Professional Documents
Culture Documents
Published by
World Scientific Publishing Co. Pte. Ltd.
5 Toh Tuck Link, Singapore 596224
USA office: 27 Warren Street, Suite 401-402, Hackensack, NJ 07601
UK office: 57 Shelton Street, Covent Garden, London WC2H 9HE
For photocopying of material in this volume, please pay a copying fee through the Copyright Clearance
Center, Inc., 222 Rosewood Drive, Danvers, MA 01923, USA. In this case permission to photocopy
is not required from the publisher.
Printed in Singapore
To the memory of Charles Steele,
eminent teacher, researcher, and mentor
This page intentionally left blank
Preface to the Revised Edition
Since the time the first edition of this book was published, the field of
biomechanics has been going through a kind of transformation. For many
researchers, the main focus has shifted from traditional biomechanics to
the rapidly growing discipline of mechanobiology. Whereas biomechanics
typically involves studies of mechanical behavior of living cells and tissues,
mechanobiology deals with the biological response to mechanical stimuli.
Since the 1980s, considerable effort has been devoted to developing new
theories for modeling mechanobiological processes, such as growth and re-
modeling. As models have become increasingly realistic, investigators have
met new challenges. In addition to nonlinear mechanical behavior, modern
models in mechanobiology increasingly include mechanical, chemical, and
genetic feedback.
If you are looking for a book on mechanobiology, this may not be the
book for you, as it addresses only nonlinear biomechanics. (Shameless plug:
my recently published book Continuum Modeling in Mechanobiology may
be helpful.) On the other hand, a firm foundation in nonlinear elasticity
is essential for developing computational models, not only in biomechanics,
but also in mechanobiology of soft tissue.
When the first edition of this book appeared, I wondered whether anyone
would buy it, including my own students. (The same is true for this revised
edition.) Over the years, more copies sold than I expected, and some people
have told me they even read it. Unfortunately, however, it never quite made
the New York Times best-seller list.
Several readers have requested solutions to the problems given at the end
of chapters, and this became my main motivation for writing this revision.
Most of the old problems have been revamped, new problems have been
added, and answers and detailed solutions to most are included. Since the
vii
viii Preface to the Revised Edition
Larry A. Taber
St. Louis, Missouri
Preface to the First Edition
ix
x Preface to the First Edition
formed and deformed bodies. This book uses and compares both methods.
A hallmark of this book is an extensive use of tensor and dyadic analysis
in general curvilinear coordinates. Some may question the need for such
complexity in this age of high-speed computers, especially in biomechanics,
where most problems of interest require computational methods, in which
case Cartesian coordinates often are sufficient. While this is true, much of
the classical and even the current literature is littered with general curvi-
linear coordinates, and so reading and understanding these papers requires
knowing the language of general tensor analysis.
The emphasis on dyadic notation was motivated by courses I took many
years ago on linear elasticity and shell theory, respectively, from Professors
George Herrmann and Charles R. Steele at Stanford University. These
courses introduced me to the systematic elegance of dyadics. Over the
years, my students and I have found that learning the intricacies of nonlin-
ear elasticity is aided by routine use of dyadics, which form a link between
direct and indicial notation. Direct notation lends insight and clarity to
the basic physical principles, while indicial notation often makes algebraic
manipulations easier (and is needed to solve problems). Judicious use of
dyadics sometimes can clarify subtle differences, as well as similarities, be-
tween the various approaches used in solving nonlinear problems.
Throughout the book, I have tried to present as many details as possible
for each derivation. While this feature may be tedious to some readers, I
know how frustrating it can be when steps are omitted or glossed over. In
some instances, however, derivations are left to exercises at the end of the
chapter. Another possible source of irritation is the way many problems
are solved using more than one approach. The purpose of doing this is to
emphasize that there is more than one way to skin a cat, as well as to bring
out some subtleties that may be missed otherwise.
Readers of this book would benefit by prior experience with linear elas-
ticity. For those readers not already familiar with the linear theory, Ap-
pendix A provides an introduction. This book can be used as a textbook
for a course on the nonlinear theory of elasticity, as a reference for a course
on biomechanics, or as a reference for researchers trying to learn the subject
on their own.
After a brief introduction in Chapter 1, Chapter 2 establishes the math-
ematical background that is needed to navigate the remainder of this book.
This chapter covers general tensor and dyadic analysis, introducing the ba-
sic concepts of changes in coordinates and reference frames. Next, Chapters
3 and 4 develop and compare the basic measures of deformation and stress,
Preface to the First Edition xi
and present the field equations of solid mechanics. Chapter 5 then discusses
constitutive theory and develops stress-strain relations for nonlinear mate-
rials with various material symmetries. Finally, Chapter 6 solves several
specific problems in soft tissue mechanics using the theory and techniques
presented in the previous chapters.
The emphasis of this book is on learning the engineering fundamentals
needed to solve nonlinear problems in biomechanics. Some mathematical
and many biological details are omitted. For more in-depth modern treat-
ments of some of these issues, I highly recommend the books by Gerhard
A. Holzapfel and Jay D. Humphrey that are listed in the bibliography.
I am greatly indebted to Ashok Ramasubramanian, who diligently
worked through most of the in-text derivations and the problems, pointing
out numerous errors. I also would like to thank Millard F. Beatty and one
of my students, Evan Zamir, who provided helpful comments on earlier ver-
sions of this book. Finally, I thank Charlene Taber for her valuable editorial
assistance. Any remaining errors are my own, and I would appreciate any
feedback.
Larry A. Taber
St. Louis, Missouri
This page intentionally left blank
Contents
1. Introduction 1
1.1 Elasticity and Pseudoelasticity . . . . . . . . . . . . . . . 1
1.2 Biomechanics and Mechanobiology . . . . . . . . . . . . . 3
xiii
xiv Contents
3. Analysis of Deformation 67
3.1 Deformation in One Dimension . . . . . . . . . . . . . . . 68
3.2 Coordinate Systems and Base Vectors . . . . . . . . . . . 69
3.3 Deformation Gradient Tensor . . . . . . . . . . . . . . . . 72
3.4 Deformation and Strain Tensors . . . . . . . . . . . . . . . 74
3.4.1 Deformation Tensors . . . . . . . . . . . . . . . . 74
3.4.2 Strain Tensors . . . . . . . . . . . . . . . . . . . . 75
3.4.3 Physical Components of Strain . . . . . . . . . . . 77
3.5 Strain-Displacement Relations . . . . . . . . . . . . . . . . 88
3.6 Geometric Measures of Deformation . . . . . . . . . . . . 91
3.6.1 Stretch . . . . . . . . . . . . . . . . . . . . . . . . 91
3.6.2 Shear . . . . . . . . . . . . . . . . . . . . . . . . . 96
3.6.3 Volume Change . . . . . . . . . . . . . . . . . . . 98
3.6.4 Area Change . . . . . . . . . . . . . . . . . . . . . 102
3.7 Principal Strains . . . . . . . . . . . . . . . . . . . . . . . 103
3.7.1 The Eigenvalue Problem . . . . . . . . . . . . . . 104
3.7.2 Strain Invariants . . . . . . . . . . . . . . . . . . . 105
3.8 Stretch and Rotation Tensors . . . . . . . . . . . . . . . . 108
3.8.1 Polar Decomposition . . . . . . . . . . . . . . . . 108
3.8.2 Principal Stretch Ratios . . . . . . . . . . . . . . . 113
3.9 Approximations . . . . . . . . . . . . . . . . . . . . . . . . 119
3.9.1 Small Displacement . . . . . . . . . . . . . . . . . 120
3.9.2 Small Deformation . . . . . . . . . . . . . . . . . . 121
3.9.3 Small Deformation and Moderate Rotation . . . . 122
3.9.4 Small Rotation . . . . . . . . . . . . . . . . . . . . 122
3.10 Deformation Rates . . . . . . . . . . . . . . . . . . . . . . 122
Contents xv
Bibliography 437
Index 447
Chapter 1
Introduction
This book deals with nonlinear analysis of elastic materials, i.e., materi-
als that deform without loss of energy. At first glance, it may seem that
limiting the discussion to elastic materials severely restricts the range of
biomechanics applications. However, the characteristic behavior of many
soft tissues suggests that nonlinear elasticity can be used as a first ap-
proximation. This behavior, called pseudoelasticity, was first recognized
by Y.C. Fung during the 1970s (Fung et al., 1979). Even today, with the
rapidly increasing power of finite element and other computational meth-
ods, researchers take advantage of this feature to justify using elasticity
theory to study the mechanics of soft tissues.
The concept of pseudoelasticity is based on the following experimental
observation. During a tensile test, a soft tissue specimen ordinarily follows
separate loading and unloading stress-strain curves, which form a hystere-
sis loop (Fig. 1.1). Such behavior is typical of viscoelastic materials. If
the loading/unloading cycle is repeated at the same rate, the curves shift,
1
2 Introduction
1
2
3
STRESS
STRAIN
Fig. 1.1. Loading and unloading curves for a typical soft tissue. After the three cycles
shown, the curves show relatively little additional change.
This is one possible approach, but separate loading and unloading con-
stitutive relations have seldom been used in practice. Rather, most re-
searchers either use viscoelasticity theory or, more commonly, simply as-
sume that the material is elastic with the constitutive relation based only
on the loading curve. This approach, although not strictly correct, has pro-
vided useful approximate results in a large number of problems. One feature
that usually cannot be ignored, however, is nonlinear behavior, which is the
focus of this book.
Early development of the nonlinear theory of elasticity is attributed
largely to R.S. Rivlin, whose work on this topic began during the 1940s.
Many of his early papers have been republished together as a collection
(Rivlin et al., 1997). Rivlin initially worked without tensors, which are a
staple of this book, but later embraced it as a convenient analytical tool.
In addition, the classic work of Truesdell and Noll (2004), which helped
put nonlinear elasticity on a firm theoretical foundation, exudes tensors.
Today, it is difficult to imagine learning this subject without the use of
tensor analysis.
1.2. Biomechanics and Mechanobiology 3
Often the terms “reference frame” and “coordinate system” are used syn-
onymously. There is, however, an important distinction. A reference
frame is attached to an observer, and it may be in motion relative to other
frames with other attached observers. In each frame, there can be one
or more coordinate systems that can be used to locate a point in that
frame. The coordinate systems in a particular frame are fixed relative to
each other.
Consider a vector a fixed in reference frame A (Fig. 2.1), which contains
a Cartesian (x, y, z) and a cylindrical polar (r, θ, z) coordinate system (pos-
sibly with different origins). Relative to the Cartesian system, a has the
components {ax , ay , az }, and, relative to the polar system, a has the compo-
nents {ar , aθ , az }. These different components describe the same vector a.
1 A stress component depends on its direction of action and the orientation of the area
on which it acts.
5
6 Vectors, Dyadics, and Tensors
z
a z
y
y
r
x q
A B
x
Fig. 2.1. Reference frames A and B with attached coordinate systems. Frame B is in
motion relative to frame A.
such as the Coriolis force, enter the analysis.) The force vector is an exam-
ple of a frame indifferent or objective quantity, while the acceleration
vector a is coordinate invariant but not frame indifferent.
Tensor analysis deals with coordinate transformation. In this chapter,
we assume that we are dealing with various coordinate systems in a single
frame of reference at a given time. Although the development is valid for
n-dimensional Euclidean space, we restrict our attention to (at most) three
dimensions.
2.2 Vectors
where the âi are the components of a with respect to the orthogonal triad
of unit vectors ei , which are tangent to the coordinate axes z 1 , z 2 , and z 3 .2
The second line of Eq. (2.1) defines the summation convention, with
2 Henceforth, unless stated otherwise, the e represent unit vectors and zi represent
i
Cartesian coordinates, a special case of the general curvilinear coordinates xi .
8 Vectors, Dyadics, and Tensors
summation implied over 1,2,3 for each repeated index in a single term.3
Because any letter can replace it without changing the meaning of an ex-
pression, a repeated index also is called a “dummy” index. For example,
aii = akk = a11 + a22 + a33 ,
while
3
ai bij = ak bkj = ai bij
i=1
gives a separate equation for each j = 1, 2, 3. Note that the number of upper
and lower indices in an equation must “balance,” i.e., after all summations
are carried out, each term must be left with the same subscripts and super-
scripts. Moreover, in general curvilinear coordinates, a single term cannot
contain, for example, three i’s or even two i’s on the same level; hence, ai bi
and aii are meaningless expressions.4 Throughout the remainder of this
book, the summation convention holds unless stated otherwise or unless
the indices are placed in parentheses, e.g., the term a(i) b(i) is not summed
over i but rather represents a separate expression for each i = 1, 2, 3.
In a Cartesian coordinate system, the ei are called base vectors. As
shown by Eq. (2.1), any vector can be expressed as a linear combination
of the base vectors. If the z i -axes and base vectors ei undergo a rigid-
body rotation to give a new coordinate system z̄ i with new base vectors ēi
(Fig. 2.2), then a can be resolved along these new base vectors as
a = ¯âi ēi , (2.2)
¯i
where â = â . A primary objective of tensor analysis is to relate compo-
i
use only subscripts. In this case, the summation convention apples to repeated subscripts
in a single term (see Appendix A).
2.2. Vectors 9
z2
z2 a = aiei = aiei
a2
z1
a2
e2 a1
e2 q e1
q
z1
e1 a1
Fig. 2.2. Components of vector a relative to two Cartesian coordinate systems, zi and
z̄i .
as “base vectors.” The gi can be considered the “natural base vectors” for a given
coordinate system.
10 Vectors, Dyadics, and Tensors
g3
x3
g2
x2
P g1
r dr
x1
r+dr
0
Fig. 2.3. Position vector r and differential position vector dr in curvilinear coordinate
system with base vectors gi .
dr = gi dxi , (2.6)
gi = r,i . (2.7)
Example 2.1 Determine the covariant base vectors for each of the follow-
ing coordinate systems:6
(a) y (b)
gq = req y x2
x P
gr = er
a x1
r
q x2 y
x ey g2
a
x,x1
g1 = ex
in skew coordinates. These sets of base vectors are illustrated in Fig. 2.4
(in two dimensions).
Here, we note several things. First, we have expressed r in terms of
the Cartesian base vectors in each case; since these vectors are constant,
this simplifies the algebra. Second, gθ is not a unit vector and it carries
the dimension of length. Third, since gx · gy = gr · gθ = 0, Cartesian and
polar coordinates are orthogonal. (Obviously, all these coordinate systems
are orthogonal relative to the z-direction.) However, g1 · g2 = sin α = 0,
indicating the nonorthogonality of skew coordinates (for α = 0). Finally,
for future reference, Eqs. (2.10) yield
r = r er (θ),
gr = r,r = er
∂er
gθ = r,θ = r .
∂θ
The first expression agrees with Eq. (2.10)1. To find ∂er /∂θ directly, we
examine the geometry (Fig. 2.5). Consider two points in the rθ-plane, P1
and P2 , that are located the same distance from the origin but are separated
by the small angle dθ. In moving from P1 to P2 , only the orientation of
er changes. To a first approximation, this change is (∂er /∂θ) dθ. The
geometry of Fig. 2.5 gives
∂er
∂θ dθ = |er | · dθ
or
∂er
∂θ = 1,
2.2. Vectors 13
der
er + dq
dq
P2 der
dq
dq
er
dq P1
r
Fig. 2.5. Geometry for the differential change in the unit vector er with θ.
the meaning. This procedure is necessary since more than two of any index is not allowed
in any single term.
14 Vectors, Dyadics, and Tensors
gi · gj = δ ji , (2.14)
where
1 for i = j
δ ji = (2.15)
0 for i = j
a2g2
g2
x2
g2
a2g2
a = aigi = aigi
a1g1 g1 x1
g1
a1g1
Fig. 2.6. Covariant and contravariant base vectors and components of a vector a in a
two-dimensional nonorthogonal coordinate system.
which is the scalar triple product of the covariant base vectors. From
√
elementary vector analysis, Eq. (2.17) implies that g represents the vol-
ume of the parallelepiped with the vectors gi as edges. The reason for the
square root is discussed later (Sec. 2.6.2). This and similar manipulations
yield
g2 × g3 g3 × g1 g1 × g2
g1 = √ , g2 = √ , g3 = √ . (2.18)
g g g
These equations show that dotting a vector with a base vector gives the
component of the same variance.
The physical meaning of Eqs. (2.20) can be seen using elementary vector
analysis (see Fig. 2.6). First, we can write gj = |gj |ej (j not summed),
√
where |gj | = gj · gj is the magnitude of gj and ej is the unit vector in
the direction of gj . Thus, aj = a · gj = |gj |(a · ej ) (j not summed), which
shows that aj represents the product of |gj | and the orthogonal projection
of a in the direction of gj . Similarly, the relation aj = a · gj = |gj |(a · ej )
(j not summed) indicates that aj is the product of |gj | and the orthogonal
projection of a in the direction of gj .
Judicious use of covariant and contravariant components and base vec-
tors leads to numerous simplifications in vector and tensor analysis.
Example 2.3 Determine the contravariant base vectors for each of the
coordinate systems of Example 2.1 (Fig. 2.4).
Solution. Recall that the scalar triple product of the vectors a, b, and
c can be written
a 1 a2 a3
a · (b × c) = b · (c × a) = c · (a × b) = b1 b2 b3 , (2.21)
c c c
1 2 3
cos θ sin θ 0
= gr · (gθ × gz ) = −r sin θ r cos θ 0 = r (cylindrical)
0 0 1
1 0 0
= g1 · (g2 × g3 ) = sin α cos α 0 = cos α (skew).
0 0 1
(2.22)
2.3. Dyadics and Tensors 17
2.3.1 Dyadics
The tensor product of two vectors a and b, denoted by ab, is called a
dyad.8 Since each of the vectors in a dyad is geometrically invariant in a
8 Some authors use the notation a ⊗ b for the tensor product.
18 Vectors, Dyadics, and Tensors
given frame of reference, it follows that the dyad also is invariant. The order
of the vectors composing a dyad is significant, and so the tensor product is
not commutative, i.e., ab = ba. The same is true of higher-order polyads,
such as abc or abcd, which also are useful in some applications. Here,
we are concerned primarily with dyads and linear combinations of dyads,
which are called dyadics. Equations written in terms of vector and tensor
components (e.g., fi = mai ) are written in indicial notation, whereas
those written without components (e.g., f = ma) are said to be written
in direct notation. As we will see later, dyadics link indicial and direct
notation.
Algebraic manipulation of dyads is straightforward so long as care is
taken to maintain the order of the component vectors. For example, the
operations
ab · c = a(b · c)
c · ab = (c · a)b (2.26)
show that the dot product of a dyad with a vector yields another vector.
Similarly, the equations
ab · cd = a(b · c)d = (b · c) ad
cd · ab = c(d · a)b = (d · a) cb (2.27)
show that the single-dot product of two dyads produces another dyad.
These equations also show that the single-dot product of two dyads is not
commutative in general. (Recall that the dot product of two vectors is
commutative.)
Two types of double-dot (scalar) product are useful in manipulating
dyads. The vertical double-dot product is defined by
ab : cd = (a · c)(b · d) = c · (ab) · d
= (c · a)(d · b) = cd : ab, (2.28)
which shows that this operation is commutative for dyads. For polyads,
however, this may not be the case, e.g.,
abc : defg = a(bc : de)fg
= a(b · d)(c · e)fg
= (b · d)(c · e)afg
= defg : abc.
Again, with care taken to maintain order, the algebra is straightforward.
2.3. Dyadics and Tensors 19
a+b = b+a
(a + b) + c = a + (b + c)
a·b = b·a
a · (b + c) = a · b + a · c
φ(a + b) = φa + φb
(φ + ψ)a = φa + ψa
a × b = −b × a
a × (b × c) = (a · c)b − (a · b)c = a · (cb − bc)
(a × b) × c = (a · c)b − (b · c)a = c · (ab − ba)
a · (b × c) = b · (c × a) = c · (a × b)
φ(ab) = (φa)b = a(φb) = φab
a(bc) = (ab)c = abc
a(b + c) = ab + ac
φ(ab + cd) = φab + φcd
2.3.2 Tensors
A linear vector function T satisfies the relations
T · (a + b) = T · a + T · b and T · (φa) = φT · a (2.30)
for any vectors a and b and scalar φ. A second-order tensor T can be de-
fined as a linear vector function that, when dotted with a vector, transforms
20 Vectors, Dyadics, and Tensors
a = T · b. (2.31)
Since the vectors a and b are invariant, T also must be invariant. A third-
order tensor is a linear vector function that transforms a vector into a
second-order tensor through a single-dot product or a second-order tensor
into a vector through a double-dot product, and so on. The equation
(ab) · (φc + ψd) = φa(b · c) + ψa(b · d)
= [φ(b · c) + ψ(b · d)]a, (2.32)
with φ and ψ being scalars, shows that a dyad ab is a linear vector function
that satisfies the definition of a second-order tensor. (Of course, the dot
products b · c and b · d also are scalars.) In this way, we can show that a
vector is a first-order tensor, a dyad is a second-order tensor, a triad is a
third-order tensor, and so on.
Any second-order tensor can be represented by a dyadic. Because they
span the space, base vectors are convenient for forming the component
dyads. In three-dimensional space, for example, we can write
T = T 11 g1 g1 + T 12 g1 g2 + T 13 g1 g3
+T 21 g2 g1 + T 22 g2 g2 + T 23 g2 g3
+T 31 g3 g1 + T 32 g3 g2 + T 33 g3 g3
= T ij gi gj , (2.33)
where the T ij are the contravariant components of the second-order tensor
T with respect to the covariant basis {gi gj }. Like a vector, a tensor can be
expressed in terms of components referred to any basis. Also like a vector,
under a change of basis in a given frame of reference, a tensor does not
change, but its components do. For a vector, the base vectors form a basis;
for a tensor, the dyads (or polyads) composed of base vectors form a basis.
In terms of natural base vectors, T has four fundamental component
representations:
The Tij are covariant components, and the T·ji and Ti·j are mixed com-
ponents of T. By convention, the first index on a tensor component is
9 In this chapter, lower-case bold letters denote first-order tensors (vectors), and upper-
associated with the first vector of the dyad, and the second index is as-
sociated with the second vector. Since T·ji = Tj·i in general, the dots in
the mixed components are needed for clarity to keep the place of the first
index. Again, we emphasize the importance of the order of the base vectors
in Eq. (2.34).
The single-dot product of a vector with a base vector was used previously
to extract the component of the vector with the same variance as the base
vector [see Eqs. (2.20)]. Analogously, vertical-double-dotting a second-order
tensor T with a dyadic basis gives the components of T with the same
variance as the basis. For example, extending Eqs. (2.20) gives10
T : gi gj = (Tk·l gk gl ) : gi gj
= Tk·l (gk · gi )(gl · gj )
= Tk·l δ ki δ jl
= Ti·j , (2.35)
T : gi gj = gi · T · gj = Tij
T : gi gj = gi · T · gj = T ij
T : gi gj = gi · T · gj = Ti·j
T : gi gj = gi · T · gj = T·ji . (2.36)
with the same result. The reader may find it instructive to try this.
22 Vectors, Dyadics, and Tensors
T+U = U+T
(T + U) + V = T + (U + V)
T · (U + V) = T · U + T · V
φ(T + U) = φT + φU
(φ + ψ)T = φT + ψT
T · (U · V) = (T · U) · V
T : U = TT : UT = U : T
T : (U + V) = T : U + T : V
T : (U · V) = U : (T · VT ) = V : (UT · T)
φ(T : U) = (φT) : U = T : (φU)
j
The subscript (gi g ), which identifies the basis, can be omitted if the ba-
sis is specified. According to convention, the first index (or base vector)
corresponds to the row and the second to the column of the matrix.
b · T · a = a · TT · b, (2.37)
where a and b are arbitrary vectors. Because the dot product of a second-
order tensor and a vector is a vector, we can write this equation in the form
2.3. Dyadics and Tensors 23
(ab)T = ba
(T + U)T = TT + UT
(TT )T = T
(T · U)T = UT · TT
(TT )−1 = (T−1 )T ≡ T−T
(T · U)−1 = U−1 · T−1
(T−1 )m = (Tm )−1 ≡ T−m
T · a = a · TT . (2.38)
gi · TT · gj = gj · T · gi → T ij = T ji
gi · TT · gj = gj · T · gi → Tij = Tji
g · T · gj = gj · T · g
i T i
→ T·ji = Tj·i . (2.41)
24 Vectors, Dyadics, and Tensors
Note that symmetry of T does not imply T·ji = T·ij . Thus, in general, the
matrix of mixed components with T·ji in the ith row and jth column is not
equal to the transpose of the matrix with T·ij in the jth row and ith column.
In other words, symmetry of the tensor T does not imply that the matrix
of mixed components is symmetric.
where T−1 is the inverse of T and I is the identity tensor, which satisfies
T · I = T. It turns out that the mixed components of I are the Kronecker
delta δ ij , and so
I = δ ij gi gj = δ ij gj gi = gi gi = gi gi , (2.43)
which also can be seen by directly substituting these relations into (2.45).
In general, the only components of A and B we need are those indicated
in Eqs. (2.47). Therefore, without confusion we set11
Aj·i ≡ Aji , Bj·i ≡ Bji (2.49)
with the understanding that dyadic operations are to be carried out with
A and B in the forms of Eqs. (2.46) or (2.48). Then, dotting Eqs. (2.47)1,2
with gk and gk , respectively, yields
11 Note, however, that the order of the base vectors corresponding to Aj and B ·i are
·i j
in the opposite order, see Eqs. (2.46). Therefore, in the matrix representation for A,
the superscript and subscript correspond to the row and column, respectively, but the
opposite is true for B.
26 Vectors, Dyadics, and Tensors
Eqs. (2.14) and (2.50) give Aij = Bji = δ ij , and Eqs. (2.43) and (2.48) show
that A = B = I, as expected.
The transformation tensors A and B are related. Since the base vectors
in any coordinate system must satisfy Eq. (2.14), then gi · gj = ḡi · ḡj = δ ji .
Thus, Eqs. (2.38) and (2.45) give
gi · gj = ḡi · ḡj = (A · gi ) · (B · gj )
= (gi · AT ) · (B · gj ) = gi · (AT · B) · gj ,
B = A−T (2.51)
with A−T ≡ (AT )−1 as defined in Table 2.3. Similarly, Eqs. (2.47) yield
or
Substituting Eqs. (2.10) for the ḡi and (2.23) for the gi yields
The tensor B can be found from Eq. (2.48) or Eq. (2.51). Here, we use
the latter equation and first compute
⎡ ⎤
cos θ sin θ 0
AT(ej ei ) = ⎣ −r sin θ r cos θ 0 ⎦ . (2.55)
0 0 1
The inverse of a tensor T can be computed from
(cof T)T
T−1 = , (2.56)
det T
where the components of cof T are the cofactors of the matrix [T·ji ], and
det T is the determinant of [T·ji ] (see Sec. 2.5.2). Thus, we have
⎡ ⎤
r cos θ − sin θ 0
1
B = A−T = (AT )−1 = ⎣ r sin θ cos θ 0 ⎦ . (2.57)
r
0 0 r (e e )
i j
These expressions for A and B can be used to check our results from
Examples 2.1 and 2.3. For instance, using Eq. (2.45)1 , we can transform
g2 = gy to ḡ2 = ḡθ by
⎡ ⎤⎡ ⎤
cos θ −r sin θ 0 0
ḡθ = A · gy = A · ey = ⎣ sin θ r cos θ 0 ⎦ ⎣ 1 ⎦
0 0 1 0
⎡ ⎤
−r sin θ
= ⎣ r cos θ ⎦ = −r sin θ ex + r cos θ ey ,
0
which agrees with gθ in (2.10).
∂xj
Aji = . (2.60)
∂ x̄i
∂ x̄i
Bji = . (2.61)
∂xj
Often the xi are known in terms of the x̄j (or vice versa), and the relations
between coordinates are difficult to invert. In this case, Eq. (2.60) can be
used to compute Aji , and then Eq. (2.51) or (2.52) gives the Bji .
Example 2.5 Use Eqs. (2.60) and (2.61) to compute the components of
the tensors A and B for a transformation from Cartesian coordinates to
cylindrical polar coordinates.
Solution. Let (x1 , x2 , x3) = (x, y, z) and (x̄1 , x̄2 , x̄3) = (r, θ, z); then
x1 = x̄1 cos x̄2 , x2 = x̄1 sin x̄2 , x3 = x̄3
1 x2
x̄1 = (x1 )2 + (x2 )2 2 , x̄2 = tan−1 , x̄3 = x3 .
x1
2.4. Coordinate Transformation 29
which agrees with Eq. (2.57) of Example 2.4. (Recall that the superscript in
Bji corresponds to the column and the subscript to the row; see footnote on
page 25.) The other components of A and B can be determined similarly
(see Problem 2.8).
With Eqs. (2.50), this and similar expressions for the other components
give
and, for simplicity, let the tensor T be represented by the dyad ab, i.e.,
T = ab.
2.5. Tensor Invariants 31
2.5.1 Trace
The trace of a dyad is defined by
tr ab ≡ a · b. (2.71)
This operation contracts the dyad ab, since inserting a dot between the
vectors lowers the order of ab by two, transforming it into a scalar. For a
second-order tensor T, the trace is defined in terms of mixed components
as
tr T ≡ tr (T·ji gi gj ) = T·ji gi · gj = T·ji δ ji = T·ii .
12 For matrices, ([A][B])T = [B]T [A]T and the usual tensor rules for balancing indices
tr I = n
tr T = T : I
tr (T · UT ) = tr (TT · U) = T : U
tr TT = tr T
tr (φT) = φ tr T
tr (T + U) = tr T + tr U
tr (T · U) = tr (U · T)
tr (T · U · V) = tr (U · V · T) = tr (V · T · U)
Thus, tr T does not change with a change of basis, and the trace is called a
scalar invariant of the tensor T. Some useful properties of the trace are
given in Table 2.4.
Here, we note that any tensor T can be decomposed as
T = TS + TA , (2.74)
where
TS = 12 (T + TT ) = TTS
TA = 12 (T − TT ) = −TTA (2.75)
tr TS = tr T, tr TA = 0. (2.76)
2.5. Tensor Invariants 33
det I = 1
det TT = det T
det(φT) = φn det T
det(T · U) = det T det U
∂ det T
= (det T) T−T
∂T
2.5.2 Determinant
The determinant of a tensor T is defined as the determinant of the matrix
of mixed components, i.e.,
−λ3 + I1 λ2 − I2 λ + I3 = 0 (2.82)
I1 = tr T
I2 = 12 [(tr T)2 − tr T2 ]
I3 = det T (2.83)
−T3 + I1 T2 − I2 T + I3 I = 0. (2.84)
2.6. Special Tensors 35
gij ≡ gi · gj . (2.86)
The scalar invariant given by Eq. (2.85) is called the metric of the space,
and the gij characterize a particular coordinate system in the space.
The quantities gij have a useful mathematical interpretation. Since
gi gj : I = gi · I · gj = gi · gj = gij , Eq. (2.36)1 shows that the gij are
the covariant components of the identity tensor. For this reason, I also
is called the metric tensor. Like any second-order tensor, I has four
representations in any coordinate system:
gi · I · gj = gi · gj = gij
gi · I · gj = gi · gj = gij
gi · I · gj = gi · gj = g·j
i
gj · I · gi = gj · gi = gj·i , (2.88)
i
g·j = gj·i = δ ij . (2.89)
Expanding the determinant in the usual manner shows that it can be writ-
ten
T = T1i T2j T3k eijk = Ti1 Tj2 Tk3 eijk (2.97)
or, equivalently,
epqr T = eijk Tpi Tqj Trk , epqr T = eijk Tip Tjq Tkr , (2.98)
where
⎧
⎨ +1 if (i, j, k) is an even permutation
eijk = eijk = −1 if (i, j, k) is an odd permutation (2.99)
⎩
0 if two or more indices are equal
gi × gj = gi gj :
.
(gi × gj ) · gk = gi gj gk .. (2.106)
and
gi × gj = gigj :
.
(gi × gj ) · gk = gi gj gk .. , (2.107)
as can be verified easily using Eq. (2.28) and its extension, the triple-dot
.
product given by abc .. def = (a · d)(b · e)(c · f ).
In Cartesian coordinates, ijk = eijk , but since the gi are not always
unit vectors, ijk = eijk in general. To find the relation between these
components, we let the unit vectors ei of Eqs. (2.100)–(2.102) be Cartesian
base vectors (ei = ei ) and set ei = ḡi and eijk = ¯ijk . Then, the permuta-
tion tensor can be written in terms of the Cartesian (barred) system and a
general (unbarred) system as
= eijk ei ej ek = ¯ijk ḡi ḡj ḡk = ijk gigj gk . (2.108)
Relative to the natural basis, the covariant components are
.
ijk = gi gj gk ..
.
= gi gj gk .. (¯lmn ḡl ḡm ḡn )
= (gi · ḡl )(gj · ḡm )(gk · ḡn ) ¯lmn
= Bil Bjm Bkn elmn , (2.109)
2.6. Special Tensors 39
in which Eq. (2.50)2 has been used. This transformation and Eq. (2.106)1,
which shows that converts a dyad into a vector via the double-dot product,
demonstrate the tensor character of .
The relation (2.109) between ijk and eijk can be simplified as follows.
First, for (i, j, k) = (1, 2, 3), comparison of Eqs. (2.97) and (2.109) reveals
that 123 = det[Bji ], with a similar result for other even permutations of
the indices. (The sign changes for odd permutations.) Next, we evaluate
det[Bji ] = det B. In terms of unbarred and barred base vectors, the identity
tensor can be written as
I = gij gi gj = ḡij ḡi ḡj . (2.110)
Extracting the covariant components and using (2.50) yields
gij = I : gi gj = (ḡkl ḡk ḡl ) : gi gj
= ḡkl (ḡk · gi )(ḡl · gj )
= Bik Bjl ḡkl . (2.111)
Therefore,
det[gij ] = det[Bik ] det[Bjl ] det[ḡkl] = (det B)2 det[ḡkl ]
or
g = (det B)2 ḡ, (2.112)
where
ijk √
= eijk / g. (2.115)
√
Note that Eq. (2.103)2 gives (g1 × g2 ) · g3 = 123 = g, which is the same
√
g found in Eq. (2.17), i.e., it represents the volume of the parallelepiped
with the gi as edges. Equations (2.115), therefore, explain the square root
in Eq. (2.17).
40 Vectors, Dyadics, and Tensors
= ai bj gi gj : . (2.116)
Second, the -δ identity is (see Problem 2.7)
ijk
irs
= δ rj δ sk − δ sj δ rk . (2.117)
QT · Q = Q · QT = I, (2.119)
Example 2.6 Consider rigid-body rotation about the z-axis of the Carte-
sian base vectors ex and ey into the new unit vectors ēx and ēy [Fig. 2.2
2.7. Physical Components 41
(page 9) with (e1 , e2 ) = (ex , ey ) and (ē1 , ē2 ) = (ēx , ēy )]. With this rotation
given by
ēx = Q · ex , ēy = Q · ey , ēz = Q · ez , (2.121)
determine the Cartesian components of Q.
Solution. In this relatively simple problem, we can write by inspection
Q = ēx ex + ēy ey + ēz ez , (2.122)
where
ēx = ex cos θ + ey sin θ
ēy = −ex sin θ + ey cos θ
ēz = ez . (2.123)
Comparing Eqs. (2.48)1 and (2.122) shows that this rotation can be consid-
ered a coordinate transformation from the unbarred to the barred system.
In matrix form, inserting Eqs. (2.123) into (2.122) yields
⎡ ⎤
cos θ − sin θ 0
Q(ei ej ) = ⎣ sin θ cos θ 0 ⎦ . (2.124)
0 0 1
Finally, Q · QT = exex + ey ey + ez ez = I can be easily verified.
are the physical components of a. Since the ei are dimensionless, the âi have
the same units as a. Moreover, âi is the length of the vector a(i)g(i) (see
Fig. 2.6, page 14). In terms of the covariant components of a, Eqs. (2.93)1
and (2.126) give
√
âi = gij aj g(ii) . (2.127)
Since b̂i and âj possess physically meaningful units, T̂·ji does also. Inserting
Eqs. (2.131) into (2.129) yields
i i g(ii) j
b̂ = T·j â ,
g(jj)
• In the matrix [T̂·ji ], the first index (i) denotes the row and the second
index (j) denotes the column. Interchanging rows and columns (making
j first and i second) gives the transpose matrix
T g(ii)
i ·i
T̂·j = Tj .
g(jj)
where Eq. (2.132) provides the right-hand side. In other words, if the
tensor T is symmetric, so is its matrix of physical components.
• Equations (2.132) give
T̂·ii = T·ii = tr T
det[T̂·ji ] = det[T·ji ] = det T. (2.133)
(The second relation is easy to show using a 2x2 matrix.) Hence, the
invariants given by Eq. (2.83) have the same values when expressed in
terms of either tensor or physical components.
• The physical components of T can be written in terms of the contravari-
ant components T ij by substituting T·ji = T ik gkj into Eq. (2.132). In
orthogonal coordinates, gkj = 0 for k = j, giving T·ji = T ij g(jj) for
i, j = 1, 2, 3 (not summed). In this case, (2.132) becomes
g(ii) √
T̂·j ≡= T g(jj)
i ij
= T ij g(ii) g(jj).
g(jj)
44 Vectors, Dyadics, and Tensors
Note that, since this is not a tensor equation, the upper and lower
indices on the two sides of the equation need not match. In terms of
natural base vectors, T can be written
√ √
T = T ij gi gj = T ij g(ii) ei g(jj)ej ,
where Eq. (2.100)1 has been used. In orthogonal coordinates, therefore,
it is convenient to redefine the physical components of the second-order
tensor T as
√
T̂ ij = T ij g(ii) g(jj), (2.134)
where the components Γijk and Γkij are called Christoffel symbols. It is
important to note that these quantities are not components of a third-order
tensor. Dotting this relation with the base vectors yields
gi ,j · gk = (Γijl gl ) · gk = Γijl δ lk
gi ,j · gk = (Γlij gl ) · gk = Γlij δ kl
or
Γijk = gi ,j · gk
Γkij = gi ,j · gk . (2.137)
gi ,j = −Γikj gk , (2.139)
which can be verified by dotting both sides of this relation with gl and
changing the indices.
Two useful properties of the Christoffel symbols warrant attention.
First, since gi = r,i , then gi ,j = r,ij = r,ji = gj ,i . Thus, from Eq. (2.136),
we find
Thus, if the metric for a coordinate system is known, this equation gives the
Γijk and then Eq. (2.142) gives the Γkij . In Cartesian coordinates, because
the base vectors are constant, all the Christoffel symbols are zero.
Example 2.7 Compute the Christoffel symbols for cylindrical polar coor-
dinates.
Solution. With (x1 , x2 , x3 ) = (r, θ, z), Eqs. (2.10) and (2.24) give the
base vectors
g1 = gr = ex cos θ + ey sin θ = er
g2 = gθ = r(−ex sin θ + ey cos θ) = reθ
g3 = gz = ez
g 1 = g r = er
g2 = gθ = r −1 eθ
g 3 = g z = ez , (2.145)
2.8. Vector and Tensor Calculus 47
and the non-zero components of the metric tensor (gij = gi ·gj , gij = gi ·gj )
are
As the only non-zero derivative of the gij is g22 ,1 = 2r, the only non-zero
Christoffel symbols given by Eq. (2.144) are
Γ221 = − 12 g22 ,1 = −r
Γ122 = Γ212 = 12 g22 ,1 = r (2.147)
gr ,θ = Γ122 g2 = r(r −1 eθ ) = eθ
= Γ212 g2 = r −1 (reθ ) = eθ ,
which agrees with the result given by taking the derivative of Eq. (2.145)1
with respect to θ.
√
Example 2.8 Compute the derivative of g = g1 · (g2 × g3 ) with respect
to xi . Express the result in terms of g and Christoffel symbols.
where the vectors have been permuted according to the scalar triple product
formula in Table 2.1 (page 19). Next, substituting Eq. (2.136) gives
√
( g),i = Γj1i gj · (g2 × g3 ) + Γj2i gj · (g3 × g1 ) + Γj3i gj · (g1 × g2 ).
48 Vectors, Dyadics, and Tensors
Since gj · (g2 × g3 ) = 0 only for j = 1, and so on for the other terms, this
expression can be written
√
( g),i = Γ11i g1 · (g2 × g3 ) + Γ22i g2 · (g3 × g1 ) + Γ33i g3 · (g1 × g2 )
√
= g (Γ11i + Γ22i + Γ33i )
or
√ √
( g),i = g Γjji. (2.149)
a,j = ai |j gi = ai |j gi , (2.150)
where
ai |j ≡ ai ,j +ak Γijk
ai |j ≡ ai ,j −ak Γkij (2.151)
From Eqs. (2.139), (2.150), and (2.151), the second derivative of the
vector a is
a,jk = (ai |j gi ),k = (ai |j ),k gi + (ai |j )gi ,k
= (ai ,j −al Γlij ),k gi − ai |j Γikl gl
= (ai ,jk −al ,k Γlij − al Γlij ,k −al |j Γlik )gi
= [ai ,jk −al ,k Γlij − al Γlij ,k −(al ,j −am Γm l
jl )Γik ]g
i
(2.152)
with a similar relation for a,jk = (ai |j gi ),k . Thus, taking higher derivatives
of a vector is straightforward, but labor-intensive. Symbolic manipulation
computer programs, however, can make tensor calculus virtually painless.
where
∂φ
dφ = da · , (2.156)
∂a
∂φ
dφ = dT : . (2.159)
∂T
Differentiation with respect to higher-order tensors follows by adding more
dots. If the differential of T is written, for example, in the form
dT = dT ij gi gj , (2.160)
2.8. Vector and Tensor Calculus 51
∂ ∂ ∂
= g i i = gi
∂a ∂a ∂ai
∂ ∂ ∂ ∂ ∂
= gi gj ij = gi gj = gi gj i = gi gj .j . (2.162)
∂T ∂T ∂Tij ∂T.j ∂Ti
Note that the base vectors and the components in each of these expressions
have the same variance.
The above observation is useful in differentiating vectors and tensors
with respect to other vectors and tensors. The basic definitions, which
follow from Eqs. (2.156) and (2.159), are
∂b ∂b
db = da · + dT :
∂a ∂T
∂U ∂U
dU = da · + dT : (2.163)
∂a ∂T
for b = b(a, T) and U = U(a, T). In these expressions, the order of the
factors is important, e.g.,
∂b ∂ ∂ ∂b
= (b) = (b) = gi gj ij . (2.164)
∂T ∂T gi gj ∂T ij ∂T
52 Vectors, Dyadics, and Tensors
dφ = dr · ∇φ, (2.165)
∂ ∂
∇≡ = gi i , (2.166)
∂r ∂x
with the invariance of ∇ indicated by the form ∂/∂r. This invariance also
can be demonstrated by the manipulation
∂ ∂ ∂
∇ = gi = δ ij gj i = (Aik Bjk )gj i
∂xi ∂x ∂x
∂xi ∂
= (Bjk gj )
∂ x̄k ∂xi
∂ ∂
= ḡk k
= ḡi i ,
∂ x̄ ∂ x̄
which involves Eqs. (2.47), (2.52), and (2.60) and the chain rule for dif-
ferentiation. Thus, ∇ has the same indicial form in both the xi and x̄i
coordinate systems. Equation (2.165) also holds if φ is replaced by a vector
or a tensor.
Example 2.9 Using the results from Example 2.3 (page 16), write the
gradient operator in Cartesian and cylindrical polar coordinates.
Solution. With (x1 , x2 , x3 ) = (x, y, z) and Eqs. (2.23), Eq. (2.166) gives
∂ ∂ ∂
∇ = ex + ey + ez (2.167)
∂x ∂y ∂z
in Cartesian coordinates. Similarly, with (x1 , x2 , x3 ) = (r, θ, z) and
Eq. (2.24), we get
∂ eθ ∂ ∂
∇ = er + + ez (2.168)
∂r r ∂θ ∂z
in cylindrical coordinates.
2.8. Vector and Tensor Calculus 53
∇φ = gi φ|i (2.169)
since ordinary and covariant differentiation of a scalar are the same. The
gradient of vector a is
∂a
∇a = gj j = gj a,j ,
∂x
and substituting Eqs. (2.150) yields
∇a = ai |j gj gi = ai |j gj gi , (2.170)
where care has been taken to maintain the order of the base vectors. This
expression shows that the covariant derivatives ai |j and ai |j are components
of the second-order tensor ∇a.
The divergence of vector a, defined by ∇ · a, can be obtained by a
contraction of the gradient, i.e., by inserting a dot between the base vectors
of (2.170). Hence,
∇ · a = ai |j gj · gi = ai |j δ ji = ai |i
= ai |j gj · gi = ai |j gij = ai |i
or
∇ · a = a i | i = ai | i , (2.171)
where the contravariant components of the metric tensor have been used
to raise one of the indices on the tensor component ai |j . The form of ai |i
appears to indicate contravariant differentiation, but this operation is of
limited use since the components dxi of dr = gi dxi needed for ∂/∂xi are
not along the coordinate curves in general. The components ai |i , therefore,
are considered shorthand for ai |j gij .
The curl of vector a is given by
∂
∇ × a = gj j × a = gj × a,j
∂x
= ai | j g j × g I ,
54 Vectors, Dyadics, and Tensors
where Eq. (2.150) has been used. With Eqs. (2.104)1 and (2.107)1, this
expression assumes the forms
∇ × a = ai | j g j g i :
= ai | j jik
g k = aj | i ijk
gk . (2.172)
This tensor has important implications that depend on the geometry of the
space. For example, interchanging j and k in Eq. (2.175) to produce ai |kj
and using (2.140) gives
≡ am R m
.ijk , (2.177)
13 The name curvature tensor stems from the physical meaning of the second derivative.
2.8. Vector and Tensor Calculus 55
∇2 φ ≡ ∇ · ∇φ. (2.178)
Application of Eqs. (2.169) and (2.171) yields
Substituting Eqs. (2.139) and (2.166) into (2.178) gives the explicit form
(see Problem 2.16)
∇2 φ = gij φ,ij −Γjki gik φ,j . (2.180)
with Eq. (2.154)1 providing Tij |k . This equation shows that the covariant
derivatives of the second-order tensor components Tij are components of
the third-order tensor ∇T.
Other operations can be a little tricky, however. Consider, for instance,
the gradient of the dot product a · b. Directly differentiating the product
14 A curved surface formed by curling an initially flat surface is an exception that allows
and thus
Note that we have used the fact that dr · ∇b is a vector in the second line
of (2.183).
Another method, which often is easier to apply, involves using the com-
ponent form of ∇ given by Eq. (2.166), which we write in shorthand form
as (Drew, 1961)
∂
∇ = gi ≡ g ∂. (2.185)
∂xi
∇(a · b) = g ∂(a · b)
= g[(∂a) · b + a · (∂b)]
= g[(∂a) · b + (∂b) · a]
= (g ∂a) · b + (g ∂b) · a
= (∇a) · b + (∇b) · a, (2.186)
Example 2.10 Compute the gradient, divergence, and curl of the dyad
ab.
Solution. With care taken to preserve order, the gradient is
∇(ab) = (g ∂)(ab)
= g[(∂a)b + a(∂b)]
= (g ∂a)b + ga ∂b.
= (∇a)b + (a∇)T b. (2.187)
Similar manipulations yield the divergence as
∇ · (ab) = (g ∂) · (ab)
= g · [(∂a)b + a(∂b)]
= (g ∂) · ab + g · a ∂b
= (g ∂ · a)b + a · g ∂b
= (∇ · a)b + a · (∇b), (2.188)
and the curl is
∇ × (ab) = (g ∂) × (ab)
= g × [(∂a)b + a(∂b)]
= (g ∂ × a)b + g × a ∂b
= (g ∂ × a)b − a × (g ∂b)
= (∇ × a)b − a × (∇b). (2.189)
∇ · a = tr ∇a
∇ · (φa) = (∇φ) · a + φ(∇ · a)
∇ · (ab) = (∇ · a)b + a · (∇b)
∇ · (T · a) = (∇ · T) · a + T : ∇a
∇ · (a × b) = (∇ × a) · b − a · (∇ × b)
∇ · (∇ × a) = 0
∇·r = 3
y
C2
y2(x)
C1 P
r t
n y1(x)
0 x
a b
where ds is the differential length along C, and the unit normal vector is
dy dx
n = t × ez = ex − e y = nx e x + ny e y . (2.193)
ds ds
For a function φ(x, y) defined in A, consider the area integral
b y2 (x)
∂φ ∂φ
dA = dy dx
A ∂y a y1 (x) ∂y
b
= {φ[x, y2 (x)] − φ[x, y1(x)]} dx
a
b
= {[φ]C2 − [φ]C1 } dx. (2.194)
a
Equations (2.195) and (2.196) are the basic relations used to develop the
integral theorems.
Multiplying Eqs. (2.195) and (2.196) by ey and ex , respectively, and
adding provides
∂φ ∂φ
(ex + ey ) dA = φ (nx ex + ny ey ) ds
A ∂x ∂y C
∂ ∂
or, since ∇ = ex + ey in two-dimensional Cartesian coordinates,
∂x ∂y
∇φ dA = nφ ds (2.197)
A C
2.8. Vector and Tensor Calculus 61
with n given by Eq. (2.193). This relation, which is written in direct no-
tation, is called the gradient theorem or Green’s theorem. In three
dimensions, it becomes
∇φ dV = nφ dA, (2.198)
V A
2.9 Problems
Unless stated otherwise, work the following problems using base vectors for
general curvilinear coordinates (gi and gi ). Lower- and upper-case bold
letters represent vectors and second-order tensors, respectively.
2.1(a) Evaluate δ ij δ ji .
(b) Show that mki jki = 2δjm . (Hint: Try using the -δ identity.)
(c) Simplify the expression ijk ai aj as much as possible.
2.2 Let
a = 3ex + ey + ez
b = ex + 2ez
c = ex − 3ey + ez ,
where the ei are Cartesian base vectors. Compute (ab) · c and (ba) · c.
2.3 Expand and simplify as much as possible the derivative
∂ i j
xx .
∂xi
2.4 For the vectors a and b and second-order tensors A and B, show the
following:
(a) (a × b) · a = 0
(b) I : A =trA
(c) A : B = 0 if A is symmetric and B is antisymmetric
(d) (AT )−1 = (A−1 )T
−1
(e) (A · B) = B−1 · A−1
2.5 If T and U are second-order tensors, write A = T · U in terms of
the following dyadic bases: (a) gi gj , (b) gi gj , and (c) gi gj . Find two
dyadic representations for each basis in terms of components of T and
U only.
2.9. Problems 63
T
2.6 Use dyadic analysis to show T : (U · V) = U : (T · V ).
2.7 Use the vector formula
a × (b × c) = (a · c)b − (a · b)c
to derive the -δ identity
ijk
irs
= δ rj δ sk − δ sj δ rk .
Hint: Working in general curvilinear coordinates, set a = gi , b = gj ,
and c = gk .
2.8 Cylindrical polar coordinates (r, θ, z) and Cartesian coordinates (x, y, z)
are related by
x = r cos θ
y = r sin θ
z = z.
(a) Determine the components of the tensors A and B for a transfor-
mation from Cartesian to cylindrical coordinates using Eqs. (2.60)
and (2.61). Write both tensors in matrix form. Hint: Be careful
when assigning rows and columns.
(b) Determine B using Eq. (2.48)2.
2.9 Spherical polar coordinates (r, θ, φ) and Cartesian coordinates (x, y, z)
are related by
x = r sin θ cos φ
y = r sin θ sin φ
z = r cos θ.
(a) For a transformation from Cartesian to spherical coordinates, use
Eq. (2.60) to find the coordinate transformation components Aji .
(b) Show that ḡi = A · gi , where ḡi = (gr , gθ , gφ), gi = (ex , ey , ez ),
and A is the coordinate transformation tensor.
(c) Relative to spherical base vectors, write the physical components
of the second-order tensor T in terms of the components T·ji and
in terms of T ij .
2.10 Parabolic cylindrical coordinates (u, v, z) and Cartesian coordinates
(x, y, z) are related by
x = 21 (u2 − v2 )
y = uv
z = z.
64 Vectors, Dyadics, and Tensors
(a) Plot curves for constant values of u and constant values of v in the
xy-plane. Plot for u, v = 1, 2, and 3. Hint: Find equations for
x(y; u) and x(y; v).
(b) Determine the base vectors ḡi and ḡi (i = u, v, w) for parabolic
coordinates in terms of Cartesian base vectors.
(c) Compute the matrices [ḡij ] and [ḡij ]. Are the parabolic coordinates
orthogonal?
(d) For transformation from Cartesian to parabolic coordinates, deter-
mine the coordinate transformation matrices [Aij ] and [Bji ] using
Eqs. (2.50).
(e) Determine [Aij ] using Eq. (2.60) and [Bji ] using (2.51).
2.11 If T, U, and V are orthogonal tensors, show that T · U · V is an or-
thogonal tensor.
2.12 Consider the vector
a = 2ex − 5ey + 3ez
and the base vectors
g1 = ex − 2ey + ez
g2 = ex − ez
g3 = 3ex − 2ey + 2ez ,
where the ei are Cartesian base vectors.
(a) Compute the gi and the natural components ai and ai of a.
(b) Suppose a tensor T has the matrix representation
⎡ ⎤
−2 5 1
T(ei ej ) = ⎣ 3 0 −2 ⎦ .
1 −1 4
Compute the principal invariants of T and the natural components
T 11 , T 23 , T22 , T13 , T2·1 , and T·21 .
2.13 Consider the eigenvalue problem for the second-order tensor
T = 2g1 g1 + g2 g2 − 5g3 g3 + 3g2 g3 + 3g3 g2 .
(a) Determine the characteristic equation for T and compute its eigen-
values.
(b) Determine the eigenvector corresponding to each eigenvalue. Show
that the eigenvectors are mutually orthogonal.
2.14 If A, B, and C are second-order tensors, show the following:
2.9. Problems 65
∂
(a) trA = I
∂A
∂
(b) trA2 = 2AT
∂A
∂ ∂A ∂B
(c) (A : B) = :B+A:
∂C ∂C ∂C
2.15 Determine the Christoffel symbols Γkij for the following coordinate sys-
tems:
(a) Spherical polar coordinates (see Problem 2.9). Then, differentiate
the base vectors gr , gθ , and gφ with respect to θ and φ by taking
straight derivatives of the gi , as well as using the Γkij .
(b) Parabolic cylindrical coordinates (Problem 2.10). Then, differenti-
ate the base vectors gu and gv with respect to u and v by taking
straight derivatives of the gi , as well as using the Γkij .
2.16 Consider the Laplacian operator defined by ∇2 = ∇ · ∇.
(a) Use Eqs. (2.139) and (2.166) to show that the Laplacian of the
scalar function σ(xi ) can be written in the form
(b) Use the equation in part (a) to derive the explicit form for ∇2 σ in
toroidal coordinates, where σ depends only on r and φ (axisymme-
try, see Appendix B).
2.17 Consider the following vectors in cylindrical polar coordinates (r, θ, z):
sin θ
a = cos θ gr − gθ + z gz
r
1 1
b= gr + 2 gθ .
r r
Using Eq. (2.150)1, compute the derivative of each vector with respect
to r and with respect to θ. Express the results in terms of the unit base
vectors er , eθ , and ez .
2.18 Derive the expression
T ij |k = T ij ,k +T lj Γikl + T il Γjkl .
Analysis of Deformation
The mechanics of hard tissues, such as bones, teeth, and horns, can be ana-
lyzed using the linear theory of elasticity, in which deformation is assumed
to be “small.” Under this condition, the distinction between the geometries
of the undeformed and the deformed body can be ignored. Soft tissues,
in contrast, often undergo “large” (finite) deformation. In this case, geo-
metric nonlinearity enters the analysis, even if the material properties
(stress-strain relations) are linear. This chapter deals only with geometry.
Nonlinear stress-strain relations characterize material nonlinearity, as
discussed in Chap. 5.
One manifestation of geometric nonlinearity is nonlinear strain-displace-
ment relations. Another can be seen through the example of a cantilever
beam of length L with an end load P (Fig. 3.1). For small deflection, the
peak bending moment is approximately P L at the support. Classical lin-
ear beam theory is based on this assumption. When the deflection is large,
however, the load moves toward the support by a significant distance, re-
ducing the bending moment. Throughout most of this chapter, we make no
restrictions on the magnitudes of deformation (i.e., strain), displacement, or
rotation. Approximations for various restrictions are examined in Sec. 3.9.
P
L
67
68 Analysis of Deformation
l = L + DL
which implies
λ = 1 + E ∗ = (1 + 2E)1/2
E = 12 (λ2 − 1). (3.5)
Equation (3.4) can be verified by substituting (3.1).
The Lagrangian strain is referred to the undeformed configuration
(length L of the bar). In some formulations, it is useful to define a strain
measure relative to the deformed configuration. For the one-dimensional
case, the Eulerian strain is defined as
2
− L2
e= . (3.6)
2 2
In terms of the other strain measures, this relation can be written
E E ∗ (1 + 12 E ∗ )
e = 12 (1 − λ−2 ) = = , (3.7)
1 + 2E (1 + E ∗ )2
which can be confirmed by substituting Eqs. (3.2), (3.4), and (3.5).
Any one of the three strain measures (E ∗ , E, e) can be used to charac-
terize a given deformation, although they differ numerically. If the defor-
mation is small (E ∗ << 1), however, then Eqs. (3.4) and (3.7) show that
E ∼= e ∼= E ∗ , i.e., all the strain measures are essentially equivalent. The
development in the following sections extends these ideas to three dimen-
sions.
b
B x3 g3
g2
G3
G2 x2
X2 u
X3 p g1
x1
P X1 r
G1
R
b o
Fig. 3.3. Coordinates and base vectors in undeformed body B and deformed body b.
Some authors [e.g., Eringen (1962; 1980)] prefer to keep the X I and
i
x coordinate systems independent of each other. In this way, the coor-
dinates can be chosen to take advantage of any symmetries in B and b.
Other authors [e.g., Green and Zerna (1968)] assume that the X I system
is imbedded in B so that it deforms with the body, and the xi system is
then taken to coincide with the X I system at all times during the defor-
mation. During the motion, each point keeps the same coordinate label,
and xi = X I are called convected coordinates. Although the flexibility
in choosing the deformed coordinate system is lost in this case, the basic
equations assume a somewhat simpler form. In this chapter, we consider
both choices.
Let R(X I , t) and r(xi , t) be the position vectors to P and p, respectively.
Then, the mapping of P into p and the inverse mapping of p into P are
described by
∂r
gI = = r,I
∂X I
∂r
gi = = r,i . (3.11)
∂xi
dR = GI dX I
dr = gI dX I = gi dxi . (3.13)
where
∂xi ∂X I
FIi ≡ , FiI ≡ . (3.15)
∂X I ∂xi
which also can be found from Eqs. (3.15) and the chain rule for differenti-
ation, or from Eq. (2.52). Physical meanings for FIi and FiI are discussed
in the next section.
∂ ∂
∇≡ = GI
∂R ∂X I
¯ ≡ ∂ = gi ∂ = gI ∂
∇ (3.18)
∂r ∂xi ∂X I
dr = F · dR = dR · FT . (3.19)
3.3. Deformation Gradient Tensor 73
N
dS ds
dR dr
F
n
B b
∂R GJ dX J
F−T = = = gI GJ δ JI = gI GI
∂r gI dX I
or, with Eqs. (3.14)1 , and (3.16)1,
F = gI GI = FIi gi GI
−1
F = GI gI = FiI GI gi . (3.22)
The expression for F shows that the FIi are actually the mixed components
F·Ii of the deformation gradient tensor. We drop the dot with the under-
standing that the deformed base vector (lower case g) always appears first
in F; in matrix form, therefore, i denotes the row and I the column.
We also could derive F = gI GI from Eqs. (3.11)1 , (3.18)1 , and (3.21)1,
i.e., FT = ∇r = GI r,I = GI gI . With this relation, it is straightforward to
show
gI = F · G I , (3.23)
74 Analysis of Deformation
¯ = (GI gI ) · (gJ ∂ ) = GI δ JI ∂ = GI ∂ ,
FT · ∇
∂X J ∂X J ∂X I
we have
¯
∇ = FT · ∇. (3.24)
The deformation gradient tensor can serve as the primary deformation vari-
able in solid mechanics problems. It is relatively easy to compute, but it
contains two features that may complicate analyses. First, as shown by
Eqs. (3.22), F is not symmetric in general, and so all nine components rela-
tive to a given basis may be unknown in some problems. Second, as shown
by Eq. (3.19), F deforms and rotates dR into dr. Since rigid-body rotation
of an element should produce no stress, certain restrictions must be placed
on how F appears in constitutive relations (see Sec. 5.2.5).1
One way around these problems is to define deformation variables based
only on changes in length, and not orientation, of line elements. There are
several ways to do this; here, we focus on the most common approaches.
the left ventricle during the cardiac cycle (Arts et al., 1984; Hansen et al., 1988; Ingels
et al., 1989; Taber et al., 1996).
3.4. Deformation and Strain Tensors 75
(3.20) give
In these equations, since (F−T · F−1 )−1 = (F−1 )−1 · (F−T )−1 = F · FT (see
Table 2.3, page 23),
C = FT · F
B = F · FT (3.26)
ds2 − dS 2 = 2 dR · E · dR
= 2 dr · e · dr, (3.27)
which extend (3.3) and (3.6) to three dimensions. In the literature, vari-
ous names are attached to these strain tensors. For instance, Green and
Saint-Venant are often associated with E, and Almansi and Hamel with e.
Equations (3.25) yield
ds2 − dS 2 = dR · C · dR − dR · dR = dR · (C − I) · dR
= dr · dr − dr · B−1 · dr = dr · (I − B−1 ) · dr
2E = C − I = FT · F − I
2e = I − B−1 = I − F−T · F−1 . (3.28)
Because E and e are tensors, the components of strain relative to any co-
ordinate system can be obtained by double-dotting with the corresponding
2 In Chap. 2, B was defined as a coordinate transformation tensor. From this point
where
These equations show that the components of the strain tensors reflect
differences between the metrics of the deformed and undeformed bodies.
Here, we also note that substituting Eqs. (3.13) and (3.30) into (3.27) gives
the alternate definitions
ds2 − dS 2 = 2EIJ dX I dX J
= 2eij dxi dxj (3.33)
for the Lagrangian and Eulerian strains.
Now, we pause to mention two important points. First, since E and e
are tensors, all the tools of tensor analysis in Chap. 2 can be used, including
the computation of principal strains and strain invariants (see Sec. 3.7). To
illustrate the second point, we examine the components of e defined by
e = eI ∗ J ∗ gI gJ .
Equations (3.12), (3.28)2 , and (3.29)2 give
2e = I − B−1 = gIJ gI gJ − GIJ gI gJ
and, therefore,
2eI ∗ J ∗ = gIJ − GIJ .
3.4. Deformation and Strain Tensors 77
Comparison with Eqs. (3.31)1 and (3.32)1 shows that EIJ = eI ∗ J ∗ . Thus,
although the strain tensors E and e are fundamentally different, the compo-
nents of E with respect to the basis {GI GJ } are equal to the components
of e with respect to the basis {gI gJ }. This equivalency of components
of different tensors in appropriate bases will play a major role in our un-
derstanding of the relations between the components of the various stress
tensors discussed in the next chapter.
Inserting this relation into Eq. (3.35) shows that the physical Lagrangian
strain components are given by
EIJ
ÊIJ = . (3.39)
G(II) G(JJ)
Example 3.2 Soft biological tissues rarely are devoid of stress in vivo. In
fact, most tissues experience ever-changing three-dimensional forces and de-
formations. To a first approximation, however, many tissues are subjected
to simple tension or compression in one, two, or three dimensions. For ex-
ample, papillary muscles stretch uniaxially to prevent the heart valves from
inverting during systole, and articular cartilage in the knee is compressed
between the femur and tibia to provide a cushion between the surfaces of
the bones. An example of multiaxial loading is that experienced by epithe-
lia, which are thin sheets of cells covering surfaces of tissues and organs.
The endothelium is a specialized epithelium that lines the inner surfaces
of blood vessels. When subjected to blood pressure, endothelia undergo
circumferential and longitudinal stretching with radial compression.
Experiments designed to determine material properties of soft tissues
commonly use uniaxial or biaxial loading protocols. Thus, simple extension
and compression of tissues is of fundamental importance in biomechanics.
3.4. Deformation and Strain Tensors 79
x1 = λ1 X 1 , x2 = λ2 X 2 , x3 = λ 3 X 3 , (3.43)
where the λI are stretch ratios in the coordinate directions. Compute the
Cartesian components of E and e in terms of the λI .
X2,x2
P p
X1
l1X1
X1,x1
R = X I eI , r = x i ei , (3.44)
where the eI = ei are unit vectors along the coordinate axes. Thus, the base
vectors are GI = R,I = eI and gi = r,i = ei , and so GIJ = eI ·eJ = δ IJ and
gij = ei · ej = δ ij . In the following, we keep subscripts and superscripts
distinct, although this is not really necessary, as eI = eI in Cartesian
coordinates.
Equations (3.18) and (3.21)1 give
FT = ∇r
∂ ∂ ∂
= e1 1
+ e2 2
+ e3 (λ1 X 1 e1 + λ2 X 2 e2 + λ3 X 3 e3 )
∂X ∂X ∂X 3
= λ1 e1 e1 + λ2 e2 e2 + λ3 e3 e3 . (3.45)
where
⎡ ⎤
a 0 0
⎣
diag [a, b, c] ≡ 0 b 0 ⎦. (3.47)
0 0 c
The use of a Cartesian basis for both the undeformed and deformed co-
ordinates allows us to employ matrix algebra without worrying about pick-
ing up extra factors that general base vectors may include. Thus, with all
components referred to the unit base vectors eI = eI = ei = ei , Eqs. (3.26)
give
[CIJ ] = [FIi ]T [FJj ] = diag [λ21 , λ22 , λ23 ]
[eij ] = gij − (B −1 )ij = 12 δ ij − (B −1 )ij
1
2
1
= diag 2 (1 − λ−2 1 −2 1 −2
1 ), 2 (1 − λ2 ), 2 (1 − λ3 ) , (3.49)
which are consistent with Eqs. (3.5) and (3.7).
g2 g2*
g1= g1*
in which
x1 = X 1 + kX 2 , x2 = X 2 , x3 = X 3 . (3.51)
Solution. (a) The deformation gradients defined in Eq. (3.15)1 are given
by the matrix
⎡ ⎤
i i 1 k 0
FI = x , I = ⎣ 0 1 0 ⎦ . (3.52)
0 0 1
F = gI G I = g1 ∗ G 1 + g2 ∗ G 2 + g3 ∗ G 3
= e1 e1 + (ke1 + e2 )e2 + e3 e3
= e1 e1 + ke1 e2 + e2 e2 + e3 e3 , (3.55)
which agrees with Eq. (3.53).
(c) As in the previous example, since the basis of FIi is the set of
Cartesian dyads {ei eI }, Eqs. (3.26) and elementary matrix operations give
⎡ ⎤⎡ ⎤ ⎡ ⎤
i T j 1 0 0 1 k 0 1 k 0
[CIJ ] = FI FJ = ⎣ k 1 0 ⎦ ⎣ 0 1 0 ⎦ = ⎣ k 1 + k 2 0 ⎦
0 0 1 0 0 1 0 0 1
⎡ ⎤⎡ ⎤ ⎡ ⎤
T 1 k 0 1 0 0 1 + k2 k 0
[Bij ] = i
FI FJj = ⎣ 0 1 0 ⎦⎣ k 1 0 ⎦ = ⎣ k 1 0 ⎦.
0 0 1 0 0 1 0 0 1
(3.56)
Note that CIJ = gIJ = gI · gJ [see Eq. (3.32)1 ] also could be computed
directly using Eqs. (3.54).
Example 3.4 Blood vessels, ureters, intestines, tracheas, and plant stems
are examples of biological tubes. Even the left ventricle has been modeled
as a thick-walled tube (Arts et al., 1979; Chadwick, 1982; Tozeren, 1983;
Guccione et al., 1991; Taber, 1991; Taber et al., 1996). Thus, deformation
of a tube, i.e., a hollow cylinder, is an important problem in biomechanics.
Consider combined extension, inflation, and torsion of a circular
tube (Fig. 3.7). For convenience, choose (X 1 , X 2 , X 3 ) = (R, Θ, Z) and
(x1 , x2 , x3 ) = (r, θ, z) as cylindrical polar coordinates for a point before and
3.4. Deformation and Strain Tensors 83
ez
eZ
y
Y z
Z
eq er
eQ eR r
R q
Q x
X
Undeformed
Deformed
associated with the undeformed and deformed positions of the element, re-
spectively. With (3.57)–(3.59), Eqs. (3.10) and (3.11) give the base vectors
G1 = R,R = eR
G2 = R,Θ = ReΘ
G3 = R,Z = eZ
g1 = r,r = er
g2 = r,θ = reθ
g3 = r,z = ez
g1 ∗ = r,R = r (R) er
g2 ∗ = r,Θ = reθ
g3 ∗ = r,Z = ψreθ + λez . (3.60)
In terms of the convected base vectors gI , we can compute the defor-
mation gradient tensor F = gI GI after determining the GI . Since the
cylindrical polar coordinates (R, Θ, Z) are orthogonal, this is a relatively
simple matter. In fact, we already have found the GI for cylindrical coor-
dinates in Example 2.3; Eqs. (2.24) give
G 1 = eR , G2 = R−1 eΘ , G 3 = eZ . (3.61)
For simplicity in this example, we write all the (orthogonal) unit vectors
with subscripts. Thus,
F = g1 ∗ G 1 + g2 ∗ G 2 + g3 ∗ G 3
= r (R) er eR + (reθ )(R−1 eΘ ) + (ψreθ + λez )eZ
= r (R) er eR + (r/R)eθ eΘ + λez eZ + ψreθ eZ , (3.62)
and the matrix of components relative to the basis {ei eI } is
⎡ ⎤
r 0 0
F(ei eI ) = ⎣ 0 r/R ψr ⎦ . (3.63)
0 0 λ
Note that, because the basis is composed of unit dyads, i.e., dyads consisting
of unit vectors, all the components of F in these equations are nondimen-
sional. In fact, as shown next, they also are the physical components of
F.
Relative to the natural basis, the tensor components of F are given by
[see Eqs. (3.15)1 and (3.22)1 ]
⎡ ⎤ ⎡ ⎤
i i r,R r,Θ r,Z r 0 0
F(gi GI ) = FI = x ,I = ⎣ θ,R θ,Θ θ,Z ⎦ = ⎣ 0 1 ψ ⎦ , (3.64)
z,R z,Θ z,Z 0 0 λ
3.4. Deformation and Strain Tensors 85
where Eqs. (3.57) have been used. Adapting Eq. (2.132) gives the physical
components
i i
g(ii)
F̂I = FI (3.65)
G(II)
The last line follows from Eq. (3.42), since the eI = eI are orthogonal unit
vectors. The ÊIJ in this expression do indeed agree with those of Eq. (3.70).
Finally, we note one last point. The matrix [EIJ ] is always symmetric
since CIJ = gIJ = gJI . Equation (3.70) shows that the matrix of physical
components also is symmetric, consistent with the discussion in Sec. 2.7.
However, the matrix of mixed components, given by EJI = EJK GKI , is
⎡ ⎤
r 2 − 1 0 0
⎢ ⎥
I 1 ⎢ r2 ψr 2 ⎥
EJ = ⎢ ⎢ − ⎥, (3.72)
2⎣
0
R2
1
R2 ⎥
⎦
0 ψr 2 λ2 + ψ 2 r 2 − 1
which is not symmetric. (Here, we have used the relation G(II) = 1/G(II)
for orthogonal coordinates.) This demonstrates that, even though E = ET ,
EJI = EIJ in general [see text following Eqs. (2.41)].
Example 3.5 If the tube of the previous example is elastic, it does not
matter whether the cylinder is extended, inflated, or twisted first (or all
three simultaneously). The final state of stress is the same in all cases, and
so we did not consider the order of the imposed deformations. If dissipa-
tion occurs, however, the path the cylinder takes to its final configuration
may affect the stresses, even though the final strain state is the same. Be-
cause viscoelastic effects often are important in biomechanics, it is useful
to illustrate how such an analysis may proceed. In this example, therefore,
consider the following two paths to the same final deformed configuration:
Path A:
1 Extension: r1 = R, θ 1 = Θ, z1 = λZ
2 Inflation: r2 = r(r1 ), θ 2 = θ 1 , z2 = z1
3 Torsion: r = r2 , θ = θ2 + ψZ = θ 2 + ψz2 /λ, z = z2
Path B:
1 Torsion: r1 = R, θ1 = Θ + ψZ, z1 = Z
2 Extension: r2 = r1 , θ 2 = θ 1 , z2 = λz1
3 Inflation: r = r(r2 ), θ = θ 2 , z = z2
In both cases, the final location of an arbitrary element is given by
Eqs. (3.57). Show that both paths produce the same total deformation
gradient tensor of (3.64).
3.4. Deformation and Strain Tensors 87
Solution. For each step in Paths A and B, the following relative defor-
mation gradients are computed [see Eq. (3.64)]:
⎡ ⎤(A) ⎡ ⎤
i (A) r1 , R r1 , Θ r1 , Z 1 0 0
FI 1 = ⎣ θ 1 ,R θ 1 ,Θ θ1 ,Z ⎦ =⎣ 0 1 0 ⎦
z1 ,R z1 ,Θ z1 ,Z 0 0 λ
⎡ ⎤(A) ⎡ ⎤
i (A) r2 ,r1 r 2 , θ1 r 2 , z1 r (R) 0 0
FI 2 = ⎣ θ 2 ,r1 θ 2 , θ1 θ 2 , z1 ⎦ =⎣ 0 1 0 ⎦
z2 ,r1 z2 ,θ1 z2 ,z1 0 0 1
⎡ ⎤(A) ⎡ ⎤
i (A) r,r2 r,θ2 r,z2 1 0 0
FI 3 = ⎣ θ,r2 θ,θ2 θ,z2 ⎦ =⎣ 0 1 ψ/λ ⎦
z,r2 z,θ2 z,z2 0 0 1
⎡ ⎤(B) ⎡ ⎤
i (B) r1 , R r1 , Θ r1 , Z 1 0 0
FI 1 = ⎣ θ 1 ,R θ 1 ,Θ θ1 ,Z ⎦ =⎣ 0 1 ψ ⎦
z1 ,R z1 ,Θ z1 ,Z 0 0 1
⎡ ⎤(B) ⎡ ⎤
i (B) r2 ,r1 r 2 , θ1 r 2 , z1 1 0 0
FI 2 = ⎣ θ 2 ,r1 θ 2 , θ1 θ 2 , z1 ⎦ =⎣ 0 1 0 ⎦
z2 ,r1 z2 ,θ1 z2 ,z1 0 0 λ
⎡ ⎤(B) ⎡ ⎤
i (B) r,r2 r,θ2 r,z2 r (R) 0 0
FI 3 = ⎣ θ,r2 θ,θ2 θ,z2 ⎦ =⎣ 0 1 0 ⎦ (3.73)
z,r2 z,θ2 z,z2 0 0 1
These matrices give the deformation gradients for each step relative to the
geometry of the previous step.
With each of these matrices, we can compute a deformation gradient
tensor Fk = (FIi )k gi GI , where k indicates a step in the deformation path.
[Note that the gi and GI given by Eqs. (3.60) remain fixed.] Then, the
successive deformations of a line element are given by
dr1 = F1 · dR
dr2 = F2 · dr1
dr = F3 · dr2
and, therefore,
dr = F3 · [F2 · (F1 · dR)] .
This equation can be written
dr = F · dR,
88 Analysis of Deformation
where
F = F3 · F2 · F1 (3.74)
is the total deformation gradient tensor. Alternatively, the chain rule gives
the equivalent expression
i ∂xi ∂xi ∂xj2 ∂xk1
FI = = j k ∂X I
= [Fji ]3 [Fkj ]2 [FIk ]1 . (3.75)
∂X I ∂x
∂x2 1
The reader may verify that substituting Eqs. (3.73) for either Path A or
Path B reproduces Eq. (3.64) (see Problem 3.5). This exercise demonstrates
that the final deformation is independent of the deformation path; but
again, if the material is inelastic, the stresses generally depend on the path.
∂r ∂u
FT = = I+ = I + ∇u
∂R ∂R
∂R ∂u
F−T = = I− ¯
= I − ∇u (3.77)
∂r ∂r
or, since IT = I,
F = (FT )T = I + (∇u)T
F−1 = (F−T )T = I − (∇u)
¯ T. (3.78)
Inserting these equations into Eqs. (3.28) yields
2E = FT · F − I
= [I + ∇u] · I + (∇u)T − I
3.5. Strain-Displacement Relations 89
2e = I − F−T · F−1
= ¯ · I − (∇u)
I − I − ∇u ¯ T
e = 12 [∇u
¯ + (∇u)
¯ T − (∇u)
¯ · (∇u)
¯ T ]. (3.80)
¯
∇u = gi (uj gj ),i = gi (uj ||i gj ) = uj ||i gi gj
= gI (uJ ∗ gJ ),I = gI (uJ ∗ ||I gJ ) = uJ ∗ ||I gI gJ , (3.84)
90 Analysis of Deformation
where the single vertical lines and double vertical lines indicate covariant
differentiation with respect to B and b, respectively. In particular, the three
relations in (3.84) involve differentiation of the base vectors GI , gi , and gI ,
respectively. Thus, Eq. (2.151)2 gives
uI |J = uI ,J −uK ΓK
IJ
ui ||j = ui ,j −uk Γ̄kij
uI ∗ ||J = uI ∗ ,J −uK ∗ Γ̄K
IJ , (3.85)
IJ = GI ,J · G
ΓK K
Γ̄kij = gi ,j · gk
IJ = gI ,J · g ,
Γ̄K K
(3.86)
last line. In Eqs. (3.88) and (3.89), gjl and gJL raised the subscripts on
ul and uL∗ , respectively. To show that gJL is needed to raise the index of
uL∗ , we consider the representations
∗
u = u I ∗ g I = u I gI .
Dotting with gJ yields
∗
(uI ∗ gI ) · gJ = (uI gI ) · gJ
or
∗ ∗
uI ∗ gIJ = uI δ JI = uJ , (3.90)
which is the desired result.
Finally, comparing Eqs. (3.81) with (3.87)–(3.89) reveals that
EIJ = 12 (uI |J + uJ |I + uK |I uK |J )
eij = 12 (ui ||j + uj ||i − uk ||i uk ||j )
∗
= 12 (uI ∗ ||J + uJ ∗ ||I − uK ∗ ||I uK ||J ). (3.91)
3.6.1 Stretch
Let dS and ds be the lengths of a differential line element before and after
deformation, respectively (see Fig. 3.4, page 73). The stretch ratio ds/dS
depends on the specified orientation of either the undeformed or the de-
formed element. In other words, we can compute a Lagrangian stretch
92 Analysis of Deformation
ratio Λ(N) for an element that occupies a specified direction N in the un-
deformed body B or an Eulerian stretch ratio λ(n) for an element that
occupies a specified direction n in the deformed body b, with N and n
being unit vectors (Fig. 3.4). In general, N and n can be chosen arbitrar-
ily in B and b, respectively. However, if N and n are bound to the same
material line element, i.e., N is convected into n during deformation, then
Λ(N) = λ(n) . On the other hand, if we choose N = n, then the material
elements associated with each of these unit vectors differ if local rotation
occurs, and so Λ(N) = λ(n) in general.
Noting the geometry of Fig. 3.4, we write
n ds = F · N dS or N dS = F−1 · n ds,
n Λ(N) = F · N = N · FT
N λ−1
(n) = F−1 · n = n · F−T . (3.93)
Λ2(N) = N · C · N
λ−2
(n) = n · B
−1
· n. (3.94)
∗ ds − dS
E(N) = = Λ(N) − 1
dS
ds − dS 1
e∗(n) = =1− . (3.95)
ds λ(n)
3.6. Geometric Measures of Deformation 93
Example 3.6 Suppose the components of the strain tensors E and e are
known in the Cartesian coordinates X I = xi . Compute the extension ratios
in terms of strain components for an element aligned with the X 1 -axis in the
undeformed body and for an element aligned with this axis in the deformed
body.
Solution. In the first case, N = e1 , and in the second case, n = e1 .
Thus, Eqs. (3.28) and (3.94) yield
Λ2(1) = e1 · C · e1 = e1 · (I + 2E) · e1 = 1 + 2E11
λ−2
(1) = e1 · B−1 · e1 = e1 · (I − 2e) · e1 = 1 − 2e11 , (3.96)
which show that Λ(1) = λ(1) in general. In this problem, the stretch ratios
depend only on normal strains. However, if either N or n is not aligned
with a coordinate axis, then shear strains would be involved. The extension
ratios, from Eqs. (3.95), are
1
∗
E(1) = Λ(1) − 1 = (1 + 2E11 ) 2 − 1
1
e∗(1) = 1 − (λ(1) )−1 = 1 − (1 − 2e11 ) 2 , (3.97)
and these equations give
∗ ∗2
E11 = E(1) + 12 E(1)
e11 = e∗(1) − 12 e∗2
(1) . (3.98)
These last relations show that if the extension ratios are small, then E11 =∼
∗ ∼ ∗
E(1) and e11 = e(1) , i.e., the linear normal strain components have the usual
physical interpretation of change in length divided by undeformed length.
Example 3.7 For the case of simple shear (see Example 3.3 and Fig. 3.6,
page 81), compute the stretch ratios for elements that are parallel to the
X 2 -axis (a) before deformation and (b) after deformation.
Solution. (a) On setting N = e2 , Eqs. (3.56)1 and (3.94)1 give the
stretch ratio
Λ2(2) = e2 · C · e2 = C22 = 1 + k 2 .
Note that
√ the geometry of Fig. 3.6 shows that the deformed length of side
AB is 1 + k 2 , in agreement with this result.
(b) Setting n = e2 in Eq. (3.94)2 gives
λ−2
(2) = e2 · B
−1
· e2 = (B −1 )22 .
94 Analysis of Deformation
λ2(2) = (1 + k 2 )−1 .
∼ λ(2) ∼
Again, we see that Λ(2) = λ(2). For k << 1, however, Λ(2) = = 1, i.e.,
to a first approximation, simple shear involves no extension in the linear
case (relative to the coordinate axes).
R = R eR , r = r er . (3.100)
not the actual solution to this problem. The true deformation field must be found by
solving the full boundary value problem of nonlinear elasticity (see Chap. 6).
3.6. Geometric Measures of Deformation 95
(a)
Fig. 3.8. (a) Annular membrane model for wound healing. With outer edge fixed, inner
edge moves inward to close the wound. (b) Stretch ratios in model after healing. (inner
edge: R = 1; outer edge: R = 2)
N2
n2
p dr2
dR2 F
P dr1
dR1
N1 n1
B b
Fig. 3.9. Change in angle (shear) between two differential line elements.
3.6.2 Shear
Shear strains characterize the change in angle, due to deformation, between
two line elements. The elements
dR1 = N1 dS1 = F−1 · dr1 = dr1 · F−T
dR2 = N2 dS2 = F−1 · dr2 = dr2 · F−T (3.105)
at a point P in B deform into the elements
dr1 = n1 ds1 = F · dR1 = dR1 · FT
dr2 = n2 ds2 = F · dR2 = dR2 · FT (3.106)
at the point p in b (Fig. 3.9), where the Ni and ni are unit vectors and F
is the local deformation gradient tensor [see Eqs. (3.19) and (3.20)].
The cosines of the angles between the undeformed and the deformed
elements are
dR1 · dR2 (dr1 · F−T ) · (F−1 · dr2 )
cos(N1 , N2 ) = N1 · N2 = = 1 1
dS1 dS2 (dr ·B−1 ·dr ) 2 (dr ·B−1 ·dr ) 2
1 1 2 2
in which Eqs. (3.25) have been used. With Eqs. (3.26), (3.94), (3.105), and
(3.106), these relations can be written
n1 · B−1 · n2
N1 · N2 = 1 1
= λ(n1 ) λ(n2 ) (n1 · B−1 · n2 )
(n1 · B−1 · n1 ) 2 (n2 · B−1 · n2 ) 2
N1 · C · N2 N1 · C · N2
n1 · n2 = 1 1 = . (3.108)
Λ(N1 ) Λ(N2 )
(N1 · C · N1 ) 2 (N2 ·C· N2 ) 2
If n1 and n2 are given, the first equation provides N1 · N2 ; if N1 and N2
are given, the second yields n1 · n2 . Of course, if there is no deformation,
then C = B = I and both equations give n1 · n2 = N1 · N2 . Finally, we
define the shears
Example 3.9 Suppose the strain components EIJ are known in Cartesian
coordinates. Compute the shear in the x1 x2 -plane in terms of the EIJ for
(a) an element that is rectangular in the undeformed body B, with edges
parallel to the X 1 and X 2 axes; and (b) an element that is rectangular in
the deformed body b, again with edges parallel to the X 1 and X 2 axes.
Solution. In the first case, N1 = e1 and N2 = e2 , while in the second
case, n1 = e1 and n2 = e2 (e1 · e2 = 0). Thus, Eqs. (3.28), (3.96), and
(3.110) yield
I + 2E 2E12
Γ(12) = e1 · − I · e2 = 1
Λ(1)Λ(2)
[(1 + 2E11 )(1 + 2E22)] 2
2e12
γ (12) = e1 · I − λ(1) λ(2)(I − 2e) · e2 = 1
[(1 − 2e11 )(1 − 2e22 )] 2
(3.111)
98 Analysis of Deformation
for cases (a) and (b), respectively. Note that the shears depend on the shear
strains, E12 and e12 , as well as the normal strains. If the normal strains
are small compared to unity, then
Γ(12) ∼
= 2E12 , γ (12) ∼
= 2e12 ,
which agree with the usual interpretation for linear shear strain compo-
nents.
Example 3.10 For simple shear (see Example 3.3 and Fig. 3.6, page 81),
Γ(12) provides a measure of the change in angle between line elements orig-
inally aligned with the X 1 and X 2 axes. Determine Γ(12) in terms of k.
Solution. With N1 = e1 and N2 = e2 , Eq. (3.110)1 gives
C C12 C12
Γ(12) = e1 · − I · e2 = = √ ,
Λ(1)Λ(2) Λ(1) Λ(2) C11 C22
where we have inserted Λ2(1) = e1 · C · e1 = C11 and Λ2(2) = e2 · C · e2 = C22.
Using Eq. (3.56)1 , we find
k
Γ(12) = √ .
1 + k2
A glance at the geometry of Fig. 3.6 reveals that Γ(12) = sin γ.
= FIi gi dX(k)
I
or
dxi(k) = FIi dX(k)
I
= xi ,I dX(k)
I
. (3.114)
3.6. Geometric Measures of Deformation 99
This relation, which involves Eq. (3.15), is an expression of the chain rule
for differentiation.
The volumes of the parallelepiped before and after deformation, respec-
tively, are
dV = dR(1) · (dR(2) × dR(3))
dv = dr(1) · (dr(2) × dr(3)). (3.115)
Substituting Eqs. (3.112) and noting Eqs. (2.103)2 and (2.115)1 gives
dV = GI · (GJ × GK ) dX(1)
I J
dX(2) K
dX(3)
√ I J K
= G eIJK dX(1) dX(2) dX(3)
dv = gi · (gj × gk ) dxi(1)dxj(2)dxk(3)
√
= g eijk dxi(1)dxj(2)dxk(3), (3.116)
where eIJK and eijk are the permutation symbols defined by Eq. (2.99),
and
G = det [GIJ ] , g = det [gij ] . (3.117)
Now, using Eqs. (3.114), (2.98)1 , and (3.116)1 in order, we can write
Eq. (3.116)2 as
√
dv = g (eijk FIi FJj FK
k I
) dX(1) J
dX(2) K
dX(3)
√ I J K
= g (eIJK j) dX(1) dX(2) dX(3)
√ √
= g (dV / G) j,
where
j ≡ det FIi . (3.118)
dv g
J≡ = j, (3.119)
dV G
which is the ratio of the deformed to the undeformed volume of the ele-
ment. An alternate form for J can be found by taking the determinant of
Eq. (3.65) and comparing the result to (3.118) and (3.119) to obtain
J = det[F̂Ii], (3.120)
where the F̂Ii are physical components of F. (Note the distinction between
j and J.)
Here, we mention two important special cases:
100 Analysis of Deformation
det C = J 2 . (3.122)
In the manipulations above, we also have used the fact that det[GIK GJK ] =
det[δIJ ] = 1, and so det[GIK ] = 1/ det[GIK ] = 1/G. Now, Eqs. (3.121) and
(3.122) give
J = det F. (3.123)
At first glance, it appears that Eqs. (2.77) and (3.118) imply j = det F,
which is generally inconsistent with Eqs. (3.119) and (3.123). Similar to the
above discussion, the explanation is that j is expressed in terms of mixed
componentsof F involving both undeformed and deformed base vectors.
The factor g/G in (3.119) links these bases. These subtle distinctions in
computing the quantities j and J are easy to overlook when reading the
literature.
Soft biological tissues usually are assumed to be incompressible. In this
case, J = 1 (dv = dV ) must be enforced as a material constraint, and the
3.6. Geometric Measures of Deformation 101
Example 3.11 Compute the dilatation ratios for the problems discussed
in Examples 3.2–3.4, i.e., uniform extension, simple shear, and axisymmet-
ric deformation of a tube.
Solution. For uniform extension, Eq. (3.45) expresses F in a Cartesian
dyadic basis. Since the components are mixed and because the undeformed
and deformed bases coincide, the definition for the determinant of a tensor
(2.77) gives
λ1 0
0
J = det F = 0 λ2 0 = λ1 λ2 λ3 .
0 0 λ3
(Here, we note that the distinctions between covariant, contravariant, and
mixed components vanish in Cartesian coordinates.) If a block of material
is stretched by λ1 = λ in the X 1 -direction and deforms freely in the other
two directions, then symmetry demands that λ2 = λ3 . If, in addition, the
material is incompressible, then
J = λλ22 = 1
−1/2
which gives λ2 = λ3 = λ .
For simple shear, Eq. (3.55) gives
1 k 0
J = det F = 0 1 0 = 1.
0 0 1
Hence, this deformation is isochoric for all values of the shear parameter k.
Lastly, combined extension, inflation, and torsion of a tube is treated in
three ways. First, Eqs. (3.64) and (3.66) give
j = det FIi = λr
G = det [GIJ ] = R2
g = det [gij ] = r 2 ,
102 Analysis of Deformation
N n
dR(2)
dr(2)
dA F
da
dR(1)
dr(1)
Next, using Eqs. (2.115)1 , (3.16)2 , and (3.114) to replace ijk , gk , dxi(1),
and dxi(2) in da gives
√
da = k K
g eijk (FK g )(FIi dX(1)
I
)(FJj dX(2)
J
)
√ i j k K I J
= g (eijk FI FJ FK )g dX(1)dX(2)
√
= g (eIJK j)gK dX(1) I J
dX(2) . (3.127)
In the last line, we have used the determinant formula (2.98)1, with j
defined by Eq. (3.118). Finally, this expression can be written in the form
da = J dA · F−1 , (3.128)
as verified by the manipulation
da = J( IJK GK dX(1) I J
dX(2) ) · (GL gL)
√
= J( G eIJK )δ K L I
L g dX(1) dX(2)
J
√
= j g eIJK gK dX(1) I J
dX(2) ,
which employs Eqs. (3.126)1 , (3.22)2 , (2.115)1, and (3.119), in this order.
This expression agrees with Eq. (3.127). Equation (3.128) is known as
Nanson’s formula.
(C − Λ2(N) I) · N = 0
(B − λ2(n) I) · n = 0. (3.133)
Solving these equations (see Sec. 2.5.3) provides the principal stretch
ratios Λ(Ni ) and λ(ni ) (i = 1, 2, 3) and the corresponding principal axes
Ni and ni . Equations (3.5)2 and (3.7) then give the principal strains
E(Ni ) = 1
2
(Λ2(Ni ) − 1)
1 −2
e(ni ) = 2 (1 − λ(ni ) ). (3.134)
3.7. Principal Strains 105
It can be shown that since C and B are symmetric tensors, their eigen-
values are real and their eigenvectors are mutually orthogonal. Also, as
n is taken in the direction of the deformed element with undeformed ori-
entation N, λ(n) = Λ(N) , and so the eigenvalues of C and B are equal.
Thus, the orthogonal principal axes Ni at the point P in the undeformed
body translate and rotate into the orthogonal principal axes ni at the point
p in the deformed body. In addition, since the eigenvalues represent the
deformation or strain components relative to the principal axes, we can
write
F = Λ1 n1 N1 + Λ2 n2 N2 + Λ3 n3 N3
C = Λ21 N1 N1 + Λ22 N2 N2 + Λ23 N3 N3
B = λ21 n1 n1 + λ22 n2 n2 + λ23 n3 n3
E = E1 N1 N1 + E2 N2 N2 + E3 N3 N3
e = e1 n1 n1 + e2 n2 n2 + e3 n3 n3 , (3.135)
where we have used the invariance properties of these tensors, along with
the definitions Ei ≡ E(Ni ) and ei ≡ e(ni ) . The relation for F follows from
Eqs. (3.93)1 and is consistent with the definitions C = FT ·F and B = F·FT .
Equations (3.135), containing tensors written in terms of their eigenvalues
and eigenvectors, are called spectral representations of these tensors.
By definition, these invariants have the same values when computed in any
coordinate system. In principal coordinates, the shear strains vanish and
the invariants become
I1 = 3 + 2(E1 + E2 + E3 )
I2 = 3 + 4(E1 + E2 + E3 ) + 4(E1 E2 + E2 E3 + E3 E1 )
I3 = (1 + 2E1 )(1 + 2E2 )(1 + 2E3 ), (3.143)
where the Ei are principal Lagrangian strains.
Example 3.12 Compute the strain invariants for Examples 3.2–3.4, i.e.,
uniform extension, simple shear, and axisymmetric deformation of a tube.
Solution. For uniform extension, Eqs. (3.48)1 and (3.142) yield (recall
that CJI = CIJ in Cartesian coordinates)
I1 = C11 + C22 + C33 = λ21 + λ22 + λ23
1
! 1 2 3 2
1 2 2 2 3 2
"
I2 = 2 (C1 + C2 + C3 ) − (C1 ) + (C2 ) + (C3 )
= C11 C22 + C22 C33 + C33 C11
= λ21 λ22 + λ22 λ23 + λ23 λ21
I3 = det CJI = λ21 λ22 λ23 . (3.144)
For simple shear, Eq. (3.56)1 gives (again, CJI = CIJ )
I1 = C11 + C22 + C33 = 3 + k 2
I2 = C11 C22 + C22 C33 + C33 C11 − (C21 )2 = 3 + k 2
I3 = det CJI = 1. (3.145)
Finally, for combined extension, inflation, and torsion of a cylinder,
Eqs. (3.72) give the mixed strain components EJI . Putting these into
Eqs. (3.142) yields
r2
I1 = 3 + 2(E11 + E22 + E33 ) = r 2 + + λ2 + ψ2 r 2
R2
I2 = 3 + 4(E11 + E22 + E33 + E11 E22 + E22 E33 + E33 E11 − E32 E23 )
r2 2 2 2 λ2 r 2
= r 2 + λ + ψ r +
R2 R2
λ2 r 2 r 2
I3 = det δ IJ + 2EJI = . (3.146)
R2
108 Analysis of Deformation
F = Θ · U = V · Θ, (3.147)
y F
1 1
4 T
x
1 O
Fig. 3.11. Uniaxial stretch without rotation (F = U = V) of a unit cube with dimensions
fixed in the y and z directions. Line elements parallel to principal axes (e.g., edges)
stretch but do not rotate. Other line elements (e.g., the diagonal) both stretch and
rotate.
(a)
dR U dr Q dr
(b)
dR Q dR V dr
ΘT · Θ = (F · U−1 )T · (F · U−1 )
= (U−T · FT ) · (F · U−1 )
= U−1 · (FT · F) · U−1
V = Θ · U · ΘT . (3.152)
The manipulations
VT = (Θ · U · ΘT )T = (ΘT )T · UT · ΘT
= Θ · U · ΘT
= V
show that V is a symmetric tensor, and the rigid-body rotations of U to
produce V in (3.152) preserve its positive definite character. To obtain the
second decomposition, we substitute for U from the first decomposition of
(3.147) to get
V = Θ · (Θ−1 · F) · ΘT
= (Θ · Θ−1 ) · F · ΘT
= I · F · Θ−1
= F · Θ−1 ,
F = Θ · U ,
3.8. Stretch and Rotation Tensors 111
where
U = Λ1 N1 N1 + Λ2 N2 N2 + Λ3 N3 N3
and Θ is a rotation tensor. Then,
C = FT · F = (Θ · U )T · (Θ · U )
= U T · (Θ T · Θ ) · U
= U · U = U 2 .
Since the eigenvalues of C are unique, the Λi must equal the Λi in
Eq. (3.150). Thus, U = U which proves uniqueness. The uniqueness
of Θ and V follows from their definitions.
Finally, we note that substituting Eq. (3.147) into (3.26) yields
C = FT · F = (Θ · U)T · (Θ · U) = (UT · ΘT ) · (Θ · U)
= UT · (ΘT · Θ) · U = UT · I · U
B = F · FT = (V · Θ) · (V · Θ)T = (V · Θ) · (ΘT · VT )
= V · (Θ · ΘT ) · VT = V · I · VT .
Because U and V are symmetric tensors, we have
C = U2 , B = V2 . (3.153)
or
Θ · ΘT = ΘT · Θ = I, (3.156)
which shows that Θ does indeed satisfy the criterion for an orthogonal
(rotation) tensor. Next, the rotated element dR̂ is stretched (or shortened)
into the element dr (Fig. 3.12b) according to
dr = V · dR̂ = V · (Θ · dR + D)
= (V · Θ) · dR + V · D
= dR · (V · Θ)T + V · D, (3.157)
dr̂ = U · dR.
dr = Θ · dr̂ + D
= (Θ · U) · dR + D = dR · (Θ · U)T + D, (3.159)
where we stipulate that the rotation tensor Θ is the same one introduced
above. Thus,
∂r
FT = = (Θ · U)T → F = Θ · U. (3.160)
∂R
Equations (3.158) and (3.160) are consistent with the polar decomposition
theorem (3.147).
In summary, the deformation gradient tensor F can be decomposed
into a deformation U followed by a rigid-body rotation Θ or a rigid-body
rotation Θ followed by a deformation V. The stretch tensors U and V
generally provide all the deformation and at least part of the rotation.
In principal directions, however, these tensors include only deformation
(stretch or shortening), with the rotation tensor Θ providing all the rotation
(see Problem 3.14). Although the rotation tensors are the same, U = V in
general. We next investigate some properties of the stretch tensors.
3.8. Stretch and Rotation Tensors 113
n = Θ · N, (3.164)
where Θ is the rotation tensor. Direct substitution into this equation veri-
fies the dyadic representation
Θ = nN. (3.165)
Now, inserting Eqs. (3.147) and (3.164) into (3.163)1 gives
Λ(N) Θ · N = (Θ · U) · N
or
Θ · (U − Λ(N) I) · N = 0.
For an arbitrary rotation Θ, this relation provides the eigenvalue problem
(U − Λ(N) I) · N = 0 (3.166)
for the principal stretch ratios Λ(N) in the unrotated principal directions
N at R in the undeformed body B. Recall, from Eq. (3.147), that the
deformation tensor U operates on dR before the rotation due to Θ.
Similarly, substituting Eq. (3.147) into (3.163)2 yields
λ(n) n = (V · Θ) · N = V · (Θ · N) = V · n,
114 Analysis of Deformation
in which Eq. (3.164) has been used. This equation gives the eigenvalue
problem
(V − λ(n) I) · n = 0 (3.167)
for the principal stretch ratios λ(n) in the rotated principal directions n at
r in the deformed body b. This is consistent with the fact that V gives a
deformation after the rotation due to Θ.
Here, we make several observations. First, comparing Eqs. (3.166) and
(3.167) with Eqs. (3.133) suggests that the eigenvalues of the strain tensors
C and B are equal to the squares of the eigenvalues (principal stretches) of
the stretch tensors U and V, respectively. Equations (3.148) and (3.150)
also are consistent with this assertion. To show that this is indeed the case,
we substitute Eqs. (3.153) into (3.133) to get
(U2 − Λ2(N) I) · N = 0
2
(V − λ2(n) I) · n = 0.
These equations can be written
(U + Λ(N) I) · (U − Λ(N) I) · N = 0
(V + λ(n) I) · (V − λ(n) I) · n = 0,
which are satisfied by Eqs. (3.166) and (3.167). In addition, since the
eigenvalues of C and B are equal, so are the eigenvalues of U and V.
Moreover, the principal directions of C and B correspond to those of U
and V, respectively. Finally, as shown by Eq. (3.164), the tensor Θ rotates
the principal directions of U (and C) into those of V (and B).
In summary, the tensors C, B, U, and V are all symmetric and positive
definite. The eigenvalues of C and B are Λ2(N) = λ2(n) , and those of U and
V are Λ(N) = λ(n) , which are the principal stretch ratios. Moreover, Θ
rotates the equal and mutually orthogonal principal directions N(i) of C
and U at point P in the undeformed body into the equal and orthogonal
principal directions n(i) of B and V at the image p of P in the deformed
body. Thus, we can extend Eqs. (3.164) and (3.165) to write
n(i) = Θ · N(i)
Θ= n(i)N(i) . (3.168)
i
Θ = ni Ni = ex ex + ey ey + ez ez = I,
and we find
U = Θ−1 · F = ΘT · F = F.
For the undeformed diagonal in Fig. 3.11 (left) defined by the vector a0 =
ex + ey , the right stretch tensor transforms a0 into the vector
a = U · a0 = λex + ey ,
which corresponds to the diagonal in the deformed cube (Fig. 3.11, right).
The rotation tensor, being the identity tensor, has no effect. In this simple
problem, therefore, U provides the entire stretch and rotation of the diag-
onal, illustrating how U can both deform and rotate line elements oriented
in non-principal directions (see also Problem 3.14).
Example 3.14 For simple shear of a cube (Example 3.3, page 80), deter-
mine the following:
(a) The eigenvalues and eigenvectors for the deformation tensors C and B.
(b) The rotation tensor Θ and the stretch tensors U and V.
Solution. (a) The eigenvalue problems are given by Eqs. (3.133). For
nontrivial solutions, we must have
det(C − Λ2(N) I) = 0
det(B − λ2(n) I) = 0 (3.169)
or
det CJI − Λ2(N) δ IJ = 0
det Bji − λ2(n) δ ij = 0. (3.170)
116 Analysis of Deformation
Because the GI and gi are Cartesian (unit) base vectors in this problem
(see Example 3.3), we can write CJI = CIJ and Bji = Bij . Thus, Eqs. (3.56)
and (3.170) give
1 − Λ2 k 0
(N)
2 2
k 1 + k − Λ(N) 0 = 0
0 0 1 − Λ(N)
2
1 + k 2 − λ2
(n) k 0
k 1 − λ2(n) 0 = 0. (3.171)
0 0 1 − λ2(n)
These characteristic equations provide the same eigenvalues, and so Λ(Ni ) =
λ(ni ) ≡ λ(i) for the ith eigenvalue. This is expected, because the λ(i) rep-
resent principal stretch ratios for an element that rotates from orientation
N(i) to orientation n(i) during deformation. We also note that expanding
(3.171)1 gives
−Λ6(N) + I1 Λ4(N) − I2 Λ2(N) + I3 = 0
with the invariants given by Eq. (3.145). This agrees with the general
characteristic equation (2.82) with λ = Λ2(N) .
For convenience, we now let
a = 1 + 14 k 2 , b = a−1 .
Then, Eqs. (3.171) yield the principal stretch ratios
λ2(1,2) = 1 + 12 k 2 ± ka, λ(3) = 1, (3.172)
and Eqs. (3.56) and (3.133) provide the corresponding principal directions
e1 + 12 k ± a e2
N(1,2) = , N(3) = e3
2 + 12 k 2 ± ka
e1 + − 12 k ± a e2
n(1,2) = , n(3) = e3 . (3.173)
2 + 12 k 2 ∓ ka
(The denominators make N(1,2) and n(1,2) unit vectors.) It is easy to show
that the N(i) are mutually orthogonal, as are the n(i).
For the special case k = 0.5, these equations give λ(1) = 1.28, λ(2) =
0.781, N1 = 0.615e1 + 0.788e2, N2 = 0.788e1 − 0.615e2, n1 = 0.788e1 +
0.615e2, and n2 = 0.615e1−0.788e2 . The eigenvectors (principal directions)
are shown in Fig. 3.13.
3.8. Stretch and Rotation Tensors 117
X2,x2
N1
n1
N2
n2
X1,x1
0
Fig. 3.13. Principal directions for simple shear of a unit cube (k = 0.5).
U = Θ−1 · F = ΘT · F
V = F · Θ−1 = F · ΘT (3.175)
since the N(i) and n(i) correspond to the directions of the principal stretch
ratios, Eqs. (3.56), (3.94), and (3.173) should give
−1
λ2(i) = N(i) · C · N(i) = n(i) · B−1 · n(i) .
r = r(R)
θ = π−Θ
z = λZ
ST
P
R p r
T
P cut glue P
a0 b
b0 a
Undeformed Everted
r = rer + zez ,
3.9. Approximations 119
we get
∂ 1 ∂ ∂
∇r = eR + eΘ + eZ [r (R)er (θ) + z(Z)ez ]
∂R R ∂Θ ∂Z
∂r r ∂er ∂θ ∂z
= eR e r + eΘ + eZ .
∂R R ∂θ ∂Θ ∂Z
The given mapping and Eq. (2.12)2 yield
F = (∇r)T = λR er eR − λΘ eθ eΘ + λZ ez eZ ,
where
∂r r
λR = , λΘ = , λZ = λ.
∂R R
Because F contains no shear terms, the base vectors ei in the deformed
configuration correspond to principal directions of strain. In most cases,
the components of F in principal coordinates are stretch ratios. Stretch
ratios, however, must always be positive, whereas the eθ eΘ component of
F is negative.
The principal stretch ratios are provided by solving the eigenvalue prob-
lem (3.169)1 for the right Cauchy-Green deformation tensor
C = FT · F = λ2R eR eR + λ2Θ eΘ eΘ + λ2Z eZ eZ .
Doing this reveals that the principal stretch ratios are the λI given above.
Consequently, in principal coordinates, the stretch tensor U = C1/2 takes
the form
U = λR eR eR + λΘ eΘ eΘ + λZ eZ eZ .
With this result, the rotation tensor is given by
Θ = F · U−1
= (λR er eR − λΘ eθ eΘ + λZ ez eZ ) · λ−1 −1 −1
R eR eR + λΘ eΘ eΘ + λZ eZ eZ
= er e R − eθ e Θ + ez e Z .
3.9 Approximations
1
E∗ = 2 (∇u)T + (∇u) . (3.178)
1
Θ∗ = 2
(∇u)T − (∇u) , (3.179)
which represents the average rotation of all line elements passing through
a point.
3.9. Approximations 121
(∇u)T = E∗ + Θ∗
∇u = E∗ − Θ∗ . (3.180)
F = I + (∇u)T ∼
= I + E∗ + Θ∗ . (3.181)
With the 1-D stretch ratio being λ = 1 + E ∗ by (3.2), this equation shows
that the deformational and rotational parts of F add, rather than multiply,
in the linear theory.
E∼
= E∗ + 12 (E∗ · Θ∗ − Θ∗ · E∗ − Θ∗ · Θ∗ ). (3.183)
giving
J∼
= 1 + EII . (3.184)
122 Analysis of Deformation
E∼
= E∗ − 12 Θ∗ · Θ∗ = E∗ + 12 Θ∗ · Θ∗T . (3.185)
E∼
= E∗ + 12 (E∗ · E∗ + E∗ · Θ∗ − Θ∗ · E∗ ). (3.186)
The state of stress in an elastic body depends only on its deformed configu-
ration, regardless of how or how fast it got there. Soft tissues generally are
viscoelastic, however, with viscous losses accompanying the deformation. In
such materials, the final stress state depends on the entire history and rate
of the deformation. Even for elastic bodies, moreover, rate formulations
often are advantageous computationally, and we will use rate equations to
develop constitutive relations for elastic materials in Chap. 5. This section
presents measures for the rate of deformation relative to a given frame of
reference. First, however, we discuss time differentiation of Lagrangian and
Eulerian quantities.
dr ∂r
v= = , (3.188)
dt ∂t
¯ Thus,
and Eq. (3.18)2 gives ∂/∂r = ∇.
dφ(r, t) ∂φ(r, t) ¯
= + v · ∇φ(r, t), (3.189)
dt ∂t
in which the first term represents the time rate of change of φ at a fixed
position r, and the second term gives the rate of change as the particle is
convected through space.
The physical meaning of the convection term in (3.189) is illustrated by
considering the following. Suppose φ has a different value but is constant at
each point in space. Then, ∂φ/∂t = 0 at each point, but a particle traveling
through space takes on different local values of φ as it passes through each
¯ = 0.
point, i.e., dφ/dt = v · ∇φ
With the definition
d ∂ ¯
≡ + v · ∇, (3.190)
dt ∂t
124 Analysis of Deformation
d ∂
= + vi ( ),i . (3.191)
dt ∂t
Consider now the time derivative of a vector. We compute, for example,
the acceleration of a particle as
dv
a= . (3.192)
dt
For the Lagrangian and Eulerian description, respectively, Eq. (3.190) gives
dv(R, t) ∂v
a= =
dt ∂t
dv(r, t) ∂v ¯
a= = + v · ∇v. (3.193)
dt ∂t
dv Dvi
a= = gi , (3.196)
dt Dt
where
D ∂
≡ + vj ( )|j (3.197)
Dt ∂t
L ≡ Ḟ · F−1 . (3.207)
∂v ¯
LT = = ∇v. (3.208)
∂r
L = vi ||j gi gj = vi ||j gi gj
∗
= vI ||J gI gJ = vI ∗ ||J gI gJ . (3.210)
The rate-of-deformation tensor (or stretch rate tensor) and the spin
tensor (or vorticity tensor) are defined to be the symmetric and antisym-
metric components of L, respectively, i.e. [see Eqs. (2.75)]
D ≡ 12 (L + LT ) = DT
Ω ≡ 12 (L − LT ) = −ΩT . (3.211)
which resemble the linear strain and rotation tensors of Eqs. (3.178) and
¯ ∼
(3.179). In fact, in the linear case with ∇ = ∇ and v ∼
= u̇, we find
D∼
= Ė∗ , Ω∼
= Θ̇∗ . (3.213)
There is, however, an important distinction. Whereas Eqs. (3.178) and
(3.179) are approximate relations when displacement gradients are small,
Eqs. (3.212) are exact for arbitrarily large velocity gradients. Also, adding
Eqs. (3.211) gives
L = D + Ω, (3.214)
Ė = FT · D · F. (3.217)
3.10. Deformation Rates 129
ė = D − e · L − LT · e. (3.218)
0 0 λż,z
With the exception of Ḟ11 , Ḟ32 , and Ḟ33 , this expression is the same as
Eq. (3.229). These other terms can be shown to be equivalent by noting
that
∂ ṙ ∂r ∂ ṙ
r,R ṙ,r = = = ṙ,R
∂r ∂R ∂R
λr θ̇,z = λr(ψ̇/λ) = rψ̇
where the chain rule and Eqs. (3.221) have been used. Putting these rela-
tions into (3.231) reproduces Eq. (3.229).
(3.233)
where modifying Eqs. (2.142) and (2.144) gives
(C)
2ΓIJK = CJK ,I +CKI ,J −CIJ ,K
(C) L
ΓIJ = C KLΓIJK . (3.234)
If the CIJ satisfy Eq. (3.233), then they also represent the components of
the deformed metric tensor, as indicated by Eq. (3.32)1. Thus, Eq. (3.233)
provides 34 = 81 compatibility conditions. Fortunately, using CIJ = CJI
and ΓK K
IJ = ΓJI reduces the number of independent and nonidentically van-
ishing equations to six, as in the linear theory [see Eqs. (A.86)]. Similar
equations can be found for EIJ (CIJ = δ IJ + 2EIJ ) and eij (in terms of
Bij ).
3.12. Problems 133
3.12 Problems
(d) Compute the stretch ratio for a fiber that originally is located at the
coordinates (-0.2,0.1,-0.1) and points in the direction of the vector
e1 + 2e3 .
(e) For the same pointas in (d), compute the dilatation ratio using the
expressions J = j g/G and J = det F.
3.4 In material Cartesian coordinates, the displacement field in a body is
u = X2 X3 e1 + X32 e2 .
Compute the strain tensors E(Xi ) and e(xi ).
3.5 In Example 3.5 (page 86), two deformation paths were considered for
extension, inflation, and torsion of a circular tube. Show that both
paths lead to the same total F as given by Eq. (3.62).
3.6 Consider simple shear of a cube (see Example 3.14, page 115).
(a) Show that the rotation tensor of Eq. (3.174) transforms the eigen-
vectors N(i) of Eqs. (3.173) into the vectors n(i).
(b) With C given by (3.56)1 and U by (3.176), confirm the relation
C = U2 .
(c) Compute the left stretch tensor V. Confirm that the relation B =
V2 is consistent with Eq. (3.56)2 .
(d) Show that the shear strains E12 and e12 vanish relative to the prin-
cipal axes of strain.
3.7 A thin membrane with middle surface lying in the xy-plane under-
goes deformation with a displacement field given by u = u(x, y)ex +
v(x, y)ey . For small deformation, use Eq. (3.183) to write the approxi-
mate 2-D strain-displacement relations in Cartesian coordinates.
3.8 In Cartesian coordinates, a deformation is defined by
x1 = X1 cos θ + X2 sin θ
x2 = −X1 sin θ + X2 cos θ
x3 = X3 .
Compute C, E, U, and Θ. Explain the result.
3.9 Relative to Cartesian coordinates, the state of strain at a point in a
body is given by
⎡ ⎤
0.5 0.3 0
E = ⎣ 0.3 0.4 −0.1 ⎦ .
0 −0.1 0.2 (e e ) i j
Determine the change in angle (in degrees) between two differential line
segments, emanating from the point, that are parallel to the vectors
2e1 + 2e2 + e3 and e1 − 5e3 in the undeformed body.
3.12. Problems 135
Analysis of Stress
The classical definition of stress is force per unit area. But which area:
the undeformed area or some other reference area? For small deformation,
the distinction is immaterial, since changes in area are negligible. For large
deformation, however, we must distinguish between the undeformed and
deformed geometries. In this case, the only physically meaningful defini-
tion for stress is force per unit deformed area, which is called true stress.
But the deformed geometry of a solid body is not known a priori, often is
difficult to measure, and may be time-dependent. Experimentalists, there-
fore, find it useful to define a Lagrangian stress or engineering stress
as the force per unit undeformed (reference) area. In addition, the gov-
erning equations sometimes have simpler forms or may be easier to solve
when written in terms of yet another type of stress (to be defined later).
We call any stress that is not the true stress a pseudostress, since it has
no physical significance. Pseudostresses always must be converted into true
stress for physical interpretation. This chapter introduces various types of
stress tensor and discusses the relations between them.
139
140 Analysis of Stress
n
N dP
dA F da
B b
Fig. 4.1. Deformation of area element dA into area element da. The contact force dP
acts on da.
Note that, even though T(N) is referred to the undeformed area dA, dP
=da, then T(N) ∼
still acts on the deformed area da (Fig. 4.1). If dA ∼ = T(n) ,
an approximation used in the linear theory (see Sec. A.3.1).
The second pseudotraction vector is defined in terms of a fictitious
(pseudo) force dP̃ acting on dA as defined by the relation
dP = F · dP̃ = dP̃ · FT . (4.5)
This transformation is analogous to the deformation of the undeformed
length element dR into the deformed element dr, i.e., dr = F · dR. In
terms of this pseudoforce, the second pseudotraction vector is
dP̃
T̃(N) (R,t) = , (4.6)
dA
and combining Eqs. (4.3)–(4.6) yields
Using the traction vectors of Eq. (4.7), we will define three stress tensors:
the Cauchy stress tensor and the first and second Piola-Kirchhoff stress
tensors. The relations between these tensors are discussed in this section,
while the relations between their various components are derived in Sec. 4.3.
(a) (b)
g3 dP
x3 dP
f n f n
x2 dv dP-1
dv da da
g2
dP-2
dP-3
x1 g1
Fig. 4.2. Differential volume element. (a) Element carved out by general curvilinear
coordinate surfaces. (b) Element in limit as volume approaches zero.
3
the x -face, for example, is n3 = −g1 × g2 /|g1 × g2 | = −g3 / g3 · g3 =
−g3 / g33 by Eq. (2.18), and in general
gi
ni = − . (4.8)
g(ii)
Thus, the vector representations of the four
faces of the tetrahedron can be
written n da and n(i)dâ(i) = −g(i)dâ(i) / g(ii) (i = 1, 2, 3).
To relate these areas, we note that they form a closed surface, and it
can be shown that the vectorial sum of the areas must vanish, i.e.,
n da + ni dâi = 0. (4.9)
This can be seen by considering the limiting case where the element col-
lapses into a triangle in the plane x3 = constant (see Fig. 4.2). In this case,
dâ1 = dâ2 = 0 and Eq. (4.9) gives n da = −ni dâi = −n3 dâ3 . Inspection
shows that n = −n3 , and so da = dâ3 as expected. Thus, Eqs. (4.8) and
(4.9) yield
n da = −ni dâi = gi dâi / g(ii). (4.10)
Setting
da = n da = gi dai (4.11)
and comparing with Eq. (4.10) shows that the covariant components of the
area vector da are related to the physical areas of the xi -faces by
dâi
dai = . (4.12)
g(ii)
4.2. Stress Tensors 143
Now, let dP and dP−i be the contact forces acting on the element faces
with unit normals n and ni respectively (Fig. 4.2b). The element also is
subjected to a body force f per unit deformed volume. Then, Newton’s
second law of motion for the element gives
dP+ dP−1 + dP−2 + dP−3 + f dv = (ρ dv)a, (4.13)
where ρ is the mass density and dv the volume of the element, and a is
the acceleration of its center of mass. The minus sign in the superscript
of dP−i indicates that the force acts on a negative face of the element,
i.e., the normal to the face makes an acute angle with the −xi -direction
(Fig. 4.2). The face of the element adjacent to this surface is positive,
with a normal making an acute angle with the +xi -direction, and so the
contact force on this adjacent area is denoted dPi .1 By Newton’s third law
of action-reaction, dP−i = −dPi and Eq. (4.13) becomes
dP− dP1 − dP2 − dP3 + f dv = ρa dv. (4.14)
The true traction vector acting on the area da is defined by Eq. (4.3), and
similarly
dP(i)
T̂i = (4.15)
dâ(i)
is the true traction vector acting on the xi -face. Substituting Eqs. (4.3)
and (4.15) into (4.14) yields
T(n) da − T̂i dâi + f dv = ρa dv, (4.16)
in which the summation convention applies to the index i.
Next, we examine the limit as the tetrahedral element shrinks to a point.
Dividing Eq. (4.16) through by da and letting da and dv approach zero gives
dâi
T(n) = T̂i (4.17)
da
since the linear dimension dv/da of the element approaches zero. Note that,
in this limit, all four faces of the tetrahedron pass through the same point.
The area ratio dâi /da is obtained by taking the dot product of Eq. (4.10)
with gj to get
dâi dâi dâj
(n da) · gj = gi · gj = δ ij = ,
g (ii) g (ii) g(jj)
1 In nonorthogonal coordinates, the normals to the element faces may not point in the
which gives
dâi
= (n · gi ) g(ii) . (4.18)
da
Combining this relation with Eq. (4.17) yields
T(n) = (n · gi ) g(ii) T̂i . (4.19)
Ti ≡ g(ii) T̂i , (4.20)
T(n) = (n · gi ) Ti = ni Ti , (4.21)
where the ni are the covariant components of n. This equation shows that
if the traction vectors Ti are known across three different planes passing
through a point, then the traction across any plane through that point can
be computed.
Next, we express the pseudotraction vector of Eq. (4.20) in the form
Ti = σ ij gj , (4.22)
T(n) = n · σ, (4.23)
where
σ = σ ij gi gj . (4.24)
Because T(n) is a true traction vector, σ is called the true stress tensor
or the Cauchy stress tensor. Of course, the Cauchy stress components
with respect to any basis can be obtained by double-dotting σ with the
appropriate dyad.
4.2. Stress Tensors 145
T(N) = N · t
T̃(N) = N · s, (4.25)
where N is the unit normal to the undeformed area element dA (Fig. 4.1).
Substituting Eqs. (4.23) and (4.25) into (4.7) yields
dP = n · σ da = N · t dA = N · s · FT dA, (4.26)
and using Eqs. (4.2) gives
dP = da · σ = dA · t = dA · s · FT . (4.27)
The undeformed and deformed areas are related by Eq. (3.128), which gives
da · σ = (J dA · F −1 ) · σ,
and thus Eq. (4.27) implies
J F−1 · σ = t = s · FT
or
σ = J −1 F · t = J −1 F · s · FT . (4.28)
This relation links the three stress tensors through the deformation gradient
tensor. Solving for the pseudostress tensors yields
t = J F−1 · σ = s · FT
s = J F−1 · σ · F−T = t · F−T . (4.29)
and disadvantages, which will become clear later. Later, we will show that
σ is a symmetric tensor (Sec. 4.7.2), i.e., σ = σ T , and so Eq. (4.29)2 gives
sT = J(F−1 · σ · F−T )T = J(F−T )T · (F−1 · σ)T
= JF−1 · (σ T · F−T ) = JF−1 · σ · F−T
= s.
Thus, s also is a symmetric tensor. However, because F is generally not
symmetric, Eq. (4.29)1 shows that t is not symmetric in general. For
solid mechanics problems involving large deformation, therefore, the second
Piola-Kirchhoff stress tensor s often is the tensor of choice for two primary
reasons: (1) It is referred to the undeformed configuration (known a priori);
and (2) it is symmetric. Once s is determined, the other stress tensors can
be computed from Eqs. (4.28) and (4.29). However, only the physical stress
components provide physically meaningful results (see Sec. 4.4).
Atluri (1984) discusses several other stress tensors that are useful in
certain situations. Equation (4.28), for example, suggests defining another
stress tensor τ = Jσ, which is called simply the Kirchhoff stress tensor,
but we do not use τ in this book. For small displacement, all these ten-
sors are essentially equal to the linear stress tensor described in Appendix
A. Indeed, Eq. (4.28) shows that σ ∼ =t∼ = s when J → 1 and F → I. We
note, however, that since F contains both deformation and rigid-body ro-
tation, stipulating small strain alone is not a sufficient condition for this
approximation.
s = sIJ GI GJ . (4.30)
Table 4.1 lists the notations used for stress components in two other popular
texts. Relations between the various stress components can be found by
substituting Eqs. (3.22) and (4.30) into (4.28). For example, we can write
σ = σ ij gi gj = J −1 F · t
= J −1 FJi gi GJ · tIj GI gj
= J −1 FJi tIj δ JI gi gj
= J −1 FIi tIj gi gj , (4.31)
which implies
σ ij = J −1 FIi tIj .
Similarly,
σ = σ ij gi gj = J −1 F · s · FT
i
= J −1 FK gi GK · sIJ GI GJ · FLj GL gj
= J −1 FK
i
FLj sIJ δ K L
I δ J gi gj
= J −1 FIi FJj sIJ gi gj (4.32)
gives
σ ij = J −1 FIi FJj sIJ .
Combining these expressions yields
which relates the stress components used by Eringen (1962; 1980) (see Table
4.1).
∗ ∗
Relations for the convected stress components σ I J can be found by
i I
setting x = X in Eq. (4.33), but it is not clear where to place aster-
isks. One way to clarify matters is to substitute Eq. (4.33) into the dyadic
representations (4.30)1 . This procedure gives, for example,
σ = σ ij gi gj = J −1 FIi tIj gi gj
∗
= J −1 FK
I KJ
t gI gJ ,
where the dummy index I has been changed to K, i and j have been set
to I and J (xi = X I ), and the asterisk indicates that the superscript J ∗ is
attached to the base vector gJ (not GJ ). Since FK I
= ∂X I /∂X K = δ IK by
Eq. (3.15), the above expression can be written
∗
J∗ ∗
σ = σI gI gJ = J −1 δ IK tKJ gI gJ
∗
= J −1 tIJ gI gJ ,
which implies
∗
J∗ ∗
σI = J −1 tIJ .
or
tIJ = sIK gK · GJ . (4.35)
This expression can be written in another form by noting r = R + u, and
so Eqs. (3.10) and (3.11) give
gI = r,I = R,I + u,I = GI + u,I
= G I + uK G K , I
= G I + uK | I G K ,
where the covariant derivative comes from Eq. (2.150). Thus,
gI · G J = G I + u K | I G K · G J
= δ JI + uK |I δ JK
= δ JI + uJ |I
and Eq. (4.35) becomes
P P
a1
x2
A1
x1
Fig. 4.3. Extension of a rectangular block.
Fig. 4.4. Inflation of a hemispherical dome. (a) Undeformed configuration. (b) Deformed
configuration.
∗ ∗
which next is written in terms of the σ I J .
Suppose a point in the undeformed shell is located at the spherical polar
coordinates X I , where X 1 , X 2 , and X 3 are the radial, meridional, and
circumferential coordinates, respectively (Fig. 4.4a). During deformation,
the undeformed base vectors GI are convected to the base vectors gI , with
g2∗ following the changing contour of the shell meridian (Fig. 4.4b).
∗
The normal to the inner surface n points along the vector −g1 , since
∗
it is normal to both g2∗ and g3∗ (Fig. 4.4b). Thus, n = −g1 / g1∗ 1∗ and
Eq. (4.44) gives
# $ # $
−g1
∗ ∗ ∗ −g 1∗
∗ ∗ · σ I J gI gJ = −p ∗ ∗
g1 1 g1 1
or
∗
J ∗ 1∗ ∗
J∗ ∗
σI δ I gJ = σ1 gJ = −p g1 .
Dotting both sides of this expression with gK yields the boundary condition
∗
K∗ ∗
On a: σ1 = −p g1 K
(K = 1, 2, 3), (4.45)
∗ ∗ ∗ ∗ ∗ ∗
which provides conditions on σ 1 1 , σ 1 2 , and σ 1 3 , i.e., on the com-
ponents of the pseudotraction vector T1 acting on the inner surface [see
Eq. (4.22)].
g(ii)
σ̂ i·j ≡= σ i·j (4.46)
g(jj)
√
σ̂ ij = σ ij g(ii) g(jj) (4.47)
Recall that the stress components are components of the traction vector,
which depends on the orientation of the plane on which it acts. Consider,
for example, the pseudotraction vector of (4.22), i.e.,
Ti = σ ij gj ,
which acts on an area element with a normal oriented in the direction of
gi .2 This equation shows that the first index of σ ij corresponds to the
plane (orthogonal to gi ) on which the stress acts, and the second index
corresponds to the direction of action (gj ). Interpreting other types of stress
components follows similar reasoning.
Consider now the dyadic representations
σ = σ ij gi gj = σ ij gi gj = σ i·j gi gj = σ ·j i
i g gj . (4.49)
ij
The relation involving σ indicates that the direction of the surface normal
(gi ) is given by raising the index of the first base vector of the dyadic, and
the stress acts in the direction of the second base vector. In fact, for all
components of the stress dyadic, changing the position of the index of the
first base vector (up or down) gives the direction of the surface normal.
Thus, σ ij acts on a surface normal to gi and points along gj , and so on.
All four sets of stress components are illustrated for two-dimensional
elements in the x1 x2 -plane (Fig. 4.5). In each case, the shape of the element
is dictated by the orientations of the gi for σ ij and σ i·j and by the gi
for σ ij and σ ·j i
i , i.e., the faces of the element are perpendicular to g and
gi , respectively. The figure suggests that the σ ij may be the easiest to
visualize, since the faces of the element align with the coordinate curves,
and the stresses point along the coordinate directions. This is one reason
we focused on contravariant stress components in the previous section.
Similar interpretations apply to components relative to the {GI } and
{gI } bases. These alternative base vectors simply carve out differently
shaped material elements. Being referred to areas and directions that are
∗ ∗
convected with the deformation, the Cauchy stress components σ I J are
advantageous in visualizing the mechanics. Of course, only the physical
components
I∗ I∗ g(II)
σ̂ ·J ∗ = σ ·J ∗ , (4.50)
g(JJ)
given by modifying Eq. (4.46), have a true physical meaning.
2 As shown in Fig. 4.2 (page 142), dP−i acts on the element face with a normal in the
direction −gi , and, therefore, dPi acts on a face with a normal in the direction +gi .
4.6. Principal Stresses 155
s22 s2. 2
s2. 1
s21
s12
s1. 1
s11
g2 s1. 2
g2
.
s 22 s22
g1
s2. 1 s21
g1 .
s12 s12
s1. 1 s11
(σ−σI) · n = 0. (4.52)
Solving this equation provides the principal Cauchy stresses σ i and the
corresponding principal directions ni (i = 1, 2, 3). Since σ is symmetric,
the eigenvalues are real and the eigenvectors are mutually orthogonal.
156 Analysis of Stress
Every particle of a solid body, as well as the entire body itself, must obey
Newton’s laws of motion. Section 4.2 explored the consequences of these
laws for a volume element that shrinks to a point. That analysis led to the
definition of the Cauchy stress tensor. Here, we use Newton’s second law
in the form of the principles of linear and angular momentum to write the
global equations of motion for a solid body. Manipulating these relations
then provides local equations of motion for an infinitesimal element.
f
n
(n)
T
v
da
r
dv
where Eq. (4.22) has been used. This relation and (4.23) give
r × T(n) = r × (n · σ) = r × (n · gi Ti ) = (n · gi ) r × Ti
= n · (gi r × Ti ).
With this equation and the divergence theorem (2.200), the first term in
Eq. (4.58) becomes
r × T(n) da = n · (gi r × Ti ) da = ∇¯ · (g r × Ti ) dv.
i (4.59)
a a v
Next, differentiating the integrand, setting r,i = gi , and using (2.137)2
yields
¯ · (g r × Ti ) ∂
∇ i = gk · (gi r × Ti )
∂xk
= gk · (gi ,k r × Ti + gi r,k × Ti + gi r × Ti ,k )
= Γkik r × Ti + δ ki gk × Ti + δ ki r × Ti ,k
= Γkik r × Ti + gi × Ti + r × Ti ,i ,
which can be simplified by noting
¯ ·σ ∂
∇ = gk · (gi Ti )
∂xk
= gk · (gi ,k Ti +gi Ti ,k )
= Γkik Ti + δ ki Ti ,k
= Γkik Ti + Ti ,i .
Combining these expressions shows that Eq. (4.59) can be written
(n) ¯ · σ)] dv.
r × T da = [gi × Ti + r × (∇ (4.60)
a v
In addition, the right-hand-side of Eq. (4.58) is
d d
(r × v) ρ dv = [(r × v) ρ dv]
dt v v dt
d d
= ρ dv (r × v) + (r × v) (ρ dv)
v dt dt
d
= ρ dv (ṙ × v + r × v̇) + (r × v) (ρ dv)
dt
v
= (r × a) ρ dv. (4.61)
v
Note that ṙ × v = 0 since ṙ = v, and d( ρ dv)/dt = 0 by conservation of
mass.
4.7. Equations of Motion 159
Substituting Eqs. (4.60) and (4.61) into (4.58) and rearranging yields
[gi × Ti + r × (∇¯ · σ + f −ρ a)] dv = 0. (4.62)
v
The term in parentheses vanishes by the equation of motion (4.57), and,
because the integral must be zero for an arbitrary volume, the principle of
angular momentum reduces to
gi × Ti = 0. (4.63)
The consequence of this result is revealed by substituting Eq. (4.22) and
using (2.103)1 to obtain
gi × Ti = gi × σ ij gj = ijk σ
ij k
g = 0,
which can be written
ij k ji k
ijk σ g =0 or jik σ g = 0.
Since jik =− ijk , adding these expressions yields
ijk (σ
ij
− σ ji)gk = 0, (4.64)
ij ji
which implies that σ = σ , i.e., the Cauchy stress tensor is symmetric
(σ = σ T ). In Sec. 4.2.2, we used this fact to show that the second Piola-
Kirchhoff stress tensor s also is symmetric.
the undeformed volume element dV . Note that, although the integrals are
taken over the undeformed body, the forces actually are applied to the
deformed body.
From here, the derivation follows that given in Sec. 4.7.1. First, with
Eqs. (4.25)1 , and (2.200), the surface integral of (4.65) becomes
T(N) dA = N · t dA = ∇ · t dV. (4.66)
A A V
∇ · t + f 0 = ρ0 a. (4.69)
Although Eqs. (4.57) and (4.69) appear similar, there are significant
differences. Since the derivatives are computed relative to the known coor-
dinates in the undeformed configuration, Eq. (4.69) is easier to solve in prin-
ciple. Moreover, with v = v(R, t), the acceleration is given by Eq. (3.193)1,
rather than (3.193)2. In contrast to σ, however, the first Piola-Kirchhoff
stress tensor t is generally not symmetric. Thus, working with the sym-
metric second Piola-Kirchhoff stress tensor s often is more convenient, and
putting Eq. (4.29)1 into (4.69) gives
T
∇ · (s · F ) +f 0 = ρ0 a. (4.70)
But this equation is more complicated than Eq. (4.69). As is becoming ap-
parent, all large-deformation formulations possess both inherent advantages
and disadvantages. While the stress and strain measures of choice vary, the
pseudostress tensor s and the Lagrangian strain tensor E are often used in
finite elasticity problems. Besides their symmetry and being referred to the
undeformed configuration, these tensors lead to a convenient form of the
material constitutive relations (see Chap. 5).
4.7. Equations of Motion 161
Example 4.3 As we have just seen, deriving the Lagrangian form of the
equation of motion from a direct application of the principle of linear mo-
mentum is straightforward. Deriving Eq. (4.69) from Eq. (4.57), or vice
versa, is a more complicated task. Show that these two forms of the equa-
tions of motion are indeed equivalent.
Solution. Beginning with Eq. (4.69), we substitute Eqs. (4.1) and (4.29)1
and use the relation
dv ρ
J= = 0 (4.71)
dV ρ
to obtain
J −1 ∇ · (J F−1 · σ) + f = ρ a. (4.72)
A glance at Eq. (4.57) reveals that our task boils down to showing that the
¯ · σ. But this is easier said than done.
first term in (4.72) is equivalent to ∇
With ∇ given by Eq. (3.18)1 , expanding this term gives
∂
J −1 ∇ · (J F−1 · σ) = J −1 GI · (J F−1 · σ)
∂X I
= J −1 GI · (J,I F−1 · σ+J F−1,I · σ + J F−1 · σ,I ).
(4.73)
In the following, we examine separately each term of this equation.
I
J,I = ∂J/∂X in the first term, we use Eq. (3.119), which
To compute
gives J = j g/G. Then, working in convected coordinates for convenience,
we can set j = 1 (see Sec. 3.6.3) to obtain
√ √ √ √
g G( g),I − g( G),I
J,I = =
G ,I G
√ √
g ( g),I ( G),I
= √ − √ . (4.74)
G g G
Specializing Eq. (2.149) gives
√ √ J
( g),I = g Γ̄
√ √ JI
( G),I = G ΓJJI , (4.75)
where the Christoffel symbols are defined by Eqs. (3.86), and Eq. (4.74)
can now be written
With the help of Eq. (3.22)2 , the first term on the right-hand-side of
Eq. (4.73) becomes
J −1 GI · (J,I F−1 · σ) = J −1 J,I GI · (GK gK ) · σ
= J −1 J,I δ IK gK · σ
= (Γ̄JJI − ΓJJI ) gI · σ. (4.77)
The second term on the right-hand-side of Eq. (4.73) requires
F−1,I = (GK gK ),I = GK ,I gK + GK gK ,I
= (ΓJKI GJ )gK − GK (Γ̄K J
JI g ),
in which Eqs. (2.136) and (2.139) have provided the derivatives of the base
vectors. Inserting this expression into the second term of (4.73) gives
J −1 GI · (J F−1,I · σ) = GI · (ΓJKI GJ gK − Γ̄K
JI GK g ) · σ
J
= (ΓJKI δ IJ gK − Γ̄K I J
JI δ K g ) · σ
= (ΓIKI gK − Γ̄IJI gJ ) · σ
= (ΓJIJ − Γ̄JIJ )gI · σ, (4.78)
where dummy indices have been redefined.
Next, using Eq. (3.18)2 in the last term of (4.73) yields
J −1 GI · (J F−1 · σ,I ) = GI · (GK gK ) · σ,I
= δ IK gK · σ,I = gI · σ,I
∂
= gI ·σ
∂X I
= ¯ · σ.
∇ (4.79)
Finally(!), substituting Eqs. (4.77)–(4.79) into (4.73) gives (since ΓJIJ =
ΓJJI )
J −1 ∇ · (J F−1 · σ) = ∇
¯ · σ, (4.80)
which proves that Eq. (4.72), or Eq. (4.69), is equivalent to Eq. (4.57).
t = tIJ GI GJ
f0 = f0J GJ
a = aJ G J (4.83)
s = sIJ GI GJ
∗
f0 = f0J GJ = f0J gJ
∗
a = aJ G J = a J g J . (4.84)
The derivation requires the gradient operators
¯ = gi ∂ = gI ∂
∇
∂xi ∂X I
∂
∇ = GI , (4.85)
∂X I
as provided by Eqs. (3.18) and the Christoffel symbols
Γ̄kij = gi , j · g k
Γ̄K
IJ = gI , J · g K
ΓK
IJ = GI ,J · GK (4.86)
which are given by Eqs. (3.86).
Consider first the divergences in Eqs. (4.81). For the first of these
expressions, Eqs. (4.82)1 and (4.85)1 give the two forms
¯ ·σ = ∂
∇ gk k · σ = gk · (σ ij gi gj ),k
∂x
∂ ∗ ∗
= gK K
· σ = gK · (σ I J gI gJ ),K
∂X
and noting Eq. (2.153) yields
∇¯ · σ = gk · (σ ij |k gi gj ) = σ ij |k δ k gj
i
= σ ij |i gj
∗
J∗ ∗
J∗
= gK · (σ I | K gI gJ ) = σ I |K δ K
I gJ
∗ ∗
= σI J
| I gJ . (4.87)
164 Analysis of Stress
σ ij |i = σ ij ,i +σ kj Γ̄iik + σ ik Γ̄jik
∗
J∗ ∗
J∗ ∗
J∗ ∗
K∗
σI |I = σ I ,I +σ K Γ̄IIK + σ I Γ̄JIK . (4.88)
Similarly, for the second of (4.81), Eqs. (2.153), (4.83)1 , and (4.85)2 give
∂
∇·t = GK · t = GK · (tIJ GI GJ ),K
∂X K
= GK · (tIJ ||K GI GJ ) = tIJ ||K δ K
I GJ
= tIJ ||I GJ , (4.89)
where
Here, the triple vertical bar indicates a mixed covariant derivative with
respect to both the undeformed and deformed configurations.
4.7. Equations of Motion 165
Substituting all these results into Eqs. (4.81) yields the equations of
motion
σ ij |i + f j = ρaj
∗
J∗ ∗ ∗
σI |I + f J = ρaJ
tIJ ||I + f0J = ρ0 aJ
∗ ∗
sIJ |||I + f0J = ρ0 aJ . (4.93)
x3, z
e3, ez
e2
x2
r eq
e1
q
x1
er
With (x1 , x2 , x3 ) = (r, θ, z), the only non-zero Christoffel symbols for cylin-
drical polar coordinates are
as given by Eqs. (2.148). Hence, expanding Eq. (4.95) yields the relations
1
σ 11 ,1 +σ 21 ,2 +σ 31 ,3 + σ 11 − rσ 22 + f 1 = ρa1
r
3
σ ,1 +σ ,2 +σ 32 ,3 + σ 12 + f 2
12 22
= ρa2
r
1 13
σ ,1 +σ ,2 +σ ,3 + σ + f 3
13 23 33
= ρa3 . (4.97)
r
(b) Since cylindrical coordinates are orthogonal, Eq. (4.47) gives the
physical stress components
√
σ̂ ij = σ ij g(ii)g(jj).
With the components of the metric tensor being [see Eqs. (2.146)]
g1 = er , g2 = reθ , g3 = ez ,
we have
fˆr = f1, fˆθ = rf 2 , fˆz = f 3
âr = a1 , âθ = ra2 , âz = a3 . (4.100)
4.8. Problems 167
Finally, after substituting (4.98) and (4.100) into (4.97) and differentiating,
we obtain the equations
∂ σ̂ rr 1 ∂ σ̂ θr ∂ σ̂ zr σ̂ rr − σ̂ θθ
+ + + + fˆr = ρâr
∂r r ∂θ ∂z r
∂ σ̂ rθ 1 ∂ σ̂ θθ ∂ σ̂ zθ 2σ̂ rθ
+ + + + fˆθ = ρâθ
∂r r ∂θ ∂z r
∂ σ̂ rz 1 ∂ σ̂ θz ∂ σ̂ zz σ̂ rz
+ + + + fˆz = ρâz , (4.101)
∂r r ∂θ ∂z r
which agree with those given in Appendix B for cylindrical coordinates.
4.8 Problems
4.1 Relative to Cartesian coordinates (x, y, z), the Cauchy stress tensor at
a point is given by the matrix
⎡ ⎤
40 30 −10
[σ ij ] = ⎣ 30 60 25 ⎦ .
−10 25 80
(a) Determine the true traction vector across a plane defined by the
relation
F = 2x + 2y + z − C = 0,
where C is a constant. Hint: A vector normal to a surface with
equation F = 0 is given by a = ∇F.
(b) The traction vector can be written in the form T(n) = σ nn n+ σ ns s,
where n and s are unit vectors normal and tangent to the plane
F = 0 (n not summed). Determine the stress components σ nn and
σ ns .
168 Analysis of Stress
(c) Determine the principal Cauchy stresses and the unit vectors defin-
ing the directions in which they act.
4.2 Direction cosines of a unit vector n relative to Cartesian axes (x, y, z)
are given by p = n · ex , q = n · ey , and r = n · ez . Suppose (p, q, r) are
known for the unit normal at a point on the boundary of a deformed
body.
√ At this point, the true surface traction vector has magnitude
3a and is oriented at the same angle relative to all three positive
coordinate axes. If σ yy = σ xz = σ yz = 0, determine the remaining
Cauchy stress components.
4.3 In Cartesian coordinates xi , the Cauchy stress tensor at a given point
is
σ = a e1 e1 − b e2 e2 + c e3 e3 + d(e1 e3 + e3 e1 ) − e(e2 e3 + e3 e2 ),
where a, b, c, d, and e are positive constants.
(a) Determine the unit normal vector for a plane parallel to the x3 -axis
on which the true traction vector is tangent to the plane.
(b) Consider a coordinate system x̄i that is obtained by rotating the xi
system through an angle θ about the x1 -axis. Compute the stress
components σ̄ 13 and σ̄ 23 relative to the rotated system.
4.4 Relative to spherical polar coordinates (r, θ, φ), the physical compo-
nents of the Cauchy stress tensor at a point in a body are
⎡ rr ⎤
σ̂ 0 0
σ (er ,eθ ,eφ ) = ⎣ 0 1 2 ⎦ .
0 2 σ̂ φφ
If the principal stresses at this point are σ̂ 1 = 1, σ̂ 2 = −1, and σ̂ 3 = 3,
find σ̂ rr and σ̂ φφ .
4.5 In Cartesian coordinates (x, y, z), the Cauchy stress distribution in a
body is given by
⎡ ⎤
xy −y 0
σ (ei ej ) = ⎣ −y x − y 0 ⎦.
3
0 0 y + 2z
Determine the physical components of σ relative to cylindrical coordi-
nates (r, θ, z).
4.6 In cylindrical coordinates, the first Piola-Kirchhoff stress tensor at the
point (r, θ, z) = (5, π/3, −2) is
t = 2er er + 3eθ eθ − eθ ez − (2/3)ez eθ ,
4.8. Problems 169
F = er er + 2eθ eθ + 3ez ez − eθ ez .
Find the Cauchy stress components σ ij , σ i·j , σ ·j
i , and σ ij (i, j = r, θ, z)
relative to the natural base vectors for cylindrical coordinates.
4.7 A circular tube with deformed inner radius a and outer radius b is
loaded by a uniform circumferential Cauchy shear stress τ 0 on its outer
surface.
(a) Determine the shear stress τ i that must be applied to the inner
surface to maintain equilibrium.
(b) Write the dyadic form of the Cauchy stress tensor σ for an arbi-
trary point in the tube in terms of the base vectors {er , eθ , ez } for
cylindrical polar coordinates.
(c) Find the Cartesian components of σ for an arbitrary point in the
tube.
4.8 In Cartesian coordinates, the deformation of a solid body from config-
uration B(X I ) to configuration b(xi ) is defined by the relations
x1 = X 1 + 0.2X 2 X 3
x2 = X 2 + 0.2(X 3 )2
x3 = X3.
where the λi are stretch ratios. The axial stretch ratio λz is constant,
and, with inflation moving a point in the wall radially from radius R to
r(R), Eq. (3.63) with ψ = 0 (no torsion) gives the other stretch ratios
dr r
λr = , λθ = .
dR R
Finally, assume the Cauchy and first Piola-Kirchhoff stress tensors are
given by
σ = σ r (r)er er + σ θ (r)eθ eθ + σ z ez ez
t = tr (R)er er + tθ (R)eθ eθ + tz ez ez .
(a) Show that the radial equilibrium equation takes the following
forms:
dσ r 1
+ (σ r − σ θ ) = 0
dr r
dtr 1
+ (tr − tθ ) = 0.
dR R
(b) Show that the two equilibrium equations are equivalent.
4.12 Consider a solid circular cylinder with deformed radius a and length d.
In Cartesian coordinates, the Cauchy stress tensor is
σ = x2 ex ex + y2 ey ey + xy(ex ey + ey ex).
Determine the physical components of the body forces and surface trac-
tions (on all surfaces) that must be applied to the cylinder to maintain
equilibrium with the given stress distribution. Express the results rel-
ative to spatial cylindrical coordinates (r, θ, z).
4.13 Consider torsion of a spherical shell about the z-axis (see Fig. B.2 in
Appendix B). Assume the Cauchy stress tensor can be written in the
dyadic form
σ = σ̂ θφ (eθ eφ + eφ eθ ),
where σ̂ θφ = σ̂ θφ (r, θ) is the only non-zero physical stress component
in spherical polar coordinates (r, θ, φ).
(a) Using ∇ ¯ · σ = 0, derive the non-trivial equilibrium equation(s) in
terms of σ̂ θφ .
(b) Using σ ij |i , derive the non-trivial equilibrium equation(s) in terms
of σ θφ .
(c) Show that the equations found in parts (a) and (b) are equivalent.
172 Analysis of Stress
4.14 A solid circular cylinder has undeformed radius a0 . Let the undeformed
and deformed configurations be described by the cylindrical coordinates
(R, Θ, Z) and (r, θ, z), respectively. Suppose a surface traction T(n) =
−pn + τ s acts on the curved surface, where p(Θ) is a pressure, τ (Θ) is
a shear stress, and n(Θ) and s(Θ) are normal and circumferential unit
vectors tangent to the deformed surface. This loading causes a point
in the cylinder located originally at the position R = ReR + ZeZ to
move to r = R(1 + b cos2 Θ)eR + ZeZ , where b is a positive constant
and the eI are unit vectors.
(a) Determine the deformation gradient tensor in the dyadic form F =
FIJ eI eJ (I, J = R, Θ, Z; summed).
(b) Write n, s, the undeformed base vectors GI and GI , and the con-
vected base vectors gI and gI in terms of the non-zero FIJ .
(c) Consider the dyadic representations
∗
J∗
σ = σ IJ GI GJ = σ I gI gJ
for the Cauchy stress tensor. Write the boundary conditions on the
∗ ∗
curved surface in terms of (i) σ IJ and (ii) σ I J .
Chapter 5
Constitutive Relations
The analyses of deformation and stress in the previous chapters are valid
for any solid body that can be represented as a continuum, regardless of the
type of material that comprises the body. Deformation and stress, however,
are linked by constitutive relations that must be found experimentally for
each type of material. In biomechanics, determining constitutive relations
is complicated by material nonlinearity, complex geometry, the composite
nature of biological tissues, and the influence of a wide range of environ-
mental variables. In general, even the functional forms of these relations
are unknown, but thermodynamic considerations place important restric-
tions on their form. This chapter considers some of these restrictions and
discusses some strategies for determining constitutive relations for soft tis-
sues.
173
174 Constitutive Relations
represents the time-rate of change of the total energy in the system, and
the right-hand-side is the rate at which energy is added to the system. The
various quantities in this equation are written below in both Eulerian and
Lagrangian forms.
Consider a body of mass density ρ0 (R) and volume V in the reference
configuration. After deformation, these quantities become ρ(r, t) and v,
respectively. The total kinetic energy for the body is
1 1
K = 2 ρ v · v dv = 2 ρ0 v · v dV, (5.2)
v V
where v(R, t) = v(r, t) is the velocity vector for a point that moves from
the undeformed location R to the deformed location r. In addition, the
total internal energy contained in the body is
U = ρu dv = ρ0 u dV, (5.3)
v V
where u is the internal energy per unit mass. Here, the internal energy
consists of thermal energy and strain energy.
The power input is the rate of work done on the body by applied loads,
which include surface and body forces. In terms of the body force vectors
in Eq. (4.1) and the surface traction vectors defined by Eqs. (4.3) and (4.4),
the total power input is
P = T(n) · v da + f · v dv
a v
(N)
= T · v dA + f0 · v dV, (5.4)
A V
in which A and a are the surface areas of the body in the undeformed and
deformed configurations, respectively.
Finally, the body gains thermal energy by heat flowing across its surface,
as well as heat generated by internal sources. Let q and q0 be the outward-
directed heat flux vectors per unit deformed and undeformed surface area,
respectively, and let r be the rate of heat production per unit mass due to
internal sources. Then, the rate of heat added to the body is
Q = − q · n da + ρr dv
a v
= − q0 · N dA + ρ0 r dV, (5.5)
A V
The various terms in Eq. (5.1) are next expressed in alternate forms so
they can be combined. First, Eq. (5.2) gives
1 d d
K̇ = 2 (ρ dv) v · v+ (v · v)ρ dv
dt dt
v
1
= 2 ( v̇ · v + v · v̇)ρ dv = v · v̇ ρ dv
v v
in which mass conservation gives d(ρ dv)/dt = 0, and substituting for ρv̇ =
ρa from the equation of motion (4.57) produces the third line. Similar
manipulations with the Lagrangian form for K in Eq. (5.2) yield
K̇ = v · (∇ · t + f 0 ) dV, (5.7)
V
¯ · σ) ∂
v · (∇ = v · ek · σ ij ei ej
∂xk
= v · (σ ij ,k δ ki ej ) = v · (σ ij ,i ej )
= (vk ek ) · (σ ij ,i ej ) = vk σ ij ,i δ kj
= vj σ ij ,i
= (vj σ ij ),i −vj ,i σ ij . (5.8)
To convert this expression back to direct notation, we note the following:
¯ · (σ · v)
∇ = ∇¯ · (σ ij ei ej · vk ek ) = ∇ ¯ · (σ ij vk δ jk ei )
∂
= ek · (σ ij vj ei ) = (σ ij vj ),k δ ki = (σ ij vj ),i
∂xk
¯ ∂
σ : ∇v = (σ ij ei ej ) : ek vl el = σ ij vl ,k (ei · ek )(ej · el )
∂xk
= σ ij vl ,k δ ik δ jl = σ ij vj ,i .
Thus, Eq. (5.8) and similar manipulations for v · (∇ · t) yield
¯ · σ)
v·(∇ = ¯ · (σ · v) − σ : ∇v
∇ ¯
v·(∇ · t) = ∇ · (t · v) − t : ∇v. (5.9)
176 Constitutive Relations
¯ = D − Ω and ∇u = FT −
In addition, Eqs. (3.212) and (3.77)1 give ∇v
I, where D is the symmetric rate-of-deformation tensor and Ω is the anti-
symmetric spin tensor. Thus, the last terms of Eqs. (5.9) can be written
¯
σ : ∇v = σ : (D − Ω) = σ : D
T
t : ∇v = t : ∇u̇ = t : Ḟ , (5.10)
The Eulerian form of the second term in Eq. (5.1), given by differenti-
ating (5.3), is
d
U̇ = (ρ dv) u + u̇ρ dv
dt
v
= u̇ρ dv. (5.12)
v
Next, the surface integrals in Eqs. (5.4) and (5.5) are converted into
volume integrals. Substituting Eqs. (4.23) and (4.25)1 into (5.4) and using
the divergence theorem (2.200) yields
P = n · σ · v da + f · v dv
a v
= ¯ · (σ · v) + f · v ] dv
[∇ (5.14)
v
= [∇ · (t · v) + f 0 · v ] dV (5.15)
V
5.1. Thermodynamics of Deformation 177
¯ ·q=0
ρu̇ − σ : D−ρr + ∇
ρ0 u̇ − t : ḞT −ρ0 r + ∇ · q0 = 0, (5.18)
which are the local forms of the first law. According to these equations, the
rate of increase in internal energy of a volume element (ρu̇ or ρ0 u̇) is equal
to the sum of the rate of work done by the stresses on the element, i.e., the
stress power (σ : D or t : ḞT ), the rate of internal heat production (ρr
or ρ0 r), and the rate of heat flow into the element (−∇ ¯ · q or −∇ · q0 ).
The stress power appears in Eqs. (5.18) in terms of the Cauchy and first
Piola-Kirchhoff stress tensors. We now derive the relation between these
two forms for the stress power. First, since σ : Ω = 0, we can write
−1
σ : D = σ : (D + Ω) = σ : L = σ : (Ḟ · F ), (5.19)
where Eqs. (3.214) and (3.207) have been substituted. Next, applying for-
mulas from Table 2.2 (page 22) with T = σ, U = Ḟ, and V = F−1 gives
−T
σ:D = Ḟ : (σ · F )
−T
= (σ · F ) : Ḟ = (σ · F−T )T : ḞT
= (F−1 · σ T ) : ḞT = (F−1 · σ) : ḞT ,
where the third line uses the symmetry property of the Cauchy stress tensor.
Finally, substituting Eq. (4.29)1 gives the result
T
σ : D = J −1 t : Ḟ . (5.20)
178 Constitutive Relations
A third form for the stress power can be written in terms of the second
Piola-Kirchhoff stress tensor s. Substituting Eq. (3.217) for D into (5.20)
yields
T −1
t : Ḟ = J σ : D = Jσ : (F−T · Ė · F ).
This expression is transformed by setting T = F−1 , UT = F−T · Ė, and
V = σ in the next-to-last formula of Table 2.2 to get
T
t : ḞT = J ( F−T · Ė) : (F−1 · σ T )
T
= J (F−1 · σ T ) : (Ė · F−1 )
= J (F−1 · σ) : (Ė · F−1 )
since σ = σ T and E = ET . Now, on setting T = Ė, U = F−1 · σ, and
VT = F−1 , the next-to-last line in Table 2.2 gives
t : ḞT = J Ė :(F−1 · σ · F−T )
= Ė : s = s : Ė, (5.21)
in which Eq. (4.29)2 has been used. Thus, Eqs. (5.20) and (5.21) give the
stress power in the three forms
ρ0 u̇ − s : Ė−ρ0 r + ∇ · q0 = 0. (5.23)
the sum of the influx of entropy through the surface of the body and the
entropy generated by internal heat sources. In mathematical terms, using
Eq. (5.5) to generalize Eq. (5.27) gives the second law in Eulerian rate form
as
d q ρr
ρη dv ≥ − · n da + dv (5.28)
dt v a T v T
and in Lagrangian form as
d q0 ρ0 r
ρ η dV ≥ − · N dA + dV. (5.29)
dt V 0 A T V T
The local forms of the second law follow by applying the divergence
theorem (2.200) to the area integrals in the above equations to obtain
ρr ¯ · q
ρη̇ − +∇ dv ≥ 0
v T T
q
ρ r 0
ρ0 η̇ − 0 + ∇ · dV ≥ 0. (5.30)
V T T
These integral relations imply the local equations
r 1 ¯ q
η̇ − + ∇ · ≥0
T ρ T
r 1 q
0
η̇ − + ∇· ≥ 0, (5.31)
T ρ0 T
r 1 ¯ 1 ¯ ≥0
η̇ − + ∇ · q− 2 q · ∇T
T ρT ρT
r 1 1
η̇ − + ∇ · q0 − q0 · ∇T ≥ 0. (5.32)
T ρ0 T ρ0 T 2
Since heat does not flow naturally from a cooler toward a warmer part
of a body, it is always true that
¯ ≤0
q · ∇T and q0 · ∇T ≤ 0,
i.e., the temperature gradient and the heat flux have opposite signs unless
¯ = 0, in which case q = 0. These terms represent the entropy intro-
∇T
duced into the body by heat conduction. This observation and Eqs. (5.32)
imply that the second law of thermodynamics also can be written in the
stronger forms
r 1 ¯
η̇ − + ∇·q ≥ 0
T ρT
r 1
η̇ −
+ ∇ · q0 ≥ 0, (5.33)
T ρ0 T
where equality holds for a reversible system
The remainder of this section uses the first six of these principles, in
the listed order, to develop general constitutive relations for a compressible
thermoelastic material. Consideration of the principle of material symmetry
is deferred to Sec. 5.3, which examines specific forms for an elastic mate-
rial in which thermal effects are ignored. The issue of incompressibility is
addressed in Sec. 5.5.
σ = σ(R, F, η, ∇η)
q = q(R, F, η, ∇η)
u = u(R, F, η, ∇η)
T = T (R, F, η, ∇η). (5.36)
a Q
p
r* r
O
b
O* A
A*
Fig. 5.1. Two reference frames (A and A∗ ) in relative motion. The point p and the
vector a are fixed in frame A, and r and r∗ are position vectors of p relative to coordinate
systems fixed in A and A∗ , respectively.
Vectors. Equation (5.38) links the two distinct position vectors r and r∗,
which locate the same point in space. We now examine how the observers
in frames A and A∗ view a single vector a that is fixed in A (Fig. 5.1).
Obviously, a does not appear to move relative to the observer in A, just
as a tree does not appear to move to a person standing on the surface of
the earth. But relative to the observer in A∗ , the vector a translates and
rotates. Translation does not change a vector, but rotation does. Thus,
even though a is not changing in frame A, the relative motion of frame A∗
makes a appear to be changing according to the relation
a∗ = Q · a (5.39)
186 Constitutive Relations
Example 5.2 Using the transformation (5.39), show that (a) the length of
a vector and the angle between two vectors fixed in frame A are objective
quantities; and (b) that velocity is not an objective quantity.
Solution. (a) Clearly, the length of a vector and the angle between two
fixed vectors are scalar invariants under (non-relativistic) changes in the
frame of reference. As measured in A, the squared length of a is ds2 = a · a.
As measured in A∗ , the squared length of a is
ds∗2 = a∗ · a∗ = (Q · a) · (Q · a)
T
= (a · QT ) · (Q · a) = a · (Q · Q) · a
= a·I·a = a·a
= ds2 , (5.40)
where Eqs. (5.37) and (5.39) have been used. Thus, as expected, the length
of a is independent of the motion of the observer.
Next, consider the angle between two vectors a and b fixed in frame A.
As seen from A∗ , Eq. (5.39) gives
a∗ = Q · a, b∗ = Q · b, (5.41)
and so
a∗ · b∗ = (Q · a) · (Q · b) = (a · QT ) · (Q · b)
T
= a · (Q · Q) · b = a · I · b
= a · b. (5.42)
∗
Thus, cos(a, b) = cos(a∗ , b ), which implies that angles are invariant under
changes of reference frame.
(b) If the velocity of a point in frame A is ṙ, then Eq. (5.38) shows that
the velocity of the point as seen from A∗ is
v∗ = ṙ∗ = Q · ṙ + Q̇ · r + ḃ
= Q · v + Q̇ · r + ḃ. (5.43)
5.2. Fundamental Constitutive Principles 187
Clearly, the velocity vector is not objective, as it does not satisfy v∗ = Q·v.
A similar calculation shows that acceleration also is not frame indifferent.
r∗1 = Q · r1 +b
r∗2 = Q · r2 +b. (5.45)
Q
a
r2* r2
r1
r1*
b O
O* A
A*
Fig. 5.2. Position vectors to the ends of vector a fixed in reference frame A, which is in
motion relative to frame A∗ .
b = T·a (5.46)
188 Constitutive Relations
Example 5.4 Suppose the governing equation for a physical system has
the form
a = A·B·b+C·c (5.49)
relative to an observer in frame A, where a, b, and c are vectors and A,
B, and C are second-order tensors. According to the principle of material
frame indifference, this equation must be seen as
a∗ = A∗ · B∗ · b∗ + C∗ · c∗ (5.50)
to an observer in frame A∗ . Show that the vector and tensor transformation
relations ensure that Eqs. (5.49) and (5.50) are equivalent.
Solution. With Eqs. (5.37), (5.39), and (5.48), Eq. (5.50) takes the form
T T
Q·a = (Q · A · Q ) · (Q · B · Q ) · (Q · b)
T
+(Q · C · Q ) · (Q · c)
T T
= Q · A · (Q · Q) · B · (Q · Q) · b
T
+Q · C · (Q · Q) · c
= Q·A·I·B·I·b+Q·C·I·c
= Q · (A · B · b + C · c).
For arbitrary Q, this relation implies Eq. (5.49), confirming its objective
character.
5.2. Fundamental Constitutive Principles 189
dr
A
dR O
O, O*
A, A*
t=0
O* A*
t>0
Material frame indifference demands that the forms of the defining rela-
tions for F be the same in all reference frames. Since dR appears the same
190 Constitutive Relations
Motion: r∗ = Q · r + b
Scalar: φ∗ = φ
Vector: a∗ = Q · a
Second-Order Tensor: T∗ = Q · T · QT
Deformation Gradient Tensor: F∗ = Q · F (5.55)
5.2. Fundamental Constitutive Principles 191
Scalars are always frame indifferent,1 but objective vectors and tensors must
either remain unaffected or transform according to the above relations.
Within a single frame of reference, equations are coordinate invariant if
they are written in tensor form. They also are frame indifferent if all their
variables transform according to Eqs. (5.55)2-5 under a change of frame.
Example 5.5 Determine whether the velocity gradient tensor L, the rate-
of-deformation tensor D, and the spin tensor Ω are objective quantities.
Solution. As seen in frame A, Eqs. (3.207) and (3.211) give
L = Ḟ · F−1
1
D = 2
(L + LT )
1
Ω = − LT ).
2 (L
Likewise, the observer in frame A∗ sees
L∗ = Ḟ∗ · (F∗ )−1
∗T
D∗ = 1 ∗
2 (L +L )
Ω∗ = 1
2
(L∗ − L∗T ).
The transformation relation F∗ = Q · F yields
Ḟ∗ = Q̇ · F + Q · Ḟ
(F∗ )−1 = F−1 · Q−1 = F−1 · QT ,
which are substituted into the above expression for L∗ to obtain
L∗ = (Q̇ · F + Q · Ḟ) · (F−1 · QT )
= Q̇ · QT + Q · L · QT .
Because of the term Q̇ · QT , we find that L is not frame indifferent.
Next, inserting this result into the relation for D∗ gives
T T
D∗ = 12 (Q̇ · QT + Q · Q̇ ) + 12 Q · (L+LT ) · Q .
Since Q is an orthogonal tensor, the first term in parentheses becomes
T d d
Q̇ · QT + Q · Q̇ = (Q · QT ) = (I) = 0.
dt dt
Thus,
D ∗ = Q · D · QT ,
which shows by (5.55)4 that D is objective. A similar analysis yields
T
Ω∗ = 12 (Q̇ · QT − Q · Q̇ ) + Q · Ω · QT ,
in which the first term generally does not disappear. Hence, Ω is not frame
indifferent.
1 Accordingly, scalars such as energy, ∇ · a, and A : B are objective quantities.
192 Constitutive Relations
Hence, Eq. (5.56) and the principle of material frame indifference provide
constitutive equations in the forms
T
σ = Θ · g(σ) (U) · Θ
T
or σ = Θ · f (σ) (C) · Θ
or σ = Θ · h(σ) (E) · ΘT . (5.60)
or
J −1 (ΘT · F) · t = g(σ) (U) · Θ .
T
t = g(t) (U) · ΘT
T
or t = f(t) (C) · Θ
T
or t = h(t) (E) · Θ . (5.62)
For the second Piola-Kirchhoff stress tensor, Eqs. (5.61) and (5.62)1 give
s = t · F−T = g(t)(U) · Θ · F−T ,
T
The last expression follows from the equation U−1 = F−1 · Θ, as verified
by the manipulations
U · U−1 = (Θ−1 · F) · (F−1 · Θ)
= Θ−1 · I · Θ = I,
where Eq. (3.147) has been used. Thus, the constitutive equations also can
be written in the forms
s = g(s)(U)
or s = f(s)(C)
or s = h(s)(E). (5.63)
These relations for the second Piola-Kirchhoff stress tensor are particularly
convenient since rotation does not appear explicitly. For small rotation
(Θ ∼
= I), Eqs. (5.60), (5.62), and (5.63) are all approximately the same.
ρ0 u̇ − s : Ė−ρ0 r + ∇ · q0 = 0
r 1
η̇ − + ∇ · q0 = 0 (5.66)
T ρ0 T
as given by Eqs. (5.23) and (5.33)2 . Eliminating r between these two equa-
tions yields
ρ0 (T η̇ − u̇) + s : Ė = 0. (5.67)
In the following, the general constitutive relations (5.64) and (5.65) are sub-
stituted into this equation to examine the cases for which (1) temperature
is an independent variable and (2) entropy is an independent variable.
Case 1: When temperature is taken as an independent variable, it is
convenient to introduce the Helmholtz free-energy function
ψ = u − Tη (5.68)
Next, substituting Eq. (5.69) into (5.70) and using the chain rule yields
∂ψ ∂ψ ∂ψ
s − ρ0 : Ė − ρ0 η + Ṫ − ρ0 · ∇Ṫ = 0.
∂E ∂T ∂∇T
For independent strain and temperature distributions, this equation implies
the relations
∂ψ
s = ρ0
∂E
∂ψ
η = −
∂T
∂ψ
= 0. (5.71)
∂∇T
196 Constitutive Relations
Thus, stress and entropy can be computed directly from ψ, and the last
equation shows that ψ is independent of the temperature gradient. The
first two of (5.71) represent the constitutive relations for a thermoelastic
material with temperature taken as an independent variable.
If the deformation is isothermal (Ṫ = 0), then ψ = ψ(R, E) and the
material behaves elastically. In this case, we define the strain-energy
density function
W (R, E) = ρ0 ψ (5.72)
per unit undeformed volume. Since the density ρ0 of the undeformed body
is independent of the deformation, Eq. (5.71)1 gives the constitutive equa-
tion
∂W
s= . (5.73)
∂E
Case 2: If entropy is taken as an independent variable, then the internal
energy is retained in the formulation. Substituting Eq. (5.65)3 into (5.67)
and using the chain rule gives
∂u ∂u ∂u
s − ρ0 : Ė + ρ0 T − η̇ − ρ0 · ∇η̇ = 0.
∂E ∂η ∂∇η
For independent strain and entropy fields, this equation yields
∂u
s = ρ0
∂E
∂u
T =
∂η
∂u
= 0. (5.74)
∂∇η
Thus, stress and temperature can be computed directly from u, and the
last equation shows that u is independent of the entropy gradient. The
first two of (5.74) represent the constitutive relations for a thermoelastic
material with entropy taken as an independent variable.
If the deformation is isentropic (η̇ = 0), then u = u(R, E) and the
material behaves elastically. In this case, we take
W (R, E) = ρ0 u, (5.75)
and Eq. (5.74)1 takes the form of Eq. (5.73).
A material possessing a constitutive relation of the form (5.73) is called
a hyperelastic material. In other words, the mechanical properties of a
hyperelastic material are characterized completely by a scalar strain-energy
5.2. Fundamental Constitutive Principles 197
∂W
σ = J −1 F · · FT
∂E
∂W ∂W
t= · FT =
∂E ∂FT
∂W
s= . (5.78)
∂E
Note that these relations are consistent with σ and s being symmetric
tensors (since E = ET ), while t is not symmetric in general.
Scalar forms of these equations can be found by direct substitution of
Eqs. (3.22), (3.30), and (4.30). Thus, Eq. (5.78)1 gives
∂W
σ kl gk gl = J −1 (FK
k
gk G K ) · GI GJ · (FLl GL gl )
∂EIJ
∂W K L
= J −1 FK
k l
FL δ δ gk gl
∂EIJ I J
∂W
= J −1 FIk FJl gk gl (5.79)
∂EIJ
198 Constitutive Relations
or
∗
J∗ ∂W
σI gI gJ = J −1 (gK GK ) · GI GJ · (GL gL )
∂EIJ
∂W K L
= J −1 δ δ gK gL
∂EIJ I J
∂W
= J −1 gI gJ ; (5.80)
∂EIJ
Eq. (5.78)2 gives
∂W
tIk GI gk = GI GJ · (FK k
G K gk )
∂EIJ
∂W
= F k δ K G I gk
∂EIJ K J
∂W
= F k G I gk (5.81)
∂EIJ J
or
∂W ∂W
tI·j GI gj = = GI g j ; (5.82)
∂(FIj GI g j) ∂FIj
∂W
σ ij = J −1 FIi FJj
∂EIJ
∗
J∗ ∂W
σI = J −1
∂EIJ
∂W
tIj = FJj
∂EIJ
∂W
tI·j =
∂FIj
∂W
sIJ = . (5.84)
∂EIJ
then we get
∂W
s12 = = 4CE12
∂E12
∂W
s21 = = 0,
∂E21
which is not correct. Some authors, therefore, replace Eq. (5.84)5 by the
equivalent expression
1 ∂W ∂W
sIJ = + , (5.88)
2 ∂EIJ ∂EJI
Thus far, we have explored the implications of six of the seven princi-
ples listed in Sec. 5.2. The seventh, the principle of material symmetry,
is used herein to deduce the form of the strain-energy density function
for orthotropic, transversely isotropic, and isotropic hyperelastic materi-
als. These types of material symmetries often are used to characterize soft
tissues. The approach used here follows Green and Adkins (1970). An
alternative method, involving theorems for tensor invariants, can be found
elsewhere (e.g., Spencer, 1984; Zheng, 1994).
This section deals with how W must depend on the strain components
for each type of material, and some specific functional forms are listed.
Section 5.7 discusses strategies for determining material parameters. It is
convenient in this section to work in local Cartesian coordinates using only
subscript notation.
5.3. Strain-Energy Density Function 201
W = W( 11 , 22 , 33 , 12 , 23 , 31 )
W = W( 11 , 22 , 33 , − 12 , 23 , − 31 ),
I3 = det(δ IJ + 2 IJ )
2 2 2
W = W( 11 , 22 , 33 , 12 , 23 , 31 , I3 ). (5.94)
¯33 = 33
2
¯12 = ( 22 − 11 ) sin θ cos θ + 12 (cos θ − sin2 θ)
¯23 = 23 cos θ − 31 sin θ
The basis for writing W in terms of physical strain components is discussed in Sec. 5.4.4.
204 Constitutive Relations
¯231 + ¯223 = 2
31 + 2
23
¯33 = 33 , (5.97)
but other combinations are more difficult to find. Fortunately, Green and
Adkins (1970) have done the job for us. Direct substitution of Eqs. (5.96)
verifies the relations
E2 = 2 12
2 2
F1 = 31 − 23
F2 = 2 31 23 , (5.99)
and Ē1 , Ē2 , F̄1 , and F̄2 are given by placing bars over all quantities in
Eqs. (5.99).
For W to be independent of rotation about the Z 3 -axis, Eqs. (5.97) and
(5.98) indicate that the strain-energy function has the form
2 2 2
W = W( 11 + 22 , 31 + 23 , 33 , E1 + E22 , F12 + F22 ,
E1 F1 + E2 F2 , E1F2 − E2 F1 ), (5.100)
which also satisfies transformation T3 of (5.91). Further manipulations
show that this expression can be written more compactly. First, Eqs. (5.99)
give
E12 + E22 = ( 11 + 22 )
2
− 4( 11 22 − 2
12 )
F12 + F22 = ( 2
31 + 2 2
23 )
2 2
E1 F1 + E2 F2 = 2 det[ IJ ] +( 11 +
22 )( 31 + 23 )
2
−2 33 ( 11 22 − 12 )
5.3. Strain-Energy Density Function 205
E12 + E22 E1 F1 + E2 F2
(E1 F2 − E2 F1 ) 2
=
E1 F1 + E2 F2 F12 + F22 ,
and comparison with the terms in (5.100) reveals that W also can be written
as
2 2 2
W = W( 11 + 22 , 31 + 23 , 33 , 11 22 − 12 , det[ IJ ]). (5.101)
Now, comparing these terms with those in the strain invariants I1 , I2 , and
I3 of Eq. (3.142) shows that the strain-energy function for a transversely
isotropic material can be simplified to6
W = W (I1 , I2 , I3 , I4 , I5 ), (5.102)
where
I4 = 33
2 2
I5 = 31 + 23 . (5.103)
I4 = e3 · E · e3
I5 = e3 · E2 · e3 − I42 , (5.104)
where μ, bi and cij are material constants and λf is the fiber stretch ratio
(λ2f = 1 + 2 33 = 1 + 2I4 ). Equation (5.105) separates the effects of the
isotropic matrix (first term) from those of the fibers (second term). In
contrast, Eq. (5.106) includes coupling between the matrix and fibers.
6 It is important to note here that any combination of invariants also is invariant.
206 Constitutive Relations
W = W (I1 , I2 , I3 ) (5.107)
where the λi are principal stretch ratios. This last expression takes advan-
tage of the fact that the strain invariants always can be written in terms
of principal strains. A disadvantage is that, except for some geometrically
symmetric problems, the eigenvalue problem for the principal strains must
be solved at each point as part of the solution procedure.
If the water in a tissue is mobile, such as in articular cartilage, then the
effects of bulk material compressibility should not be ignored. For com-
pressible materials, W must include the effects of local changes in volume,
as characterized by the strain invariant I3 . One possible form for W , given
by generalizing Eq. (5.111), is
n
n
n
W = cijk (I1 − 3)i (I2 − 3)j (I3 − 1)k , (5.113)
i=0 j=0 k=0
where the cijk are constants. Another form was proposed by Blatz and Ko
(1962), who studied the strain-energy density function for foam rubbers.
Using theoretical and experimental arguments, they suggested the function
μα 1 − 2ν −ν/(1−2ν)
W = I1 − 3 + I3 −1
2 ν
μ(1 − α) 1 − 2ν ν/(1−2ν)
+ I2 /I3 − 3 + I3 − 1 , (5.114)
2 ν
where μ, ν, and α (0 ≤ α ≤ 1) are constants. For small strain, μ becomes
the shear modulus and ν becomes Poisson’s ratio. (These parameters have
no physical meaning for large strain.) In addition, for I3 = 1, W takes
the Mooney-Rivlin form (5.109) for an incompressible rubber-like material.
Finally, setting α = 0 and ν = 0.25 yields results in good agreement with
some experimental data for foam rubber (Blatz and Ko, 1962). In this case,
Eq. (5.114) becomes
1/2
W = (μ/2)(I2 /I3 + 2I3 − 5). (5.115)
Most soft tissues stiffen with increasing strain levels. This behavior is
captured, for example, by a function of the form
where Q is given by the right-hand side of Eq. (5.109) or (5.114) for incom-
pressible or compressible tissue, respectively.
208 Constitutive Relations
2
X2 Z
X2'
X1
P
1' Z1
X
Fig. 5.4. Three coordinate systems defined at a point in an undeformed orthotropic ma-
terial (two-dimensional case). The dashed lines represent principal material directions.
where
∂X J
AJI = eI · G J =
∂Z I
J
∂X
AJI = eI · G J =
∂Z I
∂X J
AJI = GI · GJ = . (5.119)
∂X I
W = W (I1 , I2 , I3 ) (5.120)
210 Constitutive Relations
∂I2 1 ∂
= (I 2 − cKL cKL )
∂cIJ 2 ∂cIJ 1
∂I1 1 ∂cKL ∂cKL
= I1 − cKL + cKL
∂cIJ 2 ∂cIJ ∂cIJ
= I1 δ IJ − cKL δ KI δ LJ
= I1 δ IJ − cIJ. (5.125)
Computing ∂I3 /∂cIJ begins with the Cayley-Hamilton theorem (2.84) in
the form
C3 − I1 C2 + I2 C−I3 I = 0. (5.126)
In terms of components, substituting C = cIJ eI eJ and I = δ IJ eI eJ yields
cIL cLK cKJ − I1 cIK cKJ + I2 cIJ − I3 δ IJ = 0, (5.127)
and setting J = I produces the contracted form
cIL cLK cKI − I1 cIK cKI + I2 cII − 3I3 = 0.
9 Since we are working in Cartesian coordinates, the use of superscript notation is sus-
Finally, solving this equation for I3 , computing ∂I3 /∂cIJ , and noting the
above relations for the invariants and their derivatives yields
∂I3
= I2 δ IJ − I1 cIJ + cIK cKJ . (5.128)
∂cIJ
In direct notation, Eqs. (5.124), (5.125), and (5.128) become
∂I1
= I
∂C
∂I2
= I1 I − C
∂C
∂I3
= I2 I−I1 C + C2 . (5.129)
∂C
The last derivative can be put in an alternate form by first writing it as
∂I3
= C−1 · (I2 C−I1 C2 +C3 )
∂C
and then using Eq. (5.126) to get
∂I3
= I3 C−1 . (5.130)
∂C
Now, substituting Eqs. (5.129) and (5.130) into (5.123) yields
∂W
s= = 2[W1 I + W2 (I1 I − C)+W3 I3 C−1 ], (5.131)
∂E
where
∂W
Wi ≡ . (5.132)
∂Ii
W = W (I1 , I2 , I3 , I4 , I5 ). (5.142)
Equations (5.121) provide the strain invariants I1 , I2 , and I3 , while (5.103)
gives
I4 = 33
2 2
I5 = 31 + 23 , (5.143)
with the IJ being Cartesian strain components in local material coordi-
nates. Equation (5.88) yields
1 ∂W ∂W
S IJ = + (5.144)
2 ∂ IJ ∂ JI
for the corresponding Cartesian components of the second Piola-Kirchhoff
stress tensor.
Differentiating Eq. (5.142) yields
∂W ∂W ∂I1 ∂W ∂I2 ∂W ∂I3
= + +
∂ IJ ∂I1 ∂ IJ ∂I2 ∂ IJ ∂I3 ∂ IJ
∂W ∂I4 ∂W ∂I5
+ + . (5.145)
∂I4 ∂ IJ ∂I5 ∂ IJ
The first three terms of this expression are equivalent to the right-hand-side
of (5.123) and, therefore, lead to the stresses of Eq. (5.140). Thus, we only
need to deal here with the last two terms, with Eq. (5.143) giving
∂I4 ∂ 33
= = δ I3 δ J3
∂ IJ ∂ IJ
∂I5 ∂ 31 ∂ 23
= 2 31 + 23
∂ IJ ∂ IJ ∂ IJ
I J I J
= 2( 31 δ 3 δ 1 + 23 δ 2 δ 3 ). (5.146)
Next, using Eqs. (5.140), (5.145), and (5.146) transforms this relation into
∗
J∗
σI = ΦGIJ + ΨH IJ − pgIJ + (2J)−1 [W4 (δ K L L K
3 δ3 + δ3 δ3 )
K L K L L K L K I J
+2W5 ( 31 δ 3 δ 1 + 23 δ 2 δ 3 + 31 δ 3 δ 1 + 23 δ 2 δ 3 )]AK AL
In direct form, Eq. (5.136) gives the first three terms of (5.148). To deal
with the other terms, we note that Eq. (5.119)2 gives
AI3 = e3 · GI . (5.150)
With this equation, it is straightforward to show that (5.148) can be written
σ = α0 I + α1 B + α2 B2 + ΘM + ΛN, (5.151)
where
M = F · (e3 e3 ) · FT
2
N= F · (e3 eα + eα e3 ) · FT α3 . (5.152)
α=1
∗ ∗ ∗ ∗ ∗ ∗
If we take σ = σ I J gI gJ , M = M I J gI gJ , and N = N I J gI gJ , then
Eqs. (5.149) give the components of M and N (see Problem 5.7).
We again emphasize that these constitutive relations are based on the
requirement that the fibers (i.e., axis of symmetry) are aligned in the Z 3 -
direction. Otherwise, the equations would need to be modified accordingly.
5.4. Constitutive Relations for Some Common Material Symmetries 215
The analysis for an orthotropic material is similar to that used for a trans-
versely isotropic material. The strain-energy density function
2 2 2
W = W( 11 , 22 , 33 , 12 , 23 , 31 , I3 ) (5.153)
of Eq. (5.94) is substituted into (5.147). With the chain rule, the necessary
derivatives are
∂W ∂W ∂ 11 ∂W ∂ 22 ∂W ∂ 33
= + +
∂ KL ∂ 11 ∂ KL ∂ 22 ∂ KL ∂ 33 ∂ KL
∂W ∂ 212 ∂W ∂ 223 ∂W ∂ 231 ∂W ∂I3
+ + + + , (5.154)
∂ 212 ∂ KL ∂ 223 ∂ KL ∂ 231 ∂ KL ∂I3 ∂ KL
where
∂ 11
= δK L
1 δ1
∂ KL
∂ 212 ∂ 12 K L
= 2 12 =2 12 δ 1 δ 2 (5.155)
∂ KL ∂ KL
and so on, with the I3 term leading to the last term in Eq. (5.140). Substi-
tuting these and related expressions into (5.154) and the result into (5.147)
yields
I∗J∗ 1 ∂W K L ∂W K L L K
σ = δ δ + · · · + 12 2 δ 1 δ 2 + δ 1 δ 2 + · · · AIK AJL −pgIJ ,
J ∂ 11 1 1 ∂ 12
which simplifies to
∗
J∗ −1/2 ∂W I J ∂W I J ∂W I J
σI = I3 A A + A A + A A
∂ 11 1 1 ∂ 22 2 2 ∂ 33 3 3
∂W I J
+ 12 2 A1 A2 + AI2 AJ1
∂ 12
∂W I J
+ 23 2 A2 A3 + AI3 AJ2
∂ 23
∂W I J
+ 31 2 A3 A1 + A1 A3 − pgIJ .
I J
(5.156)
∂ 31
216 Constitutive Relations
J ∂W
ŝI = (5.161)
∂ ÊI J
or
J 1 ∂W ∂W
ŝI = + . (5.162)
2 ∂ ÊI J ∂ ÊJ I
−1/2 ∂W
Λ = I3
∂I5
Note again that these equations would need to be modified for fibers ori-
ented in any direction other than X 3 . For orthotropic materials, the revi-
sions required for Eq. (5.156) are similar.
5.5. Incompressibility 219
5.5 Incompressibility
¯ · v = tr D = 0.
Δ=∇ (5.173)
∂W
σ = J −1 F · · FT − p I, (5.177)
∂E
∂W
t = JF−1 · σ = · FT − Jp F−1
∂E
∂W
s = JF−1 · σ · F−T = − Jp F−1 · F−T . (5.178)
∂E
composed of neo-Hookean material with W = C(λ21 + λ22 + λ23 − 3). Is the “hydrostatic
pressure” zero, as may be expected for this case?
5.5. Incompressibility 221
∂W
σ ij = J −1 FIi FJj − pgij
∂EIJ
∗ ∗ ∂W
σ I J = J −1 sIJ = J −1 − pgIJ
∂EIJ
∂W
tIj = FJj − JpF Ij
∂EIJ
∂W
tI·j = − JpFjI . (5.180)
∂FIj
1/2
W ∗ = W − p(J − 1) = W − p(I3 − 1). (5.181)
Since J = 1, the added multiplier term does not alter the strain energy
stored in an incompressible material. With this expression, we can use
Eq. (5.78)2 to compute a first Piola-Kirchhoff stress tensor
∂W ∗ ∂W ∂J
t∗ = T
= T
−p . (5.182)
∂F ∂F ∂FT
The last term of this equation can be transformed by noting that, because
J = det F, the last formula in Table 2.5 (page 33) gives
∂J ∂ det F
= = (det F)F−T = JF−T ,
∂F ∂F
which, with J = 1, yields
∂J
= F−1 . (5.183)
∂FT
Accordingly, Eq. (5.182) becomes
∂W
t∗ = − pF−1 ,
∂FT
or, by (5.78)2 ,
∂W ∂W
t∗ = T
− pF−1 = · FT − pF−1 .
∂F ∂E
This result agrees with that for the first Piola-Kirchhoff stress tensor of
Eq. (5.178)1 with J = 1.
Thus, the constitutive relations for an incompressible material also can
be obtained by setting J = 1 and replacing W by W ∗ in Eqs. (5.78) to get
∂W ∗
σ = F· · FT
∂E
∂W ∗ ∂W ∗
t= · FT =
∂E ∂FT
∂W ∗
s= (5.184)
∂E
On setting
1
C1 = 2
(λ + 2μ)
C2 = −2μ,
where λ and μ are the Lamé constants, the above stress-strain relations
become
σ ij = λΔδij + 2μ ij (5.188)
or
σ = λΔI + 2μE, (5.189)
where
Δ = ii = tr E (5.190)
i j
is the dilatation, and E = ij e e is the strain tensor. Equations (5.188)
and (5.189) are Hooke’s law for a linear isotropic material [see Eq. (A.146)].
The two material constants λ and μ completely characterize the mechanical
behavior of a linear isotropic elastic material.
Note that Eq. (5.189) is valid for large strains if the material behavior
remains linear. In general, however, materials behave nonlinearly when
deformation becomes large. Thus, Eq. (5.189) usually is restricted to small
strains.
Determining constitutive relations for soft biological tissues has been and,
in some cases, remains a subject of considerable ongoing research. This
section discusses some of the issues involved in this process.
stants can be determined from uniaxial testing alone. On the other hand,
determining constants for linear anisotropic materials, e.g., bone, generally
requires various sets of experiments, such as tensile tests on samples cut
from different directions relative to the material axes. These experiments,
however, are still relatively straightforward.
In contrast, the functional form of W is generally not known a pri-
ori for nonlinear elastic materials, and determining mechanical properties
requires data from various loading protocols. Unless the structure is one-
dimensional, e.g., an actin microfilament, uniaxial testing alone is not ade-
quate. At the very least, biaxial testing is required (Humphrey, 2002; Sacks
and Sun, 2003).
In most studies of soft tissue, the functional form of W is postulated,
guided by microstructural considerations and the principles discussed ear-
lier in this chapter. (Some popular choices for W are given in Sec. 5.3.)
Then, for the postulated W , the boundary value problem is solved for an ex-
perimental loading protocol, and fitting theoretical results to experimental
data provides values for the unknown material coefficients.
A disadvantage of this approach is that it is difficult to know whether
the chosen form for W is correct even if it appears to fit the data well. One
reason for this is that the range of strain combinations that can be covered
experimentally is limited, and some physiologically significant deformations
may occur outside this realm for which the postulated W is not appropriate.
In biomechanics applications, however, the choice for W may not need
to be precise if it predicts approximately the correct behavior within the
deformation range of interest. Another approach is to use experimental data
to determine the functional form of W , as well as the material parameters
(Humphrey, 2002; Sacks and Sun, 2003).
Due to the complexity of the problem, most material testing of soft tis-
sues has involved relatively simple geometry and loading protocols. To
enable precise control of loading conditions, these tests often are conducted
on specimens that are cut to convenient geometries. Experiments on my-
ocardium, for example, have used thin rectangular samples cut from the
heart (Demer and Yin, 1983; Humphrey and Yin, 1987; Yin et al., 1987;
Humphrey et al., 1990a; 1990b; Humphrey, 2002; Sacks and Sun, 2003). To
facilitate biaxial stretching in the fiber and cross-fiber directions, the edges
of the specimen are cut approximately parallel to and normal to the local
5.8. Problems 227
5.8 Problems
In general, the stress, heat flux, and velocity differ slightly between the
ends of the element (Fig. 5.5).
da wq
q+ dx
q wx
V + wV dx
f
V r
wx
dx
v v + wv dx
wx
Fig. 5.5. Differential element for a bar with thermomechanical loads (Problem 5.1).
(a) By summing forces on the element, derive the 1-D differential equa-
tion of motion.
(b) Using conservation of energy, derive the 1-D form of Eq. (5.18)1 .
5.2 Consider a stress tensor τ defined by
τ = 12 (s · U + U · s),
where s is the second Piola-Kirchhoff stress tensor, and U is the right
stretch tensor. Show that the stress power can be written in the form
Ps = J −1 τ : U̇.
5.3 Determine the transformation relations for the following quantities in
frame A as viewed by an observer in frame A∗ :
(a) The strain tensors E and e as defined by Eqs. (3.28).
(b) The stress tensors σ, t, and s, as defined by Eqs. (4.23) and (4.29).
(c) The gradient operator ∇ ¯ in the deformed configuration.
Hint: Consider the relations
∂r ¯ = ∂
dr = dr∗ · ∗ , ∇
∂r ∂r
as given by Eqs. (2.163)1 and (3.18)2 .
5.4 A bar is in a state of uniaxial stress σ = σ 0 exex relative to Cartesian
coordinates (x, y, z) fixed in reference frame A. Another Cartesian
system (x̄, ȳ, z̄), fixed in reference frame A∗ , initially coincides with the
system in A. Suppose the bar rotates through an angle θ in the xy-
plane while maintaining the same uniaxial state of stress (Fig. 5.6). If
frame A∗ rotates with the bar, determine the stress dyadic σ ∗ observed
in A∗ in terms of the Cartesian base vectors ēi fixed in frame A∗ .
5.8. Problems 229
y
y
T x
x
Fig. 5.6. Rotation of a bar under constant uniaxial stress (Problem 5.4).
(a) For deformation without transverse shear (Exz = Eyz = 0), write
W as a function of the Cartesian strain components Eij relative to
the plate coordinates.
(b) Suppose the plate is in a state of plane stress (σ zz = 0) and con-
strained to undergo biaxial stretch (specified Exx and Eyy ) with-
out shear (Exy = 0). Determine the constitutive relations for the
Cauchy stress components σ xx, σ yy , and σ xy in plate coordinates.
Hint: Use incompressibility to determine Ezz and the condition
σ zz = 0 to find the Lagrange multiplier p.
(c) For c = 1 kPa, α = 0.1, β = 60◦ , and Eyy = 0, plot the stress
components from part (b) as functions of the stretch ratio Λx for
0.5 ≤ Λx ≤ 3. Do the principal axes of stress and strain coincide?
Fibers
y x-
y- E
x
Biomechanics Applications
The governing equations, along with appropriate boundary and initial con-
ditions, define a boundary value problem in the theory of elasticity. As
derived in the previous chapters, the tensor equations of nonlinear elastic-
ity consist of the following:
233
234 Biomechanics Applications
Kinematic Relations:
T
F = (∇r)
E = 12 (FT · F − I)
= 12 ∇u + (∇u) + (∇u) · (∇u)T
T
(6.1)
Stresses:
σ = J −1 F · t = J −1 F · s · FT (6.2)
Equations of Motion:
¯ · σ + f = ρa
∇
∇ · t + f 0 = ρ0 a
∇ · (s · FT )+f 0 = ρ0 a (6.3)
Constitutive Relations:
∂W
σ = J −1 F · · FT − p I
∂E
∂W
t= · FT − JpF−1
∂E
∂W
s= − JpF−1 · F−T (6.4)
∂E
Compressible material: p = 0; incompressible material: J = 1.
Incompressibility:
J = det F = 1 (6.5)
Kinematic Relations:
Fi = λ i
Ei = 12 (λ2i − 1) (6.6)
Stresses:
σ i = J −1 λi ti = J −1 λ2i si (6.7)
Equations of Motion:
Depend on the specific coordinate system.
Constitutive Relations:
λ2i ∂W λi ∂W
σi = −p = −p
J ∂Ei J ∂λi
∂W J ∂W J
ti = λi − p= − p
∂Ei λi ∂λi λi
∂W J 1 ∂W J
si = − 2p= − 2p (6.8)
∂Ei λi λi ∂λi λi
Compressible material: p = 0; incompressible material: J = 1.
Incompressibility:
J = λ1 λ2 λ3 = 1 (6.9)
In the constitutive relations, the equation
∂W ∂W ∂Ei ∂W
= = λi (i not summed)
∂λi ∂Ei ∂λi ∂Ei
has been used.
The specific equations and variables used to formulate and solve a
boundary value problem depend on the geometry, the types of loads, the
solution method (e.g., numerical or analytical), and personal preference.
Moreover, elasticity problems can be formulated either by specifying the
applied loads and computing the resulting deformation or by specifying the
deformation and computing the loads needed to produce it. The latter
method, called the inverse method, often is more convenient in solving non-
linear problems. Sometimes, however, it is not reasonable to specify the
entire deformation a priori. In such cases, a semi-inverse method may be
useful in which only part of the deformation is specified, with the rest to
be determined as part of the solution procedure.
236 Biomechanics Applications
t22
X2, x2
Fibers
t11 X1, x1
where the λi are stretch ratios. Here, we assume the λi are specified vari-
ables.
J = det F = λ1 λ2 λ3 = 1. (6.12)
For example, if λ1 and λ2 are specified, then this relation gives λ3 indepen-
dently of the forces applied in the x3 -direction. Thus, a well-posed problem
for this case requires specifying one of the surface tractions, say t3 , rather
than λ3 . (For the membrane to remain planar, t3 must be applied to both
faces normal to x3 , in opposite directions.) Then, Eq. (6.11) gives (with
J = 1)
∂W
p = λ3 − t3 , (6.13)
∂λ3
which is uniformly valid for homogeneous deformation. The other stress
components now can be determined using (6.11).
238 Biomechanics Applications
X3, x3
X2, x2
t22
Fibers
t11
X1, x1
I2 = 3 + 4(E1 + E2 + E3 + E1 E2 + E2 E3 + E3 E1 )
= λ21 λ22 + λ22 λ23 + λ23 λ21
I4 = E1 = 21 (λ21 − 1)
I5 = 0, (6.16)
W = W (I1 , I2 , I4 ). (6.17)
∂W
= λ2 [2W1 + 2W2 (λ21 + λ23 )]
∂λ2
∂W
= λ3 [2W1 + 2W2 (λ21 + λ22 )].
∂λ3
Inserting these expressions and λ3 = 1/λ1 λ2 into Eqs. (6.15) yields
1
t1 = 2λ1 1 − (W1 + λ22 W2 ) + λ1 W4
λ41 λ22
1
t2 = 2λ2 1− 2 4 (W1 + λ21 W2 ). (6.18)
λ1 λ2
The Cauchy stress components, given by (6.7), are σ i = λi ti (i not
summed).
For illustration, suppose the membrane is composed of a material with
E E
O1 O
Fig. 6.3. Stress-stretch curves for equibiaxial stretching of a transversely isotropic, in-
compressible membrane for various values of the fiber modulus (β). The fibers are
oriented in the x1 -direction.
where β correlates with the fiber stiffness. Substitution into Eqs. (6.18)
gives the stresses as functions of λ1 and λ2 . For equibiaxial stretch
(λ1 = λ2 ), t1 is plotted in Fig. 6.3 as a function of the stretch ratio for
various values of β. Since the fibers do not contribute to the stress in the
x2 -direction, the β = 0 curve for t1 also represents the curve for t2 , but for
all values of β. For relatively small strain (λ1 = λ2 < 1.05), all the solutions
nearly coincide. For large strain, however, the curves diverge significantly,
with the curve shape changing from primarily concave downward to con-
cave upward as β increases. The curves for β = 0 and 0.5 are typical for
incompressible rubber, while the curves for larger values of β are similar to
those generated by tests on most soft biological tissues (Fung, 1993).
Results also are shown for an isotropic membrane (β = 0, Fig. 6.4).
Here, the membrane is stretched in the x1 -direction, while the x2 -direction
is held fixed at λ2 = 1, 1.5, and 2. Note that t2 remains relatively constant
as λ1 increases while λ2 is held at the value 2.
O
O
O
Fig. 6.4. Stress-stretch curves for biaxial stretching of an isotropic incompressible mem-
brane (β = 0). The stretch ratio λ2 is held fixed at the three values indicated.
For small strain, the material parameter ν can be identified with Pois-
son’s ratio. For the special case ν = 1/4, the above relations become
⎡ 1
⎤
2 3
λ 1
t1 = 2C2 ⎣ 2 − 3⎦
λ1 λ1
⎡ 1
⎤
λ2 3
1⎦
t2 = 2C2 ⎣ 1 − 3 . (6.24)
λ2 λ2
O O
tiple solutions always must be kept in mind when dealing with nonlinear
problems.
Equibiaxial loading curves computed from Eqs. (6.23) for various values
of ν show that the effects of compressibility can be quite large (Fig. 6.5).
The membrane stiffness increases markedly with ν, especially as ν ap-
proaches 0.5, corresponding to an incompressible membrane.
For many years, shear deformation received relatively little attention in the
biomechanics literature. Nevertheless, shear is an important issue in nu-
merous problems. For example, torsional and transverse shear of the heart
wall play major roles in proper functioning of the left ventricle (Waldman
244 Biomechanics Applications
Fig. 6.6. Stress-stretch curves for uniaxial loading of an isotropic compressible bar for
various values of the material constant ν.
et al., 1985; Hansen et al., 1988; Moon et al., 1994; LeGrice et al., 1995,
1997).
In this section, we examine the forces required to sustain simple shear
of a homogeneous block of tissue (see Example 3.3 on page 80). As in the
stretching problem, surface tractions are assumed to be distributed uni-
formly, giving a homogeneous deformation. Hence, the equilibrium equa-
tions (6.3) are satisfied identically (with f = a = 0). The problem of simple
shear illustrates some important points in using nonorthogonal coordinate
systems to solve problems in nonlinear elasticity.
Fig. 6.7. Simple shear of a unit cube composed of transversely isotropic material.
6.3. Simple Shear of Soft Tissue 245
x1 = X 1 + kX 2 , x2 = X 2 , x3 = X 3 , (6.28)
where the parameter k defines the magnitude of the shear (Fig. 6.7). This
mapping produces the convected base vectors
as given by (3.54), and Eqs. (2.18) provide the contravariant base vectors
∗ ∗ ∗
g1 = e1 − ke2 , g 2 = e2 , g 3 = e3 . (6.30)
The directions of these base vectors are shown in Fig. 6.7. In addition, with
(3.56)1 , the Cartesian components of the Lagrangian strain tensor are
⎡ ⎤
0 k/2 0
[EIJ ] = 12 [CIJ − δ IJ ] = ⎣ k/2 k 2 /2 0 ⎦ . (6.31)
0 0 0
Note that Eq. (3.52) yields J = det F = 1 for any value of k. Thus, the
deformation described by (6.28) is isochoric, regardless of the composition
of the block.
Later, we will need components of the metric tensors. Since GI = GI =
eI ,
⎡ ⎤
1 0 0
[GIJ ] = [GIJ ] = [eI · eJ ] = ⎣ 0 1 0 ⎦, (6.32)
0 0 1
Given the strains of (6.31), all stress components can now be computed from
(6.36). On the other hand, if the block consists of compressible material,
then p = 0 and Eq. (6.36) stipulates that the functional form of W must
∗ ∗
be compatible with the plane-stress condition σ 3 3 = ∂W/∂E33 = 0 when
E33 = 0.
With W expressed in terms of the strain invariants, Eq. (5.148) gives
the constitutive relation
∗
J∗
σI = ΦGIJ + ΨH IJ − pgIJ + ΘM IJ + ΛN IJ (6.38)
M IJ = δ I2 δ J2
N IJ = (δ I2 δ J1 + δ I1 δ J2 )E12 .
∗ ∗
Surface Tractions. With the components σ I J of the Cauchy stress
tensor given above, we can compute the true traction vector on the surfaces
of the block using
T(ni ) = ni · σ, (6.42)
where ni is the unit vector normal to the deformed xi -surface. The geom-
etry of Fig. 6.7, along with Eqs. (6.30) and (6.33), gives
∗ ∗
g1 e1 − ke2 g2
n1 = ∗ ∗ = √ , n 2 = ∗ ∗ = e2 . (6.43)
g1 1 1 + k2 g2 2
Hence, we have
g1
∗ ∗ ∗ ∗ ∗
σI J
∗ ∗
σ1 J
(n1 ) 1
T = ∗ ∗ · σ I J
gI gJ = ∗ ∗ δ I gJ = ∗ ∗ gJ
g1 1 g1 1 g1 1
1 ∗ ∗ ∗ ∗
= ∗ ∗ σ 1 1 g1 ∗ + σ 1 2 g2 ∗
g 1 1
g2
∗ ∗ ∗ ∗ ∗
σI J
∗ ∗
σ2 J
T(n2 ) = ∗ ∗ · σ I J gI gJ = ∗ ∗ δ 2I gJ = ∗ ∗ gJ
g2 2 g2 2 g2 2
1 ∗ ∗ ∗ ∗
= ∗ ∗ σ 2 1 g1 ∗ + σ 2 2 g2 ∗ . (6.44)
g 2 2
Because the T(ni ) are vectors, the stress components normal and tan-
gential to the surfaces of the block can be determined by dotting with the
corresponding unit vectors (see the geometry in Fig. 6.7). On the x1 -face,
these components are
∗ ∗ ∗
(n1 ) g1 σ1 1
Tnorm = ∗ ∗ · T(n1 ) = 1∗ 1∗
g1 1 g
∗ ∗
σ1 1
=
1 + k2
g2 ∗ 1 ∗ ∗ ∗ ∗
(n1 )
Ttang =√ · T(n1 ) = ∗ ∗ σ 1 1 g2 ∗ 1 ∗ + σ 1 2 g2 ∗ 2 ∗
g2 ∗ 2 ∗ g 1 1 g2 ∗ 2 ∗
k ∗ ∗ ∗ ∗
= 2
σ 1 1 + σ1 2 , (6.45)
1+k
while on the x2 -face,
∗ ∗ ∗
(n2 ) g2 σ2 2
Tnorm = ∗ ∗ · T(n2 ) = ∗ ∗
g2 2 g2 2
∗ ∗
= σ2 2
g1 ∗ 1 ∗ ∗
(n ) 2∗ 2∗
2
Ttang = √ · T(n2 ) = σ 2 1
g 1 ∗ 1∗ + σ g 1 ∗ 2∗
g1 ∗ 1 ∗ g1 ∗ 1 ∗ g 2 ∗ 2 ∗
∗ ∗ ∗ ∗
= σ2 1
+ kσ 2 2
. (6.46)
6.3. Simple Shear of Soft Tissue 249
where
With Eqs. (3.22)2 and (6.30) and GI = eI , the inverse of the deformation
gradient tensor for this problem can be written in the form
6.3.2 Solution
Incompressible Tissue. For an incompressible material, the condi-
tion I3 = 1 renders ∂W/∂I3 indeterminate, and p becomes a Lagrange
∗ ∗
multiplier in Eqs. (6.40). Setting σ 3 3 = 0 in (6.40) yields
p = Φ + (2 + k 2 )Ψ, (6.51)
Substituting these relations into (6.45) and (6.46) yields the surface trac-
tions
(n1 ) k2
Tnorm =− [Φ + (2 + k 2 )Ψ]
1 + k2
(n )
1
k kΛ
Ttang = 2
(Φ + Ψ) +
1+k 2
(n2 )
Tnorm = −k 2 Ψ + Θ
(n )
2
Ttang = k(Φ + Ψ + Λ/2 + Θ). (6.53)
Note that, unlike in the linear theory, normal forces must be exerted on
the block to maintain simple shear. For the material considered here, the
normal stress is compressive on the x1 -face and tensile on the x2 -face. How-
ever, if the material is isotropic (β = 0), the normal force on the x2 -face
(n2 )
is zero. In fact, Eq. (6.53)3 shows that Tnorm can be compressive for cer-
tain transversely isotropic materials. For k << 1, these results reduce to
those of the linear theory, with the tangential tractions becoming equal to
∗ ∗
the shear stress σ 1 2 , and the normal stresses [O(k 2 )] can be neglected
compared to the shear stress [O(k)].
simultaneously for shear in the X 1 X 2 -plane. One material that does satisfy
these conditions consists of fibers initially aligned in the X 2 -direction that
are embedded in a matrix of the Blatz-Ko type. Here, we consider a material
characterized by
1 − 2ν −ν/(1−2ν)
2
W = C1 I1 − 3 + I3 − 1 + βI4 , (6.57)
ν
which is an extended form of Eq. (5.114) with α = 1 and reduces to
Eq. (6.19) when I3 = 1. Direct substitution shows that Eq. (6.56) is satis-
fied identically, and Eqs. (6.40) yield stress components identical to those of
Eq. (6.54). This is the expected result, because the specified deformation is
isochoric, and hence material compressibility should not affect the solution.
Papillary muscles play a vital role in the pumping efficiency of the heart.
These cylindrically shaped muscles are attached at one end to the wall of
the left or right ventricle and at the other end to the mitral or tricuspid
valve, respectively. During ventricular systole, increasing blood pressure
in the ventricles tends to push the valves upward into the atria, which
would allow backflow of blood during ejection. The papillary muscles pre-
vent this inversion by contracting and holding the valves closed. Due to
their relatively simple geometry and muscle fibers aligned along their axes,
papillary muscles have long been popular in studies of the passive and ac-
tive mechanical properties of heart muscle (Pinto and Fung, 1973; Fung,
1993; Criscione et al., 1999; Humphrey, 2002). For determining extensional
and shear properties, combined extension and torsion is a useful loading
protocol.
As a model for a papillary muscle, consider a solid circular cylinder
of undeformed radius b0 and length 0 (Fig. 6.8). The unloaded cylinder
is composed of material that is transversely isotropic relative to the axial
direction Z. The ends are subjected to tractions that exert a net twisting
moment M and normal force N , while the curved surface is traction free.
Given the angle of twist per unit undeformed length ψ and the (uniform)
axial stretch ratio λ, we want to compute M , N , and the stress distribution
in the cylinder. Note that, unlike the previous problems in this chapter,
torsional deformation of a cylinder is not homogeneous.1
1 This book considers only passive behavior of heart muscle. The mechanics of active
contraction are treated elsewhere (see, e.g., Fung, 1993; Humphrey, 2002; Taber, 2020).
252 Biomechanics Applications
Z
N
Q
b0 b
R
l0 l g3
G3 g3*
Fibers
Undeformed Deformed
Fig. 6.8. Extension and torsion of a cylindrical model for a papillary muscle loaded by
an axial force N and twisting moment M.
⎡ ⎤ ⎡ ⎤
1 0 0 1 0 0
IJ ⎢ 1 ⎥ ij ⎢ 1 ⎥
G = [GI · GJ ] = ⎣ 0 0 ⎦, g = [gi · gj ] = ⎣ 0 0 ⎦,
R2 r2
0 0 1 0 0 1
⎡ ⎤
1
⎢ ṙ 2 0 0 ⎥
IJ ⎢ ⎥
⎢ 1 ψ2 ψ ⎥
g = [gI · gJ ] = ⎢ 0 + − 2 ⎥. (6.62)
⎢ r2 λ2 λ ⎥
⎣ ψ 1 ⎦
0 − 2
λ λ2
The deformation gradient tensor can be written in several forms. With
the above relations, Eq. (3.22)1 gives
F = gI G I
= g1 ∗ G 1 + g2 ∗ G 2 + g3 ∗ G 3
r
= ṙer eR + eθ eΘ + λez eZ + ψreθ eZ
R
= ṙg1 G1 + g2 G2 + λg3 G3 + ψg2 G3 . (6.63)
Note that, with the FIi given by Eq. (3.64), the last expression is consistent
with the form F = FIi gi GI . (Recall again the distinction between the base
vectors g1∗ and g1 , etc.) For relatively simple problems such as this, many
authors find it convenient to work entirely with physical components, as
given by the third line of (6.63). For illustration, however, we stay here
with tensor components. Of course, physical components always can be
computed as desired.
Finally, Eq. (3.69) gives the Lagrangian strain components
⎡ 2 ⎤
ṙ − 1 0 0
1⎣ ⎦,
[EIJ ] = 0 r 2 − R2 ψr 2 (6.64)
2 2 2 2 2
0 ψr λ +ψ r −1
254 Biomechanics Applications
ψ 2r4
I5 = Ê13
2 2
+ Ê23 = . (6.65)
4R2
Recall that the modified invariants I4 and I5 are needed for a curvilinear
transversely isotropic material, with the physical strain components Ê13
and Ê23 given by Eq. (3.70).
where
−1/2 ∂W −1/2 ∂W 1/2 ∂W
Φ = 2 I3 , Ψ = 2 I3 , p = −2 I3
∂I1 ∂I2 ∂I3
∗ ∗
2 Green and Zerna (1968) circumvented complications in using σ I J by taking the un-
−1/2 ∂W −1/2 ∂W
Θ = I3 , Λ = I3 . (6.68)
∂I4 ∂I5
The M IJ and N IJ are functions of ÂIJ defined by Eq. (5.164). Because
the principal material axes X I coincide with the X I -axes, we find
⎡ ⎤
I 1 0 0
1 ∂X 1
ÂIJ = = δ I = ⎣ 0 R−1 0 ⎦ . (6.69)
G(J J ) ∂X J G(J J ) J
0 0 1
Hence, Eqs. (5.141)4 and (5.169)3,4 yield
⎡ 2 ⎤
r 2 2 2
⎢ R2 + λ + ψ r 0 0 ⎥
IJ ⎢ ⎢ 1 2 r 2 ⎥
⎥
=⎢ 2 2 2 ⎥
H
⎢ 0 ṙ + λ + ψ r − ψ
⎣ R 2 R 2 ⎥
2
r 2
r ⎦
0 − 2ψ ṙ 2 + 2
R R
⎡ ⎤
⎡ ⎤ 0 0 0
IJ 0 0 0 IJ ⎢ ⎢ ψr 2 ⎥ ⎥
M = ⎣ 0 0 0 ⎦, N =⎢ 0 0
2R2 ⎥ , (6.70)
⎣ 2 ⎦
0 0 1 ψr
0 0
2R2
which are valid for a cylinder with fibers oriented originally in the X 3 -
direction. The solution for an isotropic cylinder can be obtained by setting
Λ = Θ = 0. Note also that Eq. (3.70) provides the ÊI J = ÊIJ required
by N IJ . Inserting these relations into (6.67) gives
∗ ∗ r2 1
σ1 1
= Φ+ + λ2 + ψ 2 r 2 Ψ − 2 p
R2 ṙ
∗ ∗ 1 1 1 ψ2
σ2 2
= 2 Φ + 2 ṙ 2 + λ2 + ψ 2 r 2 Ψ − + p
R R r2 λ2
∗ ∗ r2 1
σ3 3
= Φ + ṙ 2 + 2 Ψ − 2 p + Θ
R λ
∗ ∗ r2 1 r2
σ2 3
= ψ − 2Ψ+ 2 p+ Λ
R λ 2R2
∗ ∗ ∗ ∗
σ1 2
= σ1 3
= 0, (6.71)
1∗ 2∗ 1∗ 3∗
which are consistent with the symmetry requirement σ =σ = 0.
Cauchy stress components relative to the {gi } basis now can be com-
puted using Eqs. (4.33) and (4.34), which give
∗
J∗
σ ij = FIi FJj σ I .
256 Biomechanics Applications
With the FIi given by the last of (6.63), this equation yields
⎡ ⎤
2 1∗ 1∗
⎢ ṙ σ 0 0 ⎥
⎢ ⎥
[σ ] = ⎢
ij
⎢ 0 σ
∗
2 2 ∗
+ 2ψσ
∗
2 3 ∗
+ψ σ
∗
2 3 3 ∗
λ(σ
∗
2 3 ∗
+ ψσ 3∗ 3∗ ⎥ .
) ⎥
⎣ ⎦
∗ ∗ ∗ ∗ ∗ ∗
0 λ(σ 2 3 + ψσ 3 3 ) λ2 σ 3 3
(6.72)
Substituting the convected stress components from (6.71) gives
11 2 r2 2 2 2
σ = ṙ Φ + +λ +ψ r Ψ −p
R2
1 2 ṙ 2 λ2 2 2 p ψ2 r2
σ 22 = + ψ Φ + + + ψ ṙ Ψ − + Λ + ψ 2 Θ
R2 R2 R2 r2 R2
r2
σ 33 = λ2 Φ + ṙ 2 + 2 Ψ + Θ − p
R
r2
σ 23 = λψ Φ + ṙ 2 Ψ + Λ + Θ
2R2
σ 12 = σ 13 = 0. (6.73)
which provide the resultant axial force N and twisting moment M applied
at the ends of the cylinder.
6.4.2 Solution
In the following, we develop explicit solutions for cylindrical specimens
composed of incompressible and compressible transversely isotropic muscle.
and the other stress components can be computed easily from the rest of
Eqs. (6.84). Substitution into (6.77) then yields the axial force and twisting
moment
1 ψ 2 b20 1 ψ2 b20
N = 2πb20 C1 λ − 2 − + C 2 1 − −
λ 4λ2 λ3 2λ3
%
C4 3 1 ψ 2 b20 ψ 4 b40
+ λ −λ+ 1+ 2 −
2 λ 4 6λ3
πb40 ψ C2 C4 2ψ2 b20
M = C1 + + λ2 − 1 + . (6.87)
λ λ 2 3λ
Note that, for an isotropic cylinder (C4 = 0), M is proportional to ψ, as
in the linear solution. Such is not the case, however, if the material is
transversely isotropic relative to the axial direction (C4 = 0).
Stress distributions for the special case of a neo-Hookean material
(C2 = C4 = 0) reveal clear differences between the linear and nonlinear
solutions (see Fig. 6.9a,b). According to the linear theory, for example, the
only non-zero stress component is the shear stress σ̂ 23 . In contrast, for large
angles of twist, the nonlinear theory predicts comparable magnitudes for
all three normal stress components (Fig. 6.9a). The results in Fig. 6.9a are
for a cylinder held at its undeformed length. As shown in Fig. 6.9b, axial
stretch increases the magnitude of the torsional shear stress σ̂ 23 . Moreover,
the distribution of axial stress σ̂ 33 , which is independent of r for stretch
without twist, becomes more and more nonuniform as the torsion increases
(Fig. 6.9a). (Note also that σ̂ 33 = σ̂ 11 when λ = 1.) Finally, the pres-
ence of axial fibers (C4 = 0) increases the magnitudes of the peak stresses
(Fig. 6.9c).
If the cylinder is held at its undeformed length (λ = 1), the axial stress
33
σ̂ is compressive at all r (Fig. 6.9a). This stress distribution produces
a net compressive force N that increases in magnitude with the twist
(Fig. 6.10).4 Hence, if N is not supplied, the cylinder elongates when
twisted. This phenomenon is known as the Poynting effect. For λ = 2, the
effect of twist on N is not nearly as large (Fig. 6.10).
Fig. 6.9. Distributions of Cauchy stress (physical components) in a solid circular cylinder
(papillary muscle) undergoing combined extension and torsion. (a,b) isotropic cylinder
with C2 = C4 = 0; (c) transversely isotropic cylinder with C2 = 0 and C4 = C1 /2.
Eqs. (6.73) still apply, but p is given by (6.68)3 rather than (6.83). If λ and
ψ are specified, then substituting the constitutive relations into the equilib-
rium equation (6.74) provides a differential equation to be solved for r(R).
In general, the resulting equation must be solved numerically, e.g., by finite
differences, which may be significantly more involved than the solution by
quadratures in the incompressible case.
A relatively simple solution, however, can be found for passive muscle
modeled as a specialized Blatz-Ko material without fibers (Carroll and
262 Biomechanics Applications
Fig. 6.10. Axial force generated in a solid isotropic circular cylinder undergoing extension
and torsion (C1 = C4 = 0). The incompressible and compressible solutions are identical
for λ = 1.
Substituting these relations into Eq. (6.74) gives the nonlinear differen-
tial equation
3 3
dR d2 R dR R
3R + − =0 (6.90)
dr dr 2 dr r
to be solved for the undeformed coordinate R(r). This equation, written
in Eulerian form for convenience, admits a solution of the form
R = cr, (6.91)
6.4. Extension and Torsion of a Papillary Muscle 263
where c is a constant. Inserting this relation into (6.89)1 and applying the
boundary condition σ 11 (b) = 0 yields
c = λ1/4 , (6.92)
and the stress components become
σ 33 = 2C2 1 − λ−5/2 (1 + λ1/2 ψ2 r 2 )
ψr
rσ 23 = 2C2
λ
σ 11 = σ 22 = σ 12 = σ 13 = 0. (6.93)
Finally, using these expressions in Eqs. (6.77), we compute the resultant
force and twisting moment
2π b0 33 1 1 ψ2 b20
N= √ σ R dR = 2πC2 b20 √ − 3 −
λ 0 λ λ 2λ3
2π b0 23 3 πC2 b40
M= σ R dR = ψ, (6.94)
λ 0 λ2
in which Eq. (6.91) has provided the transformation from r to R.
Arteries are the blood vessels that carry blood away from the heart. The
wall of an artery consists of three layers. The inner layer (intima) consists
of a single layer of epithelial cells, the middle layer (media) contains elastin
and smooth muscle, and the outer layer (adventitia) is composed mainly of
connective tissue.
Reflecting differences in function, the size and composition of arteries
change with distance from the heart. Arteries near the heart, including the
aorta, are relatively large and serve primarily as conduits of blood. The
walls of these “elastic arteries” are dominated by elastin. Downstream,
arteries gradually become smaller in diameter, and smooth muscle becomes
more prominent. The small-caliber arterioles (“muscular arteries”) regulate
the flow of blood to the tissues via smooth-muscle contraction or relaxation.
This response changes the diameter of the vessel, thereby altering resistance
to flow.
Hence, arteries are not merely passive tubes. Besides actively chang-
ing their diameter, arteries grow (change size) and remodel (change ma-
terial properties) to adapt to changes in flow and pressure. Developing
theoretical models for these processes, while accounting for heterogeneity
and viscoelasticity, is a challenging and ongoing endeavor (Rachev, 1997;
Humphrey, 2002; Taber, 2020). In this section, however, we ignore these
complications and consider arteries as thick-walled passive, homogeneous,
pseudoelastic tubes. We do include anisotropy, however.
Another feature that we include is residual stress, i.e., the stress that
remains in a body when all external loads are removed. Residual stress in
artery walls can be generated by nonuniform growth, swelling, turnover of
wall constituents, or other means. A popular method used to characterize
the magnitude of residual stress in soft tissue is to cut the tissue and mea-
sure the resulting deformation caused by the release of stress (Fung, 1990;
Humphrey, 2002; Taber, 2020). For example, when an unloaded section
of an artery is cut transmurally, the section typically springs open as cir-
cumferential residual stress is relieved (Fig. 6.12). Measuring the opening
angle φ provides an approximate way to characterize the amount of residual
6.5. Extension, Inflation, and Torsion of an Artery with Residual Stress 265
strain that was in the section before the cut. Residual stress then can be
computed if material properties are known.
Most investigators have assumed that a single radial cut is enough to
render an artery nearly free of stress. Some studies have suggested, however,
that a combination of radial and circumferential cuts may be needed (Vos-
soughi et al., 1993; Greenwald et al., 1997; Taber and Humphrey, 2001).
For simplicity, we assume here that the zero-stress state is produced by one
radial cut.
F = F2 . F1
f0
F1 F2 a b
a0 f
b0
B0 BU BL
Fig. 6.12. Artery (or left ventricle) cross section in zero-stress (B0), unloaded (BU ), and
loaded (BL) configurations.
R = R eR + Z eZ
ρ = ρ eρ + ζ e ζ
r = r er + z ez . (6.95)
In terms of a set of Cartesian base vectors, the unit vectors for each coor-
dinate system are
eR = ex cos Θ + ey sin Θ
eΘ = −ex sin Θ + ey cos Θ
eZ = ez
eρ = ex cos ϑ + ey sin ϑ
eϑ = −ex sin ϑ + ey cos ϑ
eζ = ez
6.5. Extension, Inflation, and Torsion of an Artery with Residual Stress 267
er = ex cos θ + ey sin θ
eθ = −ex sin θ + ey cos θ
ez = ez . (6.96)
The deformation F1 describes bending of a circular bar, and the map-
ping from B0 to BU is assumed to have the form
ρ = ρ(R), ϑ = πΘ/φ0 , ζ = ΛZ, (6.97)
where φ0 is the half-sector angle defined in Fig. 6.12, and Λ is the axial
stretch ratio of BU relative to B0 . Note that ϑ = 0 when Θ = 0, ϑ = π
when Θ = φ0 , and the opening angle is given by φ = 2(π − φ0 ). As in
Eqs. (3.57), combined extension, inflation, and torsion from BU to BL is
defined by the relations
r = r(ρ), θ = ϑ + ψζ, z = λζ, (6.98)
where ψ is the angle of twist per unit axial length of BU , and λ is the axial
stretch ratio in the loaded tube BL relative to the unloaded tube BU .
Equations (6.95) and (6.96) give the natural base vectors for the coor-
dinate systems in B0 and BU as
B0 : GR = R,R = eR BU : Gρ = ρ,ρ = eρ
GΘ = R,Θ = R eΘ Gϑ = ρ,ϑ = ρ eϑ
GZ = R,Z = eZ Gζ = ρ,ζ = eζ
(6.99)
G R = eR G ρ = eρ
GΘ = R−1 eΘ Gϑ = ρ−1 eϑ
G Z = eZ G ζ = eζ ,
where Eq. (2.18) is used to compute the contravariant base vectors. In
addition, with Eqs. (6.95)–(6.98), the convected base vectors are
B0 → BU : gR = ρ,R = ρ eρ
π
gΘ = ρ,Θ = ρ eϑ
φ0
gZ = ρ,Z = Λeζ
(6.100)
BU → BL : gρ = r,ρ = ṙer
gϑ = r,ϑ = reθ
gζ = r,ζ = ψreθ + λez ,
where prime and dot denote differentiation with respect to R and ρ, respec-
tively.
268 Biomechanics Applications
πρ
= ρ eρ eR + eϑ eΘ + Λeζ eZ (6.101)
φ0 R
F2 gρ G ρ + gϑ G ϑ + gζ G ζ
=
r
= ṙer eρ + eθ eϑ + λez eζ + ψr eθ eζ . (6.102)
ρ
The total deformation from B0 to BL is described by (see Fig. 6.12 and
Example 3.5, page 86)
F = F2 · F1
= λR er eR + λΘ eθ eΘ + λZ ez eZ + γeθ eZ , (6.103)
where
λR = r
πr
λΘ =
φ0 R
λZ = λΛ
γ = ψrΛ. (6.104)
Here, we have used the relation ṙρ = (∂r/∂ρ)(∂ρ/∂R) = ∂r/∂R = r .
Note that the λs in these terms represent stretch ratios only if γ = 0. Also,
if there is no residual stress in BU , then B0 and BU are identical. In this
case, φ0 = π and Λ = 1, and Eq. (6.103) reduces to Eq. (3.62) of Example
3.4, as it should.
Finally, with (6.103) and I = GR GR + GΘ GΘ + GZ GZ = eR eR +
eΘ eΘ + eZ eZ , the Lagrangian strain tensor of BL relative to B0 is
1
E = 2 (F
T
· F − I)
= ÊRR eR eR + ÊΘΘ eΘ eΘ + ÊZZ eZ eZ
+ÊΘZ eΘ eZ + ÊZΘeZ eΘ , (6.105)
where
ÊRR = 1
2
λ2R − 1
ÊΘΘ = 1
2 λ2Θ − 1
ÊZZ = 1
2
λ2Z + γ 2 − 1
ÊΘZ = ÊZΘ = 12 γλΘ . (6.106)
As indicated by Eq. (3.42), the ÊIJ are physical components of E.
6.5. Extension, Inflation, and Torsion of an Artery with Residual Stress 269
∂W
σ = F· · FT − p I. (6.107)
∂E
As in the papillary muscle problem, we take advantage of simplifications
by writing the equilibrium equations in terms of stress components relative
to the basis associated with the coordinates (r, θ, z) = (x1 , x2 , x3 ) in BL .
Since these coordinates are orthogonal, Eq. (4.48) gives
σ = σ̂ ij ei ej , (6.108)
σ̂ rr = σ̄ rr − p
σ̂ θθ = σ̄ θθ − p
σ̂ zz = σ̄ zz − p
∂W ∂W
σ̂ θz = λZ λΘ +γ
∂ ÊΘZ ∂ ÊZZ
∂W ∂W
σ̂ rθ = λR λΘ +γ
∂ ÊRΘ ∂ ÊZR
∂W
σ̂ zr = λR λZ , (6.109)
∂ ÊZR
where
∂W
σ̄ rr = λ2R
∂ ÊRR
∂W ∂W ∂W ∂W
σ̄ θθ = λ2Θ + γλΘ + + γ2
∂ ÊΘΘ ∂ ÊΘZ ∂ ÊZΘ ∂ ÊZZ
∂W
σ̄ zz = λ2Z , (6.110)
∂ ÊZZ
Note the coupling between normal stress and shear strain, as well as between
normal strain and shear stress, for a general anisotropic material.
270 Biomechanics Applications
∂ σ̂ rr σ̂ rr − σ̂ θθ
+ = 0
∂r r
∂ 2 rθ
(r σ̂ ) = 0
∂r
∂
(rσ̂ rz ) = 0. (6.111)
∂r
T(n) = n·σ
= −e1 · (σ̂ ij ei ej )
= −σ̂ ij δ 1i ej
= −σ̂ 1j ej .
r = a: σ̂ rr = −P, σ̂ rθ = σ̂ rz = 0
r = b: σ̂ rr = σ̂ rθ = σ̂ rz = 0 (6.112)
b
2
πa P + N = 2π σ̂ zz r dr
a
b
M = 2π σ̂ zθ r 2 dr. (6.113)
a
The pressure term in Eq. (6.113)1 comes from the force exerted by P on
the supports at the closed ends of the artery (Humphrey, 2002). Note also
that the equations for N and M are derived from (6.77) using the relations
between tensor and physical stress components of Eq. (4.98).
6.5. Extension, Inflation, and Torsion of an Artery with Residual Stress 271
6.5.3 Solution
Because the stresses depend only on r, Eqs. (6.111)2,3 can be integrated to
obtain
C1 C2
σ̂ rθ = 2 , σ̂ rz = , (6.114)
r r
where C1 and C2 are constants. Applying the boundary conditions (6.112)
gives C1 = C2 = 0, and thus σ̂ rθ = σ̂ rz = 0 everywhere. In the papillary
muscle problem, we used symmetry arguments to establish this result a
priori, and so the circumferential and axial equilibrium equations (6.111)2,3
were satisfied identically. Moreover, Eqs. (6.109) show that, in order for
these shear stresses to be zero, the artery must be composed of tissue with
a strain-energy density function that satisfies the conditions
∂W ∂W
= = 0. (6.115)
∂ ÊRΘ ∂ ÊZR
Otherwise, the stipulated deformation cannot be realized without applying
shear stresses to the curved surfaces of the tube.
Because the artery wall is assumed to be incompressible, the matrix
form of Eq. (6.103) gives the constraint
J = det F = λR λΘ λZ = 1.
Substituting (6.104) yields
π r dr
λΛ = 1,
φ0 R dR
which can be integrated to obtain
φ
r 2 = 0 (R2 − a20 ) + a2 , (6.116)
πλΛ
where the integration constant was found from the condition r(a0 ) = a.
If the opening angle φ, the axial stretch ratio λ, and the deformed inner
radius a are known, then the deformed outer radius can be computed using
the relation
1/2
φ0 2 2 2
b = r(b0 ) = (b − a0 ) + a . (6.117)
πλΛ 0
With r(R) now known, Eqs. (6.104) and (6.106) provide the deformation
gradient and strain components. Then, with W (ÊIJ ) specified, Eqs. (6.109)
and (6.110) give the stress components in terms of the yet unknown La-
grange multiplier p. To determine p, we substitute (6.109)1,2 into the radial
equilibrium equation (6.111)1 and integrate to get
b
θθ dr
rr
p(r) = σ̄ + σ̄ − σ̄ rr . (6.118)
r r
272 Biomechanics Applications
The upper limit on this integral is chosen to satisfy the boundary condition
σ̂ rr = σ̄ rr − p = 0 at r = b. With p(r) known, the internal pressure
required to maintain the deformation can be found using (6.118) and the
final boundary condition on the curved surfaces, σ̂ rr (a) = −P , to get
b
θθ dr
P = σ̄ − σ̄ rr . (6.119)
a r
Finally, the applied end loads M and N can be computed using Eqs. (6.113).
For a general form of W , the above integrals must be evaluated numer-
ically. Computing N , therefore, involves evaluating the above integral for
p within the integral in Eq. (6.113)1 . Although this procedure is relatively
straightforward, it may be more convenient to write the equation for N in a
form that does not involve p. This can be done by first writing Eq. (6.113)1
as
b
b
2 zz rr rr
πa P + N = 2π (σ̂ − σ̂ ) r dr + σ̂ r dr .
a a
Integrating the last integral by parts and using (6.109) and (6.111)1 yields
b 2 b b rr 2
r rr dσ̂ r
σ̂ rr r dr = σ̂ − dr
a 2 a a dr 2
b
b2 rr a2 rr σ̄ rr − σ̄ θθ r 2
= σ̂ (b) − σ̂ (a) + dr.
2 2 a r 2
Now, applying the boundary conditions σ̂ rr (a) = −P and σ̂ rr (b) = 0 and
substituting the result into the above relation for N yields
b
πa2 P + N = πa2 P + π 2σ̂ zz − σ̂ rr − σ̂ θθ r dr
a
b
= πa2 P + π 2σ̄ zz − σ̄ rr − σ̄ θθ r dr.
a
The integral in this expression does not contain p. Thus, Eq. (6.113)1 can
be replaced by
b
zz
N =π 2σ̄ − σ̄ rr − σ̄ θθ r dr. (6.120)
a
Note that the ÊRΘ and ÊZR terms have been dropped in order to satisfy
the requirement (6.115). Using experimental data, the authors determined
the following values for the material constants:
C = 11.2 kPa
a1 = 0.0499 a4 = 0.0042 a8 = 0.100
(6.122)
a2 = 1.0672 a5 = 0.0903
a3 = 0.4775 a6 = 0.0585.
Because Chuong and Fung (1986) did not examine torsion, they actually set
a8 = 0. The value a8 = 0.1 is used here only for illustration. In addition,
their measurements of vessel geometry before and after a transmural cut
yielded the following approximate parameter values:
a0 = 3.9 mm
b0 = 4.5 mm
a = 1.4 mm (unloaded)
b = 2.0 mm (unloaded)
φ = 220◦ (φ0 = 70◦ )
Λ = 1. (6.123)
For the unloaded intact artery with λ = 1, these values satisfy Eq. (6.117)
approximately.
Fig. 6.14. Pressure, axial force, and torque versus deformed inner radius in artery. Note
that, due to nonlinear coupling effects, the inflation stiffness in (a) first decreases and
then increases as the artery is stretched axially.
276 Biomechanics Applications
Fig. 6.15. Stress and strain distributions (physical components) in loaded artery with
(φ = 220◦ ) and without (φ = 0) residual stress.
strain are relatively modest (Fig. 6.15b). This behavior is caused by the
strong material nonlinearity between stress and strain, i.e., relatively small
changes in strain can produce large changes in stress. The bending that
accompanies the deformation from B0 to BU (Fig. 6.12) lowers the strain at
the inner wall and raises it at the outer wall (Fig. 6.15b, φ = 220◦), causing
corresponding but relatively large changes in stress in the loaded artery
(Fig. 6.15a). According to these results, the presence of residual stress
and strain in arteries tends to homogenize transmural stress distributions
in the loaded vessel, thereby increasing the efficiency of the artery as a
load-bearing structure (Fung, 1991).
In the heart, the left ventricle (LV) pumps blood to the tissues and organs
of the body. Due to the relatively high hydraulic resistance of the sys-
temic vascular system, the LV experiences higher pressures than the other
three chambers and, therefore, does the most work. Being composed of car-
diac muscle (myocardium), the LV responds like other muscles to increased
workload, i.e., it grows. The wall of the LV, therefore, is thicker than the
walls of the atria and right ventricle, and it becomes even thicker in response
to hypertension (high blood pressure). Because more blood is needed to
supply this greater muscle volume, the LV is more prone to myocardial in-
farction than the other heart chambers. (Infarctions usually are caused by
blockage of a coronary artery feeding the myocardium.) It is believed that
wall stress plays a role in adaptation of the heart to altered loading, as well
as in pathological conditions including infarction. For more than a century,
therefore, researchers have proposed models to compute wall stress in the
heart (Woods, 1892; Mirsky et al., 1974; Yin, 1981; McCulloch et al., 1992;
Hunter et al., 2003).
Z
(a) (b) Q cut
base R
eC
fibers eF
apex b
Fig. 6.16. Models for left ventricle. (a) Ellipsoid of revolution. (b) Thick-walled cir-
cular cylinder in approximately zero-stress configuration showing fiber orientation at
epicardium, midwall, and endocardium.
Fig. 6.17. Transmural fiber angle distributions for model of left ventricle.
6.6.2 Analysis
When subjected to internal pressure, the passive LV stretches in the longi-
tudinal and circumferential directions and, due to the helical fiber geometry,
twists about its axis of symmetry. Because this type of deformation was
studied for an artery in the previous section, we assume that the deforma-
tion of the LV model also is described by Eqs. (6.97) and (6.98). Therefore,
with the following modifications, the analysis of the artery also applies to
the LV.
First, because the apex is not constrained, we set the twisting moment
and axial force to zero, i.e., M = N = 0. With these specified loads,
Eqs. (6.113)2 and (6.120) provide two equations to be solved simultaneously
for ψ and λ. Moreover, if we specify a, as in the artery problem, then
Eq. (6.119) provides P . Otherwise, we could stipulate the value of P and
add (6.119) as a third simultaneous equation for the unknown a. In practice,
280 Biomechanics Applications
solving three nonlinear equations is not much more difficult than solving
two. Hence, we choose to specify the value of P , along with M = N = 0,
and solve for a, λ, and ψ.
Second, we need to modify the constitutive relations (6.109) to account
for the fiber geometry. For this purpose, it is convenient to express W in
terms of strain components relative to local Cartesian axes (R, F, C) in the
zero-stress (cut) configuration, where F and C represent the fiber and cross-
fiber directions, respectively. The unit vectors eF and eC are obtained by
rotating eΘ and eZ about eR through the helix angle β (Fig. 6.16b), giving
eF = eΘ cos β + eZ sin β
eC = −eΘ sin β + eZ cos β. (6.125)
Note that, as in the artery problem, the stipulated geometry demands that
ÊRΘ = ÊRZ = 0, which implies ÊRF = ÊRC = 0. Similarly, the artery
analysis shows that the Cauchy stress tensor for this problem can be written
σ = σ̂ rr er er + σ̂ θθ eθ eθ + σ̂ zz ez ez + σ̂ θz eθ ez + σ̂ zθ ez eθ (6.128)
relative to the deformed coordinates (r, θ, z). Here again, hat denotes phys-
ical components.
Substituting Eqs. (6.103) and (6.126)–(6.128), along with the relation
I = er er + eθ eθ + ez ez , into (6.107) provides the appropriate constitutive
relations. The manipulations are straightforward, but lengthy because of
the coordinate transformations involved. To make the computations more
manageable, we first note that Eq. (6.126) gives
∂W
= WRR eR eR + WF F eF eF + WCC eC eC + WF C eF eC + WCF eC eF ,
∂E
(6.129)
where
∂W
WIJ ≡ . (6.130)
∂ ÊIJ
6.6. Passive Filling of the Left Ventricle 281
Next, Eqs. (6.125) are used to write the above dyads in terms of eΘ and
eZ , i.e.,
eF eF = eΘ eΘ cos2 β + (eΘ eZ + eZ eΘ ) sin β cos β + eZ eZ sin2 β
eC eC = eΘ eΘ sin2 β − (eΘ eZ + eZ eΘ ) sin β cos β + eZ eZ cos2 β
eF eC = sin β cos β(eZ eZ − eΘ eΘ ) + eΘ eZ cos2 β − eZ eΘ sin2 β
eC eF = sin β cos β(eZ eZ − eΘ eΘ ) − eΘ eZ sin2 β + eZ eΘ cos2 β.
(6.131)
Inserting Eqs. (6.103), (6.129), and (6.131) into (6.107) reveals that all dot
products now involve the orthogonal system (eR , eΘ , eZ ), hence eliminating
these base vectors and leaving only (er , eθ , ez ), which is the basis used for
σ in (6.128). This observation allows us to express Eq. (6.107) in matrix
form without confusion.
In cylindrical coordinates, the matrix forms of Eqs. (6.103) and (6.128)
are
⎡ ⎤ ⎡ rr ⎤
λR 0 0 σ̂ 0 0
[F] = ⎣ 0 λΘ γ ⎦ , [σ] = ⎣ 0 σ̂ θθ σ̂ θz ⎦ ,
0 0 λZ 0 σ̂ zθ σ̂ zz
and putting (6.131) into (6.129) yields
⎡ ⎤
WRR 0 0
2 % %
⎢ c WF F + s2 WCC sc(WF F − WCC ) ⎥
∂W ⎢ 0 ⎥
=⎢⎢ −sc(WF C + WCF ) +c WF C − s WCF ) ⎥
2 2
∂E % % ⎥,
⎣ sc(WF F − WCC ) s2 WF F + c2 WCC ⎦
0 2 2
+c WF C − s WCF ) +sc(WF C + WCF )
(6.132)
where WF C = WCF and
c ≡ cos β, s ≡ sin β. (6.133)
Finally, in terms of these matrices, Eq. (6.107) becomes
[σ] = [σ̄] − p[I]
∂W
[σ̄] = [F] [F]T . (6.134)
∂E
For β = 0, we have eF = eΘ and eC = eZ , and these constitutive relations
reduce to Eqs. (6.109). (Recall that σ̂ rθ = σ̂ rz = 0.)
To compute the solution, we specify φ0 and take Λ = 1 as in the artery
problem. Then, with a, λ, and ψ to be determined, Eq. (6.116) gives r(R),
282 Biomechanics Applications
and (6.104) and (6.106) give the strains relative to the (R, Θ, Z) system.
Next, after the strains in the rotated (R, F, C) system are computed by
standard coordinate transformation, Eqs. (6.118) and (6.134)2 provide p(r),
then (6.134)1 yields the stress components relative to (r, θ, z). Given the
pressure P , the values of a, λ, and ψ are found by solving the system of
equations
b
zz
2σ̄ − σ̄ rr − σ̄ θθ r dr = 0
a
b
σ̂ zθ r 2 dr = 0
a
b dr
σ̄ θθ − σ̄ rr
= P, (6.135)
a r
as provided by (6.113)2 , (6.119), and (6.120) with M = N = 0.
It often is of interest to determine the load carried by the muscle fibers.
The fiber stress σ̂ ff can be computed using the relation
σ̂ ff = ef · σ · ef , (6.136)
where ef is the unit vector in the deformed fiber direction. This vector is
given by
F · eF
ef = , (6.137)
|F · eF |
in which Eqs. (6.103) and (6.125)1 provide F and eF , respectively.
Fig. 6.18. Transmural distributions of physical stresses in model for left ventricle with
cubic fiber angle distribution and opening angle φ = 50◦ . (a) Stresses for P = 2.67 kPa.
(b) Residual stresses (P = 0).
6.6.4 Results
Stress distributions are shown for a cubic distribution of fiber angle (n = 3
and β 0 = 60◦ , see Fig. 6.17) at an inflation pressure P = 2.67 kPa = 20
mmHg (Fig. 6.18a). Unlike the results for the artery (Fig. 6.15a), the stress
components in the LV generally do not vary monotonically across the wall.
Rather, the peak fiber stress occurs in the subendocardial layers of the wall.
This behavior reflects the effects of the fiber architecture. For example,
near midwall, where the fibers are oriented essentially in the circumferen-
tial direction (Fig. 6.17), the fiber and circumferential stresses are nearly
identical. Near the endocardium and epicardium, however, longitudinal
stress contributes significantly to the fiber stress.
Exhibiting similar complexity, residual stresses are an order of mag-
nitude smaller than the stresses in the loaded ventricle (Fig. 6.18b). As
discussed in the artery problem, these stresses must be self-equilibrating.
This requirement provides a valuable check on the accuracy of our solution.
The effects of fiber angle distribution and residual strain on fiber stress
are illustrated in Fig. 6.19, where various distributions of the fiber angle β
are considered. Note the following:
Fig. 6.19. Effects of fiber angle β and opening angle φ on transmural distributions of
physical fiber stress in model for left ventricle (P = 2.67 kPa).
Fig. 6.20. Effects of fiber geometry and residual stress on global behavior of model for left
ventricle. (a) Pressure-volume curves. (b) Axial stretch λ and twist ψ versus pressure.
z
f
a
P
6.7.2 Solution
With Eqs. (6.145), enforcing incompressibility yields
r 2 dr
J = det F = λr λθ λφ = = 1. (6.150)
R2 dR
Integrating this relation and using the condition r(a0 ) = a gives
r 3 = R3 − a30 + a3 . (6.151)
Hence, if the deformed inner radius a is known, the outer radius is
1/3
b = r(b0 ) = b30 − a30 + a3 . (6.152)
The Lagrange multiplier is determined by inserting (6.147) into (6.148)
and integrating. This procedure yields
b
∂W ∂W ∂W ∂W dr
p(r) = λr + λθ + λφ −2λr , (6.153)
∂λr r ∂λθ ∂λ φ ∂λr r
which satisfies the boundary condition σ̂ r (b) = 0. Substituting this ex-
pression into Eq. (6.147)1 and using the boundary condition σ̂ r (a) = −P
gives
b
∂W ∂W ∂W dr
P = λθ + λφ −2λr . (6.154)
a ∂λθ ∂λ φ ∂λr r
If the deformed inner radius a is specified, then Eqs. (6.145) and (6.151)
provide the stretch ratios as a function of r(R), and the above equations give
p(r) and P . Then, the stress components can be computed from (6.147).
If the shell wall is “thin,” the blastula can be treated as a pressur-
ized membrane. In this case, a0 /h0 >> 1 with h0 = b0 − a0 being the
undeformed wall thickness, and the above integrals can be simplified by
assuming that stress and strain vary relatively little across the wall. As the
deformed wall thickness h = b −a approaches zero, Eqs. (6.153) and (6.154)
become
∂W
p ∼ = λr
∂λr
h ∂W ∂W ∂W
P ∼ = λθ + λφ −2λr . (6.155)
a ∂λθ ∂λφ ∂λr
6.7. Blastula with Internal Pressure 289
Fig. 6.22. Pressure-radius curves for blastula model. (a) Effects of material constant α.
(b) Effects of ratio of undeformed inner radius to wall thickness. Results for a membrane
approximation also are shown.
Fig. 6.23. Computed distribution of circumferential stress across blastula wall for a =
2a0 . A stress concentration develops as the material constant α increases.
6.7.4 Results
Pressure-radius curves for three values of α are shown in Fig. 6.22a. As
α increases, the blastula stiffens considerably at large radii. For α → 0,
Eq. (6.157) reduces to the form for neo-Hookean material, i.e., W → C(I1 −
3). For this case, the pressure reaches a peak near a/a0 = 1.3, indicating a
limit-point instability. If pressure is prescribed for α = 0, then the radius
would become infinite, i.e., the blastula would burst, when the pressure is
greater than about 0.1 kPa. In contrast, the curves for larger values of α
6.8. Bending of the Embryonic Heart 291
remain stable, which should benefit the embryo. This type of analysis also
has applications in studies of aneurysm rupture (Humphrey, 2002).
For α = 0.2, Fig. 6.22b illustrates the effects of the relative wall thick-
ness. Results are shown for the “exact” thick-walled theory (6.154) and for
thin-walled membrane theory (6.155). As expected, the shell stiffens as the
shell becomes thicker (a0 /h0 decreases). Also as expected, the accuracy of
membrane theory increases dramatically for thinner shells.
As in the artery without residual stress (Fig. 6.15a, page 276), circum-
ferential stress decreases from the inner to the outer radius (Fig. 6.23). The
results shown correspond to a deformation a = 2a0 .6
As the material becomes more nonlinear (increasing α), a strong stress
concentration develops in the inner layers of the wall. In early work in car-
diac mechanics, (Mirsky, 1973) modeled the left ventricle as a thick-walled
isotropic spherical shell and found an extremely large stress concentration
near the endocardium. As Fung (1991, 1997) later pointed out, however,
these results do not make much sense physiologically, because the inner
layers of the ventricular wall would need to perform much more work than
the outer layers. Such a design could significantly decrease the efficiency
of the heart as a pump. This observation led in part to Fung’s emphasis
on the importance of residual stress (Fung, 1991), which can reduce stress
concentrations dramatically in the cardiovascular system (Fig. 6.15a).
The heart is the first functioning organ in the embryo. During the early
stages of development, a pair of membranes, on the left and right sides
of the embryo, fold and fuse at the midline to create a single, relatively
straight tube. The primitive heart tube is composed of an outer layer of
myocardium, a middle layer of extracellular matrix (cardiac jelly), and an
inner layer of endocardium. Soon thereafter, the first contractions occur
and the morphogenetic process of cardiac looping begins, as the heart
tube bends and twists into a curved tube to lay out the basic plan of the
future four-chambered pump (Manner, 2000; Taber, 2001, 2006).
Studies suggest that the forces that drive looping arise from sources both
intrinsic and extrinsic to the heart. During the later stages of looping, for
example, the body of the embryo compresses the ends of the heart tube,
6 This exceptionally large deformation is used here only for illustration. It is not likely
that blastulae actually deform this much due to internal pressure. During gastrulation,
however, regional strains may exceed 100%.
292 Biomechanics Applications
(a) (b)
Y y
b r T
a1 a2 X r1 r2 x
b
undeformed deformed
λc r2
M = r × T(eθ ) dr dz, (6.168)
−λc r1
where r is the position vector from the origin.7 We also could compute
a force resultant due to σ̂ z on the z-faces of the beam, but this is not of
interest here. Substituting Eqs. (6.158)2 and (6.167) and integrating over
z yields
r2
N = 2λc eθ σ̂ θ dr ≡ N eθ
r
1r2
M = 2λc ez r σ̂ θ dr ≡ M ez , (6.169)
r1
with N and M being the normal force and bending moment, respectively.
Note, however, that global equilibrium for the entire deformed beam de-
mands that N = 0. This observation provides a check on the accuracy of
any numerical solution.
7 Since bending moments are couples, the location of the reference point does not matter.
6.8.2 Solution
With (6.162), the incompressibility condition is
∂r
det F = λr λθ λz = kλr = 1,
∂X
which, upon integration, yields
r(X) = [(2/kλ)(X − A)]1/2 , (6.170)
a1 r22 − a2 r12
A = . (6.171)
r22 − r12
With these relations, Eqs. (6.162) provide the stretch ratios in terms of the
variables r1 , r2 , and λ.
To solve the governing equations, we first substitute Eqs. (6.164) into
(6.165) and integrate to get
r2
∂W ∂W ∂W dr
p(r) = λr + λθ − λr , (6.172)
∂λr r ∂λ θ ∂λr r
where the limits on the integral have been chosen to satisfy the boundary
condition σ̂ r (r2 ) = 0. With this expression for p, Eq. (6.164)1 and the
boundary condition σ̂ r (r1 ) = 0 yield
r2
∂W ∂W dr
λθ − λr = 0, (6.173)
r1 ∂λ θ ∂λr r
which is to be solved for r2 if r1 and λ are specified. With the stretch ratios
now known in terms of these variables, Eqs. (6.164) and (6.172) give the
stress components, and Eqs. (6.169) provide N (which should be zero) and
M.
Fig. 6.25. Beam model for looping of the embryonic heart. (a) Bending moment versus
curvature κ. The linear solution for an isotropic beam is shown for comparison. (b)
Bending stress distributions.
6.8.4 Results
In Fig. 6.25a the bending moment M is plotted as a function of the average
beam curvature
2
κ= .
r1 + r2
6.8. Bending of the Embryonic Heart 297
Also shown is the bending moment given by the linear theory for cylindrical
bending of an isotropic plate.8 For a plate, the bending moment is (Szilard,
1974)
EY Iκ
(M )lin = .
1 − ν2
EY κ(r − r0 )
(σ̂ θ )lin = ,
1 − ν2
where r0 = (a1 + a2 )/2 is the location of the middle surface of the plate.
For the isotropic beam, the linear and nonlinear solutions for M agree
until κ 1.5 mm−1 (Fig. 6.25a). For larger bending, however, the magni-
tude of M drops below the linear value.
As expected, including fibers increases M , with longitudinal fibers hav-
ing the greater effect. For this reason, Nakamura et al. (1980) speculated
that the circumferential fiber alignment observed in the heart tube facil-
itates looping by offering less resistance to bending than fibers of other
orientations.
Distributions of bending stress (σ̂ θ ) are shown for κ ≈ 5 (Fig. 6.25b).
The linear and nonlinear solutions for the isotropic beam are similar, but
note the shift in the location of the neutral axis, as defined by the point
where σ̂ θ = 0. In contrast, the transversely isotropic beams contain rela-
tively strong stress concentrations — at the inner curvature for circumfer-
ential fibers and at the outer curvature for longitudinal fibers.
The possible implications of these results for cardiac looping are not
clear. Studies have shown that mechanical stress plays a role in morpho-
genesis (Beloussov, 1998; 2015; Taber, 2020), but the relative simplicity of
this model must be kept in mind when interpreting these results. Never-
theless, this problem illustrates how models can suggest avenues for further
experimental investigation.
8 In cylindrical bending of a plate, the length of the plate in the z-direction constrains
the deformation so that λz ∼ = 1, consistent with our beam analysis. Thus, a plate solution
is more appropriate than a beam solution for comparison purposes.
298 Biomechanics Applications
6.9 Problems
6.1 A rectangular bar is stretched by forces applied only at its ends. The
bar is composed of an isotropic compressible material with the strain-
energy density function
−1/3
W = (μ/2)(I1 + 3I3 − 6),
where the Ii are strain invariants and μ is the shear modulus in the
small-strain limit. Determine the axial Cauchy stress as a function of
the stretch ratio in the axial direction.
6.2 Consider a rectangular bar composed of anisotropic incompressible ma-
terial containing fibers embedded in an isotropic matrix. With the
fibers aligned in both transverse directions (x and y), the strain-energy
density function is given by
W = C(I1 − 3) + 21 D1 (λ2x − 1)2 + 12 D2 (λ2y − 1)2 ,
where I1 is the first strain invariant, and C, D1 , and D2 are constants.
Suppose the bar is stretched in the axial (z) direction by the specified
stretch ratio λz . Taking C = D2 = 1, plot on one graph λx and λy
versus λz (0.5 ≤ λz ≤ 4) for D1 = 0.5, 1, and 2. Hint: The symmetry
condition λx = λy holds only for the case D1 = D2 .
6.3 Consider planar deformation of an incompressible rectangular mem-
brane with undeformed edges parallel to the Cartesian axes X1 and
X2 . The membrane is composed of isotropic Ogden material with
N
an b n
W = (Λ1 + Λb2n + Λb3n − 3),
n=1
b n
where an and bn are constants, and the Λi are stretch ratios in the
principal directions of strain (and stress), with Λ3 corresponding to the
transverse (thickness) direction.
(a) Compute the principal Cauchy stresses σ i in terms of Λ1 and Λ2 .
(b) Suppose the deformation is described by the mapping
x1 = λ1 X1 + kλ2 X2
x2 = λ2 X2
x3 = λ3 X3 ,
6.9. Problems 299
mapping
x1 = λ1 X1
1 3
x2 = 3 kX1 + λ2 X2
x3 = X3 ,
where Xi and xi are Cartesian coordinates, and λ1 , λ2 , and k are
positive constants.
(a) Sketch the deformed shape of the block in the x1 x2 -plane.
(b) Determine the stretch ratio for a line segment located at the center
of the cube before deformation that is parallel to the x1 -axis after
deformation.
(c) Take μ = 1, λ1 = 2, λ2 = 1.5, and k = 1. For the cases of transverse
isotropy (c = 2, d = 4) and isotropy (c = d = 0), compute the non-
zero normal and tangential Cauchy stresses acting on all surfaces,
as well as any body forces, that are needed to hold the block in
equilibrium with the specified deformation. State the values of the
surface stresses on the left and right sides of the block, as well as
on the faces normal to the X3 -axis. Plot the stresses applied to the
upper surface and any body forces as functions of X1 .
6.6 Using relations presented in Sec. 6.4, derive Eqs. (6.86) and (6.87).
6.7 Ignoring residual stress, write a computer program to solve the problem
involving a cylindrical model for the passive left ventricle discussed
in Sec. 6.6. For a baseline model, use the material properties and
parameter values given in that section; for a0 and b0 , use the unloaded
values given for a and b, respectively.
(a) Write the steps of the solution procedure, including all necessary
equations. The equations can be written in tensor or matrix form,
with reference to equations in the text for specific details. Compute
the solution for internal pressures 0 ≤ P ≤ PED , where PED = 2.67
kPa (20 mm Hg) is end-diastolic pressure.
(b) Check your calculations against the results in Figs. 6.19 and 6.20a
for φ = 0◦ (no residual stress).
(c) Examine the effects of varying the distribution of fiber angle β,
defined by (6.124), as well as the fiber stiffness by changing the
material constant a2 in (6.138). Let r be the deformed radial co-
ordinate, σ̂ ff the fiber stress (the physical Cauchy stress in the
deformed fiber direction), ψ the twist, and V0 and V the cavity
volume before and after deformation, respectively. Plot σ̂ ff vs. r
at P = PED , ψ vs. P , and P vs. V /V0 for the following cases:
6.9. Problems 301
Rigid cylinder Z z
4 T
Elastic tube R r
a
b P
Undeformed Deformed
Fig. 6.26. Axial shear of a tube (Problem 6.8). Modified from Taber (2020).
6.14 As discussed in Sec. 6.7, the cells in the early sea urchin embryo create
a spherical fluid-filled shell called the blastula. Invagination (inward
bending) soon generates a circular dimple in the shell, beginning the
process of gastrulation, which creates the primitive gut (Gilbert, 2010).
Here, we examine some of the mechanics involved in this problem.
Neglecting internal fluid pressure, consider a model for half a blastula
consisting of an unloaded hemispherical shell with inner radius a0 and
outer radius b0 that is composed of isotropic, incompressible material.
To simulate invagination, the shell is everted, i.e., it is turned inside
out. (A fully everted hemispherical shell generally stays that way when
unloaded.) In spherical coordinates, assume the everted configuration
is described by the mapping (Fig. 6.27).
f
F
q
Q r
R
r = r(R)
θ = π−Θ
φ = Φ.
(a) Working with physical components, determine the deformation gra-
dient tensor F.
(b) Derive the relation
r = (A − R3 )1/3 ,
where A is a constant. Determine the radii of the inner and outer
surfaces of the shell in the everted configuration as functions of A,
and discuss any constraints on the value of A.
(c) Suppose the strain-energy density function is
c α(I1 −3)
W = e −1 ,
α
where c and α are constants. Derive an equation to solve for A.
306 Biomechanics Applications
R
4
x1 x2 x
R1 R2
undeformed deformed
309
310 Linear Theory of Elasticity
gives a separate equation for each j = 1, 2, 3. Note also that the indices
in an equation must “balance” in that all terms must contain the same
number of non-dummy indices. Thus, ai + bij cj = 0 is a valid equation, but
aik + bij cj = 0 is not since the first term contains an extra k. Moreover,
three or more of the same index in a single term is not allowed (unless
some of the indices are placed in parentheses). The following development
assumes an elementary knowledge of vector analysis.
x3
a x3
e3 x2
e3 e2
e2 Rotation
x2
e1 e1
x1
x1
9 The summation convention introduced here is modified slightly from that used in
Chapter 2. The difference is that here, because the focus is on Cartesian coordinates,
repetition of subscripts in a term is allowed.
A.1. Mathematical Preliminaries 311
ei · ej = δ ij
ei × ej = ijk ek , (A.1)
where
1 for i = j
δ ij = (A.2)
0 for i = j
⎧
⎨ +1 if (i, j, k) is an even permutation
ijk = −1 if (i, j, k) is an odd permutation
⎩ (A.3)
0 if two or more indices are equal
is the permutation symbol. For example, Eqs. (A.1)2 and (A.3) give
e2 × e1 = 21k ek = 211 e1 + 212e2 + 213e3 = −e3 .
The permutation symbol can be expressed directly in terms of the ei by
dotting both sides of Eq. (A.1)2 by el to get
el · (ei × ej ) = el · ( ijk ek ) = ijk el · ek = ijk δ lk ,
where Eqs. (A.1) have been used. For a given l, the only term on the
right-hand-side that survives the summation over k is the term for k = l,
i.e.,
The Kronecker delta and the permutation symbol are related through the
-δ identity
a · b = ai b i . (A.7)
Note that the dummy indices in b have been changed from i to j, since more
than two of any subscript in a term is meaningless. The cross product is
a × b = (ai ei ) × (bj ej ) = ai bj (ei × ej )
or
a × b = ai b j ijk ek (A.8)
ab = ai bj ei ej . (A.9)
The tensor product of two vectors is called a dyad, and a linear combination
of dyads, e.g., the right-hand-side of Eq. (A.9), is called a dyadic. A useful
A.1. Mathematical Preliminaries 313
T = Tij ei ej , (A.11)
where the Tij are the components of T with respect to the basis {ei ej }.
Higher order tensors can be written analogously. Like a vector or dyad, a
tensor is geometrically invariant; when the basis changes, the components
of T change, but T itself does not. (A vector is a tensor of the first order.)
The double-dot product of two tensors yields a scalar. Consider, for ex-
ample, the scalar product of A = Aij ei ej and B = Bij ei ej . By Eq. (A.10),
10 In this section, lower-case bold letters denote first-order tensors (vectors), and upper-
we have
A:B = (Aij ei ej ) : (Bkl ek el )
= Aij Bkl (ei · ek )(ej · el )
= Aij Bkl δ ik δ jl
and contraction yields
I = δ ij ei ej = ei ei (A.16)
x2
x2
a2
a
a2 x1
e2 a1
q e1
e2
q
e1 a1 x1
Fig. A.2. Components of a vector a relative to two Cartesian coordinate systems, xi and
x̄i , with base vectors ei and ēi.
which indicates that the Aij represent the direction cosines between the
two sets of coordinate axes. For the geometry of Fig. A.2,
ē1 = e1 cos θ + e2 sin θ
ē2 = −e1 sin θ + e2 cos θ
ē3 = e3 , (A.19)
316 Linear Theory of Elasticity
and so
⎡ ⎤
cos θ sin θ 0
[Aij ] = [ēi · ej ] = ⎣ − sin θ cos θ 0 ⎦ . (A.20)
0 0 1
With these relations, transforming vector and tensor components is
straightforward. First, however, we note a convenient method for extracting
the components of vectors and tensors. Dotting Eq. (A.6) with ej gives
a · ej = (ai ei ) · ej = ai δ ij = aj ,
while double-dotting Eq. (A.11) with ek el yields
T : (ek el ) = (Tij ei ej ) : (ek el )
= Tij (ei · ek )(ej · el )
= Tij δ ik δ jl = Tkl .
where Eq. (A.10) has been used. Thus, we have
ai = a · e i
Tij = T : ei ej = ei · T · ej . (A.21)
āi = Aij aj
T̄ij = Aik Ajl Tkl . (A.24)
of the vector on the coordinate axes (Fig. A.2). A physical interpretation for the com-
ponents of a tensor cannot be visualized as easily.
A.1. Mathematical Preliminaries 317
∂
∇ ≡ ei . (A.25)
∂xi
∇φ = ei φ,i (A.26)
∇a = aj ,i ei ej (A.27)
∇ · a = ai , i . (A.28)
∇ × a = ei × (aj ej ),i = aj ,i ei × ej ,
and using Eq. (A.1)2 gives
∇ × a = aj , i ijk ek . (A.29)
Gradient Theorem:
∇φ dV = nφ dA or φ,i dV = ni φ dA (A.30)
V A V A
Divergence Theorem:
∇ · a dV = n · a dA or ai ,i dV = ni ai dA (A.31)
V A V A
Curl Theorem:
∇ × a dV = n × a dA or aj , i ijk dV = n i aj ijk dA
V A V A
(A.32)
Similar relations between surface and line integrals can be obtained by
replacing V by A and A by the contour C (length element ds) in these
equations.
A.2.1 Strain
Consider a body that deforms from the undeformed (reference) configura-
tion B into the deformed configuration b (Fig. A.3a). An arbitrary point
P in the body B is located by the position vector
R = x i ei , (A.33)
where the xi are the Cartesian coordinates of P . In the body b, the position
vector to p, the deformed image of P , is
r = R + u, (A.34)
A.2. Analysis of Deformation 319
Fig. A.3. (a) Geometry of deformation. (b) Differential position and displacement
vectors.
Thus,
du = dR · (∇u) = (∇u)T · dR, (A.37)
where Eq. (A.15) has been used. The dyad ∇u is called the displacement
gradient tensor. Now, inserting this expression into Eq. (A.36) gives,
since IT = I,
dr = dR · (I + ∇u)
= [I + (∇u)T ] · dR. (A.38)
This relation shows that ∇u characterizes the deformation of dR into dr.
Shear Strain. Thus far, we have shown that the extensional strain for
a line element oriented originally in the direction of N can be computed
from E. Since N is arbitrary, the extensional strain in any direction at a
point in the body can be computed from this strain tensor. A complete
description of deformation, however, also requires the shear strains, which
correspond to angle (shape) changes. Next, we show that E also contains
this information.
Consider two orthogonal line elements at P described by (Fig. A.4)
dR1 = N1 dS1 , dR2 = N2 dS2 , (A.43)
where dS1 and dS2 are the undeformed lengths of the elements oriented in
the directions of the orthogonal unit vectors N1 and N2 in the undeformed
body. According to Eq. (A.38), these elements deform into
dr1 = n1 ds1 = dR1 · (I + ∇u) = [I + (∇u)T ] · dR1
dr2 = n2 ds2 = dR2 · (I + ∇u) = [I + (∇u)T ] · dR2 (A.44)
in the directions of the unit vectors n1 and n2 , which are not orthogonal
in general (Fig. A.4).
N2
B
dR2 b n2
P p
dR1 N1 a dr2
x3 dr1
R
n1
e3 r
e2
x2
e1
x1
Fig. A.4. Geometry for two originally orthogonal line elements deforming into elements
with an enclosed angle α.
cos α = 2N1 · E · N2
1/2 1/2 , (A.45)
1 + 2E(N1 N1 ) 1 + 2E(N2N2 )
Note that Eij = Eji, and so the strain tensor is symmetric, i.e., only six of
the nine strain components are independent.
Given the Eij at a point, the deformation can be computed for elements
of any orientation. For example, the extensional strain for a fiber passing
through a point that initially is parallel to the x1 -axis is obtained by setting
N = e1 in Eq. (A.41) to get E(e1 e1 ) = e1 · E · e1 = E11 , and Eq. (A.47)
yields
2 2 2
∂u1 1 ∂u1 ∂u2 ∂u3
E11 = + + + .
∂x1 2 ∂x1 ∂x1 ∂x1
In addition, the deformed angle between fibers that are parallel to the x1
and x2 axes in the undeformed body is obtained by setting N1 = e1 and
N2 = e2 in Eq. (A.45). This procedure gives
2E12
cos α = 1/2
,
(1 + 2E11 ) (1 + 2E22 )1/2
where Eq. (A.47) provides
1 ∂u1 ∂u2 ∂u1 ∂u1 ∂u2 ∂u2 ∂u3 ∂u3
E12 = + + + + .
2 ∂x2 ∂x1 ∂x1 ∂x2 ∂x1 ∂x2 ∂x1 ∂x2
Similar relations can be obtained for the other strain components.
Example A.2 Consider the set of coordinate axes x̄i , which are obtained
by rotating the Cartesian axes xi by an angle θ about the x3 -axis (Fig. A.2,
page 315). Write the strain components Ēij relative to the x̄i system in
terms of the components Eij relative to the xi system.
Solution. One way to do this is to use Eq. (A.24)2 to write the tensor
transformation
with the Aij provided by (A.20). For illustration, rather than simply sum
over the repeated indices, we write Eq. (A.48) in matrix form. This requires
324 Linear Theory of Elasticity
Linear Strain Tensor. Equations (A.41), (A.42), and (A.45) are exact
expressions, valid for arbitrarily large deformation. If the strains (displace-
ment gradients) are small, then the nonlinear term of Eq. (A.42), i.e., the
last term in brackets, can be neglected to obtain the linear strain tensor
∗
Moreover, if we define the linear shear strain as E(N1 N2 )
= 21 (π/2−α) << 1,
then cos α = cos π/2 − 2E ∗ = sin 2E ∗ ∼ 2E ∗
= . Thus, for
(N1N2 ) (N1 N2 ) (N1 N2 )
∗ ∗
E(N1 N1 )
<< 1 and E(N2 N2 )
<< 1, Eqs. (A.41) and (A.45) yield
∗
E(NN) = N · E∗ · N
∗
E(N1 N2 )
= N1 · E∗ · N2 . (A.55)
Comparison with Eq. (A.21)2 reveals that the linear extensional (normal)
and shear strains are components of the linear strain tensor. (The shear no
longer depends on the normal strains.) Thus, in general we can write
∗
Ēij = E∗ : ēi ēj = ēi · E∗ · ēj , (A.56)
where the ēi are an orthogonal set of unit vectors of arbitrary orientation.
∗ ∗
If i = j, the Ēij are normal strain components; if i = j, the Ēij are shear
strain components.
The scalar form of the linear strain-displacement relation in Carte-
sian coordinates is obtained by neglecting the nonlinear term uk ,i uk ,j in
Eq. (A.47) to get
∗
Eij = 12 (ui ,j +uj ,i ). (A.57)
Thus,
∂er
= er ,θ = −e1 sin θ + e2 cos θ = eθ
∂θ
∂eθ
= eθ ,θ = −e1 cos θ − e2 sin θ = −er , (A.59)
∂θ
and all derivatives of the unit vectors with respect to r and z vanish.
In cylindrical polar coordinates, the gradient operator has the form (see
Section 2.8.3)
∂ 1 ∂ ∂
∇ = er + eθ + ez , (A.60)
∂r r ∂θ ∂z
A.2. Analysis of Deformation 327
A.2.2 Rotation
As Eq. (A.38) shows, the displacement gradient tensor describes the map-
ping of dR into dr. Although translation does not affect this transforma-
tion, rotation changes the vector dR and so is contained in ∇u. To separate
the effects of deformation and rigid-body motion, we write
(∇u)T = E∗ + Θ∗ , (A.64)
328 Linear Theory of Elasticity
where
E∗ = 12 [(∇u)T + ∇u].
is the linear strain tensor of Eq. (A.54) and
is the linear rotation tensor (see below). Substituting Eq. (A.64) into
(A.37) yields
du = (E∗ + Θ∗ ) · dR. (A.66)
∗
Consider now rigid-body motion. In this case, E = 0 and Eq. (A.66)
gives
du = Θ∗ · dR = ω × dR, (A.67)
where ω is the rotation vector. The magnitude of ω gives the rotation
angle, while its direction gives the rotation axis. The right-hand-side of
this expression follows from the geometry of Fig. A.3b (page 319), with dR
rotating through an angle |ω| as its tip moves a distance |du| = |ω| |dR|.
(Since ω is orthogonal to dR, |ω × dR| = |ω| |dR| sin π2 .) In terms of
components, Eq. (A.67) can be written
(Θ∗ij ei ej ) · (dxk ek ) = (ω k ek ) × (dxj ej )
or, with Eq. (A.1)2 ,
Θ∗ij dxk δ jk ei = Θ∗ij dxj ei = ωk dxj kji ei .
Thus,
which are the familiar rotation components from linear elasticity theory,
with their usual geometric interpretations. In a deformable body, Θ∗ rep-
resents the average rotation of all fibers passing through a given point.
A.2. Analysis of Deformation 329
ω = 12 ∇ × u. (A.70)
(E∗ − E ∗ I) · N = 0, (A.72)
Example A.4 Consider simple shear of a unit cube (Fig. A.6). During
deformation, each point of the cube moves parallel to the x1 -axis with the
displacement vector given by
u = kx2 e1 , (A.73)
(a) The Cartesian components of the linear strain tensor E∗ and rotation
tensor Θ∗ .
(b) The components of the nonlinear (Lagrangian) strain tensor E.
(c) The components of E relative to axes x̄i that are rotated through an
angle θ = 45◦ about the x3 -axis.
(d) The principal strains and principal directions for E∗ .
∂ ∂ ∂
∇u = e1 + e2 + e3 (kx2 e1 )
∂x1 ∂x2 ∂x3
⎡ ⎤
0 0 0
= k e2 e 1 = ⎣ k 0 0 ⎦
0 0 0
and so
⎡ ⎤
0 k 0
(∇u)T = k e1 e2 = ⎣ 0 0 0 ⎦.
0 0 0
In terms of these quantities, Eqs. (A.54) and (A.65) provide the linear strain
A.2. Analysis of Deformation 331
1 1 ∂ ∂ ∂
ω = 2∇ ×u= e1 + e2 + e3 × (kx2 e1 )
2 ∂x1 ∂x2 ∂x3
1 1
= 2 k e2 × e1 = − 2 k e3 = ω 3 e3 .
= 1
2
[k (e1 e2 + e2 e1 ) + k 2 e2 e2 ]
⎡ 1 ⎤
0 2
k 0
= ⎣ 1k 1 2
0 ⎦.
2 2k (A.75)
0 0 0
332 Linear Theory of Elasticity
x2
x2
e2 x1
e2
e1
q
e1 x1
(c) For θ = 45◦ , the unit vectors along the x̄i axes in Fig. A.7 are
ē1 = √12 (e1 + e2 ) = √12 [1, 1, 0]T
ē2 = √1 (−e1 + e2 ) = √1 [−1, 1, 0]T
2 2
ē3 = e3 = [0, 0, 1]T (A.76)
in terms of the nonrotated unit vectors ei . Thus, Eqs. (A.21) and (A.75)
give
Ē11 = ē1 · E · ē1
⎡ ⎤ ⎡ ⎤
0 k 0 1
= √1 [1, 1, 0] · 1 ⎣ k k2 0 ⎦· √1 ⎣ 1 ⎦
2 2 2
0 0 0 0
= 1
2k + 14 k 2
Ē22 = ē2 · E · ē2 = − 12 k + 14 k 2
Ē12 = ē1 · E · ē2 = 14 k 2 (A.77)
and so on.
(d) To compute the principal strains and principal directions for the
linear case, we use (A.74)1 and write Eq. (A.72) in matrix form
⎡ 1 ⎤⎡ ⎤ ⎡ ⎤
−E ∗ 2
k 0 N1 0
⎣ 1 k −E ∗ 0 ⎦ ⎣ N2 ⎦ = ⎣ 0 ⎦ . (A.78)
2
0 0 −E ∗ N3 0
A.2. Analysis of Deformation 333
Taking the determinant of the 3×3 matrix gives the characteristic equation
E ∗ (E ∗2 − 14 k 2 ) = 0,
which provides the principal strains
E1∗ = 12 k, E2∗ = − 12 k, E3∗ = 0. (A.79)
The corresponding principal directions, obtained by substituting these
equations into (A.78), are
N1 = √1 [1, 1, 0]T , N2 = √1 [−1, 1, 0]T , N3 = [0, 0, 1]T . (A.80)
2 2
Note that these directions are the same as those used in part (c) to trans-
form the nonlinear strain components [see Eq. (A.76)], i.e., parallel to the
diagonals of the undeformed cube. When linearized for k << 1, Eqs. (A.77)
agree with the principal strains of (A.79), as they should, including the van-
ishing shear strain (relative to the magnitudes of E11 and E22 ). But since
the nonlinear shear strain is not zero, the principal directions for the linear
and nonlinear problems are not the same.
∇ × (∇ × E∗ )T = 0, (A.84)
Due to the symmetry of the strain tensor, only six of these equations are in-
dependent. Moreover, using Eq. (A.3), we can show that the compatibility
conditions can be written in the alternate form
∗ ∗
Eij ,kl + Ekl ,ij − Elj∗ ,ki − Eki
∗
,lj = 0. (A.86)
A.3. Analysis of Stress 335
that B1 exerts on B2 and vice versa (Fig. A.8). If we examine the force
system on B1 alone, then the forces exerted by B2 are represented by trac-
tions T(N) acting on the cutting plane (Fig. A.8b). An infinite number
of cutting planes can pass through any given point, each with a different
traction vector. The set of tractions across all planes passing through a
point in a body defines the state of stress at that point.
e3
T(N)
f N
T-1
dV dh
T-2 e2
T-3
e1
where a is the acceleration and ρ is the mass density. The surface of the
element adjacent to the −ei surface has a unit normal ei and traction Ti ,
and by Newton’s third law of action-reaction, T−i = −Ti . Thus, Eq. (A.90)
becomes
T(N) dA − Ti dAi + f dV = ρa dV, (A.91)
in which the summation convention applies. The surface areas and volume
of the element are related by the geometric relations
dAi = dA cos(N, ei ) = dA (N · ei )
1
dV = 3 dh dA, (A.92)
where dh is the distance from the origin to the N face (Fig. A.9). Substi-
tuting Eqs. (A.92) into (A.91) gives
T(N) dA − Ti dA (N · ei ) + f ( 13 dh dA) = ρ a( 13 dh dA).
Dividing through by dA and letting dh → 0 yields
T(N) = (N · ei ) Ti , (A.93)
338 Linear Theory of Elasticity
i.e., as the volume of the element shrinks to zero, the body and inertia
forces disappear. Equation (A.93) reveals that if the traction is known on
three mutually orthogonal planes passing through a point, then it can be
computed on any arbitrarily oriented plane. Putting N = Nj ej into this
equation gives
T(N) = Ni Ti . (A.94)
T(N) = N · σ, (A.96)
where
σ = σ ij ei ej (A.97)
Here, the σ̄ ij are the components of σ with respect to any orthogonal basis
{ēi ēj }. The nine components of the stress tensor relative to any basis
completely define the state of stress at a point in a body.
T(N)
S V(NN)
V(NS)
(σ − σI) · N = 0, (A.100)
which is an eigenvalue problem similar to Eq. (A.72). Here, the three eigen-
values σ i are the principal stresses and the corresponding eigenvectors Ni
correspond to the principal directions of stress. Since σ is symmetric,
the eigenvalues are real and the eigenvectors are mutually orthogonal.
It is important to note that the principal axes of stress and strain coin-
cide in isotropic solids. This is not always the case for anisotropic materials,
however, as normal stresses may produce shear strains [see Eq. (A.144) be-
low]. In this case, an element oriented in the principal directions of stress
would undergo shear deformation.
and then deduce the local equations of motion for an infinitesimal element.
Each element of a body must satisfy these equations.
Consider first a particle of mass m in motion relative to a Newtonian
frame of reference. The displacement of the mass at time t is u(R, t) relative
to its initial position R in the reference frame. Then, the velocity and
acceleration of m are, respectively,
du
v(R, t) = = u̇
dt
dv
a(R, t) = = v̇, (A.101)
dt
where dot denotes differentiation with respect to time. The law of conser-
vation of linear momentum for the particle is
d
F= (mv), (A.102)
dt
where F is the force acting on the particle and mv is the linear momentum.
Because m is constant (ignoring relativistic effects), this relation becomes
F = ma, i.e., Newton’s second law.
Now, consider a body of mass density ρ(R, t) subjected to a traction
T(N) (R, t) acting over the surface area A and a body force f (R, t) act-
ing over the volume V (Fig. A.11). Here again, we neglect the difference
f
N
T(N)
V
dA
R
dV
∇ · σ + f = ρü. (A.105)
If a given problem is static or if inertia effects are small, i.e., the prob-
lem is quasi-static, then we can set ü = 0 and this relation becomes an
equilibrium equation.
The component form of the equation of motion can be found by substi-
tuting σ = σ ij ei ej , f = fi ei , and u = ui ei in Eq. (A.105). Then, the first
term can be written
∂
∇·σ = ek · (σ ij ei ej )
∂xk
= σ ij ,k δ ki ej = σ ij ,i ej = σ ji,j ei ,
A.3. Analysis of Stress 343
Example A.6 Suppose the stresses in the wedge of Fig. A.12 are given by
⎡ ⎤
ax21 2cx1 x2 0
σ = ⎣ 2cx1 x2 bx22 0 ⎦ (A.111)
0 0 0
relative to the Cartesian coordinates (x1 , x2 , x3 ). Compute the body forces
and surface tractions required to hold the wedge in static equilibrium with
the stipulated stress distribution.
Solution. First, consider local equilibrium. With üi = 0, Eqs. (A.106)
give the equilibrium equations
∂σ 11 ∂σ 21 ∂σ 31
+ + + f1 = 0
∂x1 ∂x2 ∂x3
∂σ 12 ∂σ 22 ∂σ 32
+ + + f2 = 0
∂x1 ∂x2 ∂x3
∂σ 13 ∂σ 23 ∂σ 33
+ + + f3 = 0, (A.112)
∂x1 ∂x2 ∂x3
and substituting (A.111) yields
f1 = −2x1 (a + c)
f2 = −2x2 (b + c)
f3 = 0. (A.113)
A.3. Analysis of Stress 345
x2
x1 + x2 = 1
x1
x3
Example A.7 Using dyadic analysis and Eq. (A.105), derive the scalar
equations of motion in cylindrical polar coordinates.
Solution. In terms of the unit vectors (er , eθ , ez ) in the (r, θ, z) direc-
tions (see Fig. A.5, page 326), the stress tensor, body force vector, and
displacement vector can be written
σ = σ rr er er + σ rθ er eθ + σ rz er ez
+σ θr eθ er + σ θθ eθ eθ + σ θz eθ ez
+σ zr ez er + σ zθ ez eθ + σ zz ez ez
f = fr er + fθ eθ + fz ez
u = u r er + u θ eθ + u z ez . (A.116)
In addition, Eqs. (A.58) give the derivatives
er ,θ = eθ , eθ ,θ = −er
(A.117)
er ,r = er ,z = eθ ,r = eθ ,z = ez ,r = ez ,θ = ez ,z = 0.
With ∇ given by Eq. (A.60), we have
∂ 1 ∂ ∂
∇·σ = er + eθ + ez · (σ rr er er + σ rθ er eθ + σ rz er ez
∂r r ∂θ ∂z
+σ θr eθ er + σ θθ eθ eθ + σ θz eθ ez + σ zr ez er + σ zθ ez eθ + σ zz ez ez )
∂σ rr ∂σ rθ ∂σ rz
= er · er er + er eθ + er ez + · · ·
∂r ∂r ∂r
eθ
+ · (σ rr er ,θ er + σ rθ er ,θ eθ + σ rz er ,θ ez
r
∂σ θr ∂σ θθ ∂σ θz
+ eθ er + eθ eθ + eθ ez
∂θ ∂θ ∂θ
+σ θr eθ er ,θ +σ θθ eθ eθ ,θ + · · · )
∂σ zr ∂σ zθ ∂σ zz
+ez · ez er + ez eθ + ez ez + · · · , (A.118)
∂z ∂z ∂z
A.4. Constitutive Relations 347
where the dots represent terms that vanish after taking derivatives and the
dot product. To complete the analysis, we simplify this expression using
Eqs. (A.117) and then insert the result and (A.116) into (A.105) to get a
relation of the form
Fr er + Fθ eθ + Fz ez = 0.
All the equations derived thus far are independent of material properties.
Elastic, viscoelastic, and plastic bodies obey the same geometric relations
and equations of motion. Stress and strain, however, are related through
the constitutive relations, which are material-dependent. Although these
equations must be determined experimentally, thermodynamic considera-
tions restrict their form.
K̇ + U̇ = P + Q, (A.120)
in which u is the internal energy per unit mass. The power input is the rate
of work done on the body. For applied surface tractions T(N) and body
forces f ,
P = T(N) · u̇ dA + f · u̇ dV, (A.123)
A V
where A is the surface area. Finally, the rate of heat added to the body is
Q=− q · N dA + ρr dV, (A.124)
A V
where q is the outward-directed heat flux vector (per unit area), r is the
rate of heat production (per unit mass) due to internal sources, and N is
the (outward-directed) unit normal to the surface.
Next, we go through a series of manipulations to express the terms of
Eq. (A.120) in alternate forms. First, Eq. (A.121) gives
d d
K̇ = 12 (ρ dV )u̇ · u̇ + (u̇ · u̇) ρ dV
dt dt
V
= 12 (u̇ · ü + ü · u̇) ρ dV = u̇ · ü ρ dV
V V
= u̇ · (∇ · σ + f ) dV, (A.125)
V
in which the second line follows from the constancy of the mass ρ dV of
an element, and substituting the equation of motion (A.105) produces the
third line. It often is convenient to develop tensor equations in component
form and then convert the result back to direct notation, especially when
the gradient operator is involved. With this procedure, the first term in
the integrand of Eq. (A.125) becomes
∂
u̇ · (∇ · σ) = u̇ · ek · σ ij ei ej
∂xk
= u̇ · (σ ij ,k δ ki ej ) = u̇ · (σ ij ,i ej )
= (u̇k ek ) · (σ ij ,i ej ) = u̇k σ ij ,i δ kj
= u̇j σ ij ,i
= (u̇j σ ij ),i −u̇j ,i σ ij . (A.126)
A.4. Constitutive Relations 349
where conservation of mass again has been used. Adding Eqs. (A.129) and
(A.130) provides a volume integral for the left-hand-side of (A.120).
The surface integrals of Eqs. (A.123) and (A.124) now are converted
into volume integrals. Substituting Eq. (A.96) into (A.123) and applying
the divergence theorem (A.31) yields
P = N · σ · u̇ dA + f · u̇ dV
A V
∗
by virtue of the chain rule and Eq. (A.12). Thus, for arbitrary Ėij , we have
∂W
σ ij = ∗ , (A.138)
∂Eij
∗
which is the constitutive relation for an elastic material. Once W (Eij )
is determined through appropriate experiments, this equation provides the
stress-strain relations.
In the case of isothermal deformation, we introduce the Helmholtz free-
energy function
Ψ = u − T η. (A.139)
1 ∗ ∗
W = 2 Cijkl Eij Ekl, (A.142)
their number can be reduced using the symmetry property of the strain ten-
sor. For example, since Eq. (A.142) can be written W = 21 CjiklEji ∗ ∗
Ekl =
1 ∗ ∗
2 C jiklE E
ij kl , then C ijkl = C jikl , and a similar argument shows that
Cijkl = Cijlk , reducing the number of independent constants to 36. Fur-
thermore, we can write W = 12 Cklij Ekl ∗ ∗
Eij , and so Cijkl = Cklij , i.e., the
matrix of 36 constants is symmetric. This observation reduces the number
of constants to 21. Thus, the mechanical properties of a general anisotropic,
linear elastic material are characterized by 21 independent elastic constants
with
Cijkl = Cjikl = Cijlk = Cklij . (A.143)
When material symmetry is present, the number of independent elastic
constants can be reduced still further (see Section 5.6). An orthotropic
material, characterized by three mutually orthogonal planes of symmetry,
has nine constants. A transversely isotropic material, which has an axis
of symmetry, has five independent constants. And an isotropic material,
with the same properties in all directions, has two constants. Note that,
for certain anisotropic materials, extensional and shear effects are coupled.
To obtain the stress-strain relations, we substitute Eq. (A.142) into
(A.138) and use (A.143) to obtain
∂ 1 ∗ ∗
σ ij = ∗ 2 Cklmn Ekl Emn
∂Eij
# $
∗ ∗
1 ∂Ekl ∗ ∗ ∂Emn
= 2 Cklmn ∗ Emn + Ekl ∂E ∗
∂Eij ij
1 ∗ ∗
= 2 Cklmn (δ ki δ lj Emn + Ekl δ mi δ nj )
1 ∗ ∗
= 2 (Cijmn Emn + Cklij Ekl )
1 ∗ ∗
= 2 (CijklEkl + CijklEkl )
or
∗
σ ij = CijklEkl , (A.144)
The governing equations, along with appropriate boundary and initial con-
ditions, define a boundary value problem in the theory of elasticity. As
derived in this chapter, the basic equations of linear isotropic elasticity
consist of the following:
Strain-Displacement Relations:
Equations of Motion:
Constitutive Relations:
σ = λ(tr E∗ ) I + 2μE∗ or ∗
σ ij = λEkk ∗
δ ij + 2μEij (A.150)
With Eq. (A.96) and the unit surface normal N = Ni ei , these conditions
can be written in the following forms:
N · σ = T̂(N) or Nj σ ij = T̂i on Aσ
u = û or ui = ûi on Au (A.151)
Here, Aσ and Au represent the portions of the surface where tractions and
displacements, respectively, are specified. In linear elasticity, the unique-
ness theorem ensures that the solution to any particular boundary value
problem is unique (Timoshenko and Goodier, 1969).
z
(a) (b)
M ey
eq
er
y uq v
a
q ur
N q u
L r
q
x ex
x
M
Fig. A.13. Torsion of a solid circular cylinder. (a) Three-dimensional view. (b) Cross
section.
u = u r er + u θ eθ + u z ez
= uex + vey + wez (A.153)
uθ = βzr, ur = uz = 0 (A.154)
relative to the cylindrical polar coordinates (r, θ, z). The variable β can
be identified as the angle of twist per unit length, which is constant for a
uniform cylinder. The geometry linking the polar and Cartesian coordinates
gives the relations (Fig. A.13b)
x = r cos θ, y = r sin θ
and
er = ex cos θ + ey sin θ
eθ = −ex sin θ + ey cos θ. (A.155)
Combining these expressions with Eqs. (A.153) and (A.154) yields the
Cartesian displacement components
If the cross section of a twisted cylinder is not circular, then our prior
symmetry arguments are not valid. In this case, the problem is not ax-
isymmetric, and warping of cross sections is possible. Here, we modify the
solution of the previous example to allow a more general deformation.
Consider a solid cylinder of length L with a constant cross section that
has a general contour C (Fig. A.14). Modifying Eqs. (A.156), we assume
that the displacement components have the form
z
(a) (b)
M
C
y
N
C
L
x
x
M
Fig. A.14. Torsion of a noncircular cylinder. (a) Three-dimensional view. (b) Cross
section.
∗ ∗ 1 ∂u ∂w ∂ψ
Exz = Ezx = + = 12 β −y
2 ∂z ∂x ∂x
∗ ∗ 1 ∂v ∂w ∂ψ
Eyz = Ezy = + = 12 β +x
2 ∂z ∂y ∂y
∗ ∗ ∗ ∗ ∗
Exx = Eyy = Ezz = Exy = Eyx = 0, (A.166)
A.5. Boundary Value Problems 359
∂2ψ ∂2ψ
+ = ∇2 ψ = 0, (A.168)
∂x2 ∂y2
M = μJ ∗ β, (A.174)
where
∗ ∂ψ ∂ψ
J =J+ x −y dA (A.175)
A ∂y ∂x
This appendix lists basic relations for some special orthogonal curvilinear
coordinate systems. Coordinates and base vectors are expressed in terms of
the Cartesian system (x, y, z). In addition, physical stress and deformation
components are used.
(x1 , x2 , x3 ) = (r, θ, z)
g1 = ex cos θ + ey sin θ = er
g2 = r(−ex sin θ + ey cos θ) = reθ
g3 = ez
Γ122 = −r
Γ212 = Γ221 = r −1
Other Γkij = 0
363
364 Special Coordinate Systems
z
y
r
q
x
Differential operators:
∂ 1 ∂ ∂
∇ = er + eθ + ez
∂r r ∂θ ∂z
∂2 1 ∂ 1 ∂2 ∂2
∇2 = 2
+ + 2 2+ 2
∂r r ∂r r ∂θ ∂z
Equations of motion:
∂ σ̂ rr 1 ∂ σ̂ θr ∂ σ̂ zr 1
+ + + (σ̂ rr − σ̂ θθ ) + fˆr = ρâr
∂r r ∂θ ∂z r
∂ σ̂ rθ 1 ∂ σ̂ θθ ∂ σ̂ zθ 2σ̂ rθ
+ + + + fˆθ = ρâθ
∂r r ∂θ ∂z r
∂ σ̂ rz 1 ∂ σ̂ θz ∂ σ̂ zz σ̂ rz
+ + + + fˆz = ρâz
∂r r ∂θ ∂z r
Deformation gradient:
⎡ ⎤
∂r 1 ∂r ∂r
⎢ ∂R R ∂Θ ∂Z ⎥
⎢ ⎥
⎢ ⎥
⎢ ∂θ r ∂θ ∂θ ⎥
F(ei ej ) = ⎢ r r ⎥ (i, j = r, θ, z)
⎢ ∂R R ∂Θ ∂Z ⎥
⎢ ⎥
⎣ ∂z 1 ∂z ∂z ⎦
∂R R ∂Θ ∂Z
B.2. Spherical Polar Coordinates 365
r
y
f
x
x = r sin θ cos φ
y = r sin θ sin φ
z = r cos θ
Differential operators:
∂ 1 ∂ 1 ∂
∇ = er + eθ + eφ
∂r r ∂θ r sin θ ∂φ
1 ∂ ∂ 1 ∂ ∂ 1 ∂2
∇2 = r2 + 2 sin θ + 2 2
r 2 ∂r ∂r r sin θ ∂θ ∂θ r sin θ ∂φ2
Equations of motion:
∂ σ̂ rr 1 ∂ σ̂ θr 1 ∂ σ̂ φr
+ +
∂r r ∂θ r sin θ ∂φ
1 rr
+ 2σ̂ − σ̂ θθ − σ̂ φφ + σ̂ θr cot φ + fˆr = ρâr
r
∂ σ̂ rθ 1 ∂ σ̂ θθ 1 ∂ σ̂ φθ
+ +
∂r r ∂θ r sin θ ∂φ
1 rθ
+ 2σ̂ + σ̂ θr + (σ̂ θθ − σ̂ φφ ) cot θ + fˆθ = ρâθ
r
rφ θφ
∂ σ̂ 1 ∂ σ̂ 1 ∂ σ̂ φφ
+ +
∂r r ∂θ r sin θ ∂φ
1
+ (2σ̂ rφ + σ̂ φr + 2σ̂ θφ cot θ + fˆφ = ρâφ
r
Deformation gradient:
⎡ ⎤
∂r 1 ∂r 1 ∂r
⎢ ⎥
⎢ ∂R R ∂Θ R sin Θ ∂Φ ⎥
⎢ ⎥
⎢ ∂θ r ∂θ r ∂θ ⎥
F(ei ej ) = ⎢ r ⎥ (i, j = r, θ, φ)
⎢ ∂R R ∂Θ R sin Θ ∂Φ ⎥
⎢ ⎥
⎣ ∂φ r sin θ ∂φ r sin θ ∂φ ⎦
r sin θ
∂R R ∂Θ R sin Θ ∂Θ
x = (b + r cos φ) cos θ
y = (b + r cos φ) sin θ
z = r sin φ
B.3. Toroidal Coordinates 367
y
b
q r
f
Answers and detailed solutions are provided below for most of the end-of-
chapter problems. In some cases, only parts of the solution are presented.
Chapter 2
369
370 Problem Solutions
and so
A = T · U = T ik U kj gi gj .
Computing A·j i using the second representation for A
ij
is left for
the reader’s enjoyment.
(c) Partial answer: A = Ai·j gi gj = T ik Ukj gi gj
2.7 Setting
a = gi , b = gj , c = gk
gives
a × (b × c) = gi × (gj × gk ) = gi × ( jkl g
l
)
ilm lmi
= jkl gm = ljk gm ,
in which Eqs. (2.103) and (2.104) have been used, and the indices
on the permutation symbols have undergone cyclic permutations. In
addition, with (2.14) we obtain
(a · c)b = (gi · gk )gj = δ ik gj
(a · b)c = (gi · gj )gk = δ ij gk .
After substitution, the given vector identity
a × (b × c) = (a · c)b − (a · b)c
becomes
ljk
lmi
gm = δ ik gj − δ ij gk .
Dotting both sides of this equation with gn yields
ljk
lmi n
δm = δ ik δ nj − δ ij δ nk .
Eliminating δ nm on the left-hand side by contraction, then replacing i
by r, n by s, and l by i gives
ijk
isr
= δ rk δ sj − δ rj δ sk .
isr
Anticyclic permutation of the indices in changes the sign to obtain
the final result
ijk
irs
= δ rj δ sk − δ rk δ sj .
2.8 The answers are provided by Eqs. (2.54) and (2.57).
Problem Solutions 373
2.9 (a) Partial answer: A13 = Arφ = −r sin θ sin φ, A31 = Aφr = cos θ.
(c) For spherical polar coordinates, Appendix B gives
grr = 1, gθθ = r 2 , gφφ = r 2 sin2 θ
with gij = 0 for i = j. Equation (2.132) yields the physical com-
ponents (1, 2, 3 = r, θ, φ)
⎡ ⎤
T·11 T·21 r −2 T·31 (r sin θ)−2
g (ii)
T̂·ji = T·ji =⎣ T·12 r 2 T·22 T·32 (sin θ)−2 ⎦ .
g(jj) 3 2 2 3 2
T·1 r sin θ T·2 sin θ T·33
To write these components in terms of T ij , we use the relation
T·ji = T ik gkj = T i1 g1j + T i2 g2j + T i3 g3j
to obtain ⎡ ⎤
T 11 T 12 T 13
T̂·ji = ⎣ T 21 r 2 T 22 r 2 T 23 r 2 ⎦.
T 31 r 2 sin θ T 32 r 2 sin θ T 33 r 2 sin2 θ
2 2
g x = g x = ex , gy = g y = ey , gz = g z = ez .
With the position vector being
r = xex + yey + zez ,
the covariant base vectors in the parabolic coordinate system are
ḡi = ∂r/∂ x̄i , i.e.,
∂r
ḡu = = uex + vey
∂u
∂r
ḡv = = −vex + uey
∂v
∂r
ḡz = = ez .
∂z
These equations yield
√
ḡ = ḡu · (ḡv × ḡz ) = u2 + v2
and the contravariant base vectors
ḡv × ḡz uex + vey
ḡu = √ =
ḡ u2 + v 2
ḡz × ḡu −vex + uey
ḡv = √ =
ḡ u2 + v 2
ḡu × ḡv
ḡz = √ = ez .
ḡ
374 Problem Solutions
(c) For the parabolic coordinates, the components of the metric tensor
are
⎡ 2 ⎤
u + v2 0 0
[ḡij ] = [ḡi · ḡj ] = ⎣ 0 u2 + v 2 0 ⎦
0 0 1
⎡ ⎤
(u2 + v2 )−1 0 0
[ḡij ] = [ḡi · ḡj ] = ⎣ 0 (u2 + v2 )−1 0 ⎦ .
0 0 1
As discussed in the footnote on page 25, the upper and lower in-
dices of Aij represent the row and column, respectively, while the
opposite is true for Bji , giving
⎡ ⎤
i ḡu · gx ḡv · gx ḡz · gx
Aj = ⎣ ḡu · gy ḡv · gy ḡz · gy ⎦
ḡu · gz ḡv · gz ḡz · gz
⎡ ⎤
u −v 0
= ⎣ v u 0 ⎦
0 0 1
⎡ u ⎤
i ḡ · gx ḡv · gx ḡz ·gx
Bj = ⎣ ḡu · gy ḡv · gy ḡz · gy ⎦
ḡu · gz ḡv · gz ḡz · gz
⎡ ⎤
u −v 0
1 ⎣ v u ⎦.
= 0
u2 + v 2 2 2
0 0 u +v
and
g2 × g3 1
g1 = √ = − (2ex + 5ey + 2ez )
g 6
g3 × g1 1
g2 = √ = (2ex − ey − 4ez )
g 6
g1 × g2 1
g3 = √ = (ex + ey + ez ).
g 3
The components of a are
In dyadic form, the first term on the right-hand side of the identity
is
∂A ∂
:B = gm gn Aij gi gj : Bkl gk gl
∂C ∂Cmn
∂Aij ∂Aij
= gm gn Bkl (gi · gk )(gj · gl ) = gm gn Bkl δ ki δ lj
∂Cmn ∂Cmn
∂Aij
= gm gn Bij .
∂Cmn
Exchanging A and and B in the above relation yields
∂B ∂B ∂Bij ij
A: = : A = gm gn A .
∂C ∂C ∂Cmn
Plugging these results into the identity demonstrates its validity.
2.15(a) For spherical coordinates, Appendix B gives the orthogonal base
vectors
gr = ex sin θ cos φ + ey sin θ sin φ + ez cos θ = er
gθ = r(ex cos θ cos φ + ey cos θ sin φ − ez sin θ) = reθ
gφ = r sin θ(−ex sin φ + ey cos φ) = r sin θ eφ ,
where the ei are unit vectors. The non-zero values of gij = gi · gj
are
grr = 1, gθθ = r 2 , gφφ = (r sin θ)2 ,
which give
grr = 1, gθθ = r −2 , gφφ = (r sin θ)−2 .
The Christoffel symbols Γijk are computed using the relation
2Γijk = gjk ,i +gki ,j −gij ,k .
For the present problem, the only non-zero derivatives of the gij are
gθθ ,r = 2r, gφφ ,r = 2r sin2 θ, gφφ ,θ = 2r 2 sin θ cos θ,
giving the non-zero Christoffel symbols
Γrθθ = r
Γθθr = −r
Γrφφ = Γφrφ = r sin2 θ
Γφφr = −r sin2 θ
Γφφθ = −r 2 sin θ cos θ
Γθφφ = Γφθφ = r 2 sin θ cos θ.
Problem Solutions 377
gr = ex cos θ + ey sin θ = er
gθ = r(−ex sin θ + ey cos θ) = reθ .
Problem Solutions 379
Chapter 3
G I = G I = gi = g i = eI = ei .
r = xi gi = xi ei = (X 1 + 2X 2 X 3 )e1 + (X 2 − 2X 1 X 3 )e2 + X 3 e3 .
g1∗ = r,1∗ = e1 − 2X 3 e2
g2∗ = r,2∗ = 2X 3 e1 + e2
g3∗ = r,3∗ = 2X 2 e1 − 2X 1 e2 + e3 ,
where r,1∗ = ∂r/∂X 1 , etc. Equations (2.17) and (2.18) give the
contravariant base vectors
∗ g2 ∗ × g3 ∗
g1 = √ ∗ = (g∗ )−1/2 [e1 − 2X 3 e2 − (2X 2 + 4X 1 X 3 )e3 ]
g
∗ g3 ∗ × g1 ∗
g2 = √ ∗ = (g∗ )−1/2 [2X 3 e1 + e2 + (2X 1 − 4X 2 X 3 )e3 ]
g
∗ g1 ∗ × g2 ∗
g3 = √ ∗ = e3 ,
g
where
√
g∗ = g1∗ · (g2∗ × g3∗ ) = 1 + 4(X 3 )2 .
Problem Solutions 383
Partial answer:
1 2
e13 = 2 (x2 − 3x3 )
e33 = − 12 [4x23 + (x2 − 3x23 )2 ]
3.6 (c) Substituting Eqs. (3.52) and (3.174) into (3.175)2 yields
⎡ ⎤ ⎡ ⎤
1 k 0 b − 12 kb 0
V = F · ΘT = ⎣ 0 1 0 ⎦ ⎣ 1 kb
2
b 0 ⎦
0 0 1 (e e ) 0 0 1 (e e )
i j i j
⎡ ⎤
b(1 + 12 k 2 ) 12 kb 0
= ⎣ 1
2 kb b 0 ⎦,
0 0 1 (e e )
i j
1 2 −1/2
where b = (1 + 4k ) .
(d) This calculation requires the strain tensors
k2 k
E = ey ey + (ex ey + ey ex )
2 2
k2 k
e = − ey ey + (ex ey + ey ex ).
2 2
3.7 Substituting u = uex + vey into Eqs. (3.178) and (3.179) yields
1
∗ u,x 2
(v,x +u,y )
E(ei ej ) = 1
2 (v,x +u,y ) v,y
0 − 12 (v,x −u,y )
Θ∗(ei ej ) = 1 .
2 (v,x −u,y ) 0
With these relations, Eq. (3.183) gives the approximate form for E.
Partial answer:
∗ ∗
Exx = Exx − Exy Θ∗xy + 21 Θ∗2
xy
Problem Solutions 385
3.9 In the undeformed body, unit vectors parallel to the given vectors are
2e1 + 2e2 + e3 e1 − 5e3
N1 = , N2 = √ .
3 26
Let Θ and θ represent the angles between the line segments before
and after deformation, respectively. With Eq. (3.108)2 , the cosines of
these angles are given by
cos Θ = N1 · N2
N1 · C · N2
cos θ = n1 · n2 = ,
Λ1 Λ2
where
C = I + 2E
Λi = (Ni · C · Ni )1/2 (i = 1, 2).
Computations using the given E yield
Δθ = θ − Θ = −11.7◦.
3.10 In terms of Eulerian quantities, Eq. (3.94)2 gives the stretch ratio
λ−2
(n) = n · B
−1
· n,
where
√
n = (ex + ey + ez )/ 3
B = F · FT .
To determine F, we use (3.78)2 , i.e.,
T
F−1 = I − (∇u)
¯ ,
where
∇¯ = ex ∂ + ey ∂ + e z ∂ .
∂x ∂y ∂z
For the given u, these relations yield
⎡ ⎤
(y/x)1/2 (x/y)1/2 0
1⎢ ⎥
F = ⎢
−1
0 y −1/2
0 ⎥.
2⎣ ⎦
(z/x)1/2 (x/z)1/2
(ei ej )
The plots should show peaks in EΘΘ coinciding with peaks in the
magnitude of shear strain.
3.14 The following table lists results for the rotations and stretch ratios:
a0 −e1 + e2 e1 e2 N1
◦ ◦ ◦
θ −23.2 0 −31.0 −16.7◦
(θ 0 )str −6.5◦ 16.7◦ −14.3◦ 0◦
(θ 0 )rot −16.7◦ −16.7◦ −16.7◦ −16.7◦
θrot −23.2◦ 0◦ −31.0◦ −16.7◦
λ 0.762 1 1.17 1.34
3.15 (a) For the given r, the matrix form of the deformation gradient
tensor is
⎡ ⎤
α1 β1 0
T
F = (∇r) = β 2⎣ α2 0 ⎦
0 0 α3 (e
i ej )
n1 = 0.583e1 − 0.812e2
n2 = −0.812e1 − 0.583e2
n 3 = e3 .
390 Problem Solutions
U = ΘT · F, V = F · ΘT .
−λ3 + I1 λ2 − I2 λ + I3 = 0.
I1 = tr C
2
I2 = 1
2 [(tr C) − tr C2 ]
I3 = det C.
A little matrix algebra shows that the above relations are con-
sistent with the usual characteristic equation of (3.133)1 given
by
3.16 (a) After inserting the given expressions for xi (Xi , t) into the position
vector r = xi ei , differentiating with respect to time yields
vL = ẋ1 = X1 X2 (1 + t)et
x1 x2 (1 + t)et
vE = .
1 + x2 tet
The equation for vE is obtained by inverting the equations for
xi (Xi , t) to get Xi (xi , t) and substituting the results into the
equation for vL .
Problem Solutions 391
¯ ∂ ∂ ∂
∇v = e1 + e2 + e3 vE (x1 , x2 , t)e1
∂x1 ∂x2 ∂x3
∂vE ∂vE
= e1 e1 + e2 e1 .
∂x1 ∂x2
From here, it is a relatively simple matter to compute D.
3.17 For r = xi ei , the given xi are
x1 = X1 + X2 t
x2 = X2 + X1 t2
x3 = X3 .
After these equations are inverted to obtain Xi (xi , t), the velocity
components vi = ẋi can be written in the Eulerian form vi (xi , t).
With this result, we get
2 2 2 2 3
dθ
=
∂θ ¯ = x1 + x2 − (3x1 + 5x2 )t + 2x1 x2 t(1 + 2t) .
+ v · ∇θ
dt ∂t 1 − t3
For the particle initially located at Xi = (2, −3, 1), these equations
give θ̇ = 58.0 at time t = 1.5. The mathematical details are left to
the reader.
3.18 Partial answer: xi = (0.256, 1.73, 4.0), v1 = 2.15, a2 = 2.48
3.19 For combined extension, inflation, and torsion of a circular tube, sub-
stituting Eq. (3.223) into (3.208) gives
LT ¯
= ∇v
∂ eθ ∂ ∂
= er + + ez (ṙer + r θ̇eθ + żez ).
∂r r ∂θ ∂z
392 Problem Solutions
er , θ = eθ , eθ ,θ = −er ,
differentiation yields
gr = er , gθ = reθ , gz = ez
g r = er , gθ = r −1 eθ , g z = ez ,
J˙ = JF−T : Ḟ.
Problem Solutions 393
Chapter 4
4.2 Answer:
a q a a
σ xx = 1− , σ xy = , σ zz =
p p p r
4.3 (a) For a plane parallel to the x3 -axis, the unit normal must have the
form
n =n1 e1 + n2 e2 ,
giving n · e3 = 0. For the given σ, the true traction vector is
T(n) = n · σ = an1 e1 − bn2 e2 + (dn1 − en2 )e3 .
If this vector is tangent to the plane, we must have
n · T(n) = an21 − bn22 = 0,
and, since n is a unit vector,
n · n = n21 + n22 = 1.
Solving these equations for n1 and n2 yields
b a
n1 = , n2 = .
a+b a+b
(b) For rotation θ about the x1 -axis, the rotated system x̄i has base
vectors
ē1 = e1
ē2 = e2 cos θ + e3 sin θ
ē3 = −e2 sin θ + e3 cos θ.
The relation σ̄ ij = ēi · σ · ēj gives
σ̄ 13 = d cos θ
σ̄ 23 = (b + c) sin θ cos θ + e(sin2 θ − cos2 θ).
4.4 The characteristic equation for the given stress matrix is
rr
σ̂ − λ 0 0
det(σ − λI) = 0 1−λ 2 = 0.
0 2 σ̂ − λ
φφ
⎡ ⎤
0.33 0 0
σ i·j = gi · σ · gj = ⎣ 0 1.11 −0.33/5 ⎦ .
0 −(0.33)(5) 0 (g j)
ig
4.8 (a) In matrix form, the second Piola-Kirchhoff stress tensor can be
written
⎡ ⎤
2X 1 −X 1 X 2 0
s = ⎣ −X 1 X 2 3X 1 + X 2 0 ⎦ .
0 0 0 (e e )
i j
where J = det F = 1.
(b) To extract the required stress components, we need the base vec-
tors GI in the undeformed configuration and the convected base
vectors gI and gI in the deformed configuration. The first two
are provided by the relations
∂R ∂r
GI = , gI = ,
∂X I ∂X I
and Eqs. (2.18) give the gI . The results are
G I = eI (I = 1, 2, 3)
g1 ∗ = e1
g2∗ = 0.2X 3 e1 + e2
g3∗ = 0.2X 2 e1 + 0.4X 3 e2 + e3
∗
g1 = e1 − 0.2X 3 e2 + [0.08(X 3)2 − 0.2X 2 ]e3
∗
g2 = e2 − 0.4X 3 e3
∗
g3 = e3 .
For X I = (−1, 2, 1), these relations yield the following compo-
nents for σ:
⎡ ⎤
−2 2 0
∗ ∗
σI J = gI · σ · gJ = ⎣ 2 −1 0 ⎦
0 0 0 (g g )
I J
398 Problem Solutions
⎡ ⎤
−1.6 1.8 0
∗
σ ·J
I∗ = gI · σ · gJ = ⎣ 1.7 −0.64 0 ⎦
0.16 0.32 0 (gI g
J)
⎡ ⎤
−1.6 1.8 0
∗
σ ·J
I = GI · σ · g J = ⎣ 2 −1 0 ⎦
0 0 0 (GI g
J)
g2
g2 V2.1
V21
g1
g1
V2.2
V22 V1.2
V 12
V11 V1.1
(c) At first glance, the above results appear to differ by the constant
(2 vs. 3) in the second term, but these equations are for two differ-
ent stress components (σ̂ θφ vs. σθφ ). For orthogonal coordinates,
Eq. (4.47) gives
√
σ̂ ij = σ ij g(ii)g(jj).
With the g(ii) provided in Appendix B, this relation becomes
σ̂ θφ σ̂ θφ
σ θφ = √ = 2 .
gθθ gφφ r sin θ
After substitution into the equilibrium equation from part (b),
differentiation yields
∂ σ̂ θφ
+ 2σ̂ θφ cot θ = 0,
∂θ
which agrees with the result from part (a).
4.14 (a) With the given position vectors
R = ReR + ZeZ
r = R(1 + b cos2 Θ)eR + ZeZ
and the gradient operator
∂ 1 ∂ ∂
∇ = eR + eΘ + eZ ,
∂R R ∂Θ ∂Z
T
the relation F = (∇r) gives deformation gradient tensor
F = (1 + b cos2 Θ)(eR eR + eΘ eΘ ) − 2b sin Θ cos ΘeR eΘ + eZ eZ .
In matrix form, this tensor can be written
⎡ ⎤
FRR FRΘ 0
F = ⎣ 0 FΘΘ 0 ⎦
0 0 FZZ (e e )
I J
⎡ 2
⎤
1 + b cos Θ −2b sin Θ cos Θ 0
= ⎣ 0 1 + b cos2 Θ 0 ⎦ .
0 0 1 (e
I eJ )
⎡ ∗ ∗ ∗ ∗ ∗ ∗ ⎤
σR R σR Θ σR Z
= ⎣ σΘ R σΘ Θ σΘ Z ⎦ .
∗ ∗ ∗ ∗ ∗ ∗
∗ ∗ ∗ ∗ ∗ ∗
σ Z R σZ Θ σZ Z (g g I J)
Problem Solutions 403
The dots represent terms that vanish after taking the dot product
(gI · gJ = δ IJ ). To convert to the unit basis, we substitute the
equations from part (b) for gI to obtain the boundary condition
⎡ ∗ ∗ ∗ ∗
⎤
FRR σ R R + a0 FRΘ σ R Θ
FRR FΘΘ ⎢ ⎢ R∗ Θ∗
⎥
⎥
⎣ a 0 F ΘΘ σ ⎦
2 2
FΘΘ + FRΘ ∗ ∗
σR Z
(eI )
⎡ ⎤ ⎡ ⎤
nR sR
⎢ ⎥ ⎢ ⎥
= −p(Θ) ⎢ ⎥ ⎢
⎣ nΘ ⎦ + τ (Θ) ⎣ sΘ ⎦
⎥
0 0
(eI ) (eI )
Chapter 5
5.1 (a) Summing forces on the element shown in Fig. 5.5 yields Newton’s
law in the form
−σ da + (σ + σ dx)da + f da dx = ρv̇ da dx,
where prime indicates differentiation with respect to x. Simplify-
ing leads to the 1-D equation of motion
σ + f = ρv̇.
(b) The kinetic and internal energies for the element are
1 2
K = 2 ρv da dx
U = ρu da dx,
where v is the average velocity and u the internal energy per unit
mass. The mechanical power input is
P = −(σ da)v + (σ + σ dx) da (v + v dx) + (f da dx)v,
where the minus sign in the first term comes from σ and v being
defined as positive in opposite directions at the left end, and the
body force acts through the average element velocity. Expanding
and neglecting terms of order greater than dx (as dx → 0) yields
P = [(σv) + fv] da dx.
Heat is added at the rate
Q = q da − (q + q dx) da + ρr da dx
= (−q + ρr) da dx.
Substituting these relations into K̇ + U̇ = P + Q yields
ρvv̇ + ρu̇ = (σv) + fv − q + ρr
or
ρu̇ = σv − q + ρr + (σ + f − ρv̇) v.
The term in parentheses vanishes by the equation of motion de-
rived in part (a), giving
ρu̇ = σv − q + ρr,
which is easily shown to be the 1-D form of (5.18)1 .
Problem Solutions 405
5.2 In terms of the given stress tensor τ , we need to show that the stress
power is given by
Ps = J −1 τ : U̇ = 12 J −1 (s · U + U · s) : U̇.
Table 2.2 provides the identities
A : (B · C) = B : (A · CT )
A : (B · C) = C : (BT · A),
which are used to modify the first and second terms, respectively, on
the right-hand side of the above expression for Ps . Setting B = s and
C = U in the first term, B = U and C = s in the second term, and
A = U̇ in both terms yields
1 −1 T
Ps = 2J [s : (U̇ · UT ) + s : (U · U̇)]
1 −1
= 2J [s : (U̇ · U + U · U̇)],
where the second line follows from U being a symmetric tensor. Next,
we use Eq. (3.153) to write the Lagrangian strain tensor as
E = 12 (C − I) = 12 (U · U − I),
and differentiating with respect to time gives
Ė = 12 (U̇ · U + U · U̇).
Comparing this equation with the last expression above for Ps shows
that
Ps = J −1 s : Ė,
which agrees with the last of (5.22), thus proving the given expression
for Ps .
5.3 (a) As viewed by observers in frames A and A∗ , respectively, the
Eulerian strain tensor is given by [see Eq. (3.28)2 ]
e = 1
2
(I − B−1 )
e∗ = 1 ∗
2
[I − (B∗ )−1 ],
where
B−1 = F−T · F−1
(B∗ )−1 = (F∗ )−T · (F∗ )−1 .
The transformation relation for F, provided by (5.55), is
F∗ = Q · F,
406 Problem Solutions
which gives
(F∗ )−1 = F−1 · Q−1 = F−1 · QT
(F∗ )−T = (F−1 · QT )T = Q · F−T .
Plugging these relations into the above equation for (B∗ )−1 yields
(B∗ )−1 = (F∗ )−T · (F∗ )−1 = (Q · F−T ) · (F−1 · QT )
= Q · B−1 · QT .
Under a change in frame, the identity tensor transforms like a
standard second-order tensor, i.e.,
I∗ = Q · I · QT .
Finally, substituting for I∗ and (B∗ )−1 into the above equation
for e∗ yields
e∗ = 12 (Q · I · QT −Q · B−1 · QT ) = Q · e · QT .
Answer for Lagrangian strain tensor: E∗ = E
(b) To determine the transformation relation for σ, we use the sym-
metry property of this tensor and the Cauchy stress formula
(4.23) to obtain
T = n · σ = σ T · n =σ · n
T∗ = σ ∗ · n∗,
where we set T(n) = T. The vectors T and n transform like any
vector, and the equation for T∗ gives
Q · T = σ ∗ · (Q · n)
or, after dotting both sides with Q−1 = QT ,
T = (QT · σ ∗ · Q) · n.
According to the Cauchy stress formula, the term in parentheses
corresponds to σ, i.e.,
σ = QT · σ ∗ · Q,
which gives the transformation relation
σ ∗ = Q · σ · QT .
5.8 (b) In terms of Cartesian components, the equation from part (a)
gives
∂W
sIJ = 2 − p(CIJ )−1
∂CIJ
∂W ∂I1 ∂W ∂I2
= 2 + − p(CIJ )−1
∂I1 ∂CIJ ∂I2 ∂CIJ
∂I1 ∂I2
= 2 b1 + b2 − p(CIJ )−1
∂CIJ ∂CIJ
for the given W . Here, (CIJ )−1 represents the IJ component of
the inverse of C, and (3.142) provides the strain invariants
I1 = CKK
1
I2 = 2 (CKK CLL − CKL CLK ).
The required derivatives are given by
∂I1 ∂CKK
= = δ IK δ JK = δ IJ
∂CIJ ∂CIJ
∂I2 1 ∂CKK ∂CLL ∂CKL
= CLL + CKK − CLK
∂CIJ 2 ∂CIJ ∂CIJ ∂CIJ
∂CLK
− CKL
∂CIJ
1
= 2 (δ IK δ JK CLL + CKK δ IL δ JL − δ IK δ JL CLK
− CKLδ IL δ JK )
1
= 2 (δ IJ CLL + CKK δ IJ − CJI − CJI )
= CKK δ IJ − CIJ ,
in which CJI = CIJ because C is symmetric. Plugging these into
the above equation for sIJ yields
sIJ = 2b1 δ IJ + 2b2 (CKK δ IJ − CIJ ) − p(CIJ )−1 .
5.9 Relative to principal axes, Eq. (3.135)4 provides the Lagrangian strain
tensor in the spectral form
3
E= Ei Ni Ni ,
i=1
where Ei and Ni are principal strains and directions, respectively.
The constitutive relation (5.78)3 for the second Piola-Kirchhoff stress
tensor yields
3 3
∂W ∂W ∂W ∂Λi
s = = Ni Ni = Ni Ni .
∂E i=1
∂E i i=1
∂Λi ∂Ei
Problem Solutions 411
Principal stretch ratios and strains are related by Eq. (3.134)1 , i.e.,
Λ2i = 1 + 2Ei ,
giving
∂Λi 1
= .
∂Ei Λi
The relations for t and σ can be found using (4.28), (4.29)1, and the
spectral representation for F. Details are left to the reader.
5.10 Answer:
1
σ2 = μ 1 −
Λ1 Λ32 Λ3
Λ1 Λ3
t2 = Λ 1 Λ 3 σ 2 , s2 = σ2
Λ2
C = FT · F, and B = F · FT , yields
−1/2
σ = I3 F · s · FT
−1/2 I1 1
= μI3 F · I · FT − F · C · FT
I3 I3
1/2 I2 −1
+ I3 − F·C ·F T
I3
−1/2 I1 1
= μI3 F · FT − F · FT · F · FT
I3 I3
1/2 I2
+ I3 − F · F−1 · F−T · FT
I3
−1/2 I1 1
= μI3 F · FT − F · FT · F · FT
I3 I3
1/2 I2
+ I3 − F · F−1 · F−T · FT
I3
# $
1 2
I2
= μ 3/2 I1 B − B + 1 − 3/2 I .
I3 I3
To obtain the final form, we use the Cayley-Hamilton theorem (2.84)
for the tensor B, i.e.,
B3 − I1 B2 + I2 B − I3 I = 0,
which gives
I1 B − B2 = I2 I − I3 B−1 .
Substituting this expression into the above equation for σ and sim-
plifying gives
−1/2 −1
σ = μ I−I3 B .
5.12 Rather than showing all details for each step, we list the equations
needed (roughly in order) for solving each part of the problem using
computer software, e.g., Matlab. All matrix components are Carte-
sian, so there is no need to specify the basis vectors for the matrices.
However, because the convected base vectors are not orthogonal, we
keep sub/superscript notation.
(a) For the specified xi (X I ), the deformation gradient tensor is
⎡ ⎤
0.8X 1 + 0.2X 2 + 0.2 0.2X 1 0
∂xi
F= =⎣ 0 0.2X 2 + 0.1 0 ⎦ ,
∂XI
0 0 0.8
Problem Solutions 413
W4 = 2cI4 , W5 = 2dI5 ,
where Wi = ∂W /∂Ii and
I1 = trC
1 2
I2 = 2 [(trC) − trC2 ]
I3 = det C
I4 = E11
2 2
I5 = E12 + E13 .
Note that I4 and I5 are modified from (5.143) for fibers oriented
in the X 1 -direction. The tensors M and N, as well as their com-
ponents, also need to be modified (see below).
Now, we are in a position to compute the stress components
using Eq. (5.148), i.e.,
∗
J∗
σI = ΦGIJ + ΨH IJ − pgIJ + ΘM IJ + ΛN IJ ,
where
−1/2
Φ = 2I3 W1
−1/2
Ψ = 2I3 W2
1/2
p = −2I3 W3
−1/2
Θ = I3 W4
−1/2
Λ = I3 W5
414 Problem Solutions
AJI = eI · GJ = δ JI .
The final result is ⎡ ⎤
9.05 0.825 0
∗ ∗
σ I J = ⎣ 0.825 −4.20 0 ⎦.
0 0 −1.50
(b) In this part, we first compute the Cauchy stress tensor in Cartesian
coordinates using
σ = α0 I + α1 B + α2 B2 + ΘM + ΛN,
where
B = F · FT .
In addition to Θ and Λ from part (a), we need the coefficients
1/2
α0 = 2I3 W3
−1/2
α1 = 2I3 (W1 + I1 W2 )
−1/2
α2 = −2I3 W2
and the tensors
M = F · (e1 e1 ) · FT
3
N = F · (e1 eα + eα e1 ) · FT Eα1 .
α=2
∗ ∗
The stress components σ I J can be determined using either
∗ ∗
σI J = gI · σ · gJ
or Eq. (2.70)2 in the form
∗ ∗
[σI J ] = [B̃]T [σ][B̃],
where (2.48) gives
B̃ = gI GI
for the transformation from Cartesian base vectors GI to con-
vected base vectors gI . (Both are expressed relative to Cartesian
coordinates.) Either way, the final result is the same as that in
part (a).
Problem Solutions 415
where
∂W
Wij ≡ .
∂Eij
Material incompressibility requires det F = Λx Λy Λz = 1, which
gives Λz = 1/ΛxΛy . Then, the plane stress condition (σ zz = 0)
yields
2c
p = 2cΛ2z = .
Λ2x Λ2y
Putting everything together, we find
2 1 2 3 2
σ xx = 2cΛx 1 − 4 2 + 4α(Λf − 1) cos β
Λx Λy
1
σ xx = 2cΛ2y 1 − 2 4 + 4α(Λ2f − 1)3 sin2 β
Λx Λy
σ xy = σ yx = 8cαΛx Λy (Λ2f − 1)3 cos β sin β.
Note that, if we set Exy = Eyx = 0 before taking derivatives of W ,
we would get σ xy = 0. This example again illustrates the hazard
of applying symmetry conditions or certain assumptions prior
to differentiating without careful consideration of the potential
consequences.
(c) Partial answer: No
Chapter 6
6.1 With I1 and I3 given by (6.16), the given strain-energy density func-
tion becomes
μ
W = [λ2x + λ2y + λ2z + 3(λxλy λz )−2/3 − 6].
2
Since p = 0 for a compressible material, Eq. (6.8)1 gives
∂W
σ i = J −1 λi (i = x, y, z; not summed).
∂λi
Problem Solutions 417
and
1
σ x = μ λ7/5
x − .
λx
6.2 The solution must satisfy the incompressibility condition
λx λy λz = 1.
Moreover, because forces are applied only in the axial (z) direction,
we must have
∂W
σ x = λx − p = 2λ2x C + D1 (λ2x − 1) − p = 0
∂λx
∂W
σ y = λy − p = 2λ2y C + D2 (λ2y − 1) − p = 0,
∂λy
which are found by substituting the given W into Eq. (6.8)1 . With
λz specified, these three equations can be solved for λx , λy , and p.
6.3 (a) As noted earlier, the principal directions of stress and strain co-
incide for an isotropic material. Therefore, for an incompressible
Ogden material (J = 1), Eq. (6.8)1 gives
∂W
σ i = Λi −p
∂Λi
N
= an Λbi n − p
n=1
with respect to the principal axes. For in-plane deformation, the
membrane has no transverse loading, and setting σ 3 = 0 yields
N
p= an Λb3n .
n=1
With this result, we find
N
σi = an Λbi n − (Λ1 Λ2 )−bn ,
n=1
⎡ ⎤
λ21 + k 2 λ22 kλ22 0
B= F·F = ⎣T
kλ22 λ22 0 ⎦,
0 0 λ23 (e
i ej )
where the ei are Cartesian base vectors. Next, with the stresses
given by the expression derived in part (a), we compute the stress
components relative to the Xi coordinates using
σ ij = ei · σ · ej
3
= σ k (ei · nk )(ej · nk ).
k=1
For the specified parameter values, we find
⎡ ⎤
8.69 1.19 0
σ = ⎣ 1.19 0.944 0 ⎦.
0 0 0 (e
i ej )
and
2 2 2 2 2
I1 = tr C = F11 + F22 + F33 + F12 + F21
I3 = det C =(F11 F22 − F12 F21 )2 F33
2
,
which are needed for the given W . The constitutive relation
(5.77)
∂W ∂W
t= → tij =
∂FT ∂Fji
gives (after setting F21 = 0)
2 2 −(1+β)
t11 = 2F11 1 − F22 F33 I3
2 2 −(1+β)
t22 = 2F22 1 − F11 F33 I3
2 −(1+β)
t12 = 2F11F22 F33 F12 I3
t21 = 2F12,
which are the only non-zero components of the first Piola-
Kirchhoff stress tensor t. To eliminate F33, we use the plane
stress condition t33 = 0 to find
F33 = (F11 F22 )−β/(1+β) = (F11 F22 )−ν/(1−ν) .
√
(b) With J = I3 , Eq. (6.2) provides the Cauchy stress tensor,
σ = J −1 F · t.
To extract the convected stress components, we need the con-
vected base vectors given by
gI = F · eI ,
which, with Eqs. (2.18), yield
g1∗ = F11 e1 , g2∗ = F12 e1 + F22 e2 , g3∗ = F33 e3
1∗ −1 −1 2∗ −1 ∗
g = F11 (e1 − F12F22 e2 ), g = F22 e2 , g3 = F33
−1
e3 .
We now can compute
∗
J∗
σI = gI · σ · gJ .
The details are left for the reader’s entertainment.
(c) Partial answer for ν = 0.25 and λ1 = 1:
∗ ∗ ∗ ∗ ∗ ∗
σ1 1
= 0.268, σ2 2
= 1.06, σ1 2
= 0.198
This can be used as a check.
420 Problem Solutions
6.7(a) Since the analysis in the text uses physical tensor components,
all hats are omitted here for convenience. For clarity, however,
because torsion causes eR and eΘ to rotate during deformation,
dyadic bases are indicated on matrices, with I, J = R, Θ, Z and
i, j = r, θ, z. Note also that the formulation in the text takes
advantage of several equations in Sec. 6.5.
First, we consider a, λ, and ψ as unknowns to be determined.
In the absence of residual stress, we set φ0 = π and Λ = 1, giving
γ = ψr and
1
2 1 2 2
2
r(R) = a + R − a0
λ
by (6.116). With this result, Eqs. (6.103)–(6.106) yield
⎡ ∂r ⎤
⎡ ⎤
FrR 0 0 0 0
⎢ ∂R ⎥
F = ⎣ 0 FθΘ FθZ ⎦ =⎢ ⎣ 0
r ⎥
ψr ⎦
0 0 FzZ (e e ) R
i J 0 0 λ (e e )
i J
and
1
E = 2 FT · F − I
⎡ 2 ⎤
FrR − 1 0 0
1⎣ 2 ⎦
= 2 0 FθΘ −1 FθΘ FθZ
2 2
0 FθΘ FθZ FzZ + FθZ − 1 (e
I eJ )
at each grid point R or r(R). Next, the Cauchy stresses are given
by
σ = σ̄ − p I,
where
∂W
σ̄ = F ·· FT = [σ̄ ij ](ei ej ) .
∂E
The matrix for ∂W/∂E relative to the basis {eI eJ } is provided by
Eq. (6.132), which requires writing W as a function of the strain
components in the fiber coordinate system. These are found using
the relations
EF F = eF ·E·eF , ECC = eC ·E·eC , EF C = ECF = eF ·E·eC ,
where
eF = eΘ cos β + eZ sin β
eC = −eΘ sin β + eZ cos β.
Problem Solutions 423
which leads to
a b
σ (1)
z r dr + σ (2)
z r dr = 0.
0 a
(k)
Substituting from part (c) and integrating yields
σz
3 1
λ − 1 a20 (C1 − C2 ) + b20 C2 + ρω2 b40 = 0,
8
which gives
1
ω2 3
λ= 1− 2 ,
ω0
428 Problem Solutions
where
8 2
ω 20 = 4 a0 (C1 − C2 ) + b20 C2 .
ρb0
According to this solution, the length of the cylinder deceases
as ω increases. This is reasonable, but the cylinder flattens com-
pletely (λ = 0) when ω = ω0 , which is not realistic. Similar
behavior occurs for inflation of a spherical neo-Hookean shell.
As shown by the curve for α = 0 in Fig. 6.22a, the pressure a
neo-Hookean shell can sustain is limited. Here, the cylinder can
sustain inertial loads only for spin rates ω < ω 0 . This unfortu-
nate behavior of neo-Hookean material becomes a serious issue in
certain cases, such as these. Other forms for W do not necessarily
have this problem (Chadwick et al., 1977).
6.10 Here, we outline one possible computational procedure to solve this
problem. After setting up a grid for R, the steps are the following:
• Given a, use Eq. (6.151) to compute the deformed coordinate r
at each grid point.
• Compute the stretch ratios at each point using (6.145). (λR also
could be computed using incompressibility.)
• Integrate (6.153) to obtain the Lagrange multiplier p at each
point. Break the integral into two parts, i.e.,
b c b
(· · · ) = (· · · ) + (· · · ), r ≤ c
r r c
b
= (· · · ), r ≥ c,
r
give
W = C[I1 − 3 + β −1 (I3−β − 1)
! "
= C λ2r + λ2θ + λ2φ + β −1 (λr λθ λφ )−2β − 1 ,
(d) The stress σ θ increases from negative at the inner surface to pos-
itive at the outer surface in the everted shell.
(e) The form of the stress distribution is consistent with those pro-
duced by bending the shell into its everted configuration. This
requires bending moments in both the θ- and φ-directions to re-
verse the curvature at each point. The peak stress increases with
the shell thickness, in agreement with intuition that thinner shells
would be easier to turn inside out.
However, since σ θ = 0 along the edge of the shell, the as-
sumed deformation is not consistent with the everted shell being
completely free of external loads. To eliminate these stresses,
equal and opposite bending stresses must be applied along the
edge, compressing the outer surface while causing the edge to
bend slightly outward. Because these cancelling stresses are self-
equilibrating, Saint-Venant’s principle suggests that their effects
would be significant only within a relatively narrow region near
the edge.
6.15 The solution to this problem follows Carroll and Horgan (1990).
(a) With the position vector in the deformed configuration given by
r = xex + yey + zez ,
we have
∂ 1 ∂ ∂
∇r = eR + eΘ + eZ [x(R)ex + kΘey + λZez ] ,
∂R R ∂Θ ∂Z
in which ∇ is written for the undeformed body in cylindrical co-
ordinates. This relation produces the deformation gradient tensor
F = (∇r)T = λx ex eR + λy ey eΘ + λz ez eZ ,
where
dx k
λx = , λy = , λz = λ
dR R
are stretch ratios.
(b) In terms of the stretch ratios, the given strain-energy density
function takes the form
μ −2
W = λx + λ−2 −2
y + λz + 2λx λy λz − 5 .
2
For the prescribed deformation, only normal stresses are present
relative to Cartesian coordinates, and Eq. (6.8)1 gives
λi ∂W
σi = (i not summed)
J ∂λi
434 Problem Solutions
J = λx λy λz .
λ3x λy λz = 1.
Armstrong CG, Lai WM, and Mow VC (1984). An analysis of the uncon-
fined compression of articular cartilage. J Biomech Eng 106:165-173.
437
438 Bibliography
Fillinger MF, Marra SP, Raghavan ML, and Kennedy FE (2003). Predic-
tion of rupture risk in abdominal aortic aneurysm during observation: Wall
stress versus diameter. J Vasc Surg 37:724-732.
Fillinger MF, Raghavan ML, Marra SP, Cronenwett JL, and Kennedy FE
(2002). In vivo analysis of mechanical wall stress and abdominal aortic
aneurysm rupture risk. J Vasc Surg 36:589-597.
Fung YC (1991). What are the residual stresses doing in our blood vessels?
Ann Biomed Eng 19:237-249.
Greenwald SE, Moore JE, Jr., Rachev A, Kane TPC, and Meister JJ (1997).
Experimental investigation of the distribution of residual strains in the
artery wall. J Biomech Eng 119:438-444.
Guccione JM, McCulloch AD, and Waldman LK (1991). Passive material
properties of intact ventricular myocardium determined from a cylindrical
model. J Biomech Eng 113:42-55.
Hansen DE, Daughters GT, Alderman EL, Ingels NB, and Miller DC (1988).
Torsional deformation of the left ventricular midwall in human hearts with
intramyocardial markers: Regional heterogeneity and sensitivity to the in-
otropic effects of abrupt rate changes. Circ Res 62:941-952.
Hashima AR, Young AA, McCulloch AD, and Waldman LK (1993). Non-
homogeneous analysis of epicardial strain distributions during acute my-
ocardial ischemia in the dog. J Biomech 26:19-35.
Holmes MH (1986). Finite deformation of soft tissue: Analysis of a mixture
model in uniaxial compression. J Biomech 108:372-381.
Holzapfel GA, Gasser TC, and Ogden RW (2000). A new constitutive
framework for arterial wall mechanics and a comparative study of material
models. J Elasticity 61:1-48.
Humphrey JD (1999). Remodeling of a collagenous tissue at fixed lengths.
J Biomech Eng 121:591-597.
Humphrey JD (2002). Cardiovascular solid mechanics: Cells, tissues, and
organs, Springer, New York
Humphrey JD and Holzapfel GA (2012). Mechanics, mechanobiology, and
modeling of human abdominal aorta and aneurysms. J Biomech 45:805-
814.
Humphrey JD and Rajagopal KR (2002). A constrained mixture model
for growth and remodeling of soft tissues. Math Models Methods Appl Sci
12:407-430.
Humphrey JD and Rajagopal KR (2003). A constrained mixture model
for arterial adaptations to a sustained step change in blood flow. Biomech
Model Mechanobiol 2:109-126.
Humphrey JD, Strumpf RK, and Yin FCP (1990a). Determination of a
constitutive relation for passive myocardium: I. A new functional form. J
Biomech Eng 112:333-339.
Bibliography 441
Huyghe JM, Arts T, van Campen DH, and Reneman RS (1992). Porous
medium finite element model of the beating left ventricle. Am J Physiol
262:H1256-H1267.
Ingels NB, Hansen DE, Daughters GT, Stinson EB, Alderman EL, and
Miller DC (1989). Relation between longitudinal, circumferential, and
oblique shortening and torsional deformation in the left ventricle of the
transplanted human heart. Circ Res 64:915-927.
LeGrice IJ, Hunter PJ, and Smaill BH (1997). Laminar structure of the
heart: A mathematical model. Am J Physiol 272:H2466-H2476.
LeGrice IJ, Smaill BH, Chai LZ, Edgar SG, Gavin JB, and Hunter PJ
(1995). Laminar structure of the heart: Ventricular myocyte arrangement
and connective tissue architecture in the dog. Am J Physiol 269:H571-
H582.
Omens JH and Fung YC (1990). Residual strain in rat left ventricle. Circ
Res 66:37-45.
Rivlin RS, Barenblatt GI, and Joseph DD (1997). Collected papers of R.S.
Rivlin, Springer, New York.
Taber LA, Sun H, Clark EB, and Keller BB (1994). Epicardial strains in
embryonic chick ventricle at stages 16 through 24. Circ Res 75:896-903.
Taber LA, Yang M, and Podszus WW (1996). Mechanics of ventricular
torsion. J Biomech 29:745-752.
Timoshenko S and Goodier JN (1969). Theory of elasticity, McGraw-Hill,
New York.
Tozeren A (1983). Static analysis of the left ventricle. J Biomech Eng
105:39-46.
Truesdell C and Noll W (2004). The non-linear field theories of mechan-
ics. In: S. A. Antman (Ed.) The non-linear field theories of mechanics.
Springer, New York.
Verbruggen SWE (2018). Mechanobiology in health and disease, Academic
Press, London.
Villarreal FJ, Lew WYW, Waldman LK, and Covell JW (1991). Transmu-
ral myocardial deformation in the ischemic canine left ventricle. Circ Res
68:368-381.
Vossoughi J, Hedjazi Z, Borris FS, Langrana NA, Friedman MH, and Grood
ES (1993). Intimal residual stress and strain in large arteries. In: Proc sum-
mer bioengineering conference. New York, ASME, pp. 434-437.
Wagenseil JE (2018). Bio-chemo-mechanics of thoracic aortic aneurysms.
Curr Opin Biomed Eng 5:50-57.
Waldman LK, Fung YC, and Covell JW (1985). Transmural myocardial
deformation in the canine left ventricle. Normal in vivo three-dimensional
finite strains. Circ Res 57:152-163.
Waldman LK, Nosan D, Villarreal F, and Covell JW (1988). Relation
between transmural deformation and local myofiber direction in canine left
ventricle. Circ Res 63:550-562.
Wang VY, Lam H, Ennis DB, Cowan BR, Young AA, and Nash MP (2009).
Modelling passive diastolic mechanics with quantitative MRI of cardiac
structure and function. Med Image Anal 13:773-784.
Whittaker P, Boughner DR, and Kloner RA (1991). Role of collagen in
acute myocardial infarct expansion. Circulation 84:2123-2134.
Woods RH (1892). A few applications of a physical theorem to membranes
in the human body in a state of tension. J Anat Physiol 26:362-370.
446 Bibliography
447
448 Index