Sei sulla pagina 1di 52

The University of Calgary

Winter 2016 BMEN525 Biomechanics of Tissues


Course Notes
Version 2016-03-06

Dr. Salvatore Federico & Dr. Elena Di Martino


salvatore.federico@ucalgary.ca, edimarti@ucalgary.ca
Contents

1 Linear Algebra 1
1.1 Vector Spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.2 Linear Dependence and Independence, Dimension, Bases . . . . . . . . . . . . . 3
1.3 Linear Maps . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.4 Scalar Product . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
1.5 Second-Order Tensors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
1.6 The Tensor Product . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
1.7 Fourth-Order Tensors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12

2 Continuum Mechanics 13
2.1 Physical Space and Bodies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
2.2 Kinematics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
2.2.1 Configuration Map and Deformation Gradient . . . . . . . . . . . . . . 14
2.2.2 Deformation and Strain Tensors . . . . . . . . . . . . . . . . . . . . . . 18
2.2.3 The Polar Decomposition Theorem . . . . . . . . . . . . . . . . . . . . . 18
2.2.4 Change of Volume and Area . . . . . . . . . . . . . . . . . . . . . . . . . 18
2.2.5 Velocity and Time Rates . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
2.3 Stress, Balance and Power . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
2.3.1 Cauchy Stress . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
2.3.2 Balance Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
2.3.3 Balance of Mechanical Power . . . . . . . . . . . . . . . . . . . . . . . . 37
2.3.4 Conjugated Stress-Strain Pairs and “Material Balance” . . . . . . . . . 38
2.4 Constitutive Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
2.4.1 Non-Linear Elasticity . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
2.4.2 Linearisation of Non-Linear Elasticity . . . . . . . . . . . . . . . . . . . 45
2.4.3 One-Dimensional Viscoelasticity . . . . . . . . . . . . . . . . . . . . . . 46
1 Chapter 1

2 Linear Algebra

3 If, as Galileo Galilei used to say, Nature speaks the language of Mathematics, we need to know
4 this language in order to understand Nature. This chapter is aimed at recalling some basic
5 concepts in Linear Algebra, i.e., vector spaces, linear maps, tensors, which will prepare to a
6 simple geometrical definition of the physical space, which is the “theatre” in which Mechanics
7 takes place. Here, with “Mechanics”, we mean Classical Mechanics, i.e., neither Relativistic
8 nor Quantum Mechanics. The mathematical notation follows international standards, and it is
9 summarised below.
∀ for every
∃ exists
∃! exists only one
A⊆B set A is contained in, or is at most equal to B
A⊂B set A is strictly contained in B
a∈A a is an element of set A
A = {a1 , a2 , ...} set defined by enumerating its elements
10
A = {a : a satisfies property P} set defined by stating the properties of its elements
A × B = {(a, b) : a ∈ A, b ∈ B} Cartesian product of sets A and B
N set of the natural numbers, i.e., N = {1, 2, 3, ...}
R set of the real numbers
C set of the complex numbers
f :A→B function f with domain A and codomain (or range) B
f : x 7→ f (x) function f mapping x into f (x) (the value of f at x)
11 In preparation for the study of Continuum Mechanics, whenever one space only is considered,
12 vectors and second-order tensors are denoted by boldface italic lowercase letters (u, v, w... and
13 a, b, c... respectively), and fourth-order tensors by a “small” blackboard bold font (A, B, C...),
14 in order to mimic the notation in the spatial picture of Mechanics. Whenever two spaces are
15 involved, quantities in the domain space are denoted with uppercase letters (vectors U , V , W ...
16 second-order tensors A, B, C... and fourth-order tensors A, B, C...) as in the material picture
17 of Mechanics, and quantities in the codomain space are denoted with lowercase letters.

1
18 1.1 Vector Spaces

19 Let V be a non-empty set, the elements of which are called vectors and denoted u, v, w,...,
20 considered together with the set of the real numbers, R, the elements of which are called, in
21 this context, scalars.
22 Let us assume that V is endowed with an internal operation, called sum, and indicated by the
23 symbol “+”,
V × V → V : (u, v) 7→ u + v, (1.1)
24 and with an external operation on R, called multiplication by scalars,

R × V → V : (λ, v) 7→ λv. (1.2)

25 The set V is a vector space on R (or real vector space, or linear space on R, or real linear space)
26 if the two operations defined above satisfy the properties below.
27 For the sum:

28 (S1) ∀u, v, w ∈ V, u + (v + w) = (u + v) + w (associative property);

29 (S2) ∃! 0V ∈ V such that, ∀v ∈ V, v + 0V = 0V + v = v (existence of the neutral element);

30 (S3) ∀v ∈ V, ∃! v 0 ∈ V such that v + v 0 = v 0 + v = 0 (existence of the opposite);

31 (S4) ∀u, v ∈ V, u + v = v + u (commutative property).

32 For the product by scalars:

33 (P1) ∀λ ∈ R and ∀u, v ∈ V, λ(u + v) = λu + λv (pseudo-distributive on V);

34 (P2) ∀λ, µ ∈ R and ∀v ∈ V, (λ + µ)v = λv + µv (pseudo-distributive on R);

35 (P3) ∀λ, µ ∈ R and ∀v ∈ V, (λµ)v = λ(µv) (pseudo-associative);

36 (P4) ∀v ∈ V, 1v = v (where 1 is the unit of R).

37 The neutral element of the sum, 0V , is called the zero vector of V, and the opposite element,
38 v 0 , is called opposite of v and is denoted by −v.
39 Vector spaces are a particular case of algebraic structure, i.e., a non-empty set, in which one
40 or more operations are defined. Here we are only concerned with real vector spaces, i.e., based
41 on the real numbers R, but exactly the same considerations can be made on complex vector
42 spaces (based on the complex numbers C), which are commonly employed, e.g., in Quantum
43 Mechanics, System Theory, Control Theory.

44 Problem 1.1. Prove or verify that Rn = R × . . . × R, the elements of which are the n-tuples
45 (x1 , ..., xn ) is a vector space on R, for every natural number n ∈ N, thereby including n = 1
46 (which means that R is a vector space on itself!).

2
47 Example 1.1. Let A be a non-empty set, and V a real vector space, let A V denote the set
48 of all functions mapping elements of A into elements of V, and let f, g, ... denote the elements
49 of A V. Relying on the fact that V is a vector space, we are allowed to define the sum of
50 elements of A V as (f + g)(x) = f (x) + g(x) and the product of elements of A V by scalars of
51 R as (λf )(x) = λ f (x). These two operations are legitimate, because the values f (x) of the
52 functions f belonging to A V are vectors of V. Indeed, f (x) + g(x) is a sum in V and λ f (x) is
53 the multiplication of λ ∈ R with the vector f (x) ∈ V. It is easy to verify that these operations
54 satisfy the properties of sum and multiplication by scalars in a vector space.
55 Problem 1.2. Consider the set C 0 ([a, b], R) of all continuous functions f : [a, b] → R. If the
56 sum of functions and the product by scalars are defined again by (f + g)(x) = f (x) + g(x) and
57 (λf )(x) = λ f (x), respectively, prove or verify that C 0 ([a, b], R) is a vector space on R.

58 1.2 Linear Dependence and Independence, Dimension, Bases


59 The concept of space dimension is crucial in all applications of Linear Algebra. First, we need
60 to introduce the definitions of linear combination, linear dependence, and linear independence.
61 Let v 1 , ..., v n ∈ V and λ1 , ..., λn ∈ R, with n ∈ N. The vector
n
X
λi v i = λ1 v 1 + ... + λn v n ∈ V (1.3)
i=1

62 is called the linear combination of the vectors v 1 , ..., v n , with coefficients λ1 , ..., λn .
63 A set of vectors v 1 , ..., v n ∈ V is called linearly dependent if, and only if there exist a set of
64 scalars λ1 , ..., λn of which at least one is non-zero, such that
n
X
λi v i = λ1 v 1 + ... + λn v n = 0. (1.4)
i=1

65 To understand what this means, let us assume that, e.g., λ1 6= 0. if this is the case, we can
66 write
n
1 X λ2 λn
v1 = − λi v i = − v 2 − ... − vn, (1.5)
λ1 λ1 λ1
i=2
67 i.e., v 1 can be expressed as linear combination of v 2 , ..., v n , and is therefore linearly dependent
68 on v 2 , ..., v n .
69 In contrast, a set of vectors v 1 , ..., v n ∈ V is called linearly independent if, and only if
n
X
λi v i = λ1 v 1 + ... + λn v n = 0 implies λ1 , ..., λn = 0, (1.6)
i=1

70 i.e., if none of the vectors can be expressed as linear combination of the others.
71 We are now ready to give the definitions of dimension and basis. A vector space V is said to
72 have dimension equal to n if, given any set of n linearly independent vectors e1 , ..., en , any
73 vector v ∈ V can be expressed as linear combination of e1 , ..., en . In this case, {ei }ni=1 is said
74 to be a basis of V, and we write
Xn
v= vi e i ≡ vi e i , (1.7)
i=1
75 where the scalars vi are called components of v in the basis {ei }ni=1 .

3
76 Remark 1.1. The components of a vector are distinct from the vector itself. A vector is
77 exclusively an element in a vector space, and can be identified by its component in a given
78 basis only once the basis is assigned! Indeed, given two bases {ei }ni=1 and {e0j }nj=1 , a vector v
79 can be expressed as v = vi ei = vj0 e0j ; it is crucial to realise that the components vi and vj0 are
80 different, but the vector itself is always the same!

Remark 1.2. In the last equation, we wrote vi ei for ni=1 vi ei : in practice, we omitted writing
P
81

82 the summation symbol, and consider that index i, which is repeated twice in the multiplicative
83 expression vi ei , is summed from 1 to n. This convention, called Einstein’s convention, is
84 customary in many applications, and saves from writing the summation symbol all the time.
85 Warning: the same index can be repeated only twice in the same multiplicative expression!

86 1.3 Linear Maps

87 Linear maps are functions between two vector spaces with the important property of linearity.
88 Let U and V be two real vector spaces. A function F : U → V is called linear if, and only if,

∀ U , Y ∈ U, and ∀λ, µ ∈ R, F (λ U + µ Y ) = λ F (U ) + µ F (Y ), (1.8)

89 i.e., if the image of a linear combination of vectors is the linear combination of the images of
90 each vector, with the same coefficients. Linear maps are also called homomorphisms (from the
91 Greek ὁμός, the same, and μορφή, form) between vector spaces, as they preserve the linear
92 structure when mapping from the domain space to the codomain space.
93 The set of all linear maps (homomorphisms) from U to V is denoted Hom(U, V), and, following
94 Example 1.1, is a vector space (since the codomain of the functions in Hom(U, V) is a vector
95 space). Since Hom(U, V) is a vector space, it follows that any linear combination of linear maps
96 in Hom(U, V), e.g., λ F + µ H, is also a linear map in Hom(U, V).
97 If the space U has dimension m and the space V has dimension n, then the space Hom(U, V)
98 has dimension n × m. If {E I }m n
I=1 is a basis of U and {ei }i=1 is a basis of V, for the linearity of
99 F ∈ Hom(U, V), we have

v = F (U ) = F (UI E I ) = UI F (E I ). (1.9)

100 Moreover, since v = F (U ) = UI F (E I ) is a vector of V, it can be written in the basis {ei }ni=1 ,
101 and we have
vi ei = [F (U )]i ei = UI [F (E I )]i ei . (1.10)
102 The scalars
FiI = [F (E I )]i (1.11)
103 are the entries of the n × m matrix [[FiI ]], called the matrix of the linear map F with respect to
104 the bases {E I }m n
I=1 in the domain and {ei }i=1 in the codomain. It is very important to note that
105 the columns of [[FiI ]] are given by the components, written in the basis {ei }ni=1 of the codomain,
106 of the images through F of the vectors {E I }m I=1 of the basis of the domain. This proves that
107 the dimension of Hom(U, V) is n × m, and allows for writing the component expression

vi = FiI UI , (1.12)

4
108 which in Engineering texts is often reported in the compact matrix formalism

{v} = [F ] {U }. (1.13)
n×1 n×m m×1

109 We shall usually omit the parentheses for the argument of a linear map, and simply juxtapose
110 the argument, i.e., we shall write
F U ≡ F (U ). (1.14)

111 Let U, V, W be real vector spaces, F ∈ Hom(U, V) and P ∈ Hom(V, W). The composition of
112 P and F is the linear map T = P F ∈ Hom(U, W) defined, for every U ∈ U, by

T U = (P F )(U ) = P (F U ), (1.15)

113 and also called the single contraction of P and F (note that v = F U ∈ V, and V is the domain
114 of P ). If the spaces U, V, W have finite dimension, and are assigned the bases {E I }m n
I=1 , {ei }i=1
p m p
115 and {α }α=1 , respectively, then the matrix of T in the bases {E I }I=1 and {α }α=1 is given by

TαI = Pαi FiI . (1.16)

116 Indeed, following the procedure seen above, we have

ω = T U = P F U = UI P F E I = FiI UI P ei = (Pαi FiI UI ) α = ωα α , (1.17)

117 which, together with


ω = T U = TαI UI α = ωα α , (1.18)
118 yields Equation (1.16). In components, we have

ωα = TαI UI = Pαi FiI UI , (1.19)

119 and, in matrix formalism


{ω} = [T ] {U } = [P ] [F ] {U }. (1.20)
p×1 p×m m×1 p×n n×m m×1

120 Given a vector space V of dimension n, a linear map a ∈ Hom(V, V), in which domain and
121 codomain are the same space, is called an endomorphism, where the prefix ἐνδο (endo) means
122 inside. The components of a will be aij if the same basis {ei }ni=1 is chosen in the domain and
123 in the codomain.
124 Analogously to the case of homomorphisms between different spaces, the composition (or single
125 contraction) of two endomorphisms a, b ∈ Hom(V, V) of a vector space V is defined as the
126 endomorphism c = a b ∈ Hom(V, V) such that, for every v ∈ V

c v = (a b)v = a(b v), (1.21)

127 where we note that b v ∈ V. In components, Equation (1.21) reads

cik vk = aij bjk vk , (1.22)

128 and therefore the matrix of c = a b is given by

cik = aij bjk , (1.23)

129 which is proved analogously to the case of the contraction of maps in different spaces.

5
130 The identity endomorphism in V is the tensor i ∈ Hom(V, V) such that, for every v ∈ V,

i v = v. (1.24)

131 For the identity, if we use the definition of components of a linear map, we have

i v = i(vj ej ) = vj i ej = vj iij ei = vi ei = v (1.25)

132 from which


vi = iij vj . (1.26)
133 Therefore, it must be 
1 if i = j,
iij = δij = (1.27)
0 if i 6= j,
134 where δij is called Kronecker delta.
135 An endomorphism a ∈ Hom(V, V) is called invertible if there exists a unique endomorphism
136 a−1 ∈ Hom(V, V), called the inverse of a, such that

a a−1 = a−1 a = i. (1.28)

137 In components,
aij (a−1 )jk = (a−1 )ip apk = δik . (1.29)

138 In the context of Continuum Mechanics, we shall also consider “uppercase” objects, such
139 as endomorphisms of the type A ∈ Hom(U, U) (in the space U of dimension m), whose
140 components we shall write AIJ , if the same basis {E I }m
I=1 is chosen in domain and codomain.
141 The “uppercase” identity in U is the linear map I ∈ Hom(U, U) defined analogously to its
142 “lowercase” counterpart, i.e., for every U ∈ U,

I U = U, (1.30)

143 with components 


1 if I = J,
IIJ = δIJ = (1.31)
0 if I 6= J.
144 In a perfectly analogous way, we also say that an “uppercase” endomorphism A ∈ Hom(U, U)
145 is invertible if there exists a unique endomorphism A−1 ∈ Hom(U, U) such that

A A−1 = A−1 A = I, (1.32)

146 which in components reads

AIJ (A−1 )JK = (A−1 )IP AP K = δIK . (1.33)

147 If the two spaces U and V have the same dimension n, a linear map F ∈ Hom(U, V) is called
148 invertible if there exists a unique linear map F ∈ Hom(V, U) such that

F F −1 = i, and F −1 F = I, (1.34)

149 which in components is

FiI (F −1 )Ij = δij , and (F −1 )Ii FiJ = δIJ . (1.35)

150 Note that domain and codomain of F −1 are switched with respect to those of F .

6
151 1.4 Scalar Product

152 We shall assume that the vector spaces we deal with are equipped with a scalar product. A
153 scalar product in a real vector space V is map

V × V → R : (v, w) 7→ v.w (1.36)

154 that is

155 1. bilinear, i.e., it is linear in each argument separately, so that

v.(λw + µz) = λ v.w + µ v.z, and (λv + µz).w = λ v.w + µ z.w, (1.37)

156 2. symmetric, i.e., it is invariant for exchange of the arguments, so that

v.w = w.v, (1.38)

157 3. positive definite, i.e.,

v.v > 0, ∀ v 6= 0V , and 0V .0V = 0. (1.39)

158 Note that we shall use notation with the low dot, i.e., v.w (Chadwick, 1976), as opposed to
159 the more common one with the centred dot v · w.
160 Two vectors v, w ∈ V are called orthogonal if their scalar product is zero, i.e., if

v.w = 0. (1.40)

161 Moreover, the positive definiteness of the scalar product allows to define the Euclidean norm

kvk = v.v. (1.41)

162 In elementary texts, the Euclidean norm is often called magnitude and indicated by |v|. We
163 emphasise that the Euclidean norm is only one of the many norms (there is an infinity of norms,
164 in fact) that can be defined in a vector space.
165 Once the scalar product is defined, we can choose to use an orthonormal basis, i.e., a basis in
166 which all vectors have norm equal to one and are all orthogonal to each other, so that

1 if i = j,
ei .ej = δij = (1.42)
0 if i 6= j,

167 where we recall that δij is called Kronecker delta.


168 When an orthonormal basis is used (and only in this case), there are some useful properties
169 that can be exploited. The scalar product of two vectors can be expressed as

v.w = (vi ei ).(wj ej ) = vi wj δij = vi wi , (1.43)

170 where we have made use of the fact that the scalar product is bilinear. Moreover, the
171 components of a vector can be obtained by the scalar product with the basis vectors, i.e.,

ei .v = ei .(vj ej ) = vj (ei .ej ) = vj δij = vi . (1.44)

7
172 In the same way, one can obtain the components of a linear map. Always keeping in mind that
173 this is valid only for orthonormal bases, for a linear map F ∈ Hom(U, V), we have

ei .(F E I ) = ei .(FjI ej ) = FjI ei .ej = FjI δij = FiI , (1.45)

174 for an endomorphism a ∈ Hom(V, V),

ei .(a ej ) = ei .(akj ek ) = akj ei .ek = akj δik = aij , (1.46)

175 and, for an endomorphism A ∈ Hom(U, U),

E I .(A E J ) = E I .(AKJ E K ) = AKJ E I .E K = AKJ δIK = AIJ . (1.47)

176 Note that we are indicating the scalar product in U and V by means of the same “low dot”.
177 In the following sections, we shall always assume to have orthonormal bases.

178 1.5 Second-Order Tensors

179 In several contexts, and particularly in Continuum Mechanics, endomorphisms, i.e., linear maps
180 from a vector space into itself, like a ∈ Hom(V, V) and A ∈ Hom(U, U) seen above, are also
181 called second-order tensors because they have two legs and we emphasise that the two legs are
182 in the same space. If U and V have the same dimension n, linear maps F ∈ Hom(U, V), which
183 map between two different spaces, are called two-point second-order tensors, as they have one
184 leg in a space and the other leg in a different space. In this section we shall introduce several
185 definitions on second-order tensors and we shall go to higher-order tensors after we introduce
186 the concept of tensor product. The etymology of the word tensor will be given in Section 2.3:
187 let us say it will be a surprise.
188 Let us start by giving the definition of transpose. There are actually two possible definitions:
189 the algebraic transpose, which prescinds from the availability of a scalar product, and the
190 metric transpose, for which, in contrast, the scalar product is necessary (see, e.g., Federico,
191 2012) coincide. When a scalar product is defined and an orthogonal basis is used, the two
192 definitions become equivalent, and we shall therefore simply speak about the transpose. We
193 shall begin from the definition of transpose for a two-point tensor, as it is more general, and
194 then particularise it to tensors.
195 Given a two-point second-order tensor F ∈ Hom(U, V) the transpose of F is the tensor F T ∈
196 Hom(V, U) such that, for every vector v ∈ V and for every vector U ∈ U,

v.(F U ) = U .(F T v), (1.48)

197 Note that domain and codomain of F T are switched with respect to those of F , and that the
198 scalar products on the left-hand side and right-hand side of Equation (1.48) are that in V and
199 that in U, respectively. In components, we have

vi FiI UI = UI (F T )Ii vi , (1.49)

200 from which, for the arbitrariness of v and U , we have

(F T )Ii = FiI , (1.50)

8
201 i.e., the matrix of the transpose F T is obtained by switching rows and columns in the matrix
202 of F .
203 Analogously, for a tensor a ∈ Hom(V, V), the transpose is the tensor aT such that, for every
204 vectors v, w ∈ V,
v.(a w) = w.(aT v). (1.51)
205 In components,
vi aij wj = wj (aT )ji vi , (1.52)
206 from which
(aT )ji = aij , (1.53)
207 i.e., again, the matrix of the transpose aT is obtained by switching rows and columns in the
208 matrix of a.
209 A tensor a ∈ Hom(V, V) is called symmetric if a = aT . In components, this means that the
210 components of a are such that aij = aji , i.e., the matrix of a is symmetric, and it has only
211 (n2 + n)/2 independent components out of the total n2 . For the case of dimension n = 3, this
212 means 6 components out of the total 9.
213 A tensor Ω ∈ Hom(V, V) is called skew-symmetric if Ω = −ΩT . In components, this reads
214 Ωij = −Ωji , i.e., the matrix of Ω is skew-symmetric, and it has only n!(n − 2)!/2! = (n2 − n)/2
215 independent components out of the total n2 . In dimension n = 3, this means 3 out of 9.
216 Any second-order tensor a ∈ Hom(V, V) can be univocally decomposed into the sum of a
217 symmetric part and a skew-symmetric part. Indeed, given the identity

a = sym(a) + skew(a) = 21 (a + aT ) + 12 (a − aT ), (1.54)

218 it is easy to verify that the tensors

sym(a) = 21 (a + aT ), skew(a) = 21 (a − aT ), (1.55)

219 are symmetric and skew-symmetric, respectively. In components, we have

[sym(a)]ij = 21 (aij + aji ), [skew(a)]ij = 21 (aij − aji ). (1.56)

220 All definitions about transpose of a tensor, symmetric and skew-symmetric tensors, and the
221 decomposition into symmetric and skew-symmetric parts that have been given for a “lowercase”
222 second-order tensor a ∈ Hom(V, V) can naturally be given for an “uppercase” second-order
223 tensor A ∈ Hom(U, U).

224 Remark 1.3. It is very important to note that, since the two-point tensor F ∈ Hom(U, V) and
225 its transpose F T ∈ Hom(V, U) have different (in fact, switched) domain and codomain, the
226 question “is a certain two-point tensor is symmetric/skew-symmetric?” is meaningless. Indeed,
227 while F contracts with an “uppercase” vector on the right and a “lowercase” one on the left, F T
228 does exactly the opposite. The question “how does one extract the symmetric/skew-symmetric
229 part of a certain two-point tensor?” is equally meaningless, as we cannot sum (or subtract) F
230 and F T because they are objects of a different type.

9
Box 1.1. Skew-Symmetric Tensors and the Cross-Product. The action of the
skew-symmetric tensor Ω on a vector u is such that

Ω u = ω × u, (1.57)

where ω is called the axial vector of Ω. Indeed, by virtue of Equation (1.57) and the
definition of skew-symmetry, we have, for every vector z,

z.(ΩT u) = u.(Ω z) = u.(ω × z) = −z.(ω × u) = −z.(Ω u), (1.58)

where we used the cyclic property of the triple product u.(ω × z). In components, the
triple product is given by
u.(ω × z) = ijk ui ωj zk , (1.59)
where ijk is the Ricci/Levi-Civita symbol, also called permutation symbol, such that

 1 for i, j, k even permutation of 1, 2, 3,
ijk = −1 for i, j, k odd permutation of 1, 2, 3, (1.60)
0 for i, j, k with repetition of 1, 2, 3.

With the Levi-Civita symbol, the relation between the components of the skew symmetric
tensor Ω and its axial vector ω is given by

Ωik = ijk ωj , (1.61)

so that the matrix of Ω is  


0 −ω3 ω2
[Ω] =  ω3 0 −ω1  . (1.62)
−ω2 ω1 0

231 A two-point tensor R ∈ Hom(U, V) is called orthogonal if it preserves the scalar product, i.e.,
232 if for every W , Y ∈ U,
(R W ).(R Y ) = W .Y , (1.63)
233 where we notice that the scalar product on the left-hand side is that in V and the scalar product
234 on the right-hand side is that in U.
235 From the definition of transpose, and the symmetry of the scalar product, we have

(R W ).(R Y ) = Y .[RT (R W )] = (RT R W ).Y . (1.64)

236 If R is orthogonal, Equations (1.63) and (1.64) yield

(RT R W ).Y = W .Y , (1.65)

237 from which, for the arbitrariness of W and Y , it must follow that

RT R = I. (1.66)

238 By comparison of Equation (1.66) with the definition of inverse (Equation (1.34)), we obtain
239 the fundamental property of orthogonal tensors,

RT = R−1 , (1.67)

10
240 i.e., the inverse of an orthogonal tensor equals its transpose. Analogously, by virtue of the first
241 of Equations (1.34) and Equation (1.67), we have

R RT = R R−1 = i. (1.68)

242 The identities in both Equations (1.66) and (1.68) obviously apply to the representing matrices.
243 From Equation (1.66) (or, alternatively, from Equation (1.68)), it follows that

det(RT R) = det(RT ) det(R) = det(I) = 1, (1.69)

244 and, since the determinant of the transpose RT is equal to that of R, we have that, for an
245 orthogonal tensor,
det(R) ± 1. (1.70)
246 Orthogonal tensors with determinant equal to +1 are called proper rotations, because they
247 preserve the orientation. In contrast, orthogonal tensors with determinant equal to −1 are
248 called improper rotations.
249 Analogously, endomorphisms, i.e., “lowercase” or “uppercase” tensors, are orthogonal if they
250 preserve the scalar product within the same space. For every w, y ∈ V,

(q w).(q y) = w.y, (1.71)

251 so that q T = q −1 , with det(q) = ±1, and, for every W , Y ∈ U,

(Q W ).(Q Y ) = W .Y , (1.72)

252 so that QT = Q−1 , with det(Q) = ±1.


253 In this course, we shall exclusively deal with proper orthogonal tensors.

254 1.6 The Tensor Product

255 Let V be a vector space, considered with an orthonormal basis {ei }ni=1 . We define the tensor
256 product of two vectors u, v ∈ V as the second-order tensor u ⊗ v (which reads “u tensor v”)
257 such that
(u ⊗ v) w = (v.w) u. (1.73)
258 We call u and v the first leg and the second leg of u ⊗ v, and we say that the tensor product
259 u ⊗ v applied onto w gives the first leg u multiplied by the scalar projection of the argument w
260 onto the second leg v. We can also say that we contract the second leg v with the argument w
261 and leave the first leg u free. We remark that the definition of tensor product given in Equation
262 (1.73) holds exclusively for the case of orthonormal bases.
263 Any second order tensor a ∈ Hom(V, V) can be expressed as the linear combination

a = aij ei ⊗ ej , (1.74)

264 where aij are the components of a with respect to the basis {ei }ni=1 in the domain and the
265 codomain. Indeed, given an arbitrary vector v ∈ V, by using the definition of the components
266 of a via the scalar product (Equation (1.44)) and the definition of tensor product (Equation
267 (1.73)), we have

a v = vj a ej = aij vj ei = aij (ej .v) ei = aij (ei ⊗ ej ) v, (1.75)

11
268 which, for the arbitrariness of v, yields Equation (1.74). This means that the tensors
269 {ei ⊗ ej }ni,j=1 constitute a tensor basis for the n2 -dimensional space Hom(V, V) of second-order
270 tensors.
271 The tensor representation of the identity is

i = ei ⊗ ei = δij ei ⊗ ej . (1.76)

272 i.e., its components are given by the Kronecker delta. This is easy to verify:

δij (ei ⊗ ej )v = δij (ej .v) ei = δij vj ei = vi ei = v = i v. (1.77)

273 Similarly, for the case of “uppercase” tensors A, we have that the tensor basis of Hom(U, U)
274 is {E I ⊗ E J }nI,J=1 , with the representation

A = AIJ E I ⊗ E J , (1.78)

275 and, for two-point tensors F , we have that the tensor basis of Hom(U, V) is {ei ⊗ E I }ni,I=1 ,
276 with the representation
F = FiI ei ⊗ E I . (1.79)

277 Problem 1.3. By using the definition of tensor product, prove that it satisfies the properties:
278

(u ⊗ v)T = (v ⊗ u), (1.80a)


a (u ⊗ v) = (a u) ⊗ v, ∀a ∈ Hom(V, V), (1.80b)
T
(u ⊗ v) a = u ⊗ (a v), ∀a ∈ Hom(V, V), (1.80c)
u ⊗ (v + w) = u ⊗ v + u ⊗ w. (1.80d)

279 Note that (v + w) ⊗ u = v ⊗ u + w ⊗ u can be proved using (1.80a) and (1.80d).

280 Problem 1.4. By using the definition of tensor product, verify that, for every second-order
281 tensor F ∈ Hom(U, V), and every vector U ∈ U, the i-th component of F U is FiI UI . Hint: use
282 F = FiI ei ⊗ E I and U = UK E K . Do the same for a second-order tensor a ∈ Hom(V, V).

283 1.7 Fourth-Order Tensors

284 Text

12
285 Chapter 2

286 Continuum Mechanics

287 Continuum Mechanics is the study of matter at scales much larger than the atomic scale.
288 At these scales, one always looks at portions of matter comprised an exceptionally high, but
289 finite number of atoms, and therefore can “pretend” that atoms do not exist, and matter is
290 continuous. Mathematically, continuity of a body means that, for every two particles however
291 close within the body, there is always another particle (and thus, by induction, an infinity
292 of particles) in between. This implies that the concept of density has to be introduced for
293 describing extensive quantities such as mass, and that one can use the tools of Differential and
294 Integral Calculus on a body, which is regarded as an open set.
295 In this chapter, we shall give a brief description of physical space as an affine space, a
296 concept derived from that of vector space (Section 2.1), introduce deformable body and its
297 configurations, as well as various measures of deformation and rates of deformation (Section
298 2.2), introduce stress and balance equations (Section 2.3), and finally introduce constitutive
299 equations, relating stress and strain in some specific materials (Section 2.4).

300 2.1 Physical Space and Bodies

301 Text

302 2.2 Kinematics

303 We must always keep in mind that deformation is an aspect of kinematics and, as such, it
304 must be studied in a manner completely independent from the causes of the motion. However,
305 in many textbooks (including, alas, the textbooks normally used in second-year Mechanics
306 of Solids courses), deformation is typically seen as “something caused by stresses”. This is a
307 distorted view, that prevents understanding deformation and strain as kinematical quantities.
308 Kinematics is indeed the study of motion and deformation regardless of their causes. Thus, the
309 study of the kinematics of continuum bodies (i.e., of motion and deformation) is completely
310 independent of stress and external forces, exactly as the kinematics of particles is completely
311 independent of forces.

13
312 2.2.1 Configuration Map and Deformation Gradient

313 Let a deformable body be represented by some convenient reference configuration B. As in


314 Section 2.1, points in the body B are denoted X, and points in the physical space S are
315 denoted x. A configuration of the body B is a map

χ : B → S : X 7→ x = χ(X). (2.1)

316 Since x = (x1 , x2 , x3 ) and X = (X1 , X2 , X3 ), we can write, in components,

χi : B → S : (X1 , X2 , X3 ) 7→ xi = χi (X1 , X2 , X3 ) = χi (X). (2.2)

317 The image of B through the configuration χ is a subset of S denoted χ(B), and is called the
318 current configuration of the body B (Figure 2.1). The map χ is called configuration map or
319 simply configuration, as already mentioned. Note that, for the moment, we are leaving time
320 out of the picture, i.e., we are considering a configuration χ at a fixed instant of time.
321 In order for the configuration map to be physically admissible, some mathematical requirements
322 must be satisfied:

323 1. The map χ must be continuous;

324 2. The map χ must be invertible;

325 3. The inverse χ−1 must be continuous;

326 4. The map χ and its inverse χ−1 must be differentiable.

327 Note that invertibility means that one is able to map the current into the reference
328 configuration, i.e., the motion of the body can be reversed. This also implies that nothing
329 “strange” happens, e.g., impedes that two material points X and X 0 are mapped into the same
330 spatial point x. A map satisfying the the four requirements above is called a diffeomorphism,
331 i.e., it is a differentiable homeomorphism (a map is a homeomorphism when it satisfies the first
332 three requirements).
333 For functions f : D ⊆ R → R (often called “real functions of a real variable”) differentiability
334 is equivalent to derivability. In contrast, as soon as the domain has at least two dimensions
335 (or, equivalently, there are at least two variables), derivability in a certain direction is a weaker
336 condition than differentiability, meaning that derivability with respect to any direction does not
337 guarantee differentiability. In the following, we shall define concepts of directional derivative
338 and differential for the configuration map χ of Continuum Mechanics, but these concept are
339 completely general.
340 The configuration map χ is said to be derivable at point X with respect to the direction
341 W ∈ TX B (we recall that this means that the vector W is attached at point X) if the limit
χ(X + hW ) − χ(X)
(∂W χ)(X) = lim , (2.3)
h→0 h
342 exists. The limit (2.3) is called directional derivative of the configuration map χ at point X in
343 the direction W ∈ TX B. It is very important to note that, since the numerator of the limit is
344 the vector χ(X + hW ) − χ(X), which is attached at point x = χ(X), the directional derivative
345 (∂W χ)(X) belongs to ∈ Tx S, i.e., is a vector attached at x = χ(X) (See again Figure 2.1).

14
Box 2.1. Partial Derivatives as Directional Derivatives. We note that, since (∂E I χ)(X) is
a vector attached at x (i.e., a vector in Tx S), its i-th component is obtained via the scalar
product with the basis vector ei (see Equation (1.44)), and equals the derivative of the
i-th component χi of χ, with respect to E I , i.e.,

ei .[(∂E I χ)(X)] = (∂E I χi )(X), (2.4)

Equation (2.4) represents nothing but the so-called partial derivative of the i-th component
χi of χ with respect to the I-th variable. Indeed, by using the definition (2.3) of directional
derivative, and by expressing the points X as X = (X1 , X2 , X3 ), we have, for the case of,
e.g., the derivative with respect to E 1 = (1, 0, 0),

χi (X + hE I ) − χi (X) χi (X1 + h, X2 , X3 ) − χi (X1 , X2 , X3 )


(∂E 1 χi )(X) = lim = lim ,
h→0 h h→0 h
(2.5)
which is the definition of partial derivative “with respect to X1 ” in the traditional
terminology (usually, in the limit above, many textbooks use ∆XI in place of h). Possible
notations for the partial derivative (∂E I χi )(X) are

∂χi
(∂E I χi )(X) ≡ χi,I (X) = (X). (2.6)
∂XI

We shall normally use χi,I (X) for the I-th partial derivative of χi at X, but will occasionally
report also the form (∂χi /∂XI )(X), which is usually used in more elementary courses.
Warning: Some books report the form FiI = ∂xi /∂XI , the use of which we strongly
discourage, as it is based on the unfortunately diffused confusion between function (χi , in
this case) and value (xi = χi (X)).

346 The configuration map χ is said to be differentiable at point X if, and only if, there exists a
347 two-point tensor F (X) ∈ Hom(TX B, Tx S) such that, for every vector W ∈ TX B,

[F (X)]W = (∂W χ)(X), (2.7)

348 i.e., if the directional derivative is a linear function of the direction W . It follows that, if a
349 function is differentiable, then it is derivable with respect to any direction W . In the context
350 of Continuum Mechanics, the differential F (X) is called the deformation gradient at point
351 X. We note that the nomenclature gradient is an abuse of terminology and is universally
352 accepted essentially because of tradition. Indeed, χ is not a vector field, but a point mapping,
353 and only tensor fields (of order zero, i.e., scalars, of order one, i.e., vectors, and higher-order
354 tensors) admit a gradient (a very advanced treatment of this topic in the context of Continuum
355 Mechanics can be found, e.g., in Marsden and Hughes, 1983). The deformation gradient F (X),
356 differential of the configuration map χ at point X, is more properly called the tangent map of
357 χ, because it maps tangent vectors attached at X into tangent vectors attached at x = χ(X),
358 and can be also denoted by (T χ)(X) ≡ F (X) ∈ Hom(TX B, Tx S).
359 Therefore, if we denote the spatial vector [F (X)]W = (∂W χ)(X) ∈ x = χ(X), directional
360 derivative of χ in direction W at X, by w, and we omit the argument X for the sake of a
361 lighter notation, we obtain
w = FW, (2.8)

15
362 and say that the spatial vector w is the push-forward of the material vector W .
363 The components FiI (X) of the deformation gradient F (X) are the partial derivatives χi,I (X).
364 Indeed, using (2.4), (2.6), (2.7) and the expression (1.45) of the components of a two-point
365 tensor when a scalar product is defined, we have that

χi,I (X) = ei .[(∂E I χ)(X)] = (∂E I χi )(X) = ei .[F (X)]E I = FiI (X). (2.9)

366 The geometrical and physical meaning of the deformation gradient F is easier to understand if
367 one thinks as the material vector W as the tangent vector at point X of some material curve
368 Γ ⊂ B passing by X (Figure 2.1). With this in mind, the push-forward w = F W is the tangent
369 at point x = χ(X) to the spatial curve γ = χ(Γ) ⊂ χ(B), which is nothing but the deformed
370 configuration of the material curve Γ, precisely as χ(B) is the deformed configuration of B. The
371 direction of w accounts for the change in direction with respect to W . The ratio kwk/kW k
372 between the norm of the deformed vector w and the norm of the referential vector W accounts
373 for the amount of stretch that has occurred along the curve Γ, at point X.

F w

W
X3
X x3
(B) ⇢ S
E3 e2 x2
E 2 X2 e3

e1 x1
E1
B
X1

Figure 2.1: A deformable body, represented by a convenient reference configuration B, and its current
configuration χ(B) ⊂ S, where χ is the configuration map, mapping material points X into spatial
points x = χ(X). The deformation gradient F pushes-forward material vectors W into spatial vectors
w = FW.

374 Since χ is a diffeomorphism, its inverse

χ−1 : χ(B) → B : x 7→ X = χ−1 (x). (2.10)

375 is also differentiable, and its differential (or tangent map) at point x is the two-point tensor
376 f (x) ∈ Hom(Tx S, TX B) defined by

[f (x)]w = (∂w χ−1 )(x), (2.11)

377 which maps the spatial vector w attached at point x into the material vector (∂w χ−1 )(x)
378 attached at X = χ−1 (x), where (∂w χ−1 )(x) is the directional derivative of χ−1 with respect to
379 w at x. Analogously to the case of F , we have that the component expression of f is

fIi (x) = χ−1


I,i (x). (2.12)

16
380 We shall now prove that the tangent map f of the inverse χ−1 is nothing but the inverse of
381 the tangent map F , i.e.,
f (x) = [F (X)]−1 (2.13)
382 We do this by noting that the composition of χ−1 and χ is the identity map in the body B,
383 i.e., the map Id : B → B : X 7→ Id(X) = X. Thus,

X = Id(X) = χ−1 (x) = χ−1 (χ(X)) = χ−1 (χ(X)), (2.14)

384 from which, for the arbitrariness of X,

Id = χ−1 ◦ χ. (2.15)

385 In components, we have


IdI = χ−1
I ◦ χ. (2.16)
386 where IdI (X) = XI . If we take the partial derivative with respect to the J-th variable, and use
387 the chain rule, we obtain
δIJ = IdI,J = [χ−1
I,j ◦ χ] χj,J , (2.17)
388 It could be of some help to note that, in the traditional notation, IdI,J = ∂XI /∂XJ = δIJ .
389 Now, by definition of F and f , Equation (2.17) becomes

δIJ = [fIj ◦ χ] FjJ , (2.18)

390 which, in component-free notation, reads

I = [f ◦ χ] F . (2.19)

391 By definition of inverse, this implies that

f = F −1 . (2.20)

392 In the following, we shall use the customary notation F −1 in place of f . Analogously to the
393 case of F , if we denote the material vector [F −1 (x)]w ∈ TX B, directional derivative of χ−1 in
394 direction w at x, by W , and we omit the argument x for the sake of a lighter notation, we
395 obtain
W = F −1 w, (2.21)
396 and say that the material vector W is the pull-back of the spatial vector w.
397 Finally, we note that, since the deformation gradient is invertible, its determinant

J = det F (2.22)

398 must be different from zero. In Section 2.2.4 we shall show that the physical meaning of J is
399 that of change of volume. Therefore, in order to prevent compenetration of matter, J must
400 be not only non-zero but also strictly positive. Indeed, if J were negative, a volume would be
401 turned “inside out”.

17
402 2.2.2 Deformation and Strain Tensors

403 2.2.3 The Polar Decomposition Theorem

404 It can be shown (Cauchy’s Polar Decomposition Theorem) that F can be decomposed as
405 the product of an orthogonal tensor, representing pure rotation, and a symmetric tensor,
406 representing pure deformation, in two equivalent ways:

F = R U, FiJ = RiI UIJ , deformation followed by a rotation, (2.23a)


F = V R, FiJ = Vij RjJ , rotation followed by a deformation. (2.23b)

407 Notice that both indices of tensor U are large, which means it is a completely material tensor,
408 and both indices of tensor V are small, which means it is a completely spatial tensor. In
409 contrast, R has, like F , the first index large and the second index small, which makes it a
410 two-point tensor. In rigid body mechanics there is no deformation: therefore U and V coincide
411 with the material and spatial identity, respectively, and F = R, i.e., a rigid body can undergo,
412 aside from translation, only rotations.

413 2.2.4 Change of Volume and Area

414 Text

415 2.2.5 Velocity and Time Rates

416 When velocity and all physical quantities related to velocity have to be defined, the
417 configuration χ must be described as a function not only of the material point X but also
418 of time t. Therefore, from now on, we shall look at the configuration as a map

χ : B × R+
0 → S : (X, t) 7→ x = χ(X, t). (2.24)

419 At a fixed instant of time t, the point map

χ( · , t) : B → χ(B, t) ⊂ S : X 7→ x = χ(X, t), (2.25)

420 is the diffeomorphism described in Section 2.2.1. At a fixed material point X, we have the curve
421 (in general, any function of a single real parameter is called a “curve”)

χ(X, · ) : R+
0 → S : t 7→ x = χ(X, t), (2.26)

422 which describes the trajectory of the material point X in time, as shown in Figure 2.2. Note,
423 in Equations (2.25) and (2.26), the use of the “dot” in function of two or more variables to
424 indicate which of the variables is not being fixed. In the former case, t is being fixed, resulting
425 in a function defined on B, i.e., a function of X only, and vice versa in the latter case.
426 It is also convenient to introduce the time map

τ : B × R+ +
0 → R0 : (X, t) 7→ t, (2.27)

427 which simply discards the X argument and behaves as the identity for the time t. Its use will
428 be shown immediately below.

18
(X, t + h) (X, t)
x = (X, t)
x0 = (X, t + h)

Figure 2.2: The trajectory described by the motion of a material point X; the point is placed at x =
χ(X, t) at time t, and at x0 = χ(X, t + h) at time t0 = t + h. The displacement after the increment of
time h is x0 − x = χ(X, t + h) − χ(X, t). The Lagrangian velocity is given by the limit of the incremental
ratio [χ(X, t + h) − χ(X, t)]/h for h → 0, and is, by definition of derivative, tangent to the trajectory
hat x.

429 Eulerian and Lagrangian Points of View

430 All physical quantities are, by nature, spatial in the sense that they are defined in the physical
431 space S, i.e., they are a function of the spatial points x, and, if they are vector or tensor
432 quantities, they are attached at point x. If the motion of a body is under study, one looks at
433 the spatial point x located in a region Ω ⊂ S and measures the physical quantities of interest
434 regardless of which specific particle is occupying point x at that particular time t. This is the
435 point of view taken by Euler.
436 However, by means of the configuration χ, one can relate a spatial point x with the particle
437 X whose placement at time t is precisely x = χ(X, t). In this way, one can track the values of
438 a physical quantity at the positions x = χ(X, t) that a particle X takes along its trajectory,
439 as time t passes. This is the point of view taken by Lagrange. Thus, if f is, e.g., an Eulerian
440 scalar field defined, at each time t, by

f ( · , t) : χ(B, t) → R : x 7→ f (x, t), (2.28)

441 its Lagrangian form is

f ◦ (χ, τ ) : B × R+
0 → R : (X, t) 7→ f (χ(X, t), τ (X, t)) = f (x, t), (2.29)

442 which is obtained by composing f with the map

(χ, τ ) : B × R+ +
0 → S × R0 : (X, t) 7→ (x, t). (2.30)

443 Indeed, the map (χ, τ ), i.e., the pair constituted by the configuration map χ and the time map
444 τ , changes the variables from (X, t) to (x, t).
445 For the case of a vector or tensor field, some extra care is needed in understanding what we
446 mean here by Lagrangian and Eulerian. Indeed, a spatial vector field is a map such that, at
447 each time t,
w( · , t) : χ(B) → T S : x 7→ w(x, t) ∈ Tx S, (2.31)
448 where Tx S is the tangent space at point x = χ(X, t), i.e., the vector space of all vectors attached
449 at x = χ(X, t), and T S is the collection of the tangent spaces at all points x ∈ S and is called
450 the tangent bundle of S. We remark that w of Equation (2.31) is Eulerian in the sense described
451 above, i.e., it is a function of the current point x. The Lagrangian form of w is

w ◦ (χ, τ ) : B × R+
0 → Tχ( · ,t) S : (X, t) 7→ w(χ(X, t), τ (X, t)) = w(x, t) ∈ Tx S, (2.32)

19
452 and it is Lagrangian in the sense that it is a function of X, but it is still spatial, because it is
453 still valued in the tangent space Tx S at the spatial point x = χ(X, t).
454 In order to obtain a form of w that is Lagrangian and material, a pull-back operation is
455 necessary, which gives
W (X, t) = F −1 (x, t) w(x, t), (2.33)
456 where W is the vector field
W : B × R+
0 → T B : (X, t) 7→ W (X, t) ∈ TX B, (2.34)
457 which is material because it is valued in the tangent space TX B, i.e., the vector space of all
458 vectors attached at the material point X (analogously to T S, T B is called the tangent bundle
459 of B). By using the map (χ, τ ), the pull-back (2.33) can be expressed without indicating the
460 arguments, as
W = [F −1 ◦ (χ, τ )] [w ◦ (χ, τ )], (2.35)
461 which in components is
WI = [(F −1 )Ii ◦ (χ, τ )] [wi ◦ (χ, τ )], (2.36)

462 Comparison of Equation (2.32) with Equations (2.35) and (2.36) clearly shows the difference
463 between what we mean by Lagrangian and material. With Lagrangian we merely indicate the
464 change of variables from (x, t) to (X, t), while the vector is still attached at the spatial point
465 x = χ(X, t), whereas with material we mean that the vector is attached at the material point
466 X. Usually, we represent a material vector or tensor field exclusively as Lagrangian, i.e., as a
467 function of X.
468 Remark 2.1. In most textbooks, Lagrangian and material are considered as synonymous, and
469 so are Eulerian and spatial. However, we prefer to maintain the distinction under the light of
470 the considerations made above.

471 Lagrangian and Eulerian Velocity

472 In particle mechanics, velocity is defined as the time derivative of the trajectory of the particle,
473 and is represented as a vector tangent to the trajectory and attached at the current position
474 of the particle. This very same definition is used in Continuum Mechanics.
475 With reference again to Figure 2.2, if we fix the point X in the configuration χ, we obtain the
476 trajectory χ(X, · ), which maps each time t into the current position x = χ(X, t) of the particle
477 X. By definition, the velocity of the particle X is given by the time derivative
χ(X, t + h) − χ(X, t)
∂t χ(X, t) = lim . (2.37)
h→0 h
478 Indeed, as shown in Figure 2.2, the numerator of the incremental ratio in Equation (2.37)
479 is the displacement, which is a vector attached at the current placement x = χ(X, t) of the
480 material point X. Moreover, when the time increment h tends to zero, the limit of the ratio of
481 the displacement χ(X, t + h) − χ(X, t) to h tends to be tangent to the trajectory of X at the
482 current placement x = χ(X, t) of X, and is precisely the velocity.
483 The velocity defined in Equation (2.37) is the Lagrangian velocity of the material point X, and
484 we must emphasise that it is a spatial vector, as it is attached at x = χ(X, t) (see Figure 2.2).
485 The Eulerian velocity is defined as the spatial vector field such that
v(x, t) = ∂t χ(X, t). (2.38)

20
486 With use of the map (χ, τ ), the relation between the Eulerian velocity v and the Lagrangian
487 velocity ∂t χ can be expressed as
v ◦ (χ, τ ) = ∂t χ, (2.39)
488 i.e., the Eulerian velocity v is the vector field such that its Lagrangian form is the Lagrangian
489 velocity ∂t χ.

490 Remark 2.2. In many textbooks, what we call Lagrangian velocity, based on the fact that it
491 is a function of X, is often named “material velocity”, although it is clearly a spatial vector
492 field. We find confusing the use of the term “material” in this case, and this is another reason
493 why we prefer to distinguish between Lagrangian and material.

494 Substantial Derivative

495 In the Eulerian point of view, one fixes the point x and, regardless of which particle X is
496 passing by that point, looks at the values of a certain physical point at that point, as time
497 passes. In this view, the regular partial time derivative suffices to study the variation of a
498 certain physical quantity in time. However, when one wants to account for the variations of
499 that physical quantities due to the fact that different particles X pass by the point x at each
500 time, an additional term arises which, together with the regular partial time derivative, gives
501 a new time derivative operator, called substantial derivative.
502 Let us start with the definition for a scalar field f , but we shall see that the definition can
503 be generalised easily to vectors and tensors. We say that the substantial derivative of f is the
504 Eulerian form of the partial time derivative of the Lagrangian form of f , i.e., we are looking
505 for the derivative operator Dt such that

∂t [f ◦ (χ, τ )] = [Dt f ] ◦ (χ, τ ). (2.40)

506 Using the chain rule, time derivative of the Lagrangian scalar field f ◦ (χ, τ ) reads

∂t [f ◦ (χ, τ )] = f,i ◦ (χ, τ ) ∂t χi + (∂t f ) ◦ (χ, τ ) ∂t τ


= f,i ◦ (χ, τ ) vi ◦ (χ, τ ) + (∂t f ) ◦ (χ, τ ) 1
= [f,i vi + ∂t f ] ◦ (χ, τ ), (2.41)

507 where we have used the identity vi ◦(χ, τ ) = ∂t χi for the i-th component of the velocity (compare
508 with Equation (2.39)) the fact that ∂t τ = 1 (coming from τ (X, t) = t), and the distributive
509 property of function composition with respect to multiplication and sum of functions in

f,i ◦ (χ, τ ) vi ◦ (χ, τ ) + (∂t f ) ◦ (χ, τ ) = [f,i vi + (∂t f )] ◦ (χ, τ ). (2.42)

510 Note that, with the traditional notation for the derivatives, the passage involving the chain
511 rule in Equation (2.41) read

∂ ∂f ∂χi ∂f ∂τ
[f ◦ (χ, τ )] = ◦ (χ, τ ) + ◦ (χ, τ ) . (2.43)
∂t ∂xi ∂t ∂t ∂t
512 The substantial derivative of f is found by comparing Equations (2.40) and (2.41) as

∂t [f ◦ (χ, τ )] = [f,i vi + ∂t f ] ◦ (χ, τ ) = Dt f ◦ (χ, τ ), (2.44)

21
513 from which

Dt f = f,i vi + ∂t f. (2.45)

514 Since f,i are the components of the gradient of f , and f,i vi = (gradf ).v, we have

Dt f = (gradf ).v + ∂t f (2.46)

515 The result in Equations (2.45) and (2.46) can be generalised to tensor fields of any order. For
516 the case of a vector field w, we have

Dt wi = wi,j vj + ∂t wi , (2.47)

517 and, considering that wi,j are the components of the gradient of w, which is a second-order
518 tensor, we have

Dt w = (gradw)v + ∂t f (2.48)

519 where the gradient gradw acts on the velocity v as a linear map. In the traditional notation
520 for the derivatives, Equations (2.45) and (2.47) read

Df ∂f ∂f Dwi ∂wi ∂wi


= vi + , = vj + , (2.49)
Dt ∂xi ∂t Dt ∂xj ∂t

521 Remark 2.3. Very often, particularly in Solid Mechanics books, the substantial derivative
522 is called “material derivative”, because it makes use of what we call the Lagrangian velocity,
523 which most textbooks call “material velocity”. We prefer not to use the terminology “material
524 derivative” because it is not applied to material quantities, but to spatial quantities and there
525 is really nothing material in it. Indeed, while we defined the substantial derivative by exploiting
526 the Lagrangian form of a field, as it is usually done in Solid Mechanics, it is actually possible
527 to use the concept of flow of the velocity to define the substantial derivative without using any
528 reference configuration. This is in fact the way in which the substantial derivative is thought
529 of in Fluid Mechanics.

Box 2.2. The Chain Rule in Most Textbooks. In many textbooks, the chain rule as applied,
e.g., in Equation (2.41) is written as

df ∂f ∂xi ∂f ∂f ∂f
f˙ = = + = vi + , (2.50)
dt ∂xi ∂t ∂t ∂xi ∂t
where df /dt is called the “total derivative” of f . This looks familiar and, at first sight,
convincing. However, notice that there is no distinction between the Lagrangian (in f˙ =
df /dt) and the Eulerian forms of f (in the remainder of the equation), which are both
called f , that there is confusion between χi and its value xi = χi (X, t), and that it is not
immediately clear why ∂f /∂t is multiplied by nothing. We find this way to write the chain
rule very confusing, but we report it because it is the formalism used in most textbooks
and in most journal articles.

22
530 Lagrangian and Eulerian Acceleration

531 The Lagrangian acceleration is the time derivative of the Lagrangian velocity

∂t2 χ = ∂t [v ◦ (χ, τ )]. (2.51)

532 The Eulerian acceleration is defined as the vector field such that its Lagrangian form is the
533 Lagrangian acceleration (compare with the case of the Eulerian velocity, Equation (2.39)), i.e.,

a ◦ (χ, τ ) = ∂t2 χ = ∂t [v ◦ (χ, τ )]. (2.52)

534 By definition of substantial derivative for a vector field (Equation (2.48)), we have

a ◦ (χ, τ ) = ∂t [v ◦ (χ, τ )] = [Dt v] ◦ (χ, τ ). (2.53)

535 and therefore


a = Dt v = (gradv) v + ∂t v, (2.54)
536 which has the component form
ai = vi,j vj + ∂t vi . (2.55)
537 In the traditional notation, Equation (2.55) reads
∂vi ∂vi
ai = vj + . (2.56)
∂xj ∂t

538 Velocity Gradient

539 We define the velocity gradient as the spatial tensor field

l = grad v, lij = vi,j . (2.57)

540 The velocity gradient l can be decomposed into its symmetric and antisymmetric parts as

l = d + w, lij = dij + wij , (2.58)

541 where (see Equation (1.55))

d = 21 (l + lT ) dij = 12 (vi,j + vj,i ), (2.59)

542 and
w = 21 (l − lT ) wij = 21 (vi,j − vj,i ) (2.60)
543 are called deformation rate and spin tensor, respectively. In the traditional notation for
544 derivatives, the component expressions of l, d and w read
   
∂vi 1 ∂vi ∂vj 1 ∂vi ∂vj
lij = , dij = + , wij = − , (2.61)
∂xj 2 ∂xj ∂xi 2 ∂xj ∂xi
545 respectively.
546 We notice that the trace of the velocity gradient l and the trace of the deformation rate d
547 are equal to each other (the spin tensor w, being skew-symmetric, is traceless) and equal the
548 divergence of the velocity v. Indeed,

tr l = lii = dii + wii = dii = tr d (2.62)

23
549 and
tr l = tr d = lii = vi,i = div v. (2.63)
550 In the traditional notation, the divergence of v reads
∂vi
div v = . (2.64)
∂xi

551 Moreover, we note that the axial vector of the spin tensor w is the vorticity ω, such that, for
552 every vector u (see Equations (1.57) and (1.61)),

w u = ω × u, wik uk = ijk ωj uk . (2.65)

553 Finally, when d = o, we say that the motion is rigid, and the vorticity ω takes the name of
554 angular velocity and, when instead w = o, which implies ω = 0, we say that the motion is
555 irrotational.

556 Time Rate of the Deformation Gradient

557 The time rate Ḟ ≡ ∂t F of the deformation gradient F can be expressed in terms of the velocity
558 gradient l. Indeed, in components, we have

ḞiJ ≡ ∂t FiJ = ∂t χi,J = (∂t χi ),J = [vi ◦ (χ, τ )],J . (2.66)

559 where we used Schwartz’ theorem on the commutativity of the second partial derivatives for
560 sufficiently regular functions (χ is a diffeomorphism and is often assumed to be even twice
561 differentiable) and the definition of spatial velocity (2.39). The derivative of the composite
562 function in Equation (2.66) is obtained by means of the chain rule, i.e.,

[vi ◦ (χ, τ )],J = vi,j ◦ (χ, τ ) χj,J + (∂t vi ) ◦ (χ, τ ) τ,J = lij ◦ (χ, τ ) FjJ . (2.67)

563 where we used τ,J = 0 (since τ (X, t) = t, τ has zero derivative with respect to anything else
564 than time), vi,j = lij and χj,J = FjJ . Therefore, by comparing Equations (2.66) and (2.67), we
565 obtain
ḞiJ = lij ◦ (χ, τ ) FjJ . (2.68)
566 and, in component-free notation,
Ḟ = l ◦ (χ, τ ) F . (2.69)
567 Following a very common abuse of notation, we can omit to write the composition by (χ, τ ) in
568 Equation (2.69), and write
Ḟ = l F . (2.70)

569 The use of Schwartz’ theorem on the commutativity of second partial derivatives in Equation
570 (2.66) is perhaps more evident in the traditional notation for derivatives:
∂FiJ ∂ ∂χi ∂ ∂χi ∂
ḞiJ ≡ = = = [vi ◦ (χ, τ )]. (2.71)
∂t ∂t ∂XJ ∂XJ ∂t ∂XJ
571 In the same notation, the derivative of the composite function vi ◦ (χ, τ ) reads
∂ ∂vi ∂χi ∂vi ∂τ
[vi ◦ (χ, τ )] = ◦ (χ, τ ) + ◦ (χ, τ ) = lij ◦ (χ, τ ) FjJ . (2.72)
∂XJ ∂xj ∂XJ ∂t ∂XJ

24
Box 2.3. Again on the Chain Rule in Most Textbooks. The chain rule (oh, what a
mistreated mathematical tool!) in Equations (2.67) and (2.73) is often written

∂vi ∂vi ∂xi ∂vi ∂t


= + = lij FjJ + 0. (2.73)
∂XJ ∂xj ∂XJ ∂t ∂XJ

Again, this looks familiar and convincing, but there is no way to distinguish the velocity
component vi seen as a function of X (what we call Lagrangian, left-hand side) and as a
function of x (what we call Eulerian, remainder of the equation). Moreover, there is the
unfortunately usual confusion between χi and its value xi = χi (X, t), and time t is treated
like a function and not like a real number in ∂t/∂XJ .

572 Time Rate of Deformation and Strain

573 The time rates of deformation and strain tensors can be found in terms of that of the
574 deformation gradient and of the velocity gradient. With the usual abuse of notation (i.e.,
575 writing l in place of l ◦ (χ, τ ), the time derivative of C is given by
T
Ċ ≡ ∂t C = (F T F )˙ = Ḟ F + F T Ḟ = F T (lT + l) F = F T (l + lT ) F
= F T (2 d) F = F T (2d)F = 2 F T d F , (2.74)

576 where we used the definition (2.59) of the deformation rate d. Therefore, the time rate of the
577 right Cauchy-Green Deformation is twice the pull-back of the deformation rate d:

Ċ = 2 F T d F , ĊIJ = 2 FiI dij FjJ . (2.75)

578 i.e., it is the pull-back of twice the deformation rate d. Since E = 21 (C − I) and İ = O, we
579 have
Ė = 12 Ċ, (2.76)
580 and thus the rate of the Green-Lagrange strain is the pull-back of d:

Ė = F T d F , ĖIJ = FiI dij FjJ . (2.77)

581 Note that, in Equations (2.74), (2.75) and (2.77), we wrote, with abuse of notation, d in place
582 of d ◦ (χ, τ ), as we had previously done in Equation (2.70).

583 Time Rate of The Volume Ratio

584 For the moment, we just report the final expression

J˙ = J (div v) ◦ (χ, τ ), (2.78)

585 or, with the usual abuse of notation,

J˙ = J div v. (2.79)

25
586 2.3 Stress, Balance and Power

587 This section is devoted to the study of stress and of the balance equations for continuum bodies,
588 and concludes with the notions of external and internal power.
589 Stress is the internal response of a continuum body to external loads and is a tensorial quantity.
590 In fact, the very word tensor was coined in 1899 by Woldemar Voigt (German physicist,
591 1850-1919), and is a sort of hybrid between the Latin words tensio (tension, stress) and vector, to
592 indicate the mathematical object necessary to describe tension, i.e., stress. This mathematical
593 object has to do with vectors, but is “not quite like a vector”, and indeed it is linear map
594 between two spaces of vectors, a type of object that today we call a tensor.
595 As a minimum, one normally needs four balance equations: mass, linear momentum, angular
596 momentum and energy, the latter expressing the First Principle of Thermodynamics. Moreover,
597 as a constraint, one should add the entropy inequality, which expresses the Second Principle
598 of Thermodynamics. Here we shall limit our study to the quasi-static case (i.e., negligible
599 accelerations) and to purely mechanical problems (i.e., no thermal power involved). Under
600 these hypotheses, Thermodynamics could be left aside, as long as the equation of the balance
601 of mechanical power is introduced, and we shall follow this approach (which is similar to that
602 followed by Bonet and Wood, 2008).

603 2.3.1 Cauchy Stress

604 Let us consider a continuum body, represented by its reference configuration B, in its current
605 configuration χ(B) ⊂ S, and, for now, let us omit indicating time. Let the body have mass
606 density ρ, and be subjected to forces per unit mass (or body forces) f , which are applied at
607 every x ∈ χ(B), and surface forces p at every x ∈ ∂χ(B), where ∂χ(B) denotes the boundary of
608 χ(B). We are interested in studying the internal actions (stresses) resulting from the action of
609 the external forces on the body. In order to do this, we need two axiomatic (i.e., self-evident)
610 statements.
611 Euler’s Separation Axiom. If the body B with current configuration χ(B) is in (dynamical
612 or statical) equilibrium and is divided into two portions B1 and B2 , which occupy the current
613 placements χ(B1 ) and χ(B2 ), surface forces must be applied at the separation surface, in order
614 to restore equilibrium in each of the portions.
615 Euler’s Stress Principle. On the separation surface between two complementary parts χ(B1 )
616 and χ(B2 ) of χ(B) there exist a surface force field t, called traction vector, which depends on
617 the point x on the separation surface and the vector n normal to the surface at x.
618 The situation described by the statements above is illustrated in Figure 2.3. We note that,
619 given the traction t(x, n), with point x regarded as a point of the separation surface on χ(B1 ),
620 the corresponding traction t(x, −n), with x regarded this time as a point of the separation
621 surface on χ(B2 ), must be equal and opposite to t(x, n), according to Newton’s third law of
622 action and reaction. Therefore, the first fundamental relation we find is

t(x, −n) = −t(x, n). (2.80)

623

624 We shall now prove that there is a linear relationship between t and n: this means that we will
625 find a tensor, which we shall call Cauchy stress tensor σ, such that t(n) = σn.

26
(B2 )
(B)

t(x, n) n

n t(x, n)

(B1 )

Figure 2.3: The current configuration χ(B) of a continuum body B, separated into the two complementary
portions χ(B1 ) and χ(B2 ), and the traction vector fields which must be applied at the separation surfaces
in order to restore equilibrium.

626 Cauchy’s Theorem (of Cauchy’s Tetrahedron). Given the current configuration χ(B) of
627 a body B, consider an infinitesimal portion of volume ∆υ, shaped as the tetrahedron in Figure
628 2.4. The facets ∆aj have normal vectors equal to −ej , whereas the facet ∆a has normal n. By
629 virtue of Euler’s stress principle, on the four facets will act traction vectors t(−ej ) and t(n),
630 respectively. Also, the tetrahedron is subjected to the body force f and attains the acceleration
631 a. Considering that the tetrahedron is infinitesimal, we can assume that the mass density, the
632 body force and the acceleration do not vary within the tetrahedron, and similarly each traction
633 vector does not vary within the corresponding facet. The dynamical equilibrium (Newton’s
634 second law) for the tetrahedron reads
ρ a ∆υ = ρ f ∆υ + t(n)∆a + t(−ej )∆aj . (2.81)

e3
t(n)

t( e2 ) n
e2

a1

e2 a

a2
a3 e1

Figure 2.4: Cauchy’s tetrahedron: for simplicity, only the traction vector at the facets with normal n
and −e2 were drawn. Figure modified from Bonet and Wood (2008).

635 By dividing by ∆a, and applying Equation (2.80) to the terms in t(−ej ), we obtain
∆υ ∆υ ∆aj
ρa = ρf + t(n) − t(ej ) . (2.82)
∆a ∆a ∆a

27
636 The ratio ∆aj /∆a is given by the cosine of the angle between ej and n, which is in turn given
637 by the scalar product of ej and n as
∆aj
= cos(ne
dj ) = ej .n = nj , (2.83)
∆a
638 i.e., it is the component of n in direction ej . Substitution into Equation (2.82) yields
∆υ ∆υ
ρa = ρf + t(n) − t(ej ) (ej .n). (2.84)
∆a ∆a
639 Now, we notice that the ratio ∆υ/∆a is an infinitesimal of order 1, as it is the ratio of an
640 infinitesimal of order 3 (a volume) and an infinitesimal of order 2 (an area). Therefore, in the
641 limit ∆υ → 0, we also have ∆υ/∆a → 0. By applying this limit to both sides of Equation
642 (2.84), we get
0 = t(n) − t(ej ) (ej .n). (2.85)
643 Now, let us write the representation of each t(ej ) in the basis {ei }3i=1 as

t(ej ) = ti (ej ) ei = σij ei , (2.86)

644 i.e., we denote with σij the i-th component ti (ej ) of the traction vector t(ej ) relative to the
645 surface with normal ej . Explicitly,

t(e1 ) = σ11 e1 + σ21 e2 + σ31 e3 , (2.87a)


t(e2 ) = σ12 e1 + σ22 e2 + σ32 e3 , (2.87b)
t(e3 ) = σ13 e1 + σ23 e2 + σ33 e3 . (2.87c)

646 Substituting Equation (2.86) into Equation (2.85) and solving for t(n), we obtain

t(n) = σij ei (ej .n), (2.88)

647 which, by definition of tensor product (Equation (1.73)), yields

t(n) = σij (ei ⊗ ej ) n. (2.89)

648 We have therefore proved that there exist a tensor σ which relates linearly the traction vector
649 t(n) to the normal n to a given area element, i.e.,

t(n) = σn. (2.90)

650 which, in components, reads


ti (n) = σij nj . (2.91)
651 Tensor σ is called Cauchy stress tensor, and is the fundamental measure of stress in Continuum
652 Mechanics. The diagonal components σ11 , σ22 , σ33 are called normal stresses, whereas the
653 off-diagonal components σij , with i 6= j, are called shear stresses. Study of the balance of
654 angular momentum (Section 2.3.2) will show that

σ = σT ⇒ σij = σji , (2.92)

655 i.e., the Cauchy stress is symmetric. Therefore, it only has six independent components out
656 of the total nine. Figure 2.5 shows the components of the stress tensor on a three-dimensional
657 and a two-dimensional “stress element”.

28
33
22

12
13 23
32
21
31

21 22 11 11
12
21
11
e3 e2
12

e1
e2 e3 22
e1

Figure 2.5: A three-dimensional (left) and a two-dimensional (right) “stress element” with the
components of the Cauchy stress tensor.

658 It is very important to notice that the two legs of the Cauchy stress have a specific meaning, as
659 we can see easily by looking at Equation (2.91). The first leg of σ is the force leg, and indeed
660 the first index of σ is the same as the index of the traction t. The second leg of σ is the area
661 leg, as it contracts with the normal n to the separation surface, and indeed the second index
662 of σ is the same as the index of the normal n. Since the Cauchy stress σ is an entirely spatial
663 tensor, so naturally are the surface traction t and the normal n, which gives the traction t the
664 meaning of current force per unit current area, and σ the meaning of true stress.

665 2.3.2 Balance Equations

666 In a certain sense, balance equations are for Continuum Mechanics the analogue of Newton’s
667 second law for particle mechanics. As mentioned in the introductory part of this section (Section
668 2.3), a thermomechanical theory needs four balance equations, i.e., mass, linear momentum,
669 angular momentum and energy, plus the entropy inequality. We shall confine ourselves to the
670 case of a purely mechanical theory, so that we will need only balance of mass, linear momentum
671 and angular momentum, and we will exploit linear momentum to find an equation of balance
672 of the purely mechanical power, which is the Continuum Mechanics analogue of the theorem
673 of the kinetic energy for particle mechanics.
674 In this section, as usual, B is a body and χ : B × R+ 0 → S is a motion of B in the physical
675 space S. Moreover, f is a scalar field on B, i.e., a function that, at every time t, depends on
676 the points x = χ(X, t) of the current configuration:

f ( · , t) : χ(B, t) → R, (2.93)

677 and R is a subset of B, with a smooth boundary ∂R, so that the current configuration of R at
678 time t is χ(R, t) and the current configuration of the boundary ∂R is ∂χ(R, t).
679 Before we describe the general form of balance equations, it is important to recall the difference
680 between extensive and intensive physical quantities. We will do this for scalar quantities, for
681 simplicity.
682 A physical quantity q defined in a body B is called extensive if it is additive for disjoint subsets.
683 This means that, if the extent of q for two disjoint subsets R1 and R2 of the body B (i.e., two

29
684 subsets with empty intersection) is q1 and q2 , respectively, then the extent of q on the union
685 R1 ∪ R2 of the two subsets is q1 + q2 . Typical examples of extensive quantities are mass, electric
686 charge, volume. In contrast, a physical quantity that does not depend on the size or the amount
687 of matter contained in the body is called intensive. The most classical example of intensive
688 quantity is the temperature.
689 In the case of a continuum body, an extensive quantity q must be associated with a density f ,
690 which is a scalar field as in Equation (2.93), so that the extent of q over the current configuration
691 χ(R, t) at time t of a region R ⊂ B is given by the integral
Z
Extent(q, χ(R, t)) = f ( · , t) dυ, (2.94)
χ(R,t)

692 which can be written as a function of time as


Z
Extent(q, χ(R, · )) = f dυ. (2.95)
χ(R, · )

693 The time derivative of the integral in Equation (2.95) is balanced by the amount of quantity q
694 that is generated in the region χ(R, t), and the flux of quantity q across the boundary ∂χ(R, t).
695 This is precisely the balance equation for the quantity q, which reads
Z Z Z
∂t f dυ = s dυ + w.n da. (2.96)
χ(R, · ) χ(R, · ) ∂χ(R, · )

696 The scalar field f , as already mentioned, is the density of the quantity q and is measured in
697 units of q per volume, the scalar field s is the source, i.e, the rate of generation of the quantity
698 q, and is measured in units of f per unit time, the vector field w is called the flux of q, and is
699 measured in units of f times velocity, and n is the normal to the boundary ∂χ(R, · ). Equation
700 (2.96) says that the amount of q in the region changes (left-hand side) because of generation
701 of q in the region (first term of the right-hand side) and flux of q through the boundary of the
702 region (second term of the right-hand side).
703 In order to make use of the general Equation (2.96), called master balance equation, it is
704 necessary to employ two important theorems: Reynold’s Transport Theorem, for the time
705 derivative of an integral over a time-dependent domain (left-hand side) and Gauss’ Divergence
706 Theorem, used to transform a surface integral (second term of the right-hand side) into a
707 volume integral. The former is proved in Box 2.4, the latter is just enunciated, without a proof,
708 in Box 2.5

30
Box 2.4. Reynold’s Transport Theorem. The derivative of the integral of a scalar field f
over a time-dependent domain χ(R, t) is given by
Z Z
 
∂t f= Dt f + f div v . (2.97)
χ(R, · ) χ(R, · )

where Dt f = [grad f ].v + ∂t f is the substantial derivative of f , defined in Section 2.2.5.


To prove the theorem, we first need to transform the integral over the time-dependent
domain χ(R, · ) into an integral over R, which is time-independent. This is achieved with
the theorem of the change of variables, i.e.,
Z Z
f= J [f ◦ (χ, τ )]. (2.98)
χ(R, · ) R

At this point, it is possible to pass the time derivative under the integral sign, and we have
Z Z
∂t f = ∂t J [f ◦ (χ, τ )]
χ(R, · ) R
Z
J˙ [f ◦ (χ, τ )] + J ∂t [f ◦ (χ, τ )] =

= (2.99)
R

Now, using the expression (2.78) of J˙ and the definition of substantial derivative (2.46),
we obtain
Z Z

∂t f= J [(div v) ◦ (χ, τ )] [f ◦ (χ, τ )] + J [(Dt f ) ◦ (χ, τ )]. (2.100)
χ(R, · ) R

At this point, the sought expression (2.97) is obtained by changing variables again in the
integral, i.e., by transforming the domain from R to forward to χ(R) and reordering the
terms in the integrand.

Box 2.5. Gauss’ Divergence Theorem. Let Ω be a subset of the physical space S with
a smooth boundary ∂Ω, n the normal to ∂Ω, and w be a vector field defined over Ω.
The Divergence Theorem enables to transform the integral of the scalar product w.n (the
component of w normal to ∂Ω) into a volume integral over the boundary Ω, i.e.,
Z Z
w.n da = div w dυ, (2.101)
∂Ω Ω

where we recall that the divergence of a vector field is defined as div w = wi,i = ∂wi /∂xi .
If w is time-dependent, and Ω happens to be the placement χ(R, t) of a region R of a
body B at time t, we have
Z Z
w.n da = div w dυ. (2.102)
∂χ(R, · ) χ(R, · )

709 By the Transport Theorem and The Divergence Theorem, the master balance equation (2.96)

31
710 becomes Z Z Z
[Dt f + f div v] dυ = s dυ + div w da, (2.103)
χ(R, · ) χ(R, · ) χ(R, · )

711 which can be written Z


[Dt f + f div v − s − div w] dυ = 0. (2.104)
χ(R, · )

712 Since the region χ(R, · ) is arbitrary, the integrand must vanish (the only function whose
713 integral is zero on any domain is the zero function), and we can localise Equation (2.104) into

Dt f + f div v − s − div w = 0, (2.105)

714 which can be rearranged into


Dt f + f div v = s + div w. (2.106)
715 Equation (2.106) is called the local form (i.e., in terms of fields, which are functions of the
716 spatial point x) of the master balance equation (2.96).
717 Using the explicit expression of the substantial derivative, the component form of Equation
718 (2.106) is
(f,i vi + ∂t f ) + f vi,i = s + wi,i , (2.107)
719 and, in the traditional notation,
 
∂f ∂f ∂vi ∂wi
vi + +f =s+ . (2.108)
∂xi ∂t ∂xi ∂xi

720 Balance of Mass

721 In this case, the density is the mass density ρ, the source term is the mass generation γ and
722 the flux term is the mass flux m:
Z Z Z
∂t ρ dυ = γ dυ + m.n da. (2.109)
χ(R, · ) χ(R, · ) ∂χ(R, · )

723 Localising, we obtain


Dt ρ + ρ div v = γ + div m. (2.110)
724 If mass generation rate γ and mass flux m vanish, the balance of mass reduces to the equation
725 of conservation of mass, also called continuity equation

Dt ρ + ρ div v = 0. (2.111)

726 Using the explicit expression of the substantial derivative, the component form of the equation
727 of conservation of mass (2.114) is

(ρ,i vi + ∂t f ) + ρ vi,i = 0, (2.112)

728 and, in the traditional notation,


 
∂ρ ∂ρ ∂vi
vi + +ρ = 0. (2.113)
∂xi ∂t ∂xi

32
729 Problem 2.1. Alternative Form of Conservation of Mass. Prove that the equation of
730 conservation of mass (2.114) has the alternative form

∂t ρ + div(ρ v) = 0, (2.114)

731 which is often preferred in Fluid Mechanics. Hint: this is easy to do in components and using
732 Leibniz’ rule.

733 Problem 2.2. Corollary of Conservation of Mass. Prove that, if conservation of mass is
734 satisfied, then, for every field f ,
Z Z
∂t ρ f dυ = ρ Dt f dυ. (2.115)
χ(R, · ) χ(R, · )

735 Hint: again, this is easy to do in components and using Leibniz’ rule.

736 Problem 2.3. Kinetic Power. Use Problem 2.2 to prove that the time derivative of the kinetic
737 energy (integral of the kinetic energy density) is
Z Z
1
∂t 2 ρ v.v dυ = ρ a.v dυ. (2.116)
χ(R, · ) χ(R, · )

738 It is possible to obtain a form of the equation of conservation of mass that that relies on the
739 theorem of the change of variables, and does not invoke the Transport Theorem. The mass
740 contained in the current configuration χ(R, · ) of region R can be expressed, by means of the
741 theorem of the change of variables, as
Z Z Z
Mass(χ(R, · )) = ρ dυ = J [ρ ◦ (χ, τ )] dV = ρR dV, (2.117)
χ(R, · ) R R

742 where we have defined the referential mass density as

ρR = J [ρ ◦ (χ, τ )]. (2.118)

743 If mass is conserved, the referential mass density ρR is a given scalar field, which is independent
744 of time. Therefore, Equation (2.118) provides the material form of conservation of mass, and
745 allows to calculate the current mass density ρ once J is known. With the usual abuse of notation,
746 one omits the composition with (χ, τ ) and simply writes

ρR = J ρ. (2.119)

747 Balance of Linear Momentum

748 In this case, the density is the density of linear momentum ρ v, the source is the external force
749 per unit volume ρ f (so that f is the external force per unit mass), and the flux will turn out
750 to be the Cauchy stress σ. Indeed, the integrand of the surface integral is the traction vector
751 t, so that the surface integral is the resultant of all surface forces, so that
Z Z Z
∂t ρ v dυ = ρ f dυ + t da. (2.120)
χ(R, · ) χ(R, · ) ∂χ(R, · )

33
752 Thus, by Cauchy’s Theorem (2.90) we obtain the form
Z Z Z
∂t ρ v dυ = ρ f dυ + σ n da, (2.121)
χ(R, · ) χ(R, · ) ∂χ(R, · )

753 in which the Cauchy stress σ is evidently the flux term, in the spirit of the master balance
754 equation (2.96). By employing the corollary of conservation of mass (2.115) and the definition
755 of acceleration (2.54) on the left-hand side and the divergence theorem on the surface integral,
756 we have Z Z Z
ρa = ρf + div σ, (2.122)
χ(R, · ) χ(R, · ) χ(R, · )

757 which can be localised into


ρ a = ρ f + div σ. (2.123)
758 In components, this reads

ρ ai = ρ (vi,j vj + ∂t vi ) = ρ fi + σij,j , (2.124)

759 and, in the traditional notation,


 
∂vi ∂vi ∂σij
ρ ai = ρ vj + = ρ fi + . (2.125)
∂xj ∂t ∂xj

760 In the quasi static case, accelerations are negligible, and the time derivative in the left-hand side
761 of Equation (2.121) can be neglected as well. Therefore, following exactly the same procedure
762 (i.e., application of the divergence theorem to the surface integral in the right-hand side), one
763 obtain the local equation of statics:

0 = ρ f + div σ. (2.126)

764 Remark 2.4. We perform the divergence of the second-order tensor σ on the second (i.e., last)
765 index (e.g., Marsden and Hughes, 1983; Epstein, 2010). Some Authors (e.g., Chadwick, 1976;
766 Ogden, 1997) prefer to perform the divergence of a second-order tensor on the first index. For
767 the case of the Cauchy stress σ, performing the divergence on the last index implies that the
768 surface traction must be written t = σ T n, i.e., ti = σji nj , which corresponds to the divergence
769 σji,j .

770 Remark 2.5. Warning: as noted by Marsden and Hughes (1983), whereas the localised form
771 of the balance equation for a vector or tensor field is always valid, the integral form is valid in
772 exclusively in Cartesian coordinates.

773 Balance of Angular Momentum

774 The balance of angular momentum states the symmetry of the Cauchy stress σ. In order to
775 work the necessary calculations out, we need to define the position vector, and show that the
776 substantial derivatives of its components give the components of the velocity, which we do in
777 Box 2.6 below.

34
Box 2.6. The Position Vector and Its Substantial Derivative. We choose an origin o ∈ S
and define the position vector as

r : S × R+
0 → To S : (x, t) 7→ r(x, t) = x − o, (2.127)

i.e., r is the function that, to every point x, associates the vector based at the chosen
origin o and pointing to x. Notice that r is not a vector field, as all its values r(x, t) are
attached at the chosen origin o. It is easy to show that the substantial derivatives of the
components of the position vector are the components of the velocity, i.e.,

Dt ri = ri,j vj + ∂t ri = δij vj + 0 = vj . (2.128)

Indeed, ri,j = δij (in the traditional notation, one has ∂ri /∂xj = ∂(xi − oi )/∂xj = δij ),
and ∂t ri = 0, since r does not depend on time explicitly.
In component-free notation, we write

Dt r = v. (2.129)

In Equation (2.129), we have a subtle abuse of notation. Indeed, we technically cannot write
the full expression of the substantial derivative for r as Dt r = (grad r)v + ∂t r because,
since r is not a vector field, it does not admit a gradient but only a tangent map, precisely
like the configuration χ. Afters a few nights of mystical torment, we decided to live with
this venial sin.

778 In the case of the balance of angular momentum, the density is the density of angular
779 momentum r × (ρ v), the source is the moment r × (ρ f ) of the forces per unit volume ρ f , and
780 the flux turns out to be the second-order tensor given by cross-product r × σ of the position
781 vector r and Cauchy stress σ, which is defined as

(r × σ)il = ijk rj σkl , (2.130)

782 where ink is the Ricci/Levi-Civita symbol defined in Equation (1.60). The integrand of the
783 surface integral is the moment r × t of the surface tractions, so that
Z Z Z
∂t r × (ρ v) dυ = r × (ρ f ) dυ + r × t da. (2.131)
χ(R, · ) χ(R, · ) ∂χ(R, · )

784 By Cauchy’s theorem (Equation (2.90),


Z Z Z
∂t r × (ρ v) dυ = r × (ρ f ) dυ + (r × σ) n da. (2.132)
χ(R, · ) χ(R, · ) ∂χ(R, · )

785 By applying the corollary of conservation of mass (2.115) and Equation (2.129), the left-hand
786 side of Equation (2.132) becomes
Z Z
 
∂t r × (ρ v) dυ = Dt r × (ρ v) + r × (ρ Dt v) dυ
χ(R, · ) χ(R, · )
Z Z
 
= v × (ρ v) + r × (ρ a) dυ = r × (ρ a) dυ (2.133)
χ(R, · ) χ(R, · )

35
787 where, in the last passage, the term v × (ρ v) vanishes identically since vectors v and ρ v are
788 parallel. Substitution of Equation (2.133) and application of the divergence theorem to the
789 surface integral of Equation (2.132) yield
Z Z Z
r × (ρ a) dυ = r × (ρ f ) dυ + div(r × σ) dυ. (2.134)
χ(R, · ) χ(R, · ) χ(R, · )

790 In coordinates, we have

[div(r × σ)]i = (ijk rj σkl ),l = ijk rj,l σkl + ijk rj σkl,l
= ijk δjl σkl + ijk rj σkl,l
= ijk σkj + ijk rj σkl,l , (2.135)

791 where we used rj,l = δjl (in the traditional notation, ∂rj /∂xl = ∂(xj − oj )/∂xl = δjl + 0).
792 Therefore,
div(r × σ) =  : σ T + r × div σ, (2.136)
793 where  : σ T is the contraction of the last two legs of the third-order Levi-Civita tensor
794  = ijk ei ⊗ ej ⊗ ek with σ T . Substituting into Equation (2.134), we obtain
Z Z Z Z
r × (ρ a) = r × (ρ f ) + r × div σ +  : σT , (2.137)
χ(R, · ) χ(R, · ) χ(R, · ) χ(R, · )

795 which, by the distributivity of the cross-product, can be rearranged into


Z Z
r × (ρ a + ρ f − div σ) dυ =  : σT , (2.138)
χ(R, · ) χ(R, · )

796 The left-hand side of Equation (2.138) vanishes by virtue of the balance of linear momentum
797 (2.123), and therefore we obtain Z
 : σ T = 0. (2.139)
χ(R, · )

798 By localising, we get to


 : σ T = 0, (2.140)
799 which in components, reads
ijk σkj = 0. (2.141)
800 This implies that the Cauchy stress is symmetric, i.e.,

σ T = σ. (2.142)

801 Indeed, the components ijk σkj of  : σ T all vanish and, explicitly, we have the three equations

123 σ32 + 132 σ23 = σ32 − σ23 = 0, (2.143a)


231 σ13 + 213 σ31 = σ13 − σ31 = 0, (2.143b)
312 σ21 + 321 σ12 = σ21 − σ12 = 0, (2.143c)

802 which imply that σij = σji , i.e., σ is symmetric.


803 Remark 2.5 applies also in this case. Therefore, while the result σ T = σ is always valid, the
804 proof is valid only in Cartesian coordinates.

36
805 2.3.3 Balance of Mechanical Power

806 By taking the scalar product of both sides of the balance of linear momentum (2.123), we
807 obtain the local form of the equation of power
ρ a.v = ρ f .v + (div σ).v. (2.144)
808 The second term on the right-hand side can be manipulated in components, by exploiting
809 Leibniz’ rule and recalling the definition (2.57) of velocity gradient, into
(div σ).v = σij,j vi = (σij vi ),j − σij vi,j = (σij vi ),j − σij lij = div(σ T v) − σ : l, (2.145)
810 where σ : l = σij lij is the double contraction of the second-order tensors σ and l. Substituting
811 this result into Equation (2.144) and integrating on the region χ(R, · ), we obtain
Z Z Z Z
T
ρ a.v dυ = ρ f .v dυ + div(σ v) dυ − σ : l dυ (2.146)
χ(R, · ) χ(R, · ) χ(R, · ) χ(R, · )

812 From Problem 2.3, we have that the left-hand side is the time derivative of the total kinetic
813 energy, i.e. (Equation (2.116))
Z Z
1
Ṫ = ∂t 2 ρ v.v dυ = ρ a.v dυ. (2.147)
χ(R, · ) χ(R, · )

814 Moreover, by the divergence theorem and Cauchy’s theorem (Equation (2.90)), the second term
815 on the right-hand side is the work of the surface tractions t acting on the boundary ∂χ(R, · )
816 of χ(R, · ), i.e.,
Z Z Z Z
T T
div(σ v) dυ = (σ v).n da = v.(σn) da = t.v da. (2.148)
χ(R, · ) ∂χ(R, · ) ∂χ(R, · ) ∂χ(R, · )

817 Substituting Equations (2.147) and (2.148) into Equation (2.146) and bringing the third term
818 of the right-hand side to the left-hand side, we obtain
Z Z Z Z
1
∂t 2 ρ v.v dυ + σ : l dυ = ρ f .v dυ + t.v da. (2.149)
χ(R, · ) χ(R, · ) χ(R, · ) ∂χ(R, · )

819 Now, the right-hand side of (2.149) is the total power of the external forces or, for short, the
820 external power, the first term being the power of the volume forces and the second being the
821 power of the surface tractions:
Z Z
Pext = ρ f .v dυ + t.v da. (2.150)
χ(R, · ) ∂χ(R, · )

822 The second term on the left-hand side of (2.149) is called stress power or internal power, and
823 is the power exerted by the internal actions, i.e., by the stress:
Z
Pint = σ : l dυ. (2.151)
χ(R, · )

824 Using Equations (2.147), (2.150) and (2.151), Equation (2.149) can be written in compact form
825 as
Ṫ + Pint = Pext , (2.152)
826 which means that the power of the external forces, Pext , is expended in part to impose an
827 acceleration to the body and thus to make its kinetic energy change, which is expressed by the
828 term Ṫ , and in part to cause a deformation in the body, the corresponding amount of power
829 being Pint .

37
830 2.3.4 Conjugated Stress-Strain Pairs and “Material Balance”

831 From the study of the definition of the internal power in Equation (2.151) it is possible to see
832 how other measures of stress, aside from the Cauchy stress σ arise.
833 We start by recalling a useful identity that is going to be employed in the following. From the
834 decomposition (2.58) of l into its symmetric part d, the deformation rate of Equation (2.59),
835 and its skew-symmetric part w, the spin tensor of Equation (2.60), we have that
σ : l = σ : (d + w) = σ : d, (2.153)
836 Indeed, σ : w vanishes identically for the the symmetry of σ (σji = σij ) and the skew symmetry
837 of w (wji = − wij ):
σ : w = σij wij = σji (− wji ) = − σji wji = − σ : w = 0, (2.154)
838 since the only number that is equal to its opposite is precisely zero.
839 We can substitute the identity (2.153) into the expression (2.151) of the internal power, and
840 obtain Z Z
Pint = σ : l dυ = σ : d dυ. (2.155)
χ(R, · ) χ(R, · )
841 Then we apply, to both the expression containing l and the expression containing d, the theorem
842 of the change of variables:
Z Z
σ : l dυ = J (σ ◦ (χ, τ )) : (l ◦ (χ, τ )) dV, (2.156a)
χ(R, · ) R
Z Z
σ : d dυ = J (σ ◦ (χ, τ )) : (d ◦ (χ, τ )) dV. (2.156b)
χ(R, · ) R

843 We say that the Cauchy stress σ is power-conjugated to the velocity gradient l or to the
844 deformation rate d.
845 From either of these expressions, we can identify the spatial tensor field τ (not to be confused
846 with the time map τ ) called Kirchhoff stress and useful in some numerical applications, such
847 that
τ ◦ (χ, τ ) = J σ ◦ (χ, τ ), (2.157)
848 which, with the usual abuse of notation, can be written
τ = J σ. (2.158)
849 Using the expression (2.157) of the Kirchhoff stress, the internal power can be expressed as
Z Z
Pint = (τ ◦ (χ, τ )) : (l ◦ (χ, τ )) dV = (τ ◦ (χ, τ )) : (d ◦ (χ, τ )) dV. (2.159)
R R

850 Also in the case of the Kirchhoff stress τ , we say that it is power-conjugated to the velocity
851 gradient l or to the deformation rate d.
852 In the subsections below, we shall show how, from the two forms (2.156) of the internal power
853 expressed in the material picture, we can find two new and fundamental measures of stress: the
854 first Piola-Kirchhoff stress P , power-conjugated to the rate Ḟ of the deformation gradient, the
855 and the second Piola-Kirchhoff stress, conjugated to the rate Ė of the Green Lagrange strain.
856 We shall also show how a so-called “material” equation of balance of linear momentum can be
857 obtained by exploiting P .

38
858 The First Piola-Kirchhoff Stress

859 Using the expression (2.69) of the rate Ḟ of the deformation gradient F , in terms of l, in
860 components, the integrand of (2.156a) can be written

J (σ ◦ (χ, τ )) : (l ◦ (χ, τ )) = J σij ◦ (χ, τ ) lij ◦ (χ, τ )


= J σij ◦ (χ, τ ) ḞiJ (F −1 )Jj ◦ (χ, τ )
= [J σij ◦ (χ, τ ) (F −T )jJ ] ḞiJ
= PiJ ḞiJ
= P : Ḟ , (2.160)

861 where
P = J σ ◦ (χ, τ ) F −T , PiJ = J σij ◦ (χ, τ ) (F −T )jJ (2.161)
862 is called the first Piola-Kirchhoff stress, or simply the first Piola stress, often written

P = J σ F −T , PiJ = J σij (F −T )jJ , (2.162)

863 with the customary abuse of notation. We note that, like F , the first Piola stress P is a
864 two-point tensor, and we say that P is the backward Piola transform of σ on its second leg,
865 which - we recall - is the area leg. The backward Piola transformation is indeed a pull-back
866 “rescaled” by the volume ratio J, which is justified by the fact that the pull-back of an area
867 element takes a term J (see Equation (4.68) in the textbook by Bonet and Wood (2008)).
868 Because of this, it turns out that P represents the nominal stress. Indeed, while the Cauchy
869 stress σ is entirely spatial and the surface traction t = σ n is a current force per unit current
870 area, the surface traction given by the first Piola stress P is a current force per unit undeformed
871 area.
872 We conclude by noting that, with the first Piola stress (2.161), the internal power reads
Z
Pint = P : Ḟ dV. (2.163)
R

873 The So-Called Material Balance of Linear Momentum

874 Note: this is has not been covered in the course. This header is just a reminder to myself.

875 The Second Piola-Kirchhoff Stress

876 Using the expression (2.77) of the rate Ė of the Green-Lagrange strain E in terms of d, in
877 components, the integrand of (2.156b) can be written

J (σ ◦ (χ, τ )) : (d ◦ (χ, τ )) = J σij ◦ (χ, τ ) dij ◦ (χ, τ )


= J σij ◦ (χ, τ ) (F −T )iI ĖIJ (F −1 )Jj ◦ (χ, τ )
= [J (F −1 )Ii ◦ (χ, τ ) σij ◦ (χ, τ ) (F −T )jJ ] ĖIJ
= SIJ ĖIJ
= S : Ė, (2.164)

39
878 where

S = J F −1 (χ, τ ) σ ◦ (χ, τ ) F −T , SIJ = J (F −1 )Ii ◦ (χ, τ ) σij ◦ (χ, τ ) (F −T )jJ (2.165)

879 is called the second Piola-Kirchhoff stress, or simply the second Piola stress, often written

S = J F −1 σ F −T , SIJ = J (F −1 )Ii σij (F −T )jJ , (2.166)

880 with the customary abuse of notation. We note that, like E, the second Piola stress S is a fully
881 material tensor, and we say that S is the full backward Piola transform of σ, i.e., forward Piola
882 transform on the second leg of S and push-forward on the first leg. Unfortunately, in contrast
883 with P , S does not have a physical meaning that is easy to grasp. However, it is a very useful
884 measure of stress in constitutive theory, and particularly in non-linear elasticity.
885 With the second Piola stress (2.165), and considering the relation (2.76) between the rate Ė of
886 the Green-Lagrange strain and the rate Ċ of the right Cauchy-Green deformation, the internal
887 power can be written in the equivalent forms
Z Z
1
Pint = S : Ė dV = 2 S : Ċ dV. (2.167)
R R

888 2.4 Constitutive Equations

889 Continuum problems are indeterminate due to the fact that the unknown variables outnumber
890 the differential equations available. Indeed, for a purely mechanical system, we have 7 equations

891 (a) 1 scalar equation for the balance of mass;

892 (b) 3 scalar equations for the balance of linear momentum;

893 (c) 3 scalar equations for the balance of angular momentum;

894 (the balance of power is found from the balance of linear momentum and therefore does not
895 add up to the count of the independent equations) and 13 unknowns

896 (a) 1 for the mass density ρ;

897 (b) 3 components χi of the configuration χ;

898 (c) 9 components σij of the Cauchy stress σ.

899 This prompts the introduction of constitutive equations, i.e., additional equations aimed at
900 closing the system of differential equations. Specific constitutive equations must be derived
901 from physical observation. For the purely mechanical case that we are considering, we need 6
902 constitutive equations for the 6 independent components of the Cauchy stress σ.

40
903 2.4.1 Non-Linear Elasticity

904 A material is called elastic if the stress can be expressed as a function of the deformation alone,
905 i.e., in terms of the first or the second Piola-Kirchhoff stresses, we have
P = P̂ (F ), S = Ŝ(E) (2.168)
906 where P̂ and Ŝ are the constitutive functions associated with P and S, respectively. Note
907 that we make P̂ depend on the deformation gradient F and Ŝ depend on the Green-Lagrange
908 strain E (alternatively, we could have Ŝ depend on the right Cauchy-Green deformation C):
909 compare with the power-conjugations in Equations (2.163) and (2.167), respectively.

910 Hyperelasticity

911 An important subclass of elastic materials are hyperelastic materials, which are defined as those
912 elastic materials such that the (material) internal power density P : Ḟ = S : Ė is an exact
913 differential in time, i.e., it is such that there exists a scalar field W : B × R+
0 → R (i.e., a scalar
914 function of X and t), such that
Ẇ = P : Ḟ = S : Ė. (2.169)
915 The scalar field W is called elastic strain energy density or elastic potential, and it is related
916 to its constitutive function in two possible ways, i.e.,
W (X, t) = W̌ (F (X, t)) = Ŵ (E(X, t)), (2.170)
917 or, omitting the arguments X and t,
W = W̌ (F ) = Ŵ (E). (2.171)
918 where we have replaced the symbol of composition in W = W̌ ◦ F and Ŵ ◦ E with regular
919 parentheses, with a slight, but accepted, abuse of notation.
920 We remark that, since W is required to be an energy, its constitutive functions cannot depend
921 explicitly on time, but only through F or E. Furthermore, we note that, if the constitutive
922 functions of W depend explicitly on the material point X, i.e.,
W (X, t) = W̌ (F (X, t), X) = Ŵ (E(X, t), X). (2.172)
923 we say that the material is inohmogeneous. In the following, we are going to omit writing the
924 explicit dependence on X, for the sake of a lighter notation.
925 With the definition in Equation (2.174), the time derivative Ẇ is calculated using the chain
926 rule and components, as
∂ W̌ ∂ Ŵ
Ẇ = (F ) ḞiJ = (E) ĖIJ , (2.173)
∂FiJ ∂EIJ
927 or, in compact symbolic notation,
∂ W̌ ∂ Ŵ
Ẇ = (F ) : Ḟ = (E) : Ė. (2.174)
∂F ∂E
928 Comparison of Equation (2.168) with Equations (2.174) and (2.173) yields the expressions of
929 the first and second Piola-Kirchhoff stresses for a hyperelastic material,
∂ W̌ ∂ W̌
P = P̂ (F ) = (F ), PiJ = P̂iJ (F ) = (F ), (2.175)
∂F ∂FiJ

41
930 and
∂ Ŵ ∂ Ŵ
S = Ŝ(E) = (E), SIJ = ŜIJ (E) = (E). (2.176)
∂E ∂EIJ
931 Note that, by recalling the power conjugation of the second Piola-Kirchhoff stress S with the
932 right Cauchy-Green deformation C, it is possible to express the constitutive functions of S
933 and the elastic potential W as a function of C. With abuse of notation, we shall denote these
934 constitutive functions with the same symbols Ŝ and Ŵ that we used for their counterparts
935 depending on E, and write
∂ Ŵ ∂ Ŵ
S = Ŝ(C) = 2 (C), SIJ = ŜIJ (C) = 2 (C). (2.177)
∂C ∂CIJ
936 The coefficient 2 in Equation (2.177) comes from the fact that C = 2 E + I, and therefore
∂ ∂
=2 . (2.178)
∂E ∂C
937 The Cauchy stress σ is found from Equations (2.175) or (2.176) (or (2.176)), by inverting
938 Equations (2.162) and (2.166), respectively, i.e.,
σ = J −1 P F T , σij = J −1 PiJ FjJ , (2.179)
939 and
σ = J −1 F S F T , σij = J −1 FiI SIJ FjJ . (2.180)
940 We say that σ is the forward Piola transform of P on its second leg (which is the area leg),
941 and the full forward Piola transform of S (i.e., forward Piola transferm on the second leg of S
942 and push-forward on the first leg of S).

943 Example 2.1. The Compressible Neo-Hookean Material. This material model was initially
944 developed in 1948 by Professor Rivlin for incompressible materials. Here, we report a form
945 suitable for the general case of compressible materials (see, e.g., Bonet and Wood, 2008):
µ λ
Ŵ (C) = (C : I − 3) − µ ln J + (ln J)2 . (2.181)
2 2
946 Note
√ that C : I = CIJ δIJ = CII = tr C is the trace of C, and that the volume ratio is J =
947 det C. It is possible to prove that, when linearised, the compressible neo-Hookean potential
948 gives the stress-strain relation of the isotropic linear elastic material. Indeed, the coefficients
949 λ and µ are called Lamé’s coefficients and are a possible pair of independent coefficients for
950 isotropic linear elasticity (much like the Young’s modulus E and the Poisson’s ratio ν). While
951 µ is nothing but the shear modulus (often denoted G in elementary Strength of Materials
952 textbooks), unfortunately λ does not have such a clear physical meaning. This is why we often
953 prefer to use the pair constituted by the bulk modulus κ and the shear modulus µ (see, e.g.,
954 Walpole, 1981, 1984; Federico, 2010, 2012). Here, however, we shall use the customary λ-µ pair.
955 Warning: Do not confuse the first Lamé’s coefficient λ with the usual symbol for the stretch.
956 In order to facilitate the evaluation of the second Piola-Kirchhoff stress S, let us first calculate
957 the derivatives of tr C = C : I and det C with respect to C in component-free notation and
958 in components. We have, with the usual abuse of notation,
∂ tr ∂(C : I)
◦C ≡ = I,
∂C ∂C
∂ tr ∂(CIJ δIJ )
◦C ≡ = δIK δJL δIJ = δKL , (2.182)
∂CKL ∂CKL

42
959 and
∂ det ∂(det C)
◦C ≡ = (det C) C −1 = J 2 C −1 ,
∂C ∂C
∂ det ∂(det C)
◦C ≡ = [C(C)]KL = (det C) (C −1 )KL = J 2 (C −1 )KL , (2.183)
∂CKL ∂CKL
960 where [C(C)] is the tensor whose components are the algebraic
√ complements of the components
961 of C. From the derivative of det C, we find that of J = det C as

∂J ∂ det C 1 1 ∂(det C) 1 1 2 −1 1
= = √ = J C = J C −1 . (2.184)
∂C ∂C 2 det C ∂C 2J 2

962 Now, from the general expression (2.177), we obtain the Second Piola-Kirchhoff stress S as
 
∂ Ŵ µ −1 1 −1 λ −1 1 −1
S=2 (C) = 2 I − µJ J C + 2 (ln J) J JC
∂C 2 2 2 2
= µ (I − C −1 ) + λ (ln J) C −1 . (2.185)

963 The Cauchy stress σ is found from Equation (2.180) as

σ = J −1 F S F T = J −1 [µ (b − i) + λ (ln J) i] , (2.186)

964 where we have used F I F T = F F T = b and F C −1 F T = F F −1 F −T F T = i, with b being


965 the left Cauchy-Green deformation and i being the spatial identity tensor.

966 Elasticity Tensors

967 Elasticity tensors are essential in numerical applications, a prime example of which is the Finite
968 Element Method, and are required to calculate time rates and/or increments of a measure of
969 stress. Of the many possible forms of elasticity tensor, we shall focus on three only, a two-point
970 elasticity tensor, a completely material one, and a completely spatial one.
971 From the first of Equations (2.168), the time rate of the first Piola-Kirchhoff stress P is
972 obtained, by use of the chain rule, as

∂ P̂ ∂ P̂iJ
Ṗ = (F ) : Ḟ = A : Ḟ , ṖiJ = (F ) ḞkL = AiJkL ḞkL , (2.187)
∂F ∂FkL
973 where tensor
∂ P̂ ∂ P̂iJ
A= (F ), AiJkL = (F ) (2.188)
∂F ∂FkL
974 is called the first elasticity tensor (first because it is related to the first Piola-Kirchhoff stress).
975 For a hyperelastic material, since Equation (2.175) holds, we have

∂ 2 W̌ ∂ 2 W̌
Ṗ = (F ) : Ḟ = A : Ḟ , ṖiJ = (F ) ḞkL = AiJkL ḞkL , (2.189)
∂F 2 ∂FiJ ∂FkL
976 with
∂ 2 W̌ ∂ 2 W̌
A= (F ), AiJkL = (F ). (2.190)
∂F 2 ∂FiJ ∂FkL

43
977 Note that, in the hyperelastic case, if W̌ is twice differentiable (as it is customarily assumed),
978 then Schwartz’ theorem ensures that the order of differentiation (with respect to FiJ and FkL )
979 can be switched, and therefore we say that the first elasticity tensor A enjoys major symmetry
980 or diagonal symmetry, i.e., in components,
∂ 2 W̌ ∂ 2 W̌
(F ) = (F ) ⇒ AiJkL = AkLiJ , (2.191)
∂FiJ ∂FkL ∂FkL ∂FiJ
981 which means that the components of A are invariant under exchange of the first pair of indices
982 with the second.
983 From the second of Equations (2.168), the time rate of the second Piola-Kirchhoff stress S is
984 obtained, by use of the chain rule, as
∂ Ŝ ∂ ŜIJ
Ṡ = (E) : Ė = C : Ė, ṠIJ = (E) ĖKL = CIJKL ĖKL , (2.192)
∂E ∂EKL
985 where tensor
∂ Ŝ ∂ ŜIJ
C= (E), CIJKL = (E) (2.193)
∂E ∂EKL
986 is called the second elasticity tensor (second because it is related to the first Piola-Kirchhoff
987 stress) or simply the material elasticity tensor. We note that, since both S = Ŝ(E and E are
988 symmetric, C enjoys minor symmetry or pair symmetry, i.e., in components
∂ ŜIJ ∂ ŜJI ∂ ŜIJ ∂ ŜJI
(E) = (E) = (E) = (E) ⇒ CIJKL = CJIKL = CIJLK = CJILK ,
∂EKL ∂EKL ∂ELK ∂ELK
(2.194)
989 which means that the components of C are invariant for exchange of the two indices in the first
990 pair and/or for exchange of the two indices in the second pair.
991 For a hyperelastic material, since Equation (2.176) holds, we have
∂ 2 Ŵ ∂ 2 Ŵ
Ṡ = (E) : Ė = C : Ė, ṠIJ = (E) ĖKL = CIJKL ĖKL , (2.195)
∂E 2 ∂EIJ ∂EKL
992 with
∂ 2 Ŵ ∂ 2 Ŵ
C= (E), CIJKL = (E). (2.196)
∂E 2 ∂EIJ ∂EKL
993 Again, in the hyperelastic case, if Ŵ is twice differentiable (as it is customarily assumed), then
994 Schwartz’ theorem ensures that the order of differentiation (with respect to FiJ and FkL ) can
995 be switched, and therefore C enjoys, aside from minor symmetry, also major symmetry, i.e., in
996 components,
∂ 2 Ŵ ∂ 2 Ŵ
(E) = (E) ⇒ CIJKL = CKLIJ . (2.197)
∂EIJ ∂EKL ∂EKL ∂EIJ
997 which means that the components of C are invariant under exchange of the first pair of indices
998 with the second.
999 Note that the material elasticity tensor C can be expressed as a function of the right
1000 Cauchy-Green deformation C, by recalling the relation (2.178) between the derivatives
1001 with respect to the Green-Lagrange strain E and with respect to the right Cauchy-Green
1002 deformation C. For an elastic material, we have
∂ Ŝ ∂ ŜIJ
C=2 (C), CIJKL = 2 (C) (2.198)
∂C ∂CKL

44
1003 and, for a hyperelastic material,

∂ 2 Ŵ ∂ 2 Ŵ
C=4 (C), CIJKL = 4 (C). (2.199)
∂C 2 ∂CIJ ∂CKL

1004 where, in Equation (2.199) the coefficient 4 comes from the fact that we are using Equation
1005 (2.178) twice.
1006 Finally, we move to the spatial elasticity tensor. This is obtained by means of a full forward
1007 Piola transformation of Equation (2.192) or, equivalently, of Equation (2.195). In components,
1008 we have

J −1 FiI ṠIJ FjJ = J −1 FiI FjJ CIJP Q EP Q


= J −1 FiI FjJ CIJKL δKP δQL EP Q ,
= J −1 FiI FjJ CIJKL (F −1 )P k FkK (F −1 )Ql FlL EP Q ,
= J −1 FiI FjJ CIJKL FkK FlL (F −1 )P k EP Q (F −1 )Ql , (2.200)

1009 where we used (F −1 )P k FkK = δKP and (F −1 )Ql FlL = δLQ . Now, we consider that
1010 (F −1 )P k EP Q (F −1 )Ql = dkl are the components of the deformation rate d = F −T E F −1
1011 (which is the inverse of Equation (2.77)), and define the spatial elasticity tensor as the tensor
1012 C with components

Cijkl = J −1 FiI FjJ CIJKL FkK FlL , (2.201)

1013 and the Truesdell rate of the Cauchy stress σ̊ as

σ̊ = J −1 F Ṡ F T , σ̊ij = J −1 FiI ṠIJ FjJ . (2.202)

1014 Note that, since S is the full backward Piola transform of σ (see Equation (2.166)), Equation
1015 (2.202) can be written

σ̊ = J −1 F (J F −1 σ F −T )˙ F T ,
 
(2.203)

1016 i.e., the Truesdell rate σ̊ of the Cauchy stress σ is obtained by first performing a full backward
1017 Piola transformation of σ and then a full forward Piola transformation of the result. Hughes
1018 and Marsden (1977) and Marsden and Hughes (1983) have elegantly proved that the Truesdell
1019 rate is expressible as a Lie derivative.
1020 From Equations (2.200), (2.201) and (2.202), we obtain that the full forward Piola
1021 transformation of Equations (2.192) or (2.195) is

σ̊ = C : d, σ̊ij = Cijkl dkl . (2.204)

1022 Finally, we note that the spatial elasticity tensor C enjoys exactly the same symmetries of the
1023 material elasticity tensor C, i.e., minor symmetry for the elastic case, and both minor and
1024 major symmetries for the hyperelastic case.

1025 2.4.2 Linearisation of Non-Linear Elasticity

1026 Text

45
1027 2.4.3 One-Dimensional Viscoelasticity

1028 Viscoelasticity is a material behaviour that has some aspect of the elasticity of solids and
1029 of the viscosity of fluids. In viscous fluids, the stress is related not to the strain, but to the
1030 strain rate (i.e., the time derivative of the strain). The combination of elasticity and viscosity
1031 can reproduce very well the time-dependent behaviour of many materials, including biological
1032 tissues. We limit our study to one-dimensional linear viscoelasticity.

1033 The Linear Viscous Damper

1034 To define the viscous behaviour, let us follow an analogy similar to that used for the case of the
1035 elastic behaviour. If the particle-mechanics equivalent of the linear elastic body is the linear
1036 spring with constitutive equation Fs = −k x, the particle-mechanics equivalent of a viscous
1037 fluid is the viscous damper or viscous dashpot. Its linear constitutive equation reads

Fd = −b ẋ, (2.205)

1038 i.e., the damping force Fd is proportional to the velocity v = ẋ, and the constant of
1039 proportionality b is called coefficient of viscous damping or simply damping coefficient.

1040 Unidimensional Linear Viscous Solid

1041 By analogy with a viscous fluid, for which the shear stress is proportional to the deformation
1042 rate (i.e., for i 6= j, σij = 2 η dij = 2 η [ 21 (vi,j − vj,i )] = η (vi,j − vj,i )), in a uni-dimensional
1043 viscous solid, the stress is related to the strain by

σ = η ,
˙ (2.206)

1044 so that the stress σ is proportional to the strain rate ,


˙ and the constant of proportionality η
1045 is called viscosity.

1046 Fundamental Linear Viscoelastic Models

1047 The two elementary models in linear viscoelasticity are given by arranging an elastic element
1048 and a viscous element either in series or in parallel.
1049 In the series model, or Maxwell model, the elastic and the viscous element are subjected to the
1050 same stress. Therefore, we have
σ σ
elastic = , ˙viscous = , ˙ = ˙elastic + ˙viscous . (2.207)
E η

1051 By summing the time derivative of the first of (2.207) with the second of (2.207), we obtain
1052 the the differential equation
1 1
˙ = σ + σ̇. (2.208)
η E
1053 Note that, if the elastic element is infinitely stiff, i.e., in the limit for 1/E → 0, we retrieve the
1054 linear viscous constitutive equation (2.206).

46
1055 In the parallel model, or Kelvin-Voigt model, the elastic and the viscous element are subjected
1056 to the same strain. Therefore,

σelastic = E , σviscous = η ,
˙ σ = σelastic + σviscous , (2.209)

1057 and, by summing the first and second of (2.209), we obtain the differential equation

σ = E  + η ,
˙ (2.210)

1058 Note that, when η → 0, we recover the unidimensional linear elastic constitutive equation
1059 σ = E .
1060 The schematic diagrams and the time-dependent stress-strain behaviour of these two
1061 elementary models are shown in Figure 2.6.
1062 The Maxwell model is able to describe the phenomenon of stress relaxation: when a constant
1063 deformation is applied instantly at time t = 0 and kept constant, the stress increases instantly
1064 to a peak value, and then decreases exponentially and asymptotically to zero. This can be
1065 captured by solving the associated homogeneous differential equation of Equation (2.208) with
1066 initial condition σ(0) = E 0 , i.e., by imposing instantaneously (t) = 0 , so that (t)
˙ = 0:
1 1
σ + σ̇ = 0, σ(0) = 0. (2.211)
η E
1067 This is a separable differential equation with solution

σ(t) = 0 exp − τt ,

(2.212)

1068 where
η
τ= , (2.213)
E
1069 is called the characteristic time of the system. Note also (Figure 2.6, left) that the stress
1070 becomes very close to the asymptotic value σ∞ = 0 at about t = 5τ .
1071 The Kelvin-Voigt model is instead able to predict the phenomenon of creep: when a constant
1072 stress is applied instantly at time t = 0, the strain increases exponentially and asymptotically
1073 to a stationary value. We can solve Equation (2.210) for σ(t) = σ0 by summing, to the solution
1074 of the associated homogeneous equation E  + η ˙ = 0,

E  + η ˙ = 0 ⇒ (t) = c exp − τt ,

(2.214)

1075 a particular integral of the non-homogeneous solution, i.e.,


σ0
(t) = , (2.215)
E
1076 and obtain
(t) = σ0 E + c exp − τt .

(2.216)
1077 By imposing the initial condition (0) = 0, we have

(t) = σ0 E 1 − exp − τt ,
 
(2.217)

1078 where, again, τ = η/E is the characteristic time. Note (Figure 2.6, right) that in this case the
1079 strain becomes very close to the asymptotic value of ∞ = σ0 /E at about t = 5τ .

47
✏ ✏
✏0 0 /E

t t

E ✏0 0

t t

E ⌘
E

Figure 2.6: Maxwell series model (left) with its prediction of stress relaxation, and Kelvin-Voigt model
(right) with its prediction of creep.

E1
E2 ⌘

Figure 2.7: The standard linear model of viscoelasticity.

1080 Remark 2.6. Notice the duality between the Maxwell and the Kelvin-Voigt models: in the
1081 former, the elements are subjected to the same stress, in the latter, to the same strain. Notice
1082 also the perfect duality between the phenomena of stress relaxation and creep.

1083 The problem of these two models is that each fails in predicting the behaviour that the other
1084 predicts very well. Also, the Maxwell model predicts that the stress relaxes to zero; this is true
1085 for some materials1 , but certainly not for all, and in particular not for biological materials.
1086 The standard linear viscoelastic model combines the Maxwell and the Kelvin-Voigt models,
1087 buy putting an elastic element in parallel with the series of an elastic and a viscous element.
1088 With reference to Figure 2.7, the standard linear model is described by
 
E2 1 1 E1
˙ = σ+ σ̇ −  , (2.218)
E1 + E2 η E2 η

1089 and is able to predict both stress relaxation and creep.


1
The next time You bake anything, try to stretch the dough and You will see that you will feel resistance at
the beginning, but this will fade to zero!

48
1090 Estimation of the Characteristic Time τ with a Stress Relaxation Test

1091 In practice, in a stress-relaxation test, the standard linear model “raises” the asymptote of the
1092 Maxwell model, by adding another linear spring, which is responsible of the value of the stress
1093 for t → ∞. Given a stress relaxation curve as in Figure 2.8, the characteristic time is found by
1094 looking at the intersection of the tangent to the stress at its peak and the asymptote. This is
1095 analogous to what one does with the Maxwell model, in which the asymptote is the t-axis.

t

Figure 2.8: Experimental determination of the characteristic time τ from a stress relaxation test.

49
1096 Bibliography

1097 Bonet, J. and Wood, R. D. 2008. Nonlinear Continuum Mechanics for Finite Element Analysis
1098 (Second Edition). Cambridge University Press, Cambridge, UK.

1099 Chadwick, P. 1976. Continuum Mechanics, Concise Theory and Problems. George Allen &
1100 Unwin Ltd., London, UK.

1101 Epstein, M. 2010. The Geometrical Language of Continuum Mechanics. Cambridge University
1102 Press, Cambridge, UK.

1103 Federico, S. 2010. Volumetric-distortional decomposition of deformation and elasticity tensor.


1104 Math. Mech. Solids, 15:672–690.

1105 Federico, S. 2012. Covariant formulation of the tensor algebra of non-linear elasticity. Int. J.
1106 Non-Linear Mech., 47:273–284.

1107 Hughes, T. J. R. and Marsden, J. E. 1977. Some applications of geometry in continuum


1108 mechanics. Rep. Math. Phys., 12:35–44.

1109 Marsden, J. E. and Hughes, T. J. R. 1983. Mathematical Foundations of Elasticity.


1110 Prentice-Hall, Englewood Cliff, NJ, USA.

1111 Ogden, R. W. 1997. Non-linear Elastic Deformations. Dover, New York, USA.

1112 Walpole, L. J. 1981. Elastic behavior of composite materials: Theoretical foundations. Adv. in
1113 Appl. Mech., 21:169–242.

1114 Walpole, L. J. 1984. Fourth-rank tensors of the thirty-two crystal classes: Multiplication tables.
1115 Proc. Roy. Soc. Lond. A, 391:149–179.

50

Potrebbero piacerti anche