q T̂(ρ, D, W) QT , (2.4)

which must hold for all proper orthogonal tensors Q. Consider now the special superposed rigid

body motion specified by (1.5). Under this special motion (2.4) becomes

T̂(ρ, D , W + Ωo) = T̂(ρ, D, W) , (2.5)

so that the response function T̂ cannot depend on W and the constitutive assumption (2.3) is

reduced to

T = T̂(ρ, D) or tij = t̂ij(ρ, dkl) , (2.6)

 - 77 -

where T̂ in (2.6) is a different function from that in (2.3). By imposing on (2.6) the invariance

requirement we arrive at

T̂(ρ+, D+) = Q T̂(ρ, D)QT . (2.7)

The restriction (2.7) must hold for all proper orthogonal tensors Q. Since the condition (2.7) is

also unaltered if Q is replaced by − Q, it follows that (2.7) holds for all orthogonal Q and not

just the proper orthogonal ones.

A viscous fluid characterized by the nonlinear constitutive equation (2.6)1 is known as the

Reiner-Rivlin fluid. We do not pursue further discussions of the general forms (2.6)1, but in the

remainder of this section consider in some detail the (Newtonian) linear viscous fluid as a special

case in which the response function T̂ is a linear function of D with coefficients which depend on

ρ.

In the discussion of the linear viscous fluid, it is convenient, in what follows, to carry out

the details of the development of the constitutive equations in terms of their tensor components.

Thus, for a linear viscous fluid, we specifiy t̂ij on the right-hand side of (2.6)2 to be linear in dkl

in the form

t̂ij(ρ, dkl) = aij + bijkl dkl , (2.8)

the coefficients aij and bijkl are functions of ρ and satisfy the symmetry conditions

aij = aji , bijkl = bjikl , bijkl = bijlk , (2.9)

where the condition (2.9)3 arises from the symmetry of dkl and is assumed without loss in gen-

erality.

The response function (2.8) must satisfy the invariance condition (2.7) which in component

form is
+
t̂ij(ρ, dkl) = Qim Qjn t̂mn (ρ, dkl) , (2.10)

where in recording the above we have also used ρ+ = ρ in the argument of the left-hand side of
(2.10). Introducing (2.8) into (2.10) we obtain the restriction
 - 78 -

aij + bijkl Qkp Qlq dpq = Qim Qjn (amn + bmnpq dpq) (2.11)

which holds for all arbitrary dkl. Considering separately the cases when dkl = 0 and dkl ≠ 0, it

follows from (2.11) that

aij = Qim Qjn amn , (2.12)

and

bijkl = Qim Qjn Qkp Qlq bmnpq ,

where in obtaining (2.12)2 we have made use of the orthogonality property of Qij. Now accord-

ing to the theorems on isotropic tensors (see the Supplement) the coefficients aij and bijkl are,

respectively, isotropic tensors of order two and four and hence must have the forms

aij = −pδij (2.13)

bijkl = λδij δkl + µδik δjl + γδil δjk ,

where p, λ, µ, γ are functions of ρ only and the use of −p instead of p in (2.13)1 is for later con-

venience and in order to conform to known expressions in the fluid dynamics literature. If we

appeal to the symmetry condition (2.9)2, then the right-hand side of (2.13)2 must be symmetric

in the pair of indices (ij) and (2.13)2 reduces to

bijkl = λδij δkl + µ(δik δjl + δil δjk) . (2.14)

It should be noted that the expression (2.14) implies a further symmetry restriction so that bijkl

satisfies

bijkl = bklij , (2.15)

in addition to (2.9)2,3. Substitution of (2.13)1 and (2.14) into (2.8) finally yields

tij = −pδij + λδij dkk + 2µdij , (2.16)

as the constitutive equation for a (Newtonian) linear viscous fluid. The scalar p in (2.16) is the

pressure and λ, µ are called the viscosity coefficients.

 - 79 -

The differential equations of motion for a (Newtonian) linear viscous fluid in terms of the

velocity field can be obtained by substituting (2.16) into Cauchy’s stress equations of motion and

leads to
.
−p,i + λdkk,i + 2µdij,j + ρbi = ρvi .

After recalling the kinematical relations

1
dkk = vk,k , dij = hh (vi,j + vj,i)
2

and making use of vj,ij = vj,ji, the above differential equations of motion become
.
−p,i + (λ + µ) vk,ki + µvi,kk + ρbi = ρvi , (2.17a)

or
.
−∇p + (λ + µ)∇(∇ . v) + µ∇2v + ρb = ρv . (2.17b)

The system of differential equations (2.17) and the mass conservation represent four scalar equa-

tions for the determination of the four unknowns, (p, v). It should be noted in the absence of

viscous affects (λ = µ = 0), (2.17) reduces to those for an inviscid fluid (see Eqs. (1.12)).
 - 80 -

3. Elastic solids: Nonlinear constitutive equations.

We consider here general constitutive equations for an elastic solid in the context of the

purely mechanical theory.

A material is defined by a constitutive assumption. Moreover, an elastic material in the

purely mechanical theory may be defined in terms of a constitutive equation for the stress. We

return to this later but first we need to establish some preliminaries.

We recall the expression for the rate of work by the body and contact forces as follows (see

Sec. 7 of Part II):

R = Rb + Rc , (3.1)

Rb = ∫ ρb . v dv , Rc = ∫ t . v da . (3.2)
P ∂P

Also, the local equations of motion (in terms of the symmetric stress tensor of Cauchy) are:
.
tki,i + ρbk = ρvk , tki = tik . (3.3)

Consider now the expression for Rc, use the divergence theorem to transform the surface

integral into a volume integral and make use of the equations of motion (3.3)1 to obtain:

Rc = ∫ t . v da = ∫ tkvkda = ∫ tkinivkda = ∫(tkivk),idv

∂P ∂P ∂P P

= ∫tki,ivkdv + ∫tkivk,idv
P P

.
= ∫ρvkvkdv − ∫ρbkvkdv + ∫tkivk,idv . (3.4)
P P P

Then, after using (3.2)1

.
Rc = ∫ρvkvkdv − Rb + ∫tkivk,idv . (3.5)
P P

From this result and (3.1) follows that

d
R = h h K(St) + ∫tkivk,idv , (3.6)
dt
P
 - 81 -

1
where K(St) = ∫ hh ρv . v dv is the kinetic energy for the material region occupied by P in the
2
P
present configuration κ .

We define an elastic body as one whose response depends on deformation gradient or on an

iiiii energy
appropriate measure of strain. We assume the existence of a strain iiiiii or a stored energy per

unit mass ψ such that

.
tkivk,i = ρ ψ , (3.7)

and we define the (total) strain energy for each part P in the present configuration by

U(S t) = ∫ ρ ψ dv . (3.8)
P

Then, by (3.6), (3.7), (3.8) the conservation of mass and the transport theorem we have

d
R(St) = h h [K(St) + U(St)] . (3.9)
dt

iiiiiii: The rate of work by contact and body forces is

We have thus proved the following theorem

equal to the sum of the rate of kinetic energy and the rate of the strain energy.

We now return to (3.7) and for an elastic body assume that the strain energy density

ψ = ψ′(F) [or ψ = ψ′(xi,A)].

Alternatively, we may assume that ψ depends on a measure of strain such as E and write
hh hh
ψ = ψ(E) or ψ = ψ(EAB) . (3.10)

Now the material derivative of ψ is

hh
. ∂ψ .
ψ = hhhhh EABxi,B , (3.11)
∂EAB

where in obtaining (3.11) we have used (from Part I) the expressions

2EAB = CAB − δAB , CAB = xi,Axi,B ,

. . .
EAB = dikxk,Axi,B , CAB = 2EAB .
 - 82 -

Introducing (3.11) into (3.7), we obtain

hh
∂ψ
tkidki = ρ hhhhh
∂EAB

or
hh
∂ψ
(tki − ρ hhhhh xi,Axk,B)dki = 0 , (3.12)
∂EAB

which must hold for all dki. Hence, we conclude that

hh hh
∂ψ ∂ψ
tki = ρxi,Axk,B hhhhh or T = ρF hhh FT . (3.13)
∂EAB ∂E

If instead of the constitutive assumption (3.10), we begin with the alternative assumption

ψ = ψ̂(C) or ψ = ψ̂(CAB) , (3.14)

then in a manner analogous to the development that led to (3.14) we deduce that

∂ψ̂ ∂ψ̂
tki = 2ρ xi,Axk,B hhhhh or T = 2ρ F hhh FT . (3.15)
∂CAB ∂C
 - 83 -

4. Elastic solids: Linear constitutive equations.

We recall that in the context of infinitesimal kinematics discussed previously (Part I, sec-

tion 7), a function f is said to be of 0(εn) as ε → 0 if there exists a nonnegative real constant C
such that | f | < Cεn as ε → 0. Moreover, for infinitesimal deformation gradients, the distinction

between Lagrangian and Eulerian descriptions disappears; and that, to within 0(ε2), the relative

displacement gradient can be written as

ui,j = eij + ωij , (4.1)

where

1 1
eij = hh (ui,j + uj,i) = 0(ε) , ωij = hh (ui,j − uj,i) = 0(ε) ,
2 2

as ε → 0 . (4.2)

We further assume that all derivatives (with respect to both coordinates and time) of kinematical

quantities, e.g., ui, eij, ωij, are also of 0(ε) as ε → 0.

iiiiiiii linear theory.

We now introduce additional assumptions in order to obtain a complete

Thus, we assume that the fixed reference configuration (which was identified with the initial
configuration) is "stress-free", i.e., in this configuration the stress vector t and the body force b

are zero. We also assume that the body force b (or its components bi) and the stress vector t (or

its components ti) -- when expressed in suitable non-dimensional form -- as well as the com-

ponents of the stress tensor tki, all are of 0(ε) as ε → 0.

Recalling that for infinitesimal motion (see Eq. (7.32) of Part 1, section 7)

J−1 = 1 − eii = 1 − 0(ε) as ε → 0 , (4.3)

from the referential form of the conservation of mass, namely ρJ = ρo, we conclude that

ρ = ρo − 0(ε) as ε → 0 . (4.4)
..
Keeping in mind that t, b, u are all of 0(ε) as ε → 0, with the help of (4.4) and after the neglect
 - 84 -

of terms of 0(ε2) as ε → 0, the equations of motion in the linear theory reduce to

.
tki,i + ρobk = ρovk , (4.5)

tki = tik , (4.6)

. .
where vi = ui is the particle velocity, a comma denotes partial differentiation and the vi on the
∂vi
right-hand side of (4.5) stands for hhh . It should be observed that each term in (4.5) is of 0(ε) as
∂t
ε → 0.

The equations of motion of the linear theory can also be obtained from the results of the

alternative derivation (see Part II, section 5). Thus, in terms of the non-symmetric Piola-

Kirchhoff stress tensor piA, the equations of motion are

.
piA,A + ρobi = ρovi , (4.7)

piAxj,A = pjAxi,A , (4.8)

and we also recall the relations

piB = xi,AsAB , sAB = sBA , (4.9)

Jtki = xk,ApiA = xk,Axi,BsAB . (4.10)

Since J = 1 + 0(ε), xi,A = δiA + ui,A = δiA + 0(ε), XA,i = δAi + 0(ε) as ε → 0, we have

piA = JtkiXA,k

= tkiδAk + 0(ε2) as ε → 0 . (4.11)

Similarly, since now piA = 0(ε) as ε → 0, (4.9)1 after using (4.11) yields

sAB = piBXA,i

= piBδAi + 0(ε2) as ε → 0 . (4.12)

It follows from (4.11) and (4.12) that if terms of 0(ε2) as ε → 0 are neglected, then piA, sAB, tij
 - 85 -

are all equal and the distinction between the Cauchy stress and the Piola-Kirchhoff stresses

disappears. Hence, the linearized version of the equations of motion (4.7) can also be written as
..
sAB,B + ρobA = ρouA . (4.13)

We now turn our attention to the linearized version of the constitutive equation discussed in

the previous section. We recall that in the context of nonlinear elasticity, the constitutive equa-

tions for the stress tensor may be expressed as

hh
∂ψ
hhhhh
tki = ρxi,Axk,B . (4.14)
∂EAB

In the linear theory discussed above the components of the stress tensor tki = 0(ε) as ε → 0.

Since the right-hand side of (4.14) must also be of 0(ε) as ε → 0 in the linear theory, for an elas-

tic body which is initially "stress-free", after linearization of (4.14), it will suffice to assume that
hh
ψ is quadratic in the infinitesimal strain eij. To justify this stipulation, we proceed to obtain the

linearized version of (4.14). Recall that in the infinitesimal theory,

J = 1 + 0(ε)

xi,A = δiA + 0(ε)

eij = EABδAiδBj + 0(ε2) .

Using these expressions, we may linearize equation (4.14) to obtain

hh hh
dψ
hhhhh I M I M I M hhhhh
dψ
tki = ρxi,Axk,B = ρo L1 − 0(ε) O LδiA + 0(ε) O LδkB + 0(ε) O
dE AB dE AB

hh hh
dψ ∂emn
∂ψ hhhhh
= ρo hhhhh δiAδkB + 0(ε ) = ρo
2 h
hhhh δiAδkB + 0(ε2)
dEAB ∂emn ∂EAB

hh
∂ψ ∂(EPQδPmδQn)
= ρo hhhhh hhhhhhhhhhhh δiAδkB
∂emn ∂EAB

hh
∂ψ ∂EPQ
= ρo hhhhh hhhhh δPmδQnδiAδkB . (4.15)
∂emn ∂EAB
 - 86 -

Since EAB is a symmetric second order tensor, we have

∂EPQ 1 I ∂EPQ ∂EPQ M 1

= hh J hhhhh + hhhhh J = hh ILδPAδQB + δPBδQA O
hhhhh M
∂EAB 2 L ∂EAB ∂EBA O 2
 - 107 -

Supplements to the Main Text

ME 185 Class Notes

 - 108 -

Supplements to the Main text in Class Notes

Rather than writing expressions in terms of components of vectors and tensors, it is some-

times convenient to use a coordinate free notation. In coordinate free notation, a vector is

referred to by the symbol v, whereas in indicial notation this vector is written in terms of its

components vi with respect to a basis ei. The values of the components vi are dependent on the

choice of basis. Similarly, a tensor may be referred to in coordinate free notation as T or in indi-

cial notation as Tij with respect to a given basis. Notice that tensor multiplication is not commu-

tative; however, the components of a tensor in indicial notation are commutative. The following

relationships between indicial and coordinate free notation are sometimes useful:

Coordinate Free Notation Indicial Notation

________________________ _________________

v (a vector) vi

A (a second order tensor) Aij

Av Aijvj

ATv Ajivj

AB AijBjk

ATB AjiBjk

ABT AijBkj

Examples from kinematics:

C = FTF CAB = FiAFiB

B = FFT Bij = FiAFjA

U2 = C UADUDB = CAB

Eigenvalues, eigenvectors, and characteristic equations.

In general, a second order tensor T operating on a vector v results in another vector (say) w

which is not necessarily in the same direction and does not necessarily have the same magnitude
 - 109 -

as v. This operation is expressed symbolically as

Tv = w or Tijvj = wi . (1)

However, if a nonzero vector v is such that when T operates on it the resulting vector w is paral-

lel to v, then v is called an eigenvector of the tensor T.

In this case, we can write

Tv = βv or Tijvj = βvi , (2)

where the scalar β is called an eigenvalue corresponding to the eigenvector v. From (2), we can

write

(T − βI)v = 0 or (Tij − βδij)vj = 0 , (3)

where I is the identity tensor. Equation (3) represents three equations which must be satisfied by

all three components of v in order that v be an eigenvector of the tensor T. According to a

theorem of linear algebra, a homogeneous system of three equations for three unknowns (such as

equation (3)), has a non-trivial solution if and only if the determinant of the matrix of

coefficients vanishes, i.e. if

det(T − βI) = 0 or det(Tij − βδij) = 0 . (4)

Equation (4) represents a cubic equation in β, which may be written as

φ(β) = − β3 + I1β2 − I2β + I3 = 0 . (5)

Equation (5) is called the characteristic equation of T and I1 , I2 , and I3 are called the principal

invariants of T. Expressions for the principal invariants can be obtained by solving the system

(4), giving
 - 110 -

I1 = trT I1 = Tii
1 1
I2 = hh (tr2T − trT2) or I2 = hh (TiiTjj − TijTji) (6)
2 2
1
I3 = detT I3 = det(Tij) = hh εijkεlmnTlTjmTkn .
6

In general equation (5) has three roots which are not necessarily real and are not necessarily dis-

tinct. However, in the special case where T is symmetric, we can prove some additional

theorems.

Theorem 1: If T is symmetric, then all the roots of the characteristic equation (5) are real.

Proof: We will prove this theorem by contradiction. Suppose first that the roots of (5) are not

real. According to a theorem of algebra, the complex roots of a polynomial with real

coefficients must occur in conjugate pairs. Therefore, if there exists a complex root β(1) of
(5), then there must also exist a root β(2) of (5) which is a conjugate of β(1). Thus β(1) and

β(2) must be of the form

β(1) = µ + iγ
β(2) = µ − iγ , (7)

where i2 = −1. The eigenvectors which correspond to the eigenvalues β(1) and β(2) are of
the form

v(1) = α + iδδ vj(1) = αj + iδj

or (8)

v(2) = α − iδδ vj(2) = αj − iδj .

Since equation (2) must hold for all eigenvectors and their corresponding eigenvalues, we

can write
 - 111 -

Tv(1) = β(1)v(1) Tijvj(1) = β(1)vi(1)

or (9)

Tv(2) = β(2)v(2) Tijvj(2) = β(2)vi(2) .

Multiply (9)1 by vi(2) and (9)2 by vi(1) to obtain (in indicial notation)

β(1)vi(1)vi(2) = Tijvj(1)vi(2)
β(2)vi(2)vi(1) = Tijvj(2)vi(1) = Tjivj(2)vi(1) = Tijvj(1)vi(2) . (10)

In obtaining (10)2, we have made use of the facts that Tij is symmetric and that i and j

are dummy variables. Subtracting (10)2 from (10)1 gives

(β(1) − β(2))vi(1)vi (2) = 0 . (11)

Substituting (7) and (8) into (11), we can write

0 = 2γi(αj + iδj) (αj − iδj) = 2γi(αjαj + δj δj) . (12)

The term in parenthesis in (12) is always greater than zero since it is the sum of two
squared real numbers. Therefore, we must have γ = 0. Equation (7) then implies that

β(1) and β(2) are both real, which is a contradiction of our initial assumption. We can
therefore conclude that if T is symmetric, all the roots of its characteristic equation are

real.

Theorem 2: The eigenvectors which correspond to distinct eigenvalues of a symmetric tensor

are orthogonal.

Proof: Suppose that v(1) and v(2) are eigenvectors of a symmetric tensor T corresponding to dis-
tinct eigenvalues β(1) and β(2), i.e. β(1) ≠ β(2). As in the previous proof, we can show that
 - 112 -

equation (11) must be satisfied. Since β(1) ≠ β(2), we must have

v(1) . v(2) = 0 or vi(1)vi(2) = 0 . (13)

Equation (13) is exactly the definition of orthogonal vectors, thus the eigenvectors v(1)
and v(2) are orthogonal.

Theorem 3: Every second order symmetric tensor has three linearly independent principal

directions.

iiiiiiiiiiiiiiiiiiiiiiiiiiiiii
Proof: For the proof of this theorem, see Advanced Engineering Mathematics, Wylie, p. 541.

Remark: In the preceding development, we can, without loss of generality, replace the eigen-

vector v with a unit vector m, where

m = v/ | v | . (14)

iiiiiiiiiiiiiiiiiii
The vector m is called the normalized eigenvector.

Using theorems 1, 2 and 3 and the previous remark, we can state that any symmetric tensor T

has three real eigenvalues β(1) , β(2) , and β(3) and three orthonormal eigenvectors

m(1) , m(2) , and m(3). It can be shown, using a theorem of linear algebra, that since T has three

linearly independent eigenvectors, it can be transformed into a diagonalized form, denoted by

Λ, using the transformation

T′ij = Λij = amianjTmn ,

where

aij = ei . m(j) and xi = aijx′j . (15)

Here the primed quantities correspond to the transformed system. It can also be shown that the
 - 113 -

diagonal members of Λ are simply the eigenvalues, so that

R (1) H
Jβ
0 0 J
(Λij) = J0 β(2) 0 J . (16)
J0 J
Q
0 β(3) P

Quadratic forms and positive definiteness.

Associated with any symmetric second order tensor T is a scalar valued function of a vector

Q(v), where v is an arbitrary vector, defined by

Q(v) = (Tv) . v = Tij vjvi = v . (Tv) . (17)

iiiiiiiiiiii
We call Q(v) a quadratic iiiiiiiiiiiii
form. The tensor T is said to be positive definite if for all

v ≠ 0 , Q(v) > 0 .

Theorem 4: A symmetric second order tensor T is positive definite if and only if all the eigen-

values of T are positive.

Proof: Again using a theorem from linear algebra, one can conclude that since the eigenvectors

m(1) , m(2) , and m(3) are orthogonal in a 3-dimensional space, then any arbitrary vector w
in that space may be represented as a linear combination of m(1) , m(2) , and m(3). Thus, we

can write

w = a1m(1) + a2m(2) + a3m(3) . (18)

The eigenvalues of T corresponding to m(1) , m(2) , and m(3) are β(1) , β(2) , and β(3), as
before, so equation (2) gives

Tm(i) = β(i)m(i) . (19)

 - 114 -

Using (18) and (19), we can write

Tw = a1β(1)m(1) + a2β(2)m(2) + a3β(3)m(3) . (20)

Substituting (20) into (17) gives

Q(w) = Tijwjwi = a 2β(1) + a 2β(2) + a 2β(3) . (21)

1 2 3

Necessity: If T is positive definite, then Q(w) > 0 for all w. Since a1 , a2 , and a3 are all real

numbers, the square of each of these terms must be positive (or zero). If w = a1m(1), then

Q(w) > 0 implies from (21) that β(1)a 2 > 0, or that β(1) > 0. Since w is arbitrary, a similar
l
argument shows that β (2)
and β (3)
are both greater than zero. Thus, if T is positive definite,

then all the eigenvalues of T are positive.

Sufficiency: If all the eigenvalues of T are greater than zero, then by (21) it follows that

Q(w) > 0 for all w ≠ 0. Therefore, T is positive definite if all the eigenvalues of T are posi-

tive.

Theorem 5: If B is a positive definite symmetric tensor, then there exists a unique symmetric

positive definite square root T of B such that T2 = B.

Remarks on the calculation of the square root of B:

Let Bik = aipakqΛpq (see (15)), where Λ is a diagonal matrix as in equation (16). Define a
h1h
tensor Λ 2pq as

R ddd H
h1h
√ (1) 0
J β 0 J
(Λ 2 )pq = J0 √βddd
(2)
0 J ,
J0 (3) J
Q
0 √βddd
P

so that
 - 115 -

h1hh1h
Λ pqΛ 2qs
2
= Λps . (22)

h1h
Defining Tij = aipajqΛ 2pq, we can show that

h1h h1h h1h h1h

TijTjk = aipajq Λ pqajmaknΛ mn = aipajqajmaknΛ pqΛ 2mn
2 2 2

h1h h1h
= aipaknδqm Λ pqΛ 2mn = aipaknΛpn = Bik
2

or

T2 = B . (23)

Theorem 6: Polar decomposition theorem (see class notes section I/3)

Any non-singular (invertible) tensor F may be uniquely expressed as

F = RU = VR , (24)

where R is orthogonal and U and V are symmetric positive definite tensors.

Proof: Existence

h1h
Define U = (F F) 2 and R = FU−1. We first show that U is symmetric postive definite.
T

To show that U2 is positive definite, consider the quadratic form

Q(v) = v . U2v = v . FTFv = Fv . Fv .

Since F is invertible, then Fv ≠ 0 if v ≠ 0. Assuming that F and v are both real, it fol-

lows that Q(v) > 0 if v = 0. Thus, U2 is positive definite. Also, since FTF is sym-
metric, then U2 must be symmetric. From theorem 5, we know that U2 must have a

unique positive definite square root U, where U is also symmetric. To show that R is

orthogonal, note that

RTR = (U−1)TFTFU−1 = U−1FTFU−1 = U−1UUU−1 = I .

 - 116 -

Using the definitions of U and R, we can write

RU = FU−1U = F .
 - 117 -

Thus, we have shown the existence of the decomposition (24).

Uniqueness

To show uniqueness of the decomposition (24), assume first that two such decomposi-

tions exist, so that F = RU = R*U*. The star is used here to denote an alternative
decomposition. Notice that

FTF = UTRTRU = U2
and

FTF = U*TR*TR*U* = U*2 ,

thus U2 = U*2. By theorem 5, the square roots of U2 and U*2 are unique, so we must

have U = U*. Notice that

0 = F − F = RU − R*U* = (R − R*)U .

Since U is non-singular, U−1 exists. Thus,

0 = (R − R*)UU−1 = R − R* ,

and so R = R*. Thus, the decomposition F = RU is unique. Similarly, one can prove
hh
the existence and uniqueness of the decomposition F = VR, where we temporarily
hh hh
allow R to be different from R. The relationship between V, U, R, and R may be

determined by observing that

hh
F = RU = (RURT)R = VR ,

where RURT is symmetric and positive definite. Since the decomposition of F is

unique, we conclude that

hh
V = RURT , R = R , and U = RTVR . (25)
 - 118 -

Extremal properties of quadratic forms.

Consider the symmetric tensor T and the quadratic form

Q(v) = v . Tv = Tijvjvi . (26)

We wish to know the extreme values of Q(v) subject to the constraint that v be a unit vector

(i.e. v . v = vi vi = 1). Using the method of Lagrange multipliers, the conditions for the

extremum of (26) are

∂
hhhh {Tijvivj − β(vivi − 1)} = 0 , (27)
∂vk

where β is the undetermined multiplier and v . v − 1 = 0 is the constraint equation. Performing

the differentiation, (27) becomes

Tijδikvj + Tijviδjk −2βviδik = 0

or

Tkjvj + Tikvi − 2βvk = 0 . (28)

Since T is symmetric, (28) becomes

Tkjvj − βvk = 0 . (29)

Therefore, Q(v) attains extreme values when v is a eigenvector of T. Using (2), it is clear that if

m is a unit eigenvector of T with corresponding eigenvalue β, then

Q(m) = m . (Tm) = β . (30)

In summary, the quadratic form obtains its extremum values when it is operating on an eigenvec-

tor. If the eigenvectors are normal, these extremum values are simply the corresponding eigen-

values.
 - 119 -

Relationship between eigenvalues of U, C, V, B, E, and D and the principal stretch.

(see class notes sections I/4 and I/5)

The principal directions are the directions in which the stretch attains an extreme value.

The principal stretch λ is the extreme value of stretch in a given principal direction. As derived

in section I/4, the eigenvalues of U and V are equal to the principal values of stretch and the

eigenvalues of C and B are equal to λ2. It was also shown in this section that the eigenvectors of
U and C are coincident and that the eigenvectors of V and B coincide.

The relative strain tensor E is given by

1
E = hh (C − I) . (31)
2

The principal values of strain, denoted by β, are equal to the eigenvalues of E. The principal

directions of strain, denoted by m, are equal to the eigenvectors of E. From (2), we have

Em = βm . (32)

Substituting (31) into (32) gives

h1h (C − I)m = βm ,
2
or

Cm = (2β + 1)m . (33)

Equation (33) implies that m is an eigenvector of C as well as of E and that 2β + 1 is an eigen-

value of C. As previously noted, the eigenvalue of C is λ2, so

1
2β + 1 = λ2 or β = hh (λ2 − 1) . (34)
2
 - 120 -

Thus, the eigenvectors of U , C , and E coincide and the principal value of strain is related to the

principal value of stretch by equation (34).

As was derived in section I/5, the eigenvalues and eigenvectors of the rate of deformation

tensor D are given by

.
λ d
Dm = hh m = h h (ln λ) m , (35)
λ dt

where m is an eigenvector of D, (m is also called the "principal direction of stretch"). It is clear

.
from the form of (35) that λ / λ (called the logarithmic rate of stretching) is an eigenvalue of D.

Example calculation

Suppose that at some instant at a particular point in the body, the deformation gradient ten-

sor F is given by

R2 1 0 H
J J
[FiA] = J1 2 0 . (36)
J
Q0 0 1 P

Notice that (36) yields a symmetric positive definite tensor. Since the polar decomposition of F
is unique, it follows that V = U = F and R = I . The principal invariants of F = U are

I1 = tr U = UAA = 5
1 1
I2 = hh (tr2U − trU2) = hh (UAAUBB − UABUBA) = 7 (37)
2 2
I3 = det U = 3 .

The characteristic equation of F is then

−β3 + 5β2 − 7β + 3 = 0 . (38)

The roots of (38), i.e. the eigenvalues of F, are

 - 121 -

β(1) = 3
β(2) = 1 (39)
β(3) = 1 .

To solve for the eigenvectors of F, substitute each of the eigenvalues into equation (2), or

Fv = βv . (40)

For β(1) = 3, equation (40) gives

(F − 3I)M(1) = 0 ,
or

− M (1) + M (1) = 0 and −2M (1) = 0 .

1 2 3

Thus, the eigenvector M(1) is given by

R1 H
= c1 JJ1 J
[Mi(1)] J
,
Q0 P

where c1 is an arbitrary constant. Similarly, for β(2) = β(3) = 1, equation (40) gives

(F − I)M(α) = 0
where α = 2,3, or

M (α) = −M (α) .
1 2

Solving for M(α) gives

R−1 H R0 H
J J
and [Mi(3)] = c3 JJ0 J
[Mi(2)] = c2 J 1 J J
.
Q 0 P Q1 P

We may take unit eigenvectors

 - 122 -

R1 H R−1 H R0 H
[Mi(1)]= hhh J1 J = hhh J 1 J , [Mi(3)] =
1 1 J J
, [Mi(2)] 0 ,
J J J J J J
√2dQ 0 P √2d Q 0P Q1 P

which form a right-handed triad. Note that every vector in the space spanned by M(2) and M(3)
is also an eigenvector with eigenvalue 1.

One can better understand the physical significance of these results by looking at the effect

of the deformation (36) upon a unit cube. The graph below gives the cross section of the cube

before and after the deformation. As expected from the preceding calculations, the principal

directions (i.e. the directions in which the stretch attains extremum values) are in the directions

of the eigenvectors M(1) , M(2) , and M(3). Line elements along M(1) are stretched to three
times the initial length. Thus, the stretch λ = 3 is equal to β(1), as expected. Line elements in the

directions M(2) and M(3) retain their initial length. Therefore, the stretch λ in these directions

equals unity, corresponding to β(2) = β(3) = 1. In this example, line elements along principal

directions undergo pure stretch as a result of the deformation; however, line elements in other

directions may undergo both stretch and rotation. This characteristic is true in general whenever

the deformation F is of the pure stretch type (i.e. whenever R = I).

 - 123 -

Supplement to Part I, Section 5

Material line elements, area elements, volume elements and their material time derivatives.

Preliminary results

Claim:

1
det A = hh εijkεlmnAliAmjAnk (1)
6

Proof: Consider an arbitrary tensor A with components Aij. Let am = Am1, bm = Am2, and

cm = Am3. Recall that

RA A A H
11 21 31
J J
a . (b × c) = ε a b c
lmn l m n = ε A A A
lmn l1 m2 n3 = det A A A
J 12 22 32 J
= det AT ≡ A . (2)
QA13 A23 A33 P

Now define Tijk by

Tijk = εlmnAliAmjAnk (3)

where by (2), T123 = A. Since l, m, and n are dummy indices, we may switch them and use

the properties of εlmn to show that

Tijk = Tjki = Tkij = − Tikj = − Tkji = − Tjik . (4)

Recalling the properties of εijk, we can conclude from (4) that Tijk must be a multiple of εijk.

>From (2) and (4), we see that

T231 = T312 = T123 = A

T321 = T213 = T132 = − A
Tijk = 0 if i = j , j = k , or i = k .

Hence, it is clear that

Tijk = A εijk . (5)

 - 124 -

1
Equate (3) and (5) and multiply the resulting equation by hh εijk, noting that εijkεijk = 6, to
6
get

1
A = det AT = hh εijkεlmnAliAmjAnk . (6)
6

Since det AT = det A for any tensor A, we have proved equation (1).

Claim:

h∂(det A)
hhhhhhh = (det A) A−T (7)
∂A

Proof: Assume that the components of A are independent, so that we can write

∂A ij
h hhh = δirδjs . (8)
∂Ars

Now differentiate (6) with respect to Ars to obtain

h∂(det A) h∂A
hhhhhhh 1
= hhh = hh εijkεlmn {δrlδsiAmjAnk + AliδrmδsjAnk + AliAmjδrnδsk}
∂Ars ∂Ars 6
1
= hh {εsjkεrmnAmjAnk + εiskεlrnAliAnk + εijsεlmrAliAmj}
6
1 1
= hh {3εsjkεrmnAmjAnk} = hh εsjkεrmnAmjAnk . (9)
6 2

Suppose now that A is non-singular (det A ≠ 0), so that there exists a tensor B with com-

ponents Bij such that AliBir = δlr. If this is the case, then (9) may be rewritten using (3) and

(5) as

h∂A
hhh = h1h ε (ε δ )A A
∂Ars 2 sjk lmn rl mj nk

1
= hh εsjkεlmn(AliBir)AmjAnk
2
1
= hh εsjk(εlmnAliAmjAnk)Bir
2
1 1
= hh εsjkεijkABir = hh (2ABsr) = ABsr . (10)
2 2
 - 125 -

Denoting B as A−T, equation (10) becomes

h∂(det A)
hhhhhhh = A A−T = (det A) A−T ,
∂A

which is identical to (7).

Material line elements

Consider a motion of a body B given by x = χ (X , t), where X is the position vector of a

certain particle in the reference configuration and x is the position vector of the same particle in

the current configuration at time t. Two neighboring material particles in the reference

configuration are located at X and X + dX. Consider a material

iiiiiiiiiiiiiiiii
line element dX in the reference

configuration at the point X and having length dS and direction M, so that dX = dS M. Under

the motion χ(X , t), the material particles X and X + dX are taken to the positions x and x + dx.

The corresponding material line element dx in the current configuration having length ds and

direction m is related to dX by

dx = F dX , or dxi = FiA dXA = xi,AdXA , (11)

where FiA = xi,A are the components of the deformation gradient F. A diagram of this motion of

the material line segment is given below. Recall from Section I/4 of the class notes that the

square of the length of a material line element in the current configuration is given by

ds2 = xi,Axi,BdXAdXB . (12)

.
hhh
Recalling that xi,A = vi,jxj,A, we can take the material derivative of (12) to get
.
hhh .
hhh .
hhh
ds2 = xi,Axi,BdXAdXB + xi,Axi,BdXAdXB

= vi,jxj,Axi,BdXAdXB + xi,Avi,jxj,BdXAdXB

= (vi,j + vj,i)xi,Axj,BdXAdXB

= 2dijxi,Axj,BdXAdXB . (13)
 - 126 -

1
This last step uses the definition of the rate of deformation tensor D = hh (L + LT). Equation (13)
2
may be rewritten using (11) as

.
hhh
ds2 = 2dijdxidxj . (14)

Material area elements

Let 1dX and 2dX represent two material line elements in the reference configuration located
at X. Suppose that the motion χ (X ,t) takes 1dX, 2dX into the elements 1dx, 2dx, respectively, at

iiiiiiiiiiiiiiiiii
x in the current configuration. Define the material area element dA with orientation N in the

reference configuration by

N dA = 1dX × 2dX (15)

so that

dAL = NLdA = N dA . eL = (1dX × 2dX) . eL = εLMN 1dXM 2dXN . (16)

The corresponding material area element in the current configuration is da with orientation n

(see Figure 2) defined by

n da = 1dx × 2dx , (17)

so that

ni da = n da . ei = (1dx × 2dx) . ei (18)

>From (18) and (11) and noting that xm,DXD,i = δmi, we can write

ni da = εijkxj,Axk,B 1dXA 2dXB = εmjkδmixj,Axk,B 1dXA 2dXB

= εmjk(xm,DXD,i)xj,Axk,B 1dXA 2dXB

= εmjkxm,Dxj,Axk,BXD,i 1dXA 2dXB .

Using (3), (5), and (16), this expression becomes

ni da = εDAB(det F)XD,i 1dXA 2dXB

 - 127 -

or

ni da = J XD,i ND dA . (19)

Assuming that F is non-singular and using (7) and (19), we can write
.
hhhh . .
h hh
ni da = J XD,i ND dA + J XD,i ND dA . (20)

Notice that

. dJ ∂J ∂FjA
J = hhh = hhhhh hhhhh = J (F−1)Aj vj,kFkA = J vj,kδkj = J vj,j . (21)
dt ∂FjA ∂t

Using (21), equation (20) becomes

.
hhhh
ni da = (J XD,ivj,j − J XD,jvj,i)ND dA

or

.
hhhh
ni da = J(vj,jXD,i − vj,iXD,j)ND dA . (22)

Material volume elements

Let 1dX, 2dX, and 3dX be three material line elements forming a right handed system
located at X in the reference configuration. Suppose that the motion χ (X , t) takes 1dX, 2dX,

and 3dX into three line elements 1dx, 2dx, and 3dx, respectively, at x in the current configuration

iiiiiiiiiiiiiiiiiiii
(see Figure 3). Define the material volume element dV in the reference configuration by

dV = 1dX . (2dX × 3dX) = εABC1dXA 2dXB 3dXC . (23)

The corresponding material volume element in the current configuration is then

dv = 1dx . (2dx × 3dx) = εijk 1dxi 2dxj 3dxk . (24)

Using equations (3), (5), (11), and (23), we can write (24) as

dv = εijkxi,Axj,Bxk,C1dXA 2dXB 3dXC = εABC(det F) 1dXA 2dXB 3dXC

 - 128 -

= J(εABC 1dXA 2dXB 3dXC)

 - 129 -

or

dv = J dV . (25)

Taking the material derivative of (25) gives

.
hh .
dv = J dV = J vj,jdV = J(div v)dV , (26)
.
where equation (21) is used for the calculation of J.

iiiiiiiiiiiiiii
A special case of the above analysis occurs for isochoric motions (i.e. motions for which

dv = dV for all volumes occupied by any material region of the body). It is clear from this
.
hh
definition that dv = 0 for isochoric motions. We see from equations (25) and (26) that the neces-

sary and sufficient condition for isochoric motion may be expressed either as

J = 1 or vi,i = div v = 0 . (27)

Summary

A summary of the basic equations derived in this appendix are given in the following:

Material line, area, and volume elements:

ds2 = xi,Axi,BdXAdXB

ni da = J XD,i ND dA

dv = J dV

Material derivatives of line, area, and volume elements:

.
hhh
ds2 = 2 dijxi,Axj,BdXAdXB

h.h
dai = J(vj,jXD,i − vj,iXD,j)dAD

.
hh
dv = J vj,jdV .
 - 130 -

Supplement to Part I, Section 7

Interpretation of infinitesimal strain measures.

Recall the following expressions:

ds2 = dxidxi = xi,Axi,BdXAdXB

dS2 = dXAdXA = XA,iXA,jdxidxj

1 1
EAB = hh (CAB − δAB) = hh (UA,B + UB,A + UM,AUM,B)
2 2

1 1
eij = hh (δij − cij) = hh (ui,j + uj,i − um,ium,j) . (1)
2 2

Also recall that the stretch λ and the extension E are defined by

ds
λ = hhh = √CABMAMB = √1
dddddddd + 2 EABMAMB
dddddddddddd (2a)
dS

= 1/√cddddd
ijmimj = 1/√1 − 2eijmimj
ddddddddd

and

ds − dS
E = hhhhhhh = λ − 1 , (2b)
dS

where

MAMA = mimi = 1 and dXA = MAdS , dxi = mids . (3)

For an infinitesimal deformation (see class notes Part I/7), we may approximate EAB by neglect-

ing term of O(ε2) as ε → 0 as

1
EAB = hh (UA,B + UB,A) = O(ε) as ε → 0 . (4)
2

The distinction between Lagrangian and Eulerian strain disappears for an infinitesimal deforma-

tion. Thus,
 - 131 -

eij = EABδiAδjB = O(ε) as ε → 0 . (5)

Hence, for an infinitesimal deformation the extension and the stretch given in (2) may be written

as

E = λ − 1 = EABMAMB + O(ε2) = eijmimj + O(ε2) (6)

and

λ = √1dddddddddddd
+ 2EABMAMB = 1 + EABMAMB + O(ε2) (7a)

or, alternatively,

λ = 1/√1ddddddddd
− 2eijmimj = 1 + eijmimj + O(ε2) . (7b)

In the following discussion, we will use e11, e22, ..., to denote the components of the infinitesimal

strain tensor, since from (5) no distinction needs to be made between Lagrangian and Eulerian

strain as ε → 0. Consider a line element which lies along the X1 axis, i.e.

dX1 = dS , dX2 = dX3 = 0 , and MA = (1, 0, 0) ,

so that from (6)

ds − dS
E = hhhhhhh = e11 . (8)
dS

That is, e11 represents the extension (the change in length divided by the original length) of a

line element lying along the X1 axis. We note that to within O(ε2), the direction mi in the current

configuration coincides with the direction MA in the reference configuration. Similar results

hold for e22 and e33. In summary: the diagonal elements of the infinitesimal strain tensor

represent the extension of line elements in the corresponding coordinate directions.

Consider two orthogonal elements which initially lie along the X1 and X2 axes, so that

dX1 = dS , dX2 = dX3 = 0 , MA = (1, 0, 0) (9)

hh h hh hh hh
dX2 = dS , dX1 = dX3 = 0 , MA = (0, 1, 0) .
 - 132 -

h
Let θ denote the angle between the corresponding deformed line elements dx and dx, whose
h
lengths are ds and ds. >From (8), neglecting terms of O(ε2) as ε → 0, we have
h h
ds = (1 + e11)dS and ds = (1 + e22)dS . (10)

Now observe that

h hh
h
dx . dx dxi dxi dXAdXB
cos θ = hhhhhhhhhh = hhhhhhh = xi,Axi,B
hhhhhhhh
h
h ds ds ds ds
| dx | | dx |
hh hh
dXAdXB MAMB
= CAB hhhhhhhh = CAB hhhhhhhhhhhhhhh
h
ds ds (1 + e11)(1 + e22)

= C12/(1 + e11)(1 + e22) = 2e12/(1 + e11 + e22) + O(ε2)

= 2e12 + O(ε2) , (11)

where we have used (1)3, (5), (9), and (10) in deriving (11).

Let φ be the amount θ differs from a right angle (i.e. φ = −θ + π/2), so that

cos θ = sin φ = φ + O(φ3) . (12)

>From (11) and (12), we see, after neglecting terms of O(ε2), that

φ
φ = 2e12 or e12 = hh . (13)
2

The angle φ is known as the i

iiii angle
shear iiiii. Similar results hold for e13 and e23. Hence: the off-

diagonal components of the infinitesimal strain tensor are seen to represent half of the change in

angle between two line elements initially along the corresponding coordinate axes.

Recall that dv=JdV, where J = det(xi,A). In the class notes Section I/7, it was shown that

neglecting terms of O(ε2) as ε → 0 gives

J = 1 + ekk . (14)
 - 133 -

Thus, dv = JdV = dV + ekkdV, or

hdv − dV
hhhhhh = ekk . (15)
dV

The invariant ekk is known as the i iiiii. Equation (15) implies that the trace of the
dilation

infinitesimal strain tensor measures the local change in volume per unit volume.
 - 134 -

Supplement to Part III, Section 2

Isotropic tensors.

Under a rotation of coordinate system, recall that the components of a vector change

according to the rule

xi = aimx′m or x′m = ajmxj , (1)

where xi are the components of a vector x with respect to one coordinate system and x′m are the

components of the same vector with respect to a different coordinate system. The coefficients

aim are components of an orthogonal tensor such that

aijaik = ajiaki = δjk and det(aij) = ± 1 . (2)

Also recall that the components Tij...k of a tensor T transform according to the rule

Tij...k = aipajq . . . akrT′pq...r , (3)

where aij’s obey equation (2).

iiiiiii if its components are invariant to rotation of the coor-

Definition: A tensor T is called isotropic

dinate system. Thus, if T is isotropic, then

Tij...k = T′ij...k ,

or from (3),

Tij...k = aipajq . . . akrTpq...r (4)

for all orthogonal aij.

Theorem 1: A scalar invariant is an isotropic tensor of order zero.

Proof: Let φ = φ̂(xi) be a scalar invariant, so that from (3) φ̂(xi) = φ̂(x′i) = φ′. Since φ = φ′, the

scalar invariant φ satisfies the definition of an isotropic tensor given by (4).

Theorem 2: The only isotropic tensor of order one is the zero vector.
 - 135 -

Proof: If ti are the components of an isotropic vector t, then from (4)

ti = ajitj (5)

for all components aji of orthogonal tensors. Consider, for instance, the choice

R−1 0 0H
R H J
a = 0 −1
Q ji P J
0 JJ .
Q 0 0 1P

With this choice for aji, (5) becomes (t1 , t2 , t3) = (−t1 , −t2 , t3), which implies that t1 = 0

and t2 = 0. Now consider the choice

R1 0 0 H
R H J
a = 0 −1 0 J .
Q ji P J J
Q0 0 −1 P

for which equation (5) becomes (t1 , t2 , t3) = (t1 , −t2 , −t3), and so we must have t3 = 0

also. Thus, if t is an isotropic tensor of order one, then t = 0.

Theorem 3: Every isotropic tensor of order two is a scalar multiple of δij.

Proof: Let Tij be the components of an isotropic tensor T, so that from (4)

Tij = aipajqTpq (6)

for all orthogonal tensor components aij. Consider now the following choices for aij and the

resulting forms of equation (6):

RT T T H R T T31 −T32 H
R 0 0 −1 H J 11 12 13 J J 33 J
R H J
a = −1
Q ij P J
0 0 J so that JT21 T22 T23 J = J T13 T11 −T12 J ,
J
J J J J
Q−T23 −T21 T22 P
Q 0 1 0P
QT31 T32 T33 P

and

RT T T H R T −T H
R0 0 −1 H 31 T32
J 11 12 13 J J 33 J
R H J J
a = 1 0 0 so that JT T T J = J−T T −T 12 J . (7)
Q ij P J J 21 22 23 13 11
J J J J
Q0 −1 0 P
QT31 T32 T33 P Q T23 −T21 T22 P
 - 136 -

In order for (7)1 and (7)2 to both be satisfied, we must have

T11 = T22 = T33 and T12 = T21 = T13 = T31 = T23 = T32 = 0 ,

or

Tij = λδij , (8)

where

λ = T11 = T22 = T33 .

Theorem 4: Every isotropic tensor of order three is a scalar multiple of the permutation tensor

εijk.

Proof: Let Tijk be an isotropic third order tensor, so that from (4) we can write

Tijk = aipajqakrTpqr (9)

for all orthogonal aij. We can represent Tijk as

RT T T T121 T122 T123 T131 T132 T133 H

J 111 112 113 | | J
R H
T = JT T T T T T T 231 232 233 J .
T T
Q ijk P 211 212 213 221 222 223
J | | J
QT311 T312 T313 T321 T322 T323 T331 T332 T333 P

Consider the following choices for aij and the resulting forms of (9):

R−T −T331 T332 −T313 −T311 T312 T323 T321 −T322 H

R 0 0 −1 H J 333 J
R H J | |
0 0 J then RTijk P= J−T133 −T131 T132
H
a = −1 −T113 −T111 T112 T123 T121 −T122 J
Q ij P J J Q
J | | J
Q T233 T231 −T232 T213 T211 −T212 −T223 −T221 T222 P
Q 0 1 0P

R T T331 T332 T313 T311 T312 T323 T321 T322 H

R0 0 1 H
J 333 | | J
R H J J R H
a = 1 0 0 then Tijk P= J T133 T131 T132 T113 T111 T112 T123 T121 T122 J
Q ij P J J Q
J | | J
Q0 1 0 P
Q T233 T231 T232 T213 T211 T212 T223 T221 T222 P

R T T332 −T331 T323 T322 −T321 −T313 −T312 T311 H

R 0 0 1H
J 333 | | J
R H J J R H
a = 0 1 0 then T = J T233 T232 −T231 T T −T −T −T T J.
Q P J
ij J Q ijk P 223 222 221 213 212 211
J | | J
Q−1 0 0 P
Q−T133 −T132 T131 −T123 −T122 T121 T113 T112 −T111 P
(10)
 - 137 -

>From equation (10), we must have

T333 = T331 = T332 = T111 = T112

= T122 = T133 = T211 = T233 = T311 = T322 = T121 = T131 = T212 =
= T232 = T313 = T323 = 0

and

T123 = T231 = T312 = −T321 = −T132 = −T213 ≡ λ ,

thus we can write

Tijk = λεijk . (11)

Sufficiency may also be shown by substituting equation (11) into equation (3), giving

aipajqakrTpqr = aipajqakrλεpqr = λ(det apq)εijk

which is equivalent to equation (9).

Theorem 5: The most general fourth order isotropic tensor has the form

Tijkl = λδijδkl + µδikδjl + γδilδjk .

Proof: Let Tijkl be the components of an isotropic tensor, so that from (4)

Tijkl = aipajqakralsTpqrs (12)

for all orthogonal aij. We proceed as in the previous proofs, choosing aij to be

R−1 0 0 H R1 0 0 H R0 0 1 H R0 0 −1 H
R H J J J J J
a = 0 −1 0 , 0 −1 0 , 1 0 0 , J0 1 0 J .
J
Q ij P J J J J J J J J
Q 0 0 1P Q0 0 −1 P Q0 1 0 P Q1 0 0 P

Using the above choices for aij in equation (12), we can show that the non-zero values of

Tijkl are related as follows:

T1111 = T2222 = T3333

T1122 = T2211 = T1133 = T3311 = T3322 = T2233 = λ
T1212 = T2121 = T1313 = T3131 = T2323 = T3232 = µ
T1221 = T2112 = T1331 = T3113 = T2332 = T3223 = γ . (13)
 - 138 -

Now consider the choice

R 1/√2
d 1/√2
d 0H
R H J
a = −1/√2d 1/√2d
Q ij P J
0 JJ .
Q 0 0 1P

If we substitute this value for aij into (12) and set i = j = k = l = 1, we find that

1
T1111 = hh (T1111 + T2211 + T2121 + T1221 + T2112 + T1212 + T1122 + T2222) ,
4

or using (13),

T1111 = λ + µ + γ . (14)

>From the restriction (13) and (14), we see that Tijkl can be represented as

Tijkl = λδijδkl + µδikδjl + γδilδjk . (15)

Sufficiency may be shown by substituting the expression for Tijkl given by equation (15)

into equation (4), yielding

aipajqakralsTpqrs = aipajqakrals(λδpqδrs + µδprδqs + γδpsδqr)

= λaipajpakralr + µaipajqakpalq + γaipajqakqalp
= λδijδkl + µδikδjl + γδilδjk = Tijkl

which is identical to equation (12). If an isotropic fourth order tensor is restricted to have

the symmetry

Tijkl = Tjikl (16)

(i.e. symmetry in the first two indices), then from equation (13) we see that γ = µ. Equation

(15) may thus be written for this special case as

Tijkl = λδijδkl + µ(δikδjl + δilδjk) . (17)

Equation (17) is the most general form for an isotropic fourth order tensor symmetric in i

and j. Notice that the representation given in equation (17) also possesses the additional

symmetries given by
 - 139 -

Tijkl = Tijlk = Tklij . (18)

 - 140 -

Supplement to Part III, Section 4

Material symmetry of elastic solids (general considerations).

For a homogeneous elastic solid, the constitutive equation of interest expresses the strain

energy per unit mass as a function of some measure of strain. Thus, let the strain energy per unit

mass ψ be given by

ψ = ψ̂(EAB) . (1)

We examine the behavior of (1) under a change of coordinates in the reference configuration

given by

XP = APQX′Q ,

where

APQ = EP . E′Q and APM AQM = AMP AMQ = δPQ . (2)

Under the transformation (2), EAB becomes

EPQ = APM AQN E′MN , (3)

where E′MN are the components of the strain tensor referred to the base vectors e′A. Since ψ is a

scalar invariant, we know that

ψ′ = ψ . (4)

With the use of (1) and (3), the strain energy function after the transformation (2) can be written

as

ψ′ = ψ̂(APM AQN E′MN) , (5a)

where as before

ψ = ψ̂(EPQ) . (5b)

Substituting (5) into (4), we conclude that

ψ̂(EPQ) = ψ̂(APM AQN E′MN) (6)

 - 141 -

hh
where ψ is in general a different function than ψ̂. It is clear from equation (6) that the response

of an elastic solid depends, in general, on the coordinates used to describe the reference

configuration. The symmetry of a homogeneous elastic solid is characterized by the set of

orthogonal coordinate transformations which leave the strain energy function form-invariant;
hh
that is, the set of APQ such that ψ is the same as ψ̂, or such that

ψ̂(EPQ) = ψ̂(E′PQ) . (7)

If equation (7) holds for all possible orthogonal coordinate transformations (i.e. for all orthogo-
iiiiiii.
nal tensors APQ), then the material is said to be isotropic

Linear elastic solids

It was shown in the class notes that, in the linear theory, the strain energy function for a

homogeneous elastic solid may be expressed in the form

1
ρoψ = hh CABMNEABEMN , (8)
2

where CABMN is a fourth order tensor of material constants which has, in general, 81 indepen-

dent components. Since EAB is a symmetric tensor, the coefficients CABMN in (8) will have the

obvious symmetries

CABMN = CBAMN = CABNM . (9)

In addition, recalling that the strain energy is a scalar invariant we may write (after changing the

dummies AB into MN and MN into AB)

1 1
ρoψ = hh CABMNEABEMN = hh CMNABEMNEAB . (10)
2 2

It then follows that CABMN must by symmetric in the pair of indices (AB) and (MN), i.e.,

CABMN = CMNAB . (11)

In summary, the fourth order tensor in (8) must satisfy the symmetries

CABMN = CBAMN = CABNM = CMNAB . (12)

 - 142 -

The symmetry of CABMN shown in (12) allows us to reduce the number of independent com-

ponents of CABMN from 81 to 21. The remaining independent components are

C1111 C1122 C1133 C1123 C1113 C1112

C2222 C2233 C2223 C2213 C2212
C3333 C3323 C3313 C3312
(13)
C2323 C2313 C2312
C1313 C1312
C1212 .

Linear elastic solids with symmetry

As mentioned previously, the symmetry of an elastic solid is characterized by the set of

orthogonal tensors APQ such that (7) holds. Substituting (8) into (7), we find

1 1
ρoψ = hh CABMNEABEMN = hh CABMNE′ABE′MN . (14)
2 2

Using (3), (14) can be written as

CABMNAAPABQAMRANSE′PQE′RS = CABMNE′ABE′MN . (15)

Switching dummy indices, (15) becomes

(CABMN − CPQRSAPAAQBARMASN)E′ABE′MN = 0 . (16)

Since (16) must hold for all E′AB and since the quantity in parenthesis in (16) is independent of

E′AB, we may conclude that

CABMN = CPQRSAPAAQBARMASN (17)

for all orthogonal APQ which lie in the set which characterizes the symmetry of the material.

Isotropic linear elastic solids

If the material is isotropic, then equation (17) must hold for all orthogonal APQ. Hence,

CABMN in an isotropic tensor which from (12) is symmetric in the first two indices. >From the

analysis given in the Appendix to Section 2 of Part III, we can represent CABMN as
 - 143 -

CABMN = λδABδMN + µ(δAMδBN + δANδBM) . (18)

Eulerian strain

In the linear theory, there is no distinction between Lagrangian and Eulerian strain as is

clear from the expression

EAB = δAiδBjeij + O(ε2) .

Substituting this into (8), we may write

1
ρoψ = hh (CABMNδAiδBjδMkδNl)eijekl + O(ε2) . (19)
2

Now define

Cijkl = CABMNδAiδBjδMkδNl , (20)

so that (19) becomes

1
ρoψ = hh Cijkl eij ekl (21)
2

If the material is isotropic, then from (18)

Cijkl = λδijδkl + µ(δikδjl + δil δjk) . (22)

We will now consider materials with various symmetries and analyze the effect of the different
symmetries on CABMN.

Case I(a) - Symmetry with respect to the X1−X2 plane

Consider a homogeneous elastic solid which in the reference configuration has material

symmetry only with respect to the X1−X2 plane. For a material with such a symmetry, the

mechanical behavior in the X3 direction is the same as that in the −X3 direction (dentoed by X′3

in Figure 1). The coordinate transformation characterizing this particular type of symmetry is

represented schematically in Figure 1 and can be expressed algebraically as

X1 = X′1 , X2 = X′2 , X3
 - 144 -

for which APQ is given by

 - 145 -

R1 0 0 H
R H J
A = 0 1 0J . (23)
Q PQ P J J
Q0 0 −1 P

Thus, a coordinate transformation which reverses the direction of the X3 axis does not alter the

strain energy function. We will show that the number of material constants needed to describe

the behavior of a material with the above symmetry is reduced from 21 to 13.

For the symmetry under consideration, equation (17) must hold for the value of APQ given

in (23), but necessarily for other choices of APQ. Substituting (23) into (17) gives the relation-

ships

C1111 = C1111 C1122 = C1122 C1133 = C1133 C1123 = −C1123

C1113 = −C1113 C1112 = C1112 C2222 = C2222 C2233 = C2233
C2223 = −C2223 C2213 = −C2213 C2212 = C2212 C3333 = C3333
C3323 = −C3323 C3313 = −C3313 C3312 = C3312 C2323 = C2323
C2313 = C2313 C2312 = −C2312 C1313 = C1313 C1312 = −C1312
C1212 = C1212

which implies that

0 = C1123 = C1113 = C2223 = C2213 = C3323 = C3313 = C1312 = C2312 .

The remaining material constants from the set (13) are

C1111 C1122 C1133 0 0 C1112

C2222 C2233 0 0 C2212
C3333 0 0 C3312
(24)
C2323 C2313 0
C1313 0
C1212 .

The material properties of an elastic solid with material symmetry with respect to the X1 − X2

plane are characterized by the 13 constants in (24).

 - 146 -

Case I(b) - Symmetry with respect to the X2−X3 plane

Consider now a solid that has material symmetry with respect to the X2−X3 plane. For such

a material, the mechanical behavior in the X1 direction is the same as the corresponding

behavior in the −X1 direction (denoted by X1 in Figure 2). The coordinate transformation for

which (17) holds, represented schematically in Figure 2, is given by

X1 = −X′1 , X2 = X′2 , X3 = X′3 . (25)

Since the material has symmetry with respect to the X2−X3 plane, a coordinate transformation

which reverses the direction of the X1 axis (such as that in (25)) will not alter the strain energy

function.

Case II - Orthotropy

iiiiiiiiii
A material which is symmetric with respect to two orthogonal planes is called orthotropic.

Consider a homogenous elastic solid which has the material symmetry properties discussed in

I(a) as well as the symmetry properties discussed in I(b) (i.e. a material which is symmetric with

respect to both the X1−X2 plane and the X2−X3 plane). We will show that the number of

material constants needed to describe the mechanical behavior of a material with the above sym-

metries is 9. For the coordinate transformation given in equation (25), APQ is given by

R−1 0 0 H
R H J
A = 0 1 0J .
Q PQ P J J
Q 0 0 1P

Substituting this choice for APQ into (17), we see that the 13 constants in (24) are related by

C1111 = C1111 , C1122 = C1122 , C1133 = C1133 , C1112 = −C1112

C2222 = C2222 , C2233 = C2233 , C2212 = −C2212
C3333 = C3333 , C3312 = −C3312
C2323 = C2323 , C2313 = −C2313
C1313 = C1313 , C1212 = C1212 ,
 - 147 -

so that

0 = C1112 = C2212 = C3312 = C2313 .

The remaining independent constants are

C1111 C1122 C1133 0 0 0

C2222 C2233 0 0 0
C3333 0 0 0
C2323 0 0 (26)

C1313 0
C1212 .

The material properties of a linear elastic solid with material symmetry with respect to the

X1−X2 plane and with respect to the X2−X3 plane are thus characterized by the 9 constants given

in (26). Notice that by comparing (17) and (26), one can see that a material with the two sym-

metries described above is also symmetric with respect to a third plane which is orthogonal with

respect to both the original planes, i.e. the X1−X3 plane.

Case III - Transverse isotropy

A homogeneous orthotropic solid that also has material symmetry in every direction lying

in a certain plane (say, the X1−X2 plane) is said to be transversely

iiiiiiiiiiiiiiiiii
isotropic. For such a material,

a coordinate transformation which alters the X1 and X2 axes by an arbitrary rotation about the X3

axis will not alter the strain energy function. This property implies that equation (17) must hold

for a transformation of the type (also represented schematically in Figure 3)

X′1 = X1 cos α + X2 sin α

X′2 = −X1 sin α + X2 cos α (27)

X′3 = X3 ,

where the angle α is arbitrary. We will show that the number of material constants needed to

describe the mechanical behavior of a material with the above symmetry is 5. The components
 - 148 -

of the orthogonal tensor APQ corresponding to (27) are

Rcos α − sin α 0 H
R H
A = J sin α cos α 0 J . (28)
Q PQ P
Q 0 0 1P

Substituting (28) and (26), since the material is also orthotropic, into (17) gives the relationships

C1111 = C2222 , C1133 = C2233 , C1313 = C2323

1
C1212 = hh (C1111 − C1122) .
2

The remaining independent compenents are

C1111 C1122 C1133 0 0 0

C1111 C1133 0 0 0
C3333 0 0 0
C1313 0 0 (29)

C1313 0
h1h (C − C1122) .
1111
2

The material properties of a linear elastic solid which is transversely isotropic are thus character-

ized by the five constants given in (29).

Case IV - Isotropic material

A homogeneous elastic solid which has material symmetry in the reference configuration

iiiiiiii Thus, equation (17) will remain valid for any

with respect to all directions is said to be isotropic.

proper orthogonal choice of APQ. We will now derive equation (18) using the symmetries which

characterize an isotropic material. We will thus show that only two material constants are

needed to describe the mechanical behavior of a linear elastic isotropic solid.

An isotropic material can be considered as a material transversely isotropic in three mutu-

ally orthogonal planes (such as the X1−X2, X1−X3, and X2−X3 planes). Thus, a coordinate

transformation which alters the X2 and X3 axes by an arbitrary rotation about the X1 axis will not
 - 149 -

alter the strain energy function (see Figure 4a). The same is true with respect to an arbitrary

rotation of the X1 and X3 axes about the X2 axes (see Figure 4b).

Thus, equation (17) must remain valid under the transformations

X′1 = X1
X′2 = X2 cos φ + X3 sin φ (30)
X′3 = −X2 sin φ + X3 cos φ

and

X′1 = X1 cos β − X3 sin β

X′2 = X2 (31)
X′3 = X3 cos β + X1 sin β ,

where φ and β are arbitrary angles, in addition to the transformation (27). For the coordinate

transformation in (30), APQ is given by

R1 0 0 H
R H J
A = 0 cos φ −sin φ J (32)
Q PQ P J J
Q0 sin φ cos φ P

and for the coordinate transformation in (31), APQ is given by

R cos β 0 sin β H
R H J
A =
Q PQ P J
0 1 0 J . (33)
J
Q−sin β 0 cos β P

>From (17), (29), (32), and (33), we can show the relationships

C1122 = C1133 , C1111 = C3333

1
C1212 = C1313 = hh (C1111 − C1122) .
2

The remaining independent constants are given by

C1111 C1122 C1122 0 0 0

C1111 C1122 0 0 0
C1111 0 0 0
µ 0 0 (34)
µ 0
 - 150 -

1
where µ = hh (C1111 − C1122). Letting λ = C1122, equation (34) may be expressed as
2

λ + 2µ λ λ 0 0 0
λ + 2µ λ 0 0 0
λ + 2µ 0 0 0
µ 0 0 (35)
µ 0
µ .

An equivalent way of writing (35) is given simply

CABMN = λδABδMN + µ(δAMδBN + δANδBM) . (36)

Notice that (36) is the same as equation (18) obtained previously.

 - 151 -

Appendix L: Some Results from Linear Algebra

The object of this appendix is to introduce the concept of a tensor in a form which is espe-

cially useful in continuum mechanics. Thus, after a rapid review and summary of basic results

from linear algebra, a tensor is regarded as a linear transformation. Many aspects of the algebra

and the calculus of tensors are subsequently discussed in coordinate-free form and frequently are

represented with reference to rectangular Cartesian basis vectors.

1. Sets.

We use the symbol X to represent a set, class, collection or family which contains elements

or objects x1,x2,.... We write this as

X = {x1,x2,...}. (1.1)

The element x is a member of X, or belongs to X, or is in X, and this is denoted by

x ∈ X. (1.2)

Let X and Y be two sets. The set X is a subset of Y if every x in X is also in Y. This rela-

tion is designated by

X ⊆ Y. (1.3)

Also Y is said to include or contain X, and we write Y ⊇ X. The empty set is contained in every

set and is designated by ∅. If

X ⊆ Y and Y ⊆ X, (1.4)

then the set Y and X are said to coincide or be equal and we write

Y = X. (1.5)
 - 152 -

If X is a subset of Y but without X being equal to Y, i.e., X ⊂ Y, then X is a proper subset

of Y and we write1

X ⊂ Y , Y ⊃ X. (1.6)

It should be clear that X ⊂ Y means that (i) if x ∈ X, then x ∈ Y and that (ii) there exists a y ∈

Y such that y ∈ X.

The union of two sets X and Y comprises all the elements of the two sets and is denoted by

X ∪ Y. (1.7)

The intersection of two sets consists only of those elements which belong to both X and Y and is

denoted by

X ∩ Y. (1.8)

The Cartesian product of two sets X and Y is the set

X × Y = {(x,y) : x ∈ X, y ∈ Y} (1.9)

of ordered pairs. For example, if X = {x1,x2,x3} and Y = {y1,y2}, then X × Y =

{(x1,y1),(x1,y2),(x2,y1),(x2,y2),(x3,y1),(x3,y2)} .

For completeness, we introduce now the following definition of a function: A function f

from the set X to the set Y is a rule which assigns to every element of X a unique element of Y.

It is often displayed as

f : X → Y , x |→ y . (1.10)

hhhhhhhhhhhhhhh
1 Our use of the symbols ⊆ and ⊂ is analogous to the use of the symbols ≤ and < to denote weak and strict ine-
qualities.
 - 153 -

The first of (1.10) indicates that the set X is transformed into Y, while the second of (1.10) indi-

cates that the element x goes to the element y. The function f, or the rule defined by (1.10)1, is

sometimes called an operation or a transformation or a mapping. This definition of a function

rules out multivalued functions.

 - 154 -

2. Vector spaces.

Consider a set V of arbitrary elements u,v,w,...,etc., and admit the following two operations

labeled as (1) and (2):

(1) Let there exist an operation, denoted by "+", which to every pair u,v assigns a unique

element u + v in V and which has the properties:

(A1) u+v=v+u (commutative property).

(A2) u + (v + w) = (u + v) + w (associative property).

(A3) There exists a zero vector o such that

v + o = v .

(A4) For every vector v there is a corresponding negative vector (−v)

such that

v + (−v) = o.

It is customary to write v + (-w) = v - w.

(2) Let there exist another operation, indicated by placing a juxtaposition of an element of

V and a real number α , which to every element v assigns a unique element αv in V and which

has the following properties:

(S1) 1v = v.

(S2) α(βv) = (αβ)v (associative property).

(S3) (α+β)v = αv + βv (distributive property for scalar addition).

(S4) α(v+w) = αv + αw (distributive property for vector addition).

 - 155 -

We then say that V is a vector space over the field of real numbers and that u,v,w,... are

vectors in V. The first operation defined under (1) above is callled vector addition with elements

u + v called the sum of u and v. The second operation defined under (2) is called multiplication

of vectors by a real number.

All the usual algebraic results can be derived from the above axioms. For example, a vec-

tor equation of the form αu = o has the solution α = 0 or u = o.

Vector sub-spaces. A sub-space of a vector space V is any nonempty subset U of V which

is such that if u and v belong to U and α is any real number, then the vectors u + v and αv also

belong to U.
 - 156 -

3. n-Dimensional vector spaces.

p
Let v1,v2,..,vp be any p vectors in a vector space V, p being some integer. The sum Σ αi vi
i=1

is called a linear combination of the p vectors vi with coefficients αi as real numbers. These vec-

tors are said to form a linearly independent set of order p if the only coefficients that satisfy the

equation

α1v1 + α2v2 + . . . + αpvp = o (3.1)

are α1 = α2 = α3 = . . . = αp = 0. A linearly dependent set is one which is not linearly

independent.

Note that every subset of vectors containing the zero vector is linearly dependent.

α1v1 + α2v2 + . . . + 1 o + αpvp = 0.

Consider the set of all systems of linearly independent vectors in the vector space V. There

are two possibilities: Either (1) there exist linearly independent systems of arbitrarily large order
p; then the vector space is said to be an infinite-dimensional space, or (2) the order of the linearly

independent sytstem is bounded. In the second case, there exists an integer n such that the order

p ≤ n and there exist linearly independent systems of order n but not of n + 1. The vector space

V is said to be a finite-dimensional vector space. The number n is termed the dimension of the

vector space and we shall use V n to designate finite n-dimensional vector space; and henceforth
in this Appendix we shall be concerned with finite dimensional vector spaces only. Let

{g1,g2,..,gn} be any such system of order n in a finite dimensional vector space V n ; it will be

called a basis of V n . Thus, we have the following definition:

A basis in a vector space V n is any linearly independent set of n vectors.

Let v be any vector in V n. The set of n + 1 vectors {v,g1,g2,..,gn} is necessarily linearly

dependent, so there exists n + 1 numbers λ,α1,α2,..,αn not all zero such that
 - 157 -

λv + α1g1 + α2g2 + . . . + αngn = o. (3.2)

If v = o, then αigi = o, (no sum on i), and hence αi = 0. For v ≠ o, the number λ must

be different from zero, for otherwise the system gi (i = 1,2,..,n) will not be linearly independent.

Equation (3.2) can, therefore, be solved for v, and there exist numbers v1,v2,...,vn such that

v = v1g1 + v2g2 + . . . + vngn . (3.3)

Thus, the vector v is expressible as a linear combination of the gi . This combination is unique;

for otherwise, there would exist another, and their difference would constitute a linear combina-

tion of the gi equal to the null vector o and with coefficients not all zero. The numbers

(v1,v2,..,vn) are called the components of v with respect to the basis {g1,g2,..,gn} . It is con-

venient to replace (3.3) by the shorter notation

v = vkgk (3.4)

in which the summation convention is used. Whenever an index appears twice in the same term,

a summation is implied over all terms by letting that index assume all its possible values, unless
the contrary is stated. Normally the convention applies to an index which appears once as a sub-

script and once as a superscript; but, as will become apparent later, in a special case it will

suffice to adopt the convention for repeated subscripts. It is convenient to recall here a theorem

the proof of which can be easily found in a book of linear algebra2. A statement of the theorem
is as follows:

For a set of vectors to constitute a basis in V n it is necessary and sufficient that every vector
in V n can be expressed in one, and only one, way as a linear combination of the vectors of that

set.

hhhhhhhhhhhhhhh
2 See a standard book on linear algebra.
 - 158 -

4. Euclidean vector spaces.

Consider a vector space of dimension n defined over the field of real numbers. Suppose

there is an operation, denoted by " . ", which to every pair of vectors u and v assigns a real

number, denoted by u . v, and which has the following properties:

(I1) u.v = v.u .

(I2) u . (v+w) = u . v + u . w .

(I3) (αu) . v = u . (αv) = α(u . v) .

(I4) u.u ≥ 0 and u . u = 0 => u = o .

The above operation (known as a rule of composition) is called the dot product or the scalar pro-

duct or the inner product of two vectors. A vector space V n obeying the rules of composition
(I1) to (I4) is a real inner product space and is called a Euclidean vector space3 E n .

The magnitude or the norm of the vector v is defined by

c c v c c = (v.v) ⁄2 .
1
(4.1)

If c c v c c = 1, the vector is said to be normalized. By (I4) c c v c c 2 is always positive if v ≠

o and it vanishes when v = o , so that its square root is real.

The norms of two vectors and the magnitude of their scalar product satisfy the Schwarz ine-

quality

hhhhhhhhhhhhhhh
3 This terminology stems from the fact that once a scalar product is admitted, all the standard theorems of
Euclid’s geometry can be established by an appeal to the properties of an inner product space.
 - 159 -

cu.vc ≤ ccucc ccvcc . (4.2)

To prove this, consider the vector αu + v where α is an arbitrary real number. Then,

c c (αu + v) c c 2 = α2 c c u c c 2 + 2αu . v + c c v c c 2 .

The left-hand side of the last expression is positive or zero for all real α and this implies

(u . v)2 ≤ c c u c c 2 c c v c c 2 ,

which is equivalent to (4.2). In view of (4.2), we define the angle between two nonzero vectors

u and v by

u.v
cos θ = hhhhhhhhh , (4.3)
ccucc ccvcc

where 0 ≤ θ ≤ π.

Vectors u and v are said to be orthogonal if

u.v = 0 . (4.4)

A set of vectors in E n is said to be orthonormal if all the vectors in the set are normalized
and mutually orthogonal. Let ei be a system of orthonormal basis in E n . It follows that ei must

satisfy the conditions

ei . ek = δik , (4.5)

where δik (i,k = 1,2,...,n) is the Kronecker delta defined by

I0ifi≠k,
δik = K (4.6)
L1ifi=k.

Every orthonormal set is necessarily linearly independent. An infinity of bases exist which are

orthonormal; and in this appendix a typical orthonormal basis will be denoted by ei . Any vector
 - 160 -

v in E n, when referred to the basis ei, can be expressed in the form

n
v = Σ viei . (4.7)
i=1

Then, by taking the scalar product of both sides of (4.7) with ek , we have

vk = v . e k . (4.8)

The numbers vk are the components of v with respect to the orthonormal basis ek and the square

of the magnitude of v is

n
ccvcc = v.v =
2
Σ vivi . (4.9)
i=1
 - 161 -

5. Points.

We denote points by boldface letters x, y, z, etc. Given any two points x and y , we associ-

ate with them a point difference y − x which is a vector v in E n and which we denote by

v = y−x (5.1)

satisfying the following rules:

(P1) (y − x) + (x − z) = (y − z) .

(P2) Given any point x and any vector v, there is a unique point y such

that y − x = v .

The point y determined in rule (P2) is denoted by

y = x + v = v + x . (5.2)

In view of (5.1) and (5.2), it is meaningful to speak of the difference of two points which is

a vector, and of the sum of a point and a vector , which is a point. In particular,

x − x = o (5.3)

holds for every point x .

The distance between two points x,y is

c x − y c = {(x − y) . (x − y)}1/2 . (5.4)

 - 162 -

The set of all points associated with a Euclidean vector space E n is called Euclidean Point

Space.4

hhhhhhhhhhhhhhh
4 Euclidean vector and point spaces are abstract concepts which at this point in our discussion are unrelated to or-
dinary vectors and points of elementary geometry. Geometrical representations of these abstract concepts will be
discussed in Section 6.
 - 163 -

6. Geometric space.

Consider the space of elementary geometry. This space is composed of elements P,Q,R,

called points. Choose O to be a reference point or origin. The directed line segments
→ →
OP, OQ,... may be associated with points x,y,... of Euclidean point space (Section 5) and the
→ →
directed line segments PQ, QR may be associated with vectors u,v,... in Euclidean vector space

(Section 4). Thus we set

→ → →
x = OP , y = OQ , z = OR ,... (6.1)

and

→ → →
u = PQ , v = QR ,... o = PP . (6.2)

The properties (A1)−(A4) of abstract vectors and the properties (P1),(P2) of points have their

immediate geometrical representation in our geometric space. Thus, the class of all line seg-
→
ments with the same direction and length as PQ, called a geometrical vector or free vector,
→
represent vectors, and directed segments such as OP, called position vectors, represent points.

For example, we have the properties

→ → → → →
PQ = −QP , PQ + QR = PR .

Next, the properties (S1)−(S4) have their geometric representation in our space of elemen-

tary geometry. Multiplication of a vector by a scalar corresponds to multiplication of a line ele-

→
ment PQ by a scalar α and is represented by a line element of length c α c c PQ c , where c PQ c is
→ →
the length of PQ in the same or opposite direction as PQ according to whether α is positive or

negative.

The rules (I1)−(I4) of Section 4 have their representation in the geometrical space. Recal-

ling also (4.3), these rules are represented by the scalar product of two geometrical vectors as

→ →
PQ . RS = c PQ c c RS c cos θ , (6.3)
 - 164 -

→ →
where θ is the angle between PQ and RS. Thus, geometrical vectors give a geometrical

representation of a Euclidean point space with its associated Euclidean vector space. Geometri-

cal space has three dimensions.

Given a geometrical space, we select an origin O and a (constant) basis {fi} of the associ-

ated vector space. These are said to constitute a coordinate system consisting of origin and a

basis (O,fi) for our geometrical space. Any point x referred to the basis fi can be expressed in
h
terms of its components xi by the formula

3
hi
x = Σ x fi , (6.4)
i=1

h
where now summation is over i = 1,2,3. The components xi are called rectilinear coordinates of

x. If (fi) is identified with a constant orthonormal basis ei, then referred to ei the point x can be

represented as

3
x = Σ xiei (6.5)
i=1

and xi are called Cartesian rectilinear coordinates of x.

We may select a different basis gi (say) at each point x in our geometric space so that

gi = gi(x). This situation arises when we consider curvilinear coordinates in geometric space.
 - 165 -

7. Indicial notation.
 - 166 -

8. Linear transformation. Tensors.

A linear transformation or tensor T is an operation which assigns to each vector v in V n

another vector Tv in V m such that5

T(v+w) = Tv + Tw , T(αv) = αTv (8.1)

for all v,w in V n. Thus, we have T: Vn → Vm .

We denote the set of all possible tensors which transform vectors in V n linearly into vec-
tors in V m by L(V n,V m) and we define the sums of two tensors and the scalar multiples of ten-

sors by

(T+S)v = Tv + Sv , (αT)v = α(Tv) . (8.2)

The transformations T + S and αT defined in (8.2) obey the rules (8.1), i.e., they are tensors.

Also the sums and scalar multiples defined in L(V n,V m) obey the rules (A1)−(A4) and (S1)−(S4).

Hence the set of all tensors constitutes a vector space. The zero element in L(V n,V m), i.e., the
element which is to be substituted for o in rules (A3) and (A4) is the transformation which

assigns the zero vector to every vector. We call this transformation the zero tensor and denote it

by O. In other words, the zero tensor O is such that

Ov = o (8.3)

for all vectors v.

The identity tensor can only be defined for transformations from V n to V n. Thus, when the
dimension m = n, we define the identity or the unit tensor I by

Iv = v (8.4)

hhhhhhhhhhhhhhh
5 This special definition is particularly useful in the present Appendix. A more general definition of a tensor is
given in Appendix C.
 - 167 -

for all vectors v in V n.

We associate with Euclidean vector spaces E n and E m a vector space of dimensions nm

denoted by E n b× E m called the tensor product space of E n and E m, and defined in the follow-

ing way: If a ∈ E n and b ∈ E m, then a b× b is an element of E n b× E m such that

(a b× b)v = a(b . v) (8.5)

for all vectors v ∈ E m. It is seen that a b× b, which sometimes is denoted as ab, obeys the rules
(8.1) and hence is a tensor, i.e., it assigns to each vector v ∈ E m the vector a(b . v) ∈ E n.

Sums and scalar multiples of such tensors can then be calculated by the rules laid down for ten-

sors in (8.2). Not all tensors generated in this way can be expressed as an element of space of

the form a b× b. The zero tensor in E n b×E n is defined by the tensor o b× o.

Consider now the tensor product space E n b× E n of dimension n2. Let ei be an orthonormal

basis in Euclidean vector space E n. Then,

ei b× ek (8.6)

is a basis in E n b× E n. To prove this, suppose αik are a set of n2 numbers such that

αikei b× ek = O .

Then,

(αikei b× ek)er = Oer = o .

With the help of (8.5) this last equation becomes

αikeiδkr = αirei = o .

Since ei is a basis in E n it follows that

 - 168 -

αik = 0 .

Hence ei b× ek are linearly independent and form a basis in E n b× E n. When u and v are vectors

in E n so that

u = uiei , v = viei ,

then

uv = u b× v = uivkei b× ek . (8.7)

Let T be a tensor in L(E n,E n) and let tik denote the components of the vector Tek with respect

to the basis ei in E n so that

Tek = tikei , tik = Tek . ei . (8.8)

Associated with each tensor T in L(E n,E n) we have the ordered array of n2 scalars tik which

combine by the rules (A1)−(A4) and (S1)−(S4) for vector spaces. Hence, L(E n,E n) is a vector

space of dimension n2. But, ei b× ek is a linear transformation in the space L(E n,E n) and hence

also forms a basis in L(E n,E n) which is identical to E n b× E n. Moreover, with the use of (8.8),
for any vector v ∈ E n we have

(T−tikei b×ek)v = Tekvk − tikeivk = o .

Hence, we conclude that

T = tikei b× ek . (8.9)

The n2 numbers tik in (8.9) are called the components (or Cartesian components) of T with

respect to the basis ei b× ek .

Similarly, we write

I = δikei b× ek (8.10)
 - 169 -

so that

Iv = v (8.11)

for all vectors v in E n. The tensor I with components δik is the unit tensor in E n b× E n or

L(E n,E n).

The system of numbers tik in (8.8) is usually regarded as an array of numbers, a matrix,

having n rows and n columns and denoted by [tik] or {tik}.

Let T be a tensor in E n b× E n. The transpose of a tensor T is a tensor TT defined by

w . Tv = v . TTw (8.12)

for all vectors v and w in E n. The transpose TT satisfies (8.1) and is a tensor, and the opera-
tion which associates TT with T is called transposition. From (8.12) it may be shown that

(TT)T = T , (T+S)T = TT + ST , (αT)T = αTT , (8.13)

so that transposition is a linear operation. Further, it follows from (8.9) and (8.12) that

[tik] = ei . Tek = ek . TTei = [tik]T ,

where [tik]T are components of the tensor TT.

A tensor T is said to be symmetric if

TT = T . (8.14)

Also, by (8.9) and (8.14) we have

tki = tik . (8.15)

Also, from (8.12) and (8.14) it follows that if T is symmetric, then

w . Tv = v . Tw (8.16)
 - 170 -

for every v,w in E n and conversely.

A tensor T is said to be skew or anti-symmetric (skew- symmetric) if

TT = −T (8.17)
or

tki = −tik . (8.18)

Every tensor T in E n b× E n can be written as

T = TS + TA , (8.19)

where

1 1
TS = hh (T+TT) − (TS)T , TA = hh (T−TT) = −(TA)T . (8.20)
2 2
 - 171 -

9. Multiplication of tensors.

Let T and S be two tensors in L(V n,V n). The product TS of two tensors T and S is the
composition of T and S , i.e., TS defined by the requirement that

(TS)v = T(Sv) (9.1)

hold for all vectors v in V n. It is seen that TS satisfies (8.1) and is a tensor in L(V n,V n). The
following rules may be established

(M1) (TS)R = T(SR) ,

(M2) T(R+S) = TR + TS ,

(M3) (R+S)T = RT+ST ,

(M4) α(TS) = (αT)S = T(αS) ,

(M5) IT = TI = T ,

where T,S,R are tensors in L(V n,V n).

For tensors T,S in L(E n,E n) it follows from (6.9) and (8.1) that

(TS)v = T(smkvkem) = timsmkvkei

for all vectors v ∈ E n so that

TS = timsmkei b× ek (9.2)

and timsmk are the components of TS with respect to the basis ei b× ej. If {tik} and {sik} are

matrix representations of the tensors T and S, respectively, then

{tik}{sik} = {timsmk} . (9.3)

 - 172 -

Note that

(TS)T = STTT . (9.4)

The commutative law is not, in general, satisfied, i.e., TS is not the same as ST.

The truth of the following formulae can be readily verified:

(a b× b)T = b b× a ,

T(a b× b) = Ta b× b ,
(9.5)
a b× (T b) = (a b× b)T ,
T

(a b× b)(c b× d) = (b . c)a b× d .

The tensor T is invertible if, for every choice of w, the equation w = Tv can be solved

uniquely for the vector v. If T is invertible, then v is unique and we write v = T−1w. The

transformation T−1 obeys the rules (8.1) and is a tensor called the inverse of T. If the inverse

T−1 exists, then

(TT−1−I)w = o , (T−1T−I)v = o

for every choice of w and every choice of v so that

TT−1 = T−1T = I , (T−1)−1 = T . (9.6)

If S and T are invertible tensors in L(V n,V n) and if w = (TS)v, it follows that

Sv = T−1w , v = S−1T−1w .

Hence, the product TS is invertible and

(TS)−1 = S−1T−1 . (9.7)

 - 173 -

If the only solution of the equation Tv = o is v = o, then T is nonsingular; otherwise it

is singular. If T is invertible, then T is nonsingular. It can be shown that a tensor T is invertible

or nonsingular if and only if det T ≠ 0.

A tensor Q is orthogonal if it is invertible and if the inverse coincides with its transpose,

i.e.,

Q−1 = QT or QQT = QTQ = I . (9.8)

A tensor Q is orthogonal if and only if it preserves inner products in the sense that

(Qu) . (Qv) = u . v (9.9)

for every vector u,v in E n.

The tensor T is positive semidefinite if it is symmetric and if v . Tv ≥ o for all vectors v in

E n. If v . Tv > o when v ≠ o, then T is positive definite.

An operation tr which assigns to each tensor T in E n b× E n a number tr T is a trace if

tr(S+T) = tr S + tr T , tr(αT) = α tr T , (9.10)

tr a b× b = a . b , (9.11)

for all vectors a,b in E n and all scalars α. The tr operation is a linear function and the follow-
ing consequences are clear:

tr T = tikei . ek = tii . (9.12)

Also,

tr TT = tr T , tr(TS) = tr(ST) ,
(9.13)
tr I = n , tr T = tr T ,
S
tr T A
= 0 ,
 - 174 -

where n in (9.12)3 is the dimension of the space.

If one defines the inner product of two tensors T,S to be the number

T . S = tr(TST) , (9.14)

then the rules (I1)−(I5) are satisfied. Hence with tr(STT) as the inner product, the space of all

tensors is an inner product space. We define the magnitude of T as

| T | = √tr(TT
dddddd
T
) . (9.15)
 - 175 -

10. Change of basis.

h
Let ei and ei be two arbitrary orthonormal bases in E n. Each vector in one basis may be

expressed in terms of those of the other basis by a nonsingular transformation. Thus

h
ei = Aei = akiek ,
(10.1)
hh h h h hh
−1
ei = A ei = akiek , A = A ,

where

h . hhh
ek Aei = δik = ei . Aek ,
(10.2)
h h hh
aki = ei . ek = aik , A = A . T

Hence, A−1 = AT and A is orthogonal.

If v is an arbitary vector and

hh
v = viei = viei , (10.3)

h
then the components of v in the two bases ei and ei are related by the equations

h h
vi = akivk , vi = aikvk . (10.4)

Also

hh hh
v . v = vivi = vivi , u . v = uivi = uivi , (10.5)

and the scalar product has the same form in terms of the components ui,vi as it has in terms of
h h
the components ui,vi and is called a scalar invariant.

Again, if T is an arbitrary tensor in E n b× E n and

h h h
T = tikei b× ek = tikei b× ek , (10.6)
 - 176 -

h
then the components of T in the two bases ei and ei are related by

h h
tik = ariasktrs , tik = airakstrs . (10.7)

If T and S are two tensors in E n b× E n and

h h h
S = sikei b× ek = sikei b× ek ,

then

hh
T .S = tirsir = tirsir ,
(10.8)
hh
T . T = tirtir = tirtir

are scalar invariant forms.

 - 177 -

11. Point, vector and tensor functions.

The distance between two points x,y is defined in (6.4). The distance vanishes if and only

if x = y. This idea of distance can be used to define limits, convergence, continuity, etc. For

example, we say that

lim xn = x if lim | xn−x | = O . (11.1)

n→∞ n→∞

.
Suppose z(t) is a function of a real variable t called a point function. The derivative z(t),

if it exists, on an open subset of the real numbers is defined by

. dz z(t+∆t) − z(t)
z(t) = hhh = lim hhhhhhhhhhhh . (11.2)
dt ∆t→∞ ∆t

Such a derivative is a vector-valued function.

In a similar way we may define limits, derivatives, etc., of vector- and tensor-valued func-

tions v,T of t using the appropriate definitions (1.2) and (9.15) for distance. The derivative of a

vector-valued function of t is a vector function and the derivative of a tensor-valued function is a

tensor function. The usual rules of differential calculus are easily extended so as to apply to

point, vector or tensor functions. For example, if T = T(t) and S = S(t) are tensors, then

.
hhh . .
TS = TS + TS , (11.3)

where, of course, the correct order must be maintained for the tensors.

If Q = Q(t) is an orthogonal tensor function, then from (9.8) and (11.3),

. .
hhh
QQ + QQT = O .
T
(11.4)
. hhh
Since QT = QT it follows that

. .
Ω = Q QT = − Q QT (11.5)
 - 178 -

is a skew tensor.

If φ(x) is a scalar function of x defined on some open region U of Euclidean point space,

then φ is differentiable on U if there is a vector field w(x) on U such that

| φ(x) − φ( x) − w( x) . (x − x) |
o o o
lim hhhhhhhhhhhhhhhhhhhhhhhhhhh = 0 (11.6)
x→ox | x−ox |

for every x in U. If this is so, w is unique. We call it the gradient of φ and write
o

∂φ
w = grad φ(x) = hhh . (11.7)
∂x

Alternatively, the gradient of φ(x) may be defined by means of

φ(x+βu) − φ(x) d ∂φ
lim hhhhhhhhhhhhh = hhh φ(x + βu) c = hhh . u = w . u (11.8)
β→0 β dβ ∂x
c β=0

for all vectors u ∈ E n. It may be shown that this definition of gradient is equivalent to that
given in (11.6).

Sometimes we use the symbol ∇ to denote the gradient operator:

∂ ∂
∇ = hhh = er h hh (11.9)
∂x ∂xr

if x = xrer and er is an orthonormal basis. From (11.6) it follows that

∂φ ∂φ
w = grad φ(x) = hhh = ∇φ = ei hhh . (11.10)
∂x ∂xi

If v(x) is a vector function of points x in some open region U of Euclidean space, then

v(x) is said to be a vector field. The gradient of a vector field v(x) is a tensor field T(x)

defined by
 - 179 -

| v(x) − v(ox) − T(ox)(x−ox) |

lim hhhhhhhhhhhhhhhhhhhhhhhhhh = o (11.11)
x→ox | x−ox |

for every ox in U. We write this as

∂v
T = grad v(x) = hhh . (11.12)
∂x

If v = viei, then it follows from (11.12) that

∂vi
T = hhhh ei b× ek (11.13)
∂xk

∂vi
so that hhhh are the components of the grad v with respect to the orthonormal basis ei. It can
∂xk

be shown that

{grad v(x)}Tc = grad{c.v(x)} (11.14)

for every constant vector c. Equation (11.14) could be taken as the defining equation for grad

v(x).

The divergence of the vector field v(x) is a scalar field defined by

∂v ∂vi
div v(x) = tr{grad v(x)} = ∇ . v = ei . hhh = hhh . (11.15)
∂xi ∂xi

The divergence of a tensor field T(x) is the vector field div T(x) for which

c . div T(x) = div{TT(x)c} (11.16)

for every constant vector c. When

T = tijei b× ej , (11.17)
 - 180 -

it follows that

∂tij
div T(x) = hhh ei . (11.18)
∂xj

If P is a bounded region whose boundary ∂P is sufficiently well-behaved, then by applica-

tion of the divergence theorem (Green’s theorem) to a vector v and a tensor T we have

∫ div v(x)dV(x) = ∫ v(x) . n(x)dA(x) , (11.19)

P ∂P
∫ div T(x)dV(x) = ∫ T(x)n(x)dA(x) , (11.20)
P ∂P
where n is the outward unit normal to ∂P, the integrations are over the volume P and the boun-

dary surface ∂P and dV and dA represent elements of volume and area, respectively. In com-

ponent form, (11.19) and (11.20) read:

∫ vi,i dv = ∫ vi ni da , (11.21)
P ∂P
∫ tij,j dv = ∫ tij nj da . (11.22)
P ∂P
 - 181 -

12. Vector product. Axial vectors. The curl operator.

Consider a Euclidean vector space E3 of three dimensions. If u,v,w are any vectors in E3 ,

the vector product of u and v is a vector in E3 denoted by u × v and defined by the properties:

(V1) w.(u×v) = u.(v×w) = v.(w×u) and is a scalar denoted by

[uvw] and called the scalar triple product ,

(V2) u × v = −v × u ,

(V3) | u×v | = | u | | v | sin θ ,

where θ is defined in (4.3). From (V1) and (V2), it follows that

u×u = o , u.(u×v) = 0 , v.(u×v) = 0 (12.1)

so that u × v is orthogonal to both u and v. It can be shown that three nonzero vectors u,v,w

are linearly independent if and only if their scalar triple product [u,v,w] = u.(v×w) = (u×v).w.

The following properties may be deduced from the definitions:

(αu) × v = α(u×v) = u × (αv) ,

u × (v+w) = u × v + u × w ,

(u+v) × w = u × w + v × w , (12.2)

u × (v×w) = v(u.w) − w(u.v) = (v b× w − w b× v)u ,

[u + x,v,w] = [uvw] + [xvw] .

The above definition of vector product does not restrict ei to be either right-handed or left-

handed basis.
 - 182 -

Let e1,e2,e3 be an orthonormal system. Then, in view of (V3) and (12.1),

e1 × e2 = ± e3 . (12.3)

A right-handed orthonormal sytem is one for which

e1 × e2 = e3 (12.4)

and a left-handed orthonormal system corresponds to the choice of the - sign in (12.3). For a

right-handed orthonormal system it follows that

[e1e2e3] = 1 , ei × ej = eijkek , eijk = [eiejek] , (12.5)

where the components eijk are known as the permutation symbol or the components of the alter-

nating tensor. Also, for any two vectors

u = uiei , v = vjej ,
(12.6)
u × v = eijkuivjek .

Let TA be a skew tensor in E3 so that

1
TA = tikei b× ek = hh tik[ei b× ek − ek b× ei] . (12.7)
2

Hence, if u is an arbitrary vector in E3

1 1
TAu = hh tik[ei(u.ek) − ek(u.ei)] = hh u × [tikei × ek] , (12.8)
2 2

if we use (12.2)4. Hence

TAu = tA × u , (12.9)
 - 183 -

where the vector

1
tA = hh tkiei × ek . (12.10)
2

Let ei be a right-handed orthonormal basis. Then, referred to ei, TA reads

TA = tijei b× ej (12.11)

and the axial vector tA with components tkA assumes the form

1
tA = hh tjieijkek = tkA ek . (12.12)
2

Alternatively

tji = eijktkA . (12.13)

If

∂v ∂v
TA = hhh − ( hhh )T , (12.14)
∂x ∂x

the corresponding axial vector tA is called the curl of v and written as

tA = curl v . (12.15)

In a right-handed orthonormal system

∂vj
curl v = eijk hhh ek . (12.16)
∂xi
 - 184 -

References for Appendix L

Gel’fand, I. M. 1961 Lectures on linear algebra (translated by A. Shenitzer from the 2nd Rus-
sian edition). Dover, New York.

Halmos, P. R. 1974 Finite dimensional vector spaces. Springer-Verlag, New York.

Hoffman, K. & Kunze, R. 1971 Linear algebra (2nd edition). Prentice-Hall, Inc., Englewood
Cliffs, New Jersey.

Stecher, M. 1988 Linear algebra. Harper & Row, New York.