You are on page 1of 105

See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/226089555

A Primer in Elasticity

Article  in  Journal of Elasticity · January 2000


DOI: 10.1023/A:1007672721487

CITATIONS READS

72 965

1 author:

Paolo Podio-Guidugli
Accademia Nazionale dei Lincei
263 PUBLICATIONS   3,173 CITATIONS   

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Non-Classical Thermoelasticity View project

Blending together Continuum Mechanics and Molecular Dynamics View project

All content following this page was uploaded by Paolo Podio-Guidugli on 19 February 2014.

The user has requested enhancement of the downloaded file.


Journal of Elasticity 58: 1–104, 2000.
P. Podio-Guidugli, A Primer in Elasticity. 1
© 2000 Kluwer Academic Publishers. Printed in the Netherlands.

CHAPTER I
Strain

1. Deformation. Displacement
Let E be a 3-dimensional Euclidean space, and let V be the vector space associated
with E. We distinguish a point p ∈ E both from its position vector

p(p) := (p − o) ∈ V

with respect to a chosen origin o ∈ E and from any triplet (ξ 1 , ξ 2 , ξ 3 ) ∈ IR3 of


coordinates that we may use to label p. Moreover, we endow V with the usual
inner product structure, and orient it in one of the two possible manners. It then
makes sense to consider the inner product a · b and the cross product a × b of
two elements a, b ∈ V; in particular, we define the length of a vector a to be |a| =
(a · a)1/2 , and denote by U := {v ∈ V | |v| = 1} the sphere of all vectors having
unit length. When needed or simply convenient, we think of E as equipped with a
Cartesian frame {o; c1 , c2 , c3 } with orthogonal basis vectors ci ∈ U (i = 1, 2, 3);
the Cartesian components of a vector v ∈ V are then vi := v · ci and, in particular,
the triplet (p1 , p2 , p3 ) ∈ IR3 , pi := p(p) · ci , of components of the position vector
are the Cartesian coordinates of a point p ∈ E.
We identify a continuous body with one of its possible shapes, a (bounded,
connected) regular region " of E,# and we call " the reference shape. A mapping
p $→ f (p) from the closure " of " onto a closed region of E is a deformation
of the body if it is one-to-one, smooth (typically, C 1 (") ∩ C 0 ("),## and locally
# Here we refer to the notion of a regular region as introduced in [15] (vid. also [11, Section 5]. For
our present purposes it suffices to say, briefly and roughly, that " is a regular region if it coincides
with the interior of its closure and, moreover, its boundary has a well-defined outer unit normal
everywhere, except perhaps for a finite number of corners and edges.
## To specify the continuity class of a tensor field ! over a region R, the most frequently used
notations are the componentwise notation Ψij ...k ∈ C(R) and ! ∈ (C(R))d a power notation that
keeps explicit track of the dimension d of the vector space where ! lives. We dislike both these
notations, the former because it is not intrinsic, the latter because it is redundant. We write instead
! ∈ C(R), and let our reader be reminded of the dimension of the space where the field ! takes
its values by choosing in a systematic manner font, case, and style, of the letter denoting the field in
question.
2 P. PODIO-GUIDUGLI

orientation-preserving in the sense that


det ∇f (p) > 0, p∈" (1.1)
(here we have written det for the determinant, and ∇ for the gradient operator). We
call f (") the deformed shape of the body. For i the identity of E, the mapping
from " into V defined by
u := f − i (1.2)
is the displacement associated with the deformation f ; we say that u(p) =
f (p) − p is the displacement of a point p of the body, from the reference to
the deformed shape. We denote the deformation gradient and the displacement
gradient, respectively, by
F = ∇f, H = ∇u, (1.3)
so that
F = I + H, (1.4)
where I = ∇i is the identity of V. The values H(p) of the displacement gradient
are elements of Lin, the set of all linear transformations of V into itself (the space of
all second-order tensors), whereas, due to condition (1.1), the deformation gradient
takes values in
Lin+ := {F ∈ Lin | det F > 0}.#
Global invertibility is the most difficult a priori requirement to meet in order to
guarantee that the solutions of the boundary value problems of continuum mechan-
ics, in the absence of self-contact, fracture, healing etc., are indeed deformations.
However, global invertibility is not an issue when, as we now do, one aims at a
local analysis of deformation.
Let p ∈ ", e ∈ U be fixed, and let a deformation f be given. By the material
fiber through p in the direction e we mean the ordered pair (p, e); the image of that
# The determinant is the real-valued function on Lin defined by the property

(det A)a × b · c = (Aa × Ab) · Ac, (∗)

for all A ∈ Lin, and for all a, b, c ∈ V such that a × b · c )= 0. Other relevant properties of the
determinant are:

det(αA) = α 3 det A, det AT = det A, det (AB) = (det A)(det B), (∗∗)

for α ∈ IR, A, B ∈ Lin.


Formulae (∗) and (∗∗)2 have a different status: the former holds only when the underlying vector
space is 3-dimensional, the latter holds in arbitrary dimension (for another example, take (∗∗)1 and
(∗∗)3 ). It is important to note once and for all that, as a rule with only advertised exceptions, we shall
not pursue generality per se whenever simplifications ensue from the use of definitions and properties
that are typical of the dimension 3.
STRAIN 3

Figure 1.

fiber in the deformation f is the ordered pair (f (p), F(p)e). This terminology is
easily motivated if we consider the Taylor expansion
! "
f (p + αe) − f (p) = F(p) (p + αe) − p + o(α)

of the C 1 -mapping α $→ f (p + αe) (Figure 1); and recall the notion of directional
derivative of f in the direction e:
f (p + αe) − f (p)
∂e f (p) := lim (1.5)
α$→0 α
to write

∂e f (p) = F(p)e. (1.6)

Thus, we may write (f (p), ∂e f (p)) for the image of the fiber (p, e) in the de-
formation f . Basically, as we shall now show, the local analysis of a deformation
consists in the reiterated use of formula (1.6).

(i) Change in length. The notion of fiber makes precise the somewhat vague
notion of oriented line element. The change in length δl(e) of a fiber (p, e) is the
length of its image, |Fe|, minus its length, |e| = 1, divided by its length:

δl(e) := |Fe| − 1. (1.7)

Note that in stating this definition we have left tacit the dependence on p, as we
shall generally do in what follows. We call δl(e) the local change in length in the
direction e; and we call

λ(e) := 1 + δl(e) = |Fe| (1.8)

the stretch of the fiber (p, e).


4 P. PODIO-GUIDUGLI

(ii) Change in area. Consider now two noncollinear fibers through p, say, (p, e1 )
and (p, e2 ). In the reference shape those fibers determine the oriented surface
element of normal
e1 × e2
nR = , (1.9)
|e1 × e2 |
whose image under f has normal
Fe1 × Fe2
n= (1.10)
|Fe1 × Fe2 |
(Figure 2). For a referential surface of normal nR , the local change in area is defined
as
|Fe1 × Fe2 | −| e1 × e2 |
δa(nR ) := . (1.11)
|e1 × e2 |

Figure 2.

To write the last formula in a more compact form, some further algebraic no-
tions are useful. Let the collections of symmetric and skew elements of Lin be
denoted by Sym and Skw, respectively, so that
Lin = Sym ⊕ Skw, (1.12)
i.e., for each A ∈ Lin,
A = sym A + skw A, (1.13)1
# $ # $
2 sym A := A + AT ∈ Sym, 2 skw A := A − AT ∈ Skw; (1.13)2

here AT is the transpose of A, defined by


a · Ab = AT a · b, a, b ∈ V. (1.14)
For each w ∈ V there is one W ∈ Skw such that
Wv = w × v, v ∈ V; (1.15)
STRAIN 5

conversely, the relation (1.15) associates a unique axial vector w with each skew
tensor W. Given A ∈ Lin, the cofactor A∗ of A is the unique element of Lin such
that, whenever w ∈ V and W ∈ Skw obey (1.15), A∗ w and AWAT obey it as well:
# $
AWAT v = (A∗ w) × v, v ∈ V.
From this definition, it follows that
A∗ (a × b) = Aa × Ab, a, b ∈ V; (1.16)
moreover, if A ∈ Lin+ , then
# $−1 # −1 $T
A∗ = (det A)A−T , A−T = AT = A , (1.17)
where A−1 is the inverse of A.
With (1.16) we can write (1.10) as
F∗ nR
n= , (1.18)
|F∗ nR |
and (1.11) as
% %
δa(nR ) = %F∗ nR % − 1. (1.19)

REMARK. Formula (1.18) shows that the normal n to the image surface of a
material surface of normal nR through a point p ∈ " is determined by the action
at p of F∗ on nR , the geometric object characterizing the material surface element in
question. In this respect that formula is more significant than (1.10), which involves
any two noncollinear fibers generating the undeformed surface; the latter formula,
however, makes it evident – in the light of (1.6) – that only tangential derivatives
count in determining n.

(iii) Change in volume. Let now (p, e1 ), (p, e2 ) and (p, e3 ) be three noncoplanar
fibers along the edges through p of a parallelepiped volume element in the reference
shape, whose image under the deformation f is determined by (f (p), F(p)e1 ),
(f (p), F(p)e2 ), (f (p), F(p)e3 ) (Figure 3).

Figure 3.
6 P. PODIO-GUIDUGLI

The local change in volume is then


Fe1 · Fe2 × Fe3 − e1 · e2 × e3
δv := , (1.20)
e1 · e2 × e3
or rather, in view of (1.16), (1.17) and (1.14),
δv = det F − 1.# (1.21)

(iv) Change in angle. Since the cosine map is uniquely invertible on [0, π ], the
angle between two nonnull vectors a, b is well-defined by
a·b
ϑ = cos−1 . (1.22)
|a| |b|
Let (p, e1 ), (p, e2 ) be two fibers at an angle ϑR = cos−1 (e1 · e2 ) in the reference
shape (Figure 4); their change in angle in a deformation f is defined to be
Fe1 · Fe2
δϑ(e1 , e2 ) := cos−1 (e1 · e2 ) − cos−1 . (1.23)
|Fe1 | |Fe2 |

Figure 4.

EXERCISES

1. Show that, in a deformation,


% %
δa(nR ) = δv ⇔ %F−T nR % = 1. (1.24)
2. Show that the angle ϕ between the image under a deformation f of the
material fiber (p, e) and the normal to the image of a material surface through p of
normal e is
det F
ϕ = cos−1 , (1.25)
|Fe| |F∗ e|
and give a geometrical interpretation of this result.
# Alternatively, (1.21) follows from (1.20) in view of (∗) in the footnote on p. 2.
STRAIN 7

2. Rigid Deformations. Pure Strains


A deformation f is homogeneous if it has constant gradient in ", i.e., if it has the
form

f (p) = fo + Fo (p − po ) (2.1)

(so that ∇f (p) ≡ Fo in ", and f (po ) = fo ); in view of (1.2) the displacement
corresponding to (2.1) is

u(p) = uo + Ho (p − po ), (2.2)

with uo = u(po ) = fo − po and Ho = Fo − I. We say that a homogeneous


deformation leaves the point po fixed if fo = po , or, equivalently, if uo = 0.
A homogeneous deformation is rigid if it has the form (2.1) with Fo ∈ Rot, with
& '
Rot := R ∈ Lin+ | RT R = RRT = I ;

in particular, a rigid deformation is a translation if Ho = 0, a rotation (about po ) if


uo = 0.
Gradients of rotations compose a maximal subgroup of the orthogonal group
Orth of Lin, with
& '
Orth := Q ∈ Lin | QT Q = QQT = I .

Thus, Orth is the collection of all elements Q of Lin whose transpose QT and
inverse Q−1 are equal; it can be represented as the direct product of Rot and the
two-element group {I, −I} consisting of the identity and the central reflection −I.
Alternatively, Orth can be defined as the collection of all second-order tensors that
preserve the inner product of vectors:

Q ∈ Orth ⇔ Qa · Qb = a · b, a, b ∈ V. (2.3)

Thus, in particular, the distance |p − q| of any two points p, q ∈ " is preserved


in a rigid deformation, and conversely. Moreover, with (1.7), (2.3) implies that
δl(e) ≡ 0, and, with (1.23), that δϑ(e1 , e2 ) ≡ 0 in a rigid deformation. Finally, by
the definition of Rot and (1.17) we have that, for R ∈ Rot,

det R = 1, R∗ = R, (2.4)

and we see from (1.19) and (1.21) that the area and volume are unchanged in a
rigid deformation.
The following representation formula for a typical element R of Rot reflects the
physical expectation that a rotation is completely characterized by an axis and an
angle (cf. [12, p. 49])

R(w, ϕ) = I + sin ϕW + (1 − cos ϕ)W, (2.5)


8 P. PODIO-GUIDUGLI

where w ∈ U determines the axis and ϕ ∈ ]−π, π [ the angle of rotation, and where
W ∈ Skw is associated with w by (1.15). With the help of a bit more algebra we
can use (2.5) to describe the action of a rotation tensor on a vector.
For a, b ∈ V, the dyadic product of a and b is the element a ⊗ b ∈ Lin defined
by
(a ⊗ b)v := (b · v)a, v ∈ V. (2.6)
Let span(a) be the line spanned by the first factor of the dyadic product, and let
{b}⊥ be the plane obtained as the orthogonal complement of the second factor.
Then, for each v ∈ V fixed, it follows from definition (2.6) that the whole subspace
{v + u | u ∈ {b}⊥} of V is mapped into one point of span(a). In particular, for
w ∈ U, P(w) := w ⊗ w projects V orthogonally onto the line spanned by w,
whereas the complementary projector (I−P(w)) maps V onto the plane orthogonal
to that line (Exercise 2). Moreover, for W the skew tensor associated with w,
W2 = −(I − w ⊗ w), W3 = −W.# (2.7)
Fix v ∈ V, and use (2.5) to look at the image of v under R(w, ϕ) as the sum of
three vectors:
R(w, ϕ)[v] = v + sin ϕWv + (1 − cos ϕ)W2 v. (2.8)
It follows from (2.7) that, as Figure 5 suggests, Wv is orthogonal to both v and
W2 v; moreover, whenever v is chosen in the plane orthogonal to the rotation axis
w (as is done in Figure 5), v and W2 v are parallel with opposite directions.
Now let the inner product of Lin be defined in terms of the trace function## by
# $
A · B := tr ABT , A, B ∈ Lin, (2.9)

Figure 5.
# Thus, W4 = −W2 , W5 = W, etc.
## The trace is the linear function on Lin characterized by the following property:

tr(a ⊗ b) = a · b, a, b ∈ V.
Other relevant properties of the trace function are:
tr A = tr AT , tr(AB) = tr(BA), A, B ∈ Lin. (cont.)
STRAIN 9

and, with slight abuse of notation, let

|A| := (A · A)1/2 . (2.10)

A small rotation is a rotation of a small angle or, alternatively, a rotation whose gra-
dient differs little from the identity. We formalize this by introducing as a smallness
parameter

ε := |R − I|. (2.11)

Note that, by (2.5), (2.7) and (2.10),

ε = 2(1 − cos ϕ)1/2, (2.12)

so that the parameter ε is small if and only if (the absolute value |ϕ| of) the rotation
angle is small; moreover,
# $
sin ϕ = O(ε), 1 − cos ϕ = O ε 2 . (2.13)

From the representation formula (2.8) we have that


# $ ! "
R(w, ϕ) − I = ϕW + O ε 2 , skw R(w, ϕ) − I = (sin ϕ)W; (2.14)

it follows that, for a rigid deformation, the first-order approximations of the dis-
placement gradient and of the skew part of the latter coincide. This motivates the
terminology infinitesimal rigid displacement for a vector field on " that admits the
representation

u(p) = uo + Wo (p − po ), Wo ∈ Skw. (2.15)


A pure strain is a homogeneous deformation whose gradient is a positive# ten-
sor. Since, for each F ∈ Lin+ there are a unique rotation R and two uniquely
determined positive tensors U, V such that the following factorizations of F hold:

F = RU = VR,## (2.16)
any homogeneous deformation f leaving a point po fixed may be regarded as the
composition, in the appropriate order, of two pure strains d and s leaving po fixed
(cont.) In dimension 3, the trace function can be defined in a manner that resembles the definition of
determinant in footnote# on p. 2, namely,

(tr A)a × b · c = (Aa × b) · c + (a × Ab) · c + (a × b) · Ac (∗)

for all A ∈ Lin, and for all a, b, c ∈ V such that a × b · c )= 0.


# A symmetric tensor S is said to be nonnegative if S · v ⊗ v ! 0 for all v ∈ V, positive if the
inequality is strict for all vectors except 0.
## This factorization result is a corollary of the so-called polar decomposition theorem (vid. [11,
Section 83; 18], which states that for each F ∈ Lin there are Q ∈ Orth and exactly two nonnegative
tensors U, V such that F = QU = VQ.
10 P. PODIO-GUIDUGLI

and a rotation r about po :


f =r ◦d =s ◦r (2.17)
(Exercise 13).

EXERCISES

1. Validate the following alternative definition of Rot:


& '
Rot := R ∈ Lin\{0} | R = R∗ .

2. Let P be an orthogonal projector on V , i.e., a second-order tensor such that


both P ∈ Sym and P2 = P. Prove that P must have one of the following four
representations: I, O; p ⊗ p, I − p ⊗ p, for p ∈ U.
3. For a, b ∈ V, prove that
det(a ⊗ b) = 0, (a ⊗ b)∗ = 0; (2.18)
(a ⊗ b)T = b ⊗ a; (b ⊗ a − a ⊗ b)v = (a × b) × v, v ∈ V. (2.19)

4. Let a, b, c be arbitrarily chosen in V . Show that


a × (b × c) = (a · c)b − (a · b)c = (b ⊗ c − c ⊗ b)[a]
= ((c · a)I − c ⊗ a)[b] = −((b · a)I − b ⊗ a)[c]; (2.20)
(a × b) × c − a × (b × c) = (a × c) × b
= (a · b)c − (b · c)a = (c ⊗ a − a ⊗ c)[b]; (2.21)

(I − a ⊗ a)∗ = (1 − a · a)I + a ⊗ a. (2.22)


In particular, observe that, for a ∈ U, the cofactor of the orthogonal projector (I −
a ⊗ a) is the complementary projector
( a ⊗ a; and that, moreover,
( since for c1 , c2 , c3
an orthonormal set in V, I = 3i=1 ci ⊗ ci , we have that 3i=1 (I − ci ⊗ ci )∗ = I.
5. It follows from definitions (2.6) and (2.9) that
a · Ab = A · a ⊗ b, a, b ∈ V and A ∈ Lin; (2.23)
(2.9) and (2.21) together allow for the definition of the Cartesian components of a
second-order tensor A:
Aij := A · ci ⊗ cj , i, j = 1, 2, 3. (2.24)
Show that
(a ⊗ b)ij = ai bj , A · B = Aij Bij . (2.25)
6. Let S ∈ Sym and W ∈ Skw. Show that
S · W = 0; S · A = S · sym A, W · A = W · skw A, A ∈ Lin. (2.26)
STRAIN 11

7. Show that the relation (1.15) associates with the orthonormal vectors ci the
skew tensors Wi = −(ci+1 ⊗ci+2 −ci+2 ⊗ci+1 ) (i = 1, 2 or 3, modulo 3). Moreover,
show that these tensors form an orthogonal basis for Skw, so that, in particular,
the skew tensor W associated with the vector w = wi ci has the representation
W = wi Wi .
8. Show that, for W, Z ∈ Skw and w, z the corresponding axial vectors,
WZ = z ⊗ w − (w · z)I; (2.27)
so that, in particular,
WZ − ZW = z ⊗ w − w ⊗ z, W · Z = 2w · z and |W|2 = 2|w|2 .
9. By a direct use of the definition of cofactor, show that
# ∗ $T
A = (AT )∗ , (AB)∗ = A∗ B∗ (2.28)
for all A ∈ Lin.
10. Let A ∈ Lin, a, b ∈ V be arbitrarily chosen. Prove the following identity
# $
Au × (a ⊗ b)v + (a ⊗ b)u × Av = (A + a ⊗ b)∗ − A∗ (u × v) (2.29)
for all u, v ∈ V.
11. For a, b, c, and d arbitrary elements of V, show that
(a ⊗ b + c ⊗ d)∗ = (a × c) ⊗ (b × d); (2.30)
so that, in particular, if a, b, c are orthonormal, then
b ⊗ a = −(a ⊗ b + c ⊗ c)∗ . (2.31)
12. Show that, for W ∈ Skw and w its axial vector,
W∗ = w ⊗ w. (2.32)
Alternatively, deduce (2.30) from (2.28).
13. Show that, in relation (2.17), the deformations r, d, and s have the form
r(p) = po + Ro (p − po ),
d(p) = po + Uo (p − po ), s(p) = po + Vo (p − po ) (2.33)
respectively, with Ro Uo = Vo Ro = Fo and Fo the gradient of f .

3. Strain Measures
Formulae (1.7), (1.19), (1.21) and (1.23) are the essence of the exact, local analysis
of deformation; remarkably, they also directly suggest how to measure strain.
12 P. PODIO-GUIDUGLI

If we take (1.14) and (2.21) into account, we can write (1.7) in the form
δl(e) = (C · e ⊗ e)1/2 − 1, C := FT F. (3.1)
The tensor C is a local and exact strain measure in the sense that, as (3.1) shows, its
component in the direction e determines – with no approximation – the deformed
length of a material fiber having that direction and, moreover, each one of the
remaining basic formulae (1.19), (1.21) and (1.23) can be written in terms of C (cf.
Exercises 1–3). There are many other strain measures; one frequently used in the
mechanics of solids is
1# $ 1
D := FT F − I = (C − I). (3.2)
2 2
For H related to F as in (1.4) let
1# $
E := H + HT , (3.3)
2
so that
1
D = E + HT H, C = I + 2E + HT H. (3.4)
2
A reasonable requirement in the definition of a strain measure is that it have
constant value over the collection of rigid deformations: one quickly verifies that
C = I, and D = 0, at the identity deformation and, more generally, at any rigid de-
formation. Formulae (3.4) make clear that neither the tensor E nor any other linear
construct based on the deformation gradient could ever meet such a requirement
exactly: at the identity deformation, e.g., E(p) ≡ 0 in ", whereas in a typical rigid
deformation
# $
E(p) ≡ sym(R − I) = (1 − cos ϕ)W2 = O ε 2 (3.5)
(cf. (2.14)). As we shall see in the next section, E does measure small strains; for
this reason it is called the infinitesimal strain tensor.

EXERCISES

1. Show that
δa(nR ) = (C∗ · nR ⊗ nR )1/2 − 1. (3.6)
2. Show that
δv = (det C)1/2 − 1. (3.7)
3. Show that
C · e1 ⊗ e2
δϑ(e1 , e2 ) = ϑR − cos−1 . (3.8)
(C · e1 ⊗ e1 )1/2 (C · e2 ⊗ e2 )1/2
STRAIN 13

In particular, confirm that one can read from (3.8) that


π Cij
δϑ(ci , cj ) = − cos−1 ) , i )= j. (3.9)
2 Cii Cjj
4.
(i) Show that
(αA)∗ = α 2 A∗ , α ∈ IR, A ∈ Lin; (3.10)
det(A + B) = det A + A∗ · B + A · B∗ + det B, A, B ∈ Lin. (3.11)

(ii) Prove that the local volume change has the following expression in terms of
the strain measure D:

δv = (1 + 2tr D + 4tr D∗ + 8det D)1/2 − 1. (3.12)

(iii) More generally, prove that, for A ∈ Lin,

3
*
det(I + αA) = 1 + α n ιn (A), (3.13)
n=1

where ιn (A) denotes the nth orthogonal invariant of A:

ι1 (A) := tr A, 2ι2 (A) := (tr A)2 − tr A2 , ι3 (A) := det A. (3.14)

5. Let W be the skew tensor associated with the unit vector w, and let α ∈ IR.
Show that
det(I + αW) = 1 + α 2 , (3.15)
and that
# $−1 # $
(I + αW)−1 = 1 + α 2 I − αW + α 2 w ⊗ w . (3.16)
6. Show that
% %2
ι2 (C) = %F∗ % . (3.17)
7. Confirm that none of the formulae of Section 1 for the changes in length,
area, etc., as well as none of the strain measures introduced in this section, actually
involves the rotation factor R of the deformation gradient F, introduced by the
factorization (2.16).
8. Given a symmetric second-order tensor S, the proper pairs of S are the
solutions (σ, s) in IR × U of the equation
(S − σ I)s = 0. (3.18)
14 P. PODIO-GUIDUGLI

Use (1.8), (3.1), and (3.18) to show that, in a deformation of gradient F, each proper
number of the strain measure C = FT F at a point is the square of a principal stretch,
i.e., of the stretch of a material fiber having the direction of a proper vector of C
corresponding to that proper number.

4. Small Strain
Intuitively, a deformation f is small at a point of " if its gradient F at that point
differs little from the identity. Just as we have done for rotations, we introduce the
smallness parameter
ε := |F − I| = |H|; (4.1)
with this definition, a deformation is (locally) small if its displacement gradient is
small.
Directly from (3.4)1 we deduce that
# $
D = E + O ε2 , (4.2)
a formula that assigns the infinitesimal strain tensor E a position as the linear
approximation, in the sense of (4.1), of the exact strain measure D. We shall write
(4.2) in the form
D∼
= E, (4.3)
and consistently use the symbol ∼ = to mean that equality holds to within O(ε 2 )
terms. This result, as well as all other approximate formulae used to measure strain
in a small deformation, may be obtained by the following formal procedure:
(i) for ε regarded now as a scaling parameter, write

H(ε) = εH, F(ε) = I + εH; (4.4)

(ii) insert (4.4) in the exact formula to be linearized, and obtain a smooth mapping
in ε: e.g., insert (4.4) into (3.2)–(3.41 ) to get

1# T $ 1
D(ε) = F (ε)F(ε) − I = εE + ε 2 HT H; (4.5)
2 2

(iii) compute the linear approximation about ε = 0 of the nonlinear mapping


obtained in step (ii), and then scale back the resulting relation by letting ε =
1: for the mapping in (4.5), (4.3) obtains because

D(1) ∼
= D(0) + D1 (0) = E (4.6)

(here a prime denotes differentiation with respect to ε).


STRAIN 15

As another example of application of this modus operandi we derive approx-


imate formulae for the local change in length and the stretch of a fiber. Inserting
(4.4) in (1.7) we have
% %
δl(e; ε) = %(I + εH)e% − 1, (4.7)
and thus, as
(I + εH)e
δl 1 (e; ε) = · He, δl 1 (e; 0) = e · He, (4.8)
|(I + εH)e|
it follows that
δl(e) ∼
= E · e ⊗ e, λ(e) ∼
= 1 + E · e ⊗ e.# (4.9)
The linear theory of strain regards these last relations as exact; therefore, within
that theory, we shall replace ∼
= by =.
Our next task is to linearize the expressions (1.19) and (1.21) for the local
changes in area and volume. We begin with the latter, for which we use a formula
known to Euler:
∂F (det F) = F∗ , (4.10)
to see that
# $1
δv 1 (ε) = det F(ε) = F∗ (ε) · F1 (ε), δv 1 (0) = I · H, (4.11)
and, hence, that
δv ∼
= E · I = tr E. (4.12)

REMARK. The local invertibility condition (1.1), a mandatory requirement in


the exact analysis of deformation (in fact, a definitory character of a deformation
within that theory), evaporates in the linear analysis, because
det F = 1 + O(|F − I|).

As to area, it follows from (1.17) that


FT F∗ = (det F)I; (4.13)
# Here we have used the fact that the first derivative of the modulus of a vector is the corresponding
unit vector:
v
∂v |v| = ;
|v|
and the fact that A · a ⊗ a = (sym A) · a ⊗ a for all A ∈ Lin and a ∈ V. We also record here for later
use the formula for the second derivative of the modulus:
+ v , + v v ,
∂v (∂v |v|) = ∂v = |v|−1 I − ⊗ .
|v| |v| |v|
16 P. PODIO-GUIDUGLI

differentiating the composition of (4.4)2 , with (4.13) we have that


# T $1 ∗ # $1 # $1
F (ε) F (ε) + FT (ε) F∗ (ε) = det F(ε) I, (4.14)
and thus, in view of (4.11),
# ∗ $1
F (0) = (I · H)I − HT . (4.15)
With (4.15), we obtain
F∗ (ε)nR # ∗ $1
δa 1 (nR ; ε) = · F (ε) nR , δa 1 (nR ; 0)
|F∗ (ε)nR |
# $
= nR · (I · H)I − HT nR , (4.16)
and conclude that
δa(nR ) ∼
= E · (I − nR ⊗ nR ). (4.17)
To complete our construction of the linear theory of strain, it remains for us to
linearize the formula (1.23) for the change in angle of two fibers at an angle ϑR =
cos−1 (e1 · e2 ) in the reference shape. Inserting (4.4) into (1.23), and differentiating
with respect to ε, we obtain
- .
−1 F(ε)e1 · F(ε)e2
sin cos δϑ 1 (e1 , e2 ; ε)
|F(ε)e1 | |F(ε)e2 |
- .
% %−1 F(ε)e 1 F(ε)e 1 F(ε)e2
= %F(ε)e1 % I− ⊗ F1 (ε)e1 ·
|F(ε)e1 | |F(ε)e1 | |F(ε)e2 |
- .
% %−1 F(ε)e2 F(ε)e2 F(ε)e1
+%F(ε)e2 % I− ⊗ F1 (ε)e2 · , (4.18)
|F(ε)e2 | |F(ε)e2 | |F(ε)e1 |

1 ## $
δϑ 1 (e1 , e2 ; 0) = H + HT · e1 ⊗ e2
sin ϑR
$
−(cos ϑR )H · (e1 ⊗ e1 + e2 ⊗ e2 ) (4.19)
(cf. the last footnote). Thus, in particular,
δϑ(e1 , e2 ) ∼
= 2E · e1 ⊗ e2 , e1 · e2 = 0. (4.20)

EXERCISES

1. Linearize (3.1) and (3.6)–(3.8).


2. Prove that
F∗ ∼
= (1 + tr H)I − HT . (4.21)
STRAIN 17

3. Show that linearization of (1.18) yields the following relation between the
current and the reference unit normal to a fixed material plane:
n∼
= nR − (I − nR ⊗ nR )HT nR . (4.22)

5. Simple Deformations
We are now in a position to consider homogeneous displacement fields (Section 2),
as well as the accompanying strain fields, from the point of view of the linear theory
of strain.
We write (2.2) as
u(p) − u(po ) = Ho (p − po ) = Wo (p − po ) + Eo (p − po ), (5.1)1
Wo = skw Ho , Eo = sym H. (5.1)2
These relations are meaningful no matter how small the displacement gradient may
be. Not so their interpretation: to say that in (5.1) the displacement from po is split
into the (infinitesimal) rigid displacement Wo (p−po ) and the (infinitesimal) purely
deformational displacement Eo (p − po ) it is necessary to agree that |Ho | is small
(and hence so are both |Wo | and |Eo |).
We call a homogeneous deformation simple whenever the associated displace-
ment field is purely deformational. Within the framework of the linear theory of
strain, simple deformations play the same role as pure strains in the exact theory.
There are three basic types of simple deformations:
(i) extension of amount α in the direction e:
# $
u(p) = α e · (p − po ) e, Eo = αe ⊗ e; (5.2)

(ii) shear of amount β with respect to the orthogonal directions e1 , e2 :


!# $ # $ "
u(p) = β e1 · (p − po ) e2 + e2 · (p − po ) e1 ,
Eo = β[e1 ⊗ e2 + e2 ⊗ e1 ]; (5.3)

(iii) dilatation of amount γ :

u(p) = γ (p − po ), Eo = γ I (5.4)

(in these formulae α, β, and γ are given real numbers, whose absolute value
equals, or is proportional to, the value of the smallness parameter defined by
(4.1)).
The decomposition (1.12) of the space Lin into the direct sum of its subspaces
Skw and Sym is the algebraic substance of the decomposition (5.1) of homoge-
neous displacements into rigid and purely deformational parts. Similarly, a decom-
position of Sym;
Sym = Sph ⊕ Dev, (5.5)1
18 P. PODIO-GUIDUGLI
/ 0
1
Sph := A ∈ Sym | A = (tr A)I) , Dev := {A ∈ Sym | tr A = 0}, (5.5)2
3

allows us to resolve every simple deformation into a dilatation and an isochoric#


combination of three extensions and three shears. To see this, let
1
sph A := (I · A)I, dev A := A − sph A (5.6)
3
be the complementary orthogonal projectors of Sym onto Sph and Dev, respec-
tively, and let c1 , c2 , c3 be the orthonormal set of vectors introduced in Section 1.
Then, we can write every given pure strain u(p) = Eo (p − po ) in the form:

u(p) = (sph Eo )(p − po ) + (dev Eo )(p − po ), (5.7)

with
3
* # $
(dev Eo )(p − po ) = (dev Eo )ii ci · (p − po ) ci
i=1
!# $ $ "
+(dev Eo )12 c1 · (p − po ) c2 + (c2 · (p − po ) c1
!# $ $ "
+(dev Eo )23 c2 · (p − po ) c3 + (c3 · (p − po ) c2
!# $ # $ "
+(dev Eo )13 c1 · (p − po ) c3 + c3 · (p − po ) c1 , (5.8)

where (1/3) tr Eo is the amount of the dilatation, and the extensions and shears
have amounts
1
(dev Eo )ii = (Eo )ii − tr Eo , (5.9)
3
and

(dev Eo )ij = (Eo )ij , (5.10)

respectively.
Various other decompositions of this sort are possible. One that yields a trans-
parent kinematical interpretation of the Cartesian components of the infinitesimal
strain tensor E is based on the following consequences of formulae (5.9) and (5.20):

δl(ci ) = E · ci ⊗ ci = Eii (index i unsummed), (5.11)

δϑ(ci , cj ) = 2E · ci ⊗ cj = 2Eij (i )= j ). (5.12)


# A deformation f is termed isochoric at a point p ∈ " if volume is preserved under f at that
point:

δv(p) = det∇f (p) − 1 ∼


= tr(∇f (p) − I) = 0.
STRAIN 19

Thus, when a deformation is studied within the linear theory of strain, diagonal
components of E measure changes in length, and off-diagonal components measure
changes in angle, of material fibers along the coordinate axes. In the light of (5.11)
and (5.12), a simple deformation may be accomplished by a sequence, in any order,
of three extensions of amount Eii in the direction of the coordinate axes, and three
shears of amount Eij in the coordinate planes.

EXERCISES

1. Let A ∈ Sym. Prove that, if the mapping v $→ A · v ⊗ v on U has constant


value, then A ∈ Sph.
2. Let a, b, c ∈ V be such that a × b = c, but otherwise arbitrary. For A ∈ Lin,
show that

(det A)a × b · c = AT A∗ · c ⊗ c, c ∈ V. (5.13)

3. Let A ∈ Lin. Prove that

AT A∗ = A∗ AT = (det A)I, (5.14)

and that formula (1.17)1 for A∗ holds whenever, in addition, A is invertible.

6. Divergence Identities
The validity of a suitable divergence lemma underlies a number of fundamental
developments in continuum mechanics, such as, e.g., the construction of a notion
of stress, or the weak formulation of boundary-value problems. The more general
the basic divergence lemma the broader the scope of the resulting theory; for the
purpose of this book the following standard version suffices.
Let R be a (bounded and connected) regular region of E, with boundary ∂R of
outer normal n, and let v(p) be a vector field on R of class C 1 (R) ∩ C 0 (R). Then,
1 1
∇v = v⊗n (6.1)
R ∂R

(Exercise 1).# This theorem has many relevant corollaries.


(i) Choose v(p) = ϕ(p)a in (6.1), with ϕ a smooth scalar field, and a an arbitrary
constant vector. Consider the identity

∇(ϕu) = ϕ∇u + u ⊗ ∇ϕ, (6.2)

# For brevity, here and in what follows we let the integration measure be suggested by the
indicated integration domain.
20 P. PODIO-GUIDUGLI

which holds for ϕ as above and u smooth. Then (6.1) reduces to


1 1
∇ϕ = ϕn (6.3)
R ∂R

(Exercise 3).
(ii) Since

Div v := tr(∇v) = I · ∇v, (6.4)

taking the trace of (6.1) we obtain the familiar divergence theorem for a vector
field:
1 1
Div v = v · n. (6.5)
R ∂R

(iii) In (6.5), choose v(p) = AT (p)a, with a an arbitrary constant vector and A a
smooth second-order tensor field over R. Then, in view of the identity
# $
Div AT u = u · Div A + A · ∇u,# (6.6)

(6.5) gives the divergence theorem for a second-order tensor field:


1 1
Div A = An. (6.7)
R ∂R

(iv) Take the skew-symmetric part of (1):


1 1
! T
"
∇v − (∇v) = [v ⊗ n − n ⊗ v]. (6.8)
R ∂R

Recall that the curl operator for a vector field v is defined to be the axial vector
of 2skw(∇v):
# $
(Curl v) × a = ∇v − (∇v)T a, a ∈ V, (6.9)

((cf. (1.15)); and that, by (2.17)2 , n × v is the axial vector of 2skw(v ⊗ n).
Then, (6.8) may be given the form of the curl theorem:
# The definition (6.4) of the divergence operator for a vector field yields the following definition
of the divergence operator for a tensor field:

Div(AT u) =: u · Div A,

for each vector field u with constant value.


STRAIN 21
1 1
Curl v = n × v. (6.10)
R ∂R

(v) Take the symmetric part of (6.1), and define the mean value of a second-order
tensor field A over R to be the volume average
1
1
A := A. (6.11)
volR R

Then, by definition (3.3),


1
1
E(v) = (v ⊗ n + n ⊗ v), (6.12)
2volR ∂R

a result that may be interpreted as follows: given a displacement field v over


a region R, the mean strain depends only on the boundary values of the
displacement.#
We can use (6.12) in two ways: to evaluate the average value of the strain field,
when displacements are given over the whole boundary; and conversely, when we
have information on E, to obtain restrictions that the accompanying displacement
field must satisfy at the boundary.
The first situation occurs when the so-called displacement equilibrium problem
is studied (Section 20). Take, e.g., R to be a ball of radius ρ, and let u(p) =
un(p), p ∈ ∂R; or, take R to be a circular cylinder with its axis in the direction e,
cross-sectional radius ρ and length λ, and let u(p) = 0 on the bases, u(p) = un(p)
on the rest of the boundary. Then, it is not difficult to show that
u u
E(u) = I, E(u) = (I − e ⊗ e), (6.13)
ρ ρ
respectively (Exercises 7 and 8).
An example of the second situation obtains when a uniform internal constraint
prevails in a continuous body. As we shall see in detail in Section 17, a linear inter-
nal constraint is a local condition, of a constitutive nature, that restricts the choice
of possible strains to some subspace of Sym; typically, such a condition consists
in assigning a constraint tensor field V(p) ∈ Sym over ", the closure of the body’s
reference shape, and requiring that the material be capable only of deformations
compatible with
# $
V(p) · E u(p) = 0, p ∈ ". (6.14)
# Both the basic lemma (6.1) and its consequence (6.12) do not depend on any smallness as-
sumption on either the field v or its gradient; however, as remarked in Section 5, it is only with
the framework of linear kinematics that E(v) admits interpretation as an exact measure of the strain
accompanying v.
22 P. PODIO-GUIDUGLI

Whenever the field V has constant value over ", we have from (6.12) and (6.14)
the following necessary condition on the prescription of displacement data at the
boundary:
1
u · Vn = 0. (6.15)
∂"
As an example, consider the constraint of incompressibility, for which V ≡ I
over " (cf. (4.12)); then (6.15) requires that the normal component of the pre-
scribed boundary displacements have null average, a condition equivalent to con-
serving the volume of ".

EXERCISES

1. Let a Cartesian frame be chosen. Show that, for v(p) a smooth vector field,
(∇v)ij = vi,j , ∇v = v,i ⊗ci , (6.16)
where (·),i := ∂(·)/∂pi . Use (6.16) to arrive at the following versions of (6.1):
1 1 1 1
vi,j = vi nj , v,i = ni v. (6.17)
R ∂R R ∂R
2. Establish the following consequences of (6.1):
1
(vol R)I = (p − po ) ⊗ n; (6.18)
∂R
1
n = 0. (6.19)
∂R

3. Show that, for R the right cylinder in Figure 6 and for ϕ(p) = ψ(p1 ), (6.3)
yields
1 β
ψ 1 (p1 ) dp1 = ψ(β) − ψ(α),
α

Figure 6.
STRAIN 23

a version of the Fundamental Theorem of Integral Calculus.


4. Verify the curl theorem in the case of an infinitesimal rigid displacement field.
5. Prove that
(i) if V denotes the skew tensor field associated with the vector field v, then

Div V = −Curl v; (6.20)

(ii) if v(p) ∈ U, p ∈ R, then

(∇v)v = −v × Curl v. (6.21)

6. Let v be a vector field, and let A be a symmetric-valued tensor field over a


region R. Show that, if both v and A are smooth, then
1 1 1
An · v = (Div A) · v + A · sym(∇v). (6.22)
∂R R

Use (6.22) to prove that (6.15) holds for a smooth, divergenceless constraint field
V(p) over ".
7. Let Un be the n-dimensional unit sphere in the vector space V n . Prove that
1
meas Un
n⊗n = IV n , (6.23)
Un n
where IV n is the identity of V (n) .
8. Let Eo ∈ Sym be given, and let u(p) = Eo (p − o), p ∈ ∂R. By two succes-
sive applications of (6.1), show that 2
E = Eo .
25

CHAPTER II
Stress

7. Forces. Balances
Once a body " and a deformation f of " have been fixed, we confront the problem
of modelling the accompanying mechanical interaction, both between parts of "
and between " and its environment. The simplest types of mechanical interactions
are described by forces. Various notions of a system of forces have been developed
to abstract from experience, and so reflect one or another set of prejudices about
the body and the environment under examination, both separately and when they
are paired. The world around us offers examples of surface actions on a body, such
as the contact action of the wind on a sail; and volume actions at a distance, such
as gravity. The creation of a notion of system of forces that efficiently accounts for
such surface–contact and volume–distance interactions, as well as the creation of
the related notion of stress, is essentially due to Euler and Cauchy.#
Formally, a Cauchy system of forces for a body " undergoing a deformation f
is the assignment of two vector fields: a surface-force field

s : f (") × U → V, (7.1)1

with the mapping s(·, n) smooth for each n ∈ U (typically, s(·, n) ∈ C 1 (f (")) ∩
C 0 (f (")) ); and a volume-force field

b : f (") → V, (7.1)2

with the mapping b continuous over f (").##


At a boundary point q of f ("), and for n(q) the outer unit normal to ∂f (")
at q, s(q, n(q)) is called the applied traction, and interpreted as the contact force,
per unit area of ∂f (") at q, exerted upon the body in its deformed shape by its
# Cauchy’s papers on the subject are dated 1823 and 1827, cf. [35, Section 200].
## We note that our definition of a deformation guarantees that, pointwise,

∂f (") = f (∂"), f (") = f (").

Thus, in particular, a deformation never entails creation or deletion of boundary points.


26 P. PODIO-GUIDUGLI

Figure 7.

Figure 8.

environment across their common boundary (Figure 7). At an interior point q of


f ("), for each oriented surface S through q of normal n at q, s(q, n) is called the
stress vector, and interpreted as the contact force, per unit area, exerted across S
upon the material on the negative side of S by the material on the positive side
(Figure 8). #
In addition to the contact force on the boundary, the environment exerts upon
the deformed body a distance force b(q) per unit volume at each interior point q
of f (").## On each body2 part 3 ⊂ 2" the total contact force and the total distance
force are the integrals ∂f (3) s and f (3) b, respectively.

REMARKS.
1. A body part 3 is not simply a subset, but rather a subbody, of ", in that it
is required that 3 can be regarded as a continuous body in itself;‡ accordingly, a
body is the maximal element of the collection of its parts.
# Thus, if S 1 is another surface through q that shares normal and orientation with S, then the stress
vector at q is the same for S and for S 1 .
## The physical dimensions of surface and volume force–densities are, respectively, [s] = F L−2
(F = force, L = length) and [b] = F L−3 .
‡ The notion of a continuous body we use is presented in the beginning of Section 1.
STRESS 27

2. Forces concentrated at points or distributed over curves, as well as mutual


body forces, do not fall within Cauchy’s framework as presented here; moreover,
Cauchy’s hypothesis that contact interactions consist simply of a contact force
depending only on the normal has been generalized in various ways.
3. Contact forces are typical of continuum mechanics. In Cauchy’s scheme the
contact actions between parts of the body and between the body and its environ-
ment are modelled in the same manner because the surface bounding a body, just as
the surface bounding any body part, is regarded as a geometrical boundary lacking
any particular mechanical substance. Consequently, within that scheme, splitting a
given material system in one or another (body, environment) pair is a mental oper-
ation that should not influence the modelling of the contact interactions among the
objects in the system: it would be unreasonable that the contact actions of adjacent
parts on any fixed part of a body were to change, when that part is regarded as a
body in itself.

Given a part 3 of a body ", a deformation f , and a system of forces (s, b), the
resultant force and the resultant moment (about the origin o) on 3 are defined to
be, respectively,
1 1
r(3, f ) := s+ b,
∂f (3) f (3)
1 1
mo (3, f ) := (q − o) × s + (q − o) × b. (7.2)
∂f (3) f (3)

Clearly,
mz (3, f ) = mo (3, f ) + (o − z) × r(3, f ), (7.3)
whence the well-known property that the resultant moment is independent of the
origin if and only if the resultant force vanishes.
A system of forces for a body " undergoing a deformation f is balanced if it
satisfies Euler’s axioms, i.e., both the resultant force and the resultant moment on
each part 3 ⊂ " vanish:
r(3, f ) = 0, mo (3, f ) = 0, 3 ⊂ "; (7.4)
when (7.4) hold, f is said to be an equilibrium deformation, and f (") an equilib-
rium shape, for ".
Euler’s axioms are central to mechanics in general, and continuum mechanics
in particular. They are general statements that concern all bodies in the theory, and
have therefore a different nature from constitutive assumptions, such as specifica-
tions of a system of forces for each admissible deformation, that concern only the
material class whose mechanical response they serve to define.

REMARK 4. The final quantification plays an essential role in determining the


consequences of relations (7.4). Roughly speaking, the richer the collection of parts
28 P. PODIO-GUIDUGLI

the stronger the balance axioms. For a single rigid body ", there is only one part,
and (7.4) take their weakest form

r(", f ) = 0, mo (", f ) = 0;

however, as is well known, this statement is generally insufficient to characterize


equilibrium of a system of rigid bodies, which has more than one and finitely many
parts, so that (7.4) have finitely many consequences; finally, (7.4) have maximal
strength in continuum mechanics, where the collection of parts of a body has the
cardinality of the continuum, and (7.4) imply pointwise balance statements.

When, rather than an equilibrium deformation, a motion is considered, Euler’s


axioms postulate the basic connections between forces and the motion itself.
Let a motion be regarded as a smooth one-parameter family of deformations
m = {ft | t ∈ T ⊂ IR} that preserve mass in the sense that, if ρR (p) is a positive-
valued scalar field assigning the mass density per unit volume of the body in its
reference shape ", then, for each value of the time parameter t in the interval T ,
# $−1
ρ(q) := det ∇ft (p) ρR (p), q = ft (p), (7.5)

gives the current mass density, i.e., the mass density per unit volume of ft ("), the
deformed shape at time t.#
With a view to deriving evolutionary versions of the balances (7.4), we first split
the distance force b into an inertial part

b(in) := −ρv. , (7.6)

with v(p, t) := ft (p) the velocity, and a noninertial part

b(ni) := b − b(in) .## (7.7)

Then, we make use of definitions (7.2) to rewrite the balance axioms (7.4) as
evolution equations: for each part 3 ⊂ ",
. mo (3, ft ) = a.o (3, ft ),
r(3, ft ) = l (3, ft ), (7.8)
# The mass, a positive scalar, is the single constitutive character of a mass point. Likewise, for a
continuous body, the mass density mapping should be regarded as a constitutive specification.
## We may interpret the noninertial part of the volume–force field as the interaction of the given
body with the other bodies within the world W on which we concentrate our interest, say, the
solar system; and the inertial part as an overall manifestation of the interaction of the given body
with the outside of W, say, the fixed stars. The notion of inertial distance forces, as well as more
generally our statement of Euler’s axioms in a dynamical context (relations (7.8)–(7.10)), may be
made unequivocal by the assumption, here left tacit, that there is an inertial observer who measures
and correlates forces and motions. More sophisticated approaches to the issue of inertia in continuum
theories have been developed (vid. [24] and the literature quoted therein), but we shall not need them
in this writing.
STRESS 29

with
1 1
r(3, ft ) = s+ b(ni) ,
∂ft (3) ft (3)
1 1
mo (3, ft ) = (q − o) × s + (q − o) × b(ni) , (7.9)
∂ft (3) ft (3)

and with l and ao , the linear and the angular momentum of part 3 in the given
motion, defined by
1 1
l(3, ft ) := ρv, ao (3, ft ) := (q − o) × (ρv). (7.10)
ft (3) ft (3)

Note that in (7.10) v is seen as a vector field over the trajectory


& '
T := (q, t) | q ∈ ft ("), t ∈ T
.
of the motion m, namely, v(q, t) = ft (ft−1 (q)). Note also that, in view of (7.5),
(7.10) can be equivalently written as
1 1
l(3, ft ) = ρR v, ao (3, ft ) = (q − o) × (ρR v), (7.11)
3 3

and, moreover,
-1 .. 1
ρv = ρv. ,
- 1ft (3) ft (3)
.. 1
(q − o) × (ρv) = (q − o) × (ρv. ).
ft (3) ft (3)

In their version (7.8)–(7.10), Euler’s axioms are usually referred to as the balance
laws of linear and angular momentum.
Given a body ", a deformation f , and a system of forces (s, b), and given a
vector field w over f ("), the power expended (on a part 3 of ", when " under-
goes the deformation f ) by the system of forces (s, b) for the “velocity field” w is
defined to be
1 1
P (3, f )[w] := s·w+ b · w. (7.12)
∂f (3) f (3)

In particular, let
u(q) = uo + wo × (q − qo ), q ∈ f ("), (7.13)
be an infinitesimal rigid displacement of the deformed shape (any rigid velocity
field over f ("), vid. Exercise 1), and consider P (3, f )[u], the power expended
by the system of forces (s, b) for each infinitesimal rigid displacement as above. It
follows from (7.2) and (7.13) that
P (3, f )[u] = r(3, f ) · uo + mqo (3, f ) · wo . (7.14)
30 P. PODIO-GUIDUGLI

We then have the following characterization result: a system of forces is balanced if


and only if the power expended on each part 3 ⊂ " vanishes in each infinitesimal
rigid displacement of f (").#

EXERCISES

1. A rigid motion of a body " is a one-parameter family of rigid deformations


(Section 2) of ":
q = ft (p) = fo (t) + R(t)(p − po ), R(t) ∈ Rot, (p, t) ∈ " × T .
Show that, for each fixed t ∈ T , the corresponding velocity field over ft (") can be
given the form
v(q) = vo + W[q − qo ], (7.15)
with
.# $
v(q, t) = ft ft−1 (q) ,
.
vo (t) = fo (t),
.
W(t) = R (t)RT (t) ∈ Skw, qo = fo (t).
(7.16)

2. Show that, for Wo the skew tensor corresponding to the vector wo in the sense
of (1.15), one has
-1 1 .
mqo (3, f ) · wo = −2 (q − qo ) ⊗ s + (q − qo ) ⊗ b · Wo .
∂f (3) f (3)
(7.17)

8. Stress. Dynamical Processes


The statement (7.4) of force and moment balances, as well as the characterization
of these balance laws in terms of virtual power, is both global and local, in that it
involves integrals, but over parts of arbitrary size and shape. Beside this partwise
statement, a pointwise statement of force and moment balances becomes possible
once a notion of stress has been built, and is in fact one of the assertions of a famous
theorem of Cauchy.##
The main thrust of the Cauchy theorem is a constructive proof of the basic fact
that, for each fixed q ∈ f ("), the mapping s(q, ·) has the representation
s(q, n) = T(q)n, n ∈ U, (8.1)
# In other words, Euler’s axioms may be restated under form of a single assertion concerning
the power expended in any infinitesimal displacement from the deformed shape; such a reductio ad
unum of the basic principles has been a long standing quest in the history of statics (vid. [3]).
## This theorem has been generalized in many respects; vid. [12, Section 14] for a modern proof
of the classical results in the form needed here.
STRESS 31

in terms of T(·), a tensor-valued field over f (") having the smoothness stipulated
for s(·, n). The pointwise equivalent of (7.4)1 is
div T + b = 0 in f ("); (8.2)1
granted (8.2)1 , the pointwise equivalent of (7.4)2 is
T ∈ Sym in f ("). (8.2)2
The tensor T is called the Cauchy stress; for q a typical interior point of f ("),
the stress vector on a surface through q oriented by n(q) is furnished by the action
of T on n, as specified by (8.1); similarly, the applied traction at a typical point of
∂f (") is given by the action of T on the exterior normal to ∂f (") at that point.

REMARK. When the global balance laws are given the evolutionary form (7.8),
Cauchy’s argument yields
div T + b(ni) = ρv. (8.3)
in the place of (8.2)1 , while (8.2)2 remains the same.

Let a motion m = {ft | t ∈ T ⊂ IR} be given, and let T be the corresponding


trajectory. A pair consisting of m and a symmetric-valued tensor field T defined
over T , with T(·, t) smooth for each t ∈ T , is called a dynamical process. Given a
dynamical process, the Cauchy theorem permits us to construct the corresponding
balanced system of forces
(s = Tn, b = −div T)
at each time. Conversely, given a motion m, the relation (8.1) between the surface
traction and the stress tensor implies that, for each ft ∈ m,
s(q, ci ) = T(q)ci , q ∈ ft ("),
for all triplets of orthonormal vectors ci , so that
T(q) = s(q, ci ) ⊗ ci , q ∈ ft ("); (8.4)
hence, there is exactly one dynamical process corresponding to a given t-parametr-
ized family of balanced systems of forces. A constitutive assumption is the pre-
scription of a class of dynamical processes; or, in view of the above discussion,
the prescription of a class of motions and the accompanying balanced systems of
forces. In our present context, the role of a constitutive prescription is to single out
a specific class of continuous bodies of Cauchy type. The constitutive assumptions
typical of linearly elastic bodies will be discussed at length in the next chapter.

REMARK. A study of the constitutive theory of Cauchy bodies falls beyond


the scope of this book. Suffice it here to consider two examples of constitutive
32 P. PODIO-GUIDUGLI

assumptions, one restricting the type of motions, the other restricting the type of
stress fields that compose the admissible dynamical processes: first, incompress-
ible bodies can only perform isochoric (volume-preserving) motions (i.e., for an
incompressible Cauchy body a typical deformation ft satisfies det(∇ft (p)) = 1 at
each place p and time t); second, the stress field in perfect fluid–bodies can only
be a pressure (i.e., T(q, t) = −π(q, t)I for each (q, t) belonging to the trajectory
of any possible motion). Needless to say, these two assumptions are mutually com-
patible, and can therefore be coupled to define a constitutive class that idealizes the
behavior of such real materials as water.

9. Simple Equilibrium Solutions. Normal and Shear Forces


Given a region R ⊂ E and a vector field s over ∂R, consider a symmetric-valued,
smooth tensor field T over R such that, for n the exterior normal to ∂R,
div T = 0 in R, Tn = s in ∂R. (9.1)
Problem (9.1) is largely underdetermined, and accordingly has many solutions.
However, it is not difficult to check that some simple constant solutions of (9.1) are
uniquely determined by the boundary datum. We shall now list the most significant
of those solutions, each of which we shall interpret as the stress field over the
deformed region occupied by a body in an equilibrium deformation maintained
solely by the surface load s.

(i) Pressure. Let R be a ball of E, and let s = −τ n over ∂R, with τ a positive
constant (Figure 9a).
Then, equation (9.1)2 is equivalent to the algebraic condition
Tn = −τ n, for all n ∈ U; (9.2)
hence, the stress in the whole region R is a uniform pressure, T = −τ I.

(ii) Tension. Let R be a circular cylinder, uniformly loaded on the end faces only,
where s = ±τ e and τ is a positive constant (Figure 9b).
Now problem (9.1) is quickly seen to be equivalent to solving the algebraic
system
Te = τ e, Tf = 0, for all f orthogonal to e, (9.3)
so that R is in a state of uniaxial (or pure) tension T = τ e ⊗ e.#

(iii) Shear. This time, let R be a cube under uniformly distributed grazing loads
on two pairs of opposite faces (Figure 9c).
# For τ < 0, we speak of a state of uniform traction in case (i), of uniaxial (or pure) compression
in case (ii).
STRESS 33

(a)

(b)

(c)
Figure 9.

The algebraic system for the stress in R is


Te1 = τ e2 , Te2 = τ e1 , Te3 = 0, (9.4)
and the unique solution of (9.4) – in fact a direct application of relation (8.4) – is
the pure shear T = τ (e1 ⊗ e2 + e2 ⊗ e1 ).
The stress vector is orthogonal to each oriented plane in a state of pressure;
parallel to a plane of normal e1 (e2 ) in a state of shear. For a general stress state it
is often expedient to resolve the stress vector relative to a plane of normal n into
its orthogonal and parallel components; these are defined by, respectively,
(n ⊗ n)[Tn] = (n · Tn)n, (9.5)1
34 P. PODIO-GUIDUGLI

(I − n ⊗ n)[Tn] = Tn − (n · Tn)n, (9.5)2

and called the normal and the shear force (per unit area) on the plane oriented
by n. The scalar Tnn = (n · Tn) is called the normal stress in the direction n; for
m a unit vector orthogonal to n, the scalar Tnm = m · Tn is called the tangential
(or shearing) stress in the direction m of the plane perpendicular to n; for n, n1 ,
n2 three orthonormal vectors, the shear force on the plane oriented by n has the
following expression in terms of tangential stresses:
(I − n ⊗ n)[Tn] = Tnn1 n1 + Tnn2 n2 . (9.6)
How severely a material body is stressed at a point where the stress field has
value T may be inferred from the extreme values at that point of the normal-stress
mapping
n $→ n · Tn (9.7)
(Exercise 2); or from the maximum value of the mapping
(n, m) $→ |m · Tn|, m · n = 0, (9.8)
delivering the absolute value of the tangential stress (Exercise 3).#

EXERCISES

1. A symmetric second-order tensor A has the following representation with


respect to an orthonormal basis {c1 , c2 , c3 }:
3
*
A = Aii ci ⊗ ci + A12 (c1 ⊗ c2 + c2 ⊗ c1 )
i=1
+ A13 (c1 ⊗ c3 + c3 ⊗ c1 ) + A23 (c2 ⊗ c3 + c3 ⊗ c2 ). (9.9)
In particular, for (αi , ai ), i = 1, 2, 3, the proper pairs of A, the spectral represen-
tation of A is
3
*
A= αi ai ⊗ ai (9.10)
i=1

(cf., e.g., [13, pp. 156–158]). Use (9.9) and (9.10) to show that problems (9.2)–(9.4)
have only the solutions indicated.
# A failure criterion based on the assumption that failure is the effect of exceeding the maximum
supportable normal stress was first explicitly proposed by Galileo in his Discorsi, published in 1638
[3, Part I, pp. 179–183]. More than two centuries later, H. Tresca associated failure with exceeding
the maximum supportable tangential stress [2, pp. 647–651]. Both Galileo’s and Tresca’s criteria are
applicable, of course to essentially disjoint classes of real materials.
STRESS 35

2. With no loss of generality, let the proper numbers τi of a symmetric tensor T


be ordered as follows: τ1 ! τ2 ! τ3 , with ti , i = 1, 2, 3, the corresponding proper
vectors. Show that
τ1 ! n · Tn ! τ3 , n ∈ U, (9.11)
so that, in particular, the maximum (minimum) value of the normal-stress mapping
(9.7) is attained for n = t1 (n = t3 ).
3. For T etc. as in the preceding exercise, show that the mapping (9.8) and the
tangential-stress mapping
(n, m) $→ m · Tn, m · n = 0, (9.12)
have the same maximum value, namely,
1
(m · Tn)max = |m · Tn|max = (τ1 − τ3 ), (9.13)
2
which is attained for
1 1
n = √ (t3 + t1 ), m = √ (t3 − t1 ). (9.14)
2 2
For what pairs (n, m) do the mappings (9.8) and (9.12) attain their minimum
values? (Vid. [4].)
4. Again for T as in the preceding exercises, show that the mapping
% %
n $→ %Tn − (n · Tn)n% (9.15)
takes its minimum value zero if and only if n is one of the proper vectors of T.
5. Let T = τ1 t ⊗ t + τ2 (I − t ⊗ t), t ∈ U, and let σ ∈ [τ1 , τ2 ] be given. Show
that, for all unit vectors n such that n · Tn = σ , the mapping (9.15) has constant
value τ = (σ − τ1 )1/2(τ2 − σ )1/2.

10. Alternative Forms of the Basic Balance Laws


We have already given, following Cauchy, pointwise equivalents of the basic bal-
ance laws. An alternative form of the law of resultant moment (7.4)2 is easily
arrived at if we take into account the fact that, for all a, b ∈ V,
a×b=0 ⇔ a ⊗ b ∈ Sym, (10.1)
and therefore write (7.4)2 in the form
Mo (3, f ) ∈ Sym, for all parts 3 ⊂ ", (10.2)
where Mo , the astatic tensor about the origin o, is defined by
1 1
Mo (3, f ) := (q − o) ⊗ s + (q − o) ⊗ b (10.3)
∂f (3) f (3)
36 P. PODIO-GUIDUGLI

(Exercise 1). The resultant moment is therefore balanced if and only if the astatic
tensor is symmetric.
We now express both (10.2) and the force balance (7.4)1 in forms involving
integrals over the reference shape.
Let a system of forces (s, b) be given for the body " as deformed by f . For
each point q of the deformed shape, and for each oriented plane through q of
normal n, consider the corresponding pair in the reference shape, consisting of
the point p = f −1 (q) and the plane of normal
FT (f −1 (q))n
nR = (10.4)
|FT (f −1 (q))n|
(Exercise 2). For such (q, n) and (p, nR ) we set
% %
sR (p, nR ) := %F∗ (p)nR %s(q, n), (10.5)1
# $
bR (p) := det F(p) b(q). (10.5)2

With these definitions we can portray the surface–force and volume–force fields
s, b in the reference shape.# Choose 3 ⊂ ", q ∈ ∂f (3), and let n = n(q) be the
exterior normal to ∂f (3) at q; moreover, let da, dA be the differentials of the area
measure in the deformed and reference shapes. Then (10.5)1 , and (1.19) imply that
the surface forces at q and p are equal:
s da = sR dA. (10.6)
It follows from (10.6) and (7.2)1 that (7.4)1 may be given the form
1 1
r(3, f ) = sR + bR = 0, 3 ∈ ". (10.7)
∂3 3

Likewise, we may write (10.2) as


1 1
Mo (3, f ) = (q − o) ⊗ sR + (q − o) ⊗ bR ∈ Sym, 3 ⊂ ". (10.8)
∂3 3

One may ask whether a representation analogous to (8.1) holds for the refer-
ential surface–force field sR , i.e., whether there is a tensor field TR over " that
delivers the value of sR at (p, nR ) by acting linearly on nR :
sR (p, nR ) = TR (p)nR , (p, nR ) ∈ " × U. (10.9)
Proceding formally, we may answer this question affirmatively by combining
(10.4)1 with (8.1) and (1.18):
% % % %
sR = %F∗ nR %s = %F∗ nR %Tn = (TF∗ )nR = TR nR , TR = TF∗ ; (10.10)
# Conversely, given (s , b ), with s (·, u) ∈ C 1 (")∩C 0 (") for all u ∈ U, and b ∈ C 0 ("), for
R R R R
each (sufficiently smooth) deformation f we can make use of (10.5) to generate a system of forces
for the body " as deformed by f .
STRESS 37

precisely, the Piola stress field TR is defined by


# $
TR (p) := T f (p) F∗ (p), p ∈ ". (10.11)
The existence of the Piola stress and the validity of (10.9) can also be established
as direct consequences of (10.7), just as is done starting from (7.4)1 for the ex-
istence of the Cauchy stress and the validity of (8.1). Of course, again in the
manner of Cauchy, local equivalents of (10.7) and (10.8) can be found; these read,
respectively,
Div TR + bR = 0 in ", (10.12)1
and
TR FT ∈ Sym in ".# (10.12)2

An exact stress measure often used in nonlinear solid mechanics and, in partic-
ular, in the mechanics of elastic plates and shells, is the Cosserat stress measure,
which is defined as follows in terms of Piola’s:
TR := F−1 TR .##
3 (10.13)
The corresponding surface–force field is
s̃R (p, nR ) = 3
TR (p)nR , (p, nR ) ∈ " × U, (10.14)
and specifies the pull-back to the reference shape of the surface force:
s̃R dA = F−1 s da. (10.15)
In terms of the Cosserat stress the balance of linear and angular momentum read
pointwise as, respectively,
# $
Div 3TR + (∇u)3 TR + bR = 0 and 3 TR ∈ Sym in ". (10.16)

EXERCISES

1. For mo , Mo defined as in (7.2)2 and (10.3), respectively, show that mo is the


axial vector of −2 Skw Mo (cf. also Exercise 7.2).
# Note that the operator Div (div) involves differentiation with respect to the space variables in the
reference (deformed) shape (Exercise 3). Note also that, due to (1.17)1 , the relations (10.12)2 and
(8.2)2 are equivalent.
## The Cosserat stress is called the second Piola–Kirchhoff stress by those who call first Piola–
Kirchhoff stress the Piola stress; the use of the denomination Piola–Cosserat stress for the Cosserat
stress is also documented. Resting on the historical notes found in [35, Section 210], we find unnec-
essary to take away from either Piola or the Cosserat brothers to give to Kirchhoff, who deserves his
fame for other achievements; and we find that the qualifier Piola–Cosserat is an arbitrary artifact.
38 P. PODIO-GUIDUGLI

2. Show that (1.18) implies that


F−T nR
n= , (10.17)
|FT nR |
and compare (10.17) and (10.4) to recognize that they have identical structure,
modulo a role interchange of the reference and deformed shapes (f " f −1 , (∇f =
F) " (gradf −1 = F−1 ), nR " n).
3. Show that
div T = (det F)−1 Div TR . (10.18)

11. Power. Stress Power


For any given collection of bodies, the notion of power expended on a typical body
part in a motion establishes the basic mechanical duality between the kinematical
and dynamical descriptions of the bodies in that collection.# For continuous bod-
ies of Cauchy type – just as for Newton’s mass–points – the relevant kinematical
descriptor is velocity, and the conjugate dynamical descriptor is force. The power
expended on a part 3 ⊂ " by the system of forces (s, b) in a motion m is defined
to be
1 1
P (3, ft )[v] := s·v+ b · v, (11.1)
∂ft (3) ft (3)
.
with v = f the velocity field associated with m (cf. the definition (7.12)). For a
t
system of forces satisfying the first of Euler’s axioms, the Cauchy relations (8.1)
and (8.2)1 , together with the divergence lemma (6.5), imply that
1 1 1
s·v+ b·v= T · grad v; (11.2)
∂ft (3) ft (3) ft (3)

conversely, if (11.2) is supposed to hold for every velocity field v, then the system
of forces (s, b) satisfies both (8.1) and (8.2)1 . More generally, given a dynamical
process (m, (T: T → Sym)), the construct
1
P (3) :=
(s)
T · grad v (11.3)
ft (3)

is called the stress power of the body part 3.## We recognize in the integrand
of (11.3) a duality between the Cauchy stress T and the spatial velocity-gradient
# In Greek, κ"ινησισ (kinesis) ≡ motion and δυναµισ
" (dunamis) ≡ force.
## It can be shown (vid., e.g., [24]) that the stress power is null for all rigid velocity fields over
arbitrary parts ft (3) of the deformed shape of a Cauchy body (such velocity fields have the form
specified by (7.15) of Exercise 7.1) if and only if T ∈ Sym, a condition that we know from our
discussion in Section 8 to be the pointwise equivalent of the second of Euler’s axioms.
STRESS 39

grad v (or, equivalently, the symmetric part of the latter). Other forms of (11.3) sug-
gest other pairs of dual stress and strain measures. The following forms involving
Piola and Cosserat stresses are important in the mechanics of solids:
1 1 1
. 3 1 3 .
P (s) (3) = TR · F = TR · D = TR · C (11.4)
3 3 2 3

where D, C are the strain tensors defined by (3.2).

REMARK. We give two examples that illustrate the mechanical significance of


the stress power.

(i) Incompressible, linearly viscous fluids are characterized by a constitutive re-


lation of the form
T = −π I + µ sym(grad v), div v = 0, µ > 0, (11.5)

with π a pressure field that is not constitutively specified, and µ the viscosity.
With (11.5), (11.3) becomes
1
P (3) =
(s)
µ|grad v|2 > 0, (11.6)
ft (3)

a measure for the mechanical rate of dissipation of this material class in a


motion.#
(ii) A class of elastic materials is often specified by assigning a real-valued map-
σ (F) over Lin+ that determines the Piola stress by differentiation:
ping 4
4
TR (F) = ∂F4
σ (F); (11.7)

4
σ (F) is interpreted as the stored energy per unit referential volume in a defor-
. σ (F)). , and hence,
mation of gradient F. Clearly, in this case, TR · F = (4
-1 ..
P (3) =
(s)
4
σ (F) , (11.8)
3

so that the total stored energy of any part of such an elastic body is conserved
over any time interval.##

While each dynamical process is purely dissipative in a linearly viscous fluid


body and purely conservative in an elastic body, for an arbitrary material class one
expects the stress power to break up into a dissipated part and a conserved part.

# Note that the pressure field π that maintains the incompressibility constraint (11.5) is
2
powerless.
## A completely analogous result holds when the stored energy is viewed as a mapping 3
σ (D), such
σ (D(F)) = σ (F) and 3
that 3 TR (D) = ∂D3σ (D).
40 P. PODIO-GUIDUGLI

EXERCISE

1. Prove that
.
grad v = F F−1 , (11.9)
and hence that (11.4)1 follows from (11.3).

12. Exact and Linearized Equilibrium Theories


In the preceding chapter we have given the linear analysis of deformation and strain
a presentation motivated and made consistent with the corresponding exact analysis
by the systematic use of an explicit linearization procedure. In this section we
discuss the relationships between the exact and linear analyses of equilibrium in
the mechanics of Cauchy solids, to the extent such a discussion can have without
making any constitutive choice. The linear theory of equilibrium of linearly elastic
solids will be dealt with in our fourth and last chapter.
We find it expedient to begin by introducing the concept of a stress field that
balances certain given loads when the body under study has a given shape.
A system of loads for a body " is a pair (so , bo ) of a surface-load vector field so ,
defined over ∂" and a volume-load vector field bo defined over ". Given a system
of loads (so , bo ), consider a tensor field S over " that satisfies both
Div S + bo = 0 in ", (12.1)
and
SnR = so in ∂". (12.2)
If there is a deformation f of " for which
S(∇f )T ∈ Sym in ", (12.3)
we say that S is a stress field that balances the loads in the deformed shape deter-
mined by f . In particular, S balances the given loads in the reference shape when
it satisfies (12.1)–(12.3) for f = i, the identity transformation, namely, when
S ∈ Sym in ".# (12.4)
With the use of the divergence theorem (6.7) it is not difficult to see that a necessary
consequence of (12.1) and (12.2) is
1 1
r(") = so + bo = 0; (12.5)
∂" "
# While we here confine attention to shape-independent systems of loads, to describe realistic
body–environment interactions it is often important to study loads that depend on shape (vid. [22, 26],
and the literature quoted therein).
STRESS 41

moreover, (12.1) and (12.2) together with (12.3) imply


1 1
# $ # $
Mo (", f ) = f (p) − o ⊗ so + f (p) − o ⊗ bo ∈ Sym, (12.6)
∂" "

whereas with (12.4) they imply


1 1
Mo (", i) = (p − o) ⊗ so + (p − o) ⊗ bo ∈ Sym. (12.7)
∂" "

Our second step is to supply an exact formulation of an equilibrium problem


in the mechanics of solids. To be definite and make our discussion easier, we con-
centrate on the case when tractions are assigned on the whole boundary. Formally,
such a problem is posed when
(i) a reference shape " and a system of loads (so , bo ) are given;
(ii) the constitutive behavior of the continuous body occupying " is specified by
assigning the class of its possible dynamical processes (Section 8);
(iii) a deformation f and a balanced Cauchy system of forces (s, b) for the pair
", f are sought such that, for (sR , bR ) defined as in (10.5),
# $
sR p, nR (p) = so (p), p ∈ ∂", (12.8)1
bR (p) = bo (p), p ∈ ". (12.8)2

In other words,
(iii)1 a deformation f and a Cauchy stress field T over " are sought such that
the corresponding Piola stress field TR = T(∇f )∗ satisfies the boundary
condition (12.2) and the balance laws (12.1), (12.3).# Granted this interpre-
tation, condition (12.5) is a solvability condition restricting the assignment
of the data " and (so , bo ). The nature of condition (12.6) is different: since
it involves also the equilibrium deformation f , it amounts to a consistency
condition between data and solution, a condition that can be checked only a
posteriori.
The linearized formulation of this equilibrium problem differs from the exact
formulation because “no distinction is made between the reference and the de-
formed shape” when the basic balance laws are stated. This is jargon to say that
(12.3) is formally replaced by (12.4), and that, consequently, the data/solution con-
sistency condition (12.6) is replaced by (12.7), a necessary condition of solvability
on the data that supplements (12.5).
Consider the smallness parameter

ε1 = sup |∇u|. (12.9)


"

# Compare (12.1) and (12.3) with (10.12) and (10.12) , respectively. That (12.2) is satisfied
1 2
follows from (10.9) and (12.8)1 .
42 P. PODIO-GUIDUGLI

For S independent of ε1 , (12.4) may be regarded as the zeroth-order approximation


in ε1 of (12.3). Indeed, in the light of the developments in Section 4 (in particular,
Exercise 2), we have from the last of (10.10) and (4.21) that
# $
TR = TF∗ = T (1 + ε1 tr H)I − ε1 HT + · · · = T + O(ε1 ).
In other words, if the linearization, typical of kinematics, implied by the choice
(12.9) of a smallness parameter is performed, then the Piola and Cauchy stress
measures coalesce. For this reason, when we deal with the linear theory, we shall
need not distinguish between these two stress measures,# and indicate stress by the
letter S.
A manipulation of (12.6) helps us to evaluate further the position of the lin-
earized versus the exact formulation of equilibrium problems in solid mechanics.
As
# $
f (p) − o = (p − o) + f (p) − p = (p − o) + u(p), (12.10)
(12.6) can be written as
1 1
Mo (", f ) = Mo (", i) + u ⊗ so + u ⊗ bo ∈ Sym. (12.11)
∂" "

Now, as indicated, the condition (12.7) is an exact consequence of (12.1), (12.2)


and (12.4); but, as (12.11) makes evident, it can also be regarded as the zeroth-order
approximation of (12.6) with respect to the smallness parameter
ε2 = λ−1
o sup |u| (12.12)
"

(here λo is some characteristic length in the problem, the diameter of ", say).
Thus, the linearized theory of equilibrium is got from the exact one when terms
of order O(ε2 ) are neglected. Similarly, the linearized theory of deformation finds
its position with respect to the exact theory when it is interpreted as the theory
that results when terms of order O(ε1 ) are neglected. We may argue that, what-
ever the constitutive choices, a consistent linearization of the exact descriptions of
strain and stress states in a body should always be pursued in terms of a smallness
parameter involving both the displacement field and its gradient:
ε = λ−1
o sup |u| + sup |∇u|. (12.13)
" "

REMARKS. 1. With the use of the linearization parameter in (12.13), the stress
power introduced in the preceding section takes the form
1 1
# .$
P (3) =
(s)
S · sym ∇u = S · E(u. ).
3 3
# Or others (Exercise 1).
STRESS 43

2. In the linearized formulation of motion problems, the linear momentum


retains the expression (7.11)1 , whereas the angular momentum, rather than as in
(7.11)2 , is written as
1
ao (3, ft ) = (p − o) × (ρR v); (12.14)
3

clearly, (12.14) may be regarded as the finite-order approximation of (7.11)2 with


respect to the smallness parameter
% %
ε2 = λ−1
o sup %ut (p)%, ut (p) = ft (p) − p. (12.15)
t ∈T ,p∈"

The linearized balance law of angular momentum has the form


1 1 -1 ..
(p − o) × so + (p − o) × bo = (p − o) × (ρR v) , (12.16)
∂" " "
where (so , bo ) is a (possibly time-dependent) pair of surface and volume loads
for ".

Given a body " and a system of loads (so , bo ) for it satisfying (12.5), both an
exact and a linearized equilibrium problem can be formulated with those data. As
failure to satisfy (12.7) implies nonexistence of solutions to the linearized problem,
we may say that the minimal consistency requirement between an exact and a lin-
earized problem with the same data is that (12.7) hold.# If that is the case, (12.11)
reduces to
1 1
u ⊗ so + u ⊗ bo ∈ Sym, (12.17)
∂" "

a relation that allows for qualitative estimates of the solution that are independent
of the constitutive law (Exercise 2). On the other hand, if (12.7) does not hold, a
rotation
r(p) = o + R(p − o), R ∈ Rot, (12.18)
of the reference shape can always be found such that
Mo (", r) ∈ Sym. (12.19)
# The general question of consistency between exact and linearized 3-dimensional elastostatics
was posed by A. Signorini in the thirties. By means of formal asymptotic expansions of the type
already introduced by the Cosserat brothers, Signorini found consistency conditions, as well as
instances of inconsistency, that were later variously generalized, and given both mechanical and
geometrical interpretation.
Second- and higher-order “corrections” of the predictions of the linear theory either are or may be
regarded as by-products of Signorini’s approach; instead, the plethora of rod and shell theories that
start off with various combinations of smallness assumptions for displacements, rotations and strains
does not directly fall within Signorini’s approximation scheme because the latter, being devised to
deal with general 3-dimensional situations, does not involve any idea of a tempered scaling to account
for the peculiar thinness of one or another 2- or 1-dimensional structural model.
44 P. PODIO-GUIDUGLI

To see this, note that

Mo (", r) = RMo (", i); (12.20)

thus, (12.19) is satisfied whenever R is chosen to be the transpose of the orthogonal


factor Q (or of its negative, if det Q = −1) in any of the polar decompositions of
Mo (", i) (Exercise 3). As

Mo (", r) = Mo (r("), i), (12.21)

we see that a given system of loads can always be balanced: if not in the original
reference shape, in the shape reached under a rotation r as above, when the latter
shape is taken as the reference shape (Exercises 4 and 5).

REMARK 3. Suppose that the surface loads so are assigned only on a portion ∂2 "
of the boundary, whereas displacements uo are assigned on the remaining portion
∂1 " = ∂"\∂2 ", ∂1 " ∩ ∂2 " = ∅. Then (12.11) is replaced by
1 1 1
Mo (", f ) = Mo (", i) + uo ⊗ SnR + u ⊗ so + u ⊗ bo ∈ Sym,
∂1 " ∂2 " "
(12.22)
1 1 1
Mo (", i) = (p − o) ⊗ SnR + (p − o) ⊗ so + (p − o) ⊗ bo . (12.23)
∂1 " ∂2 " "

Now, even if generally the pointwise distribution of equilibrium tractions over ∂1 "
is unknown, and such is the pointwise distribution of equilibrium displacements
over ∂2 ", it may still happen that we have sufficient a priori information to con-
clude that Mo (", i) ∈ Sym, and so derive integral qualitative estimates, of either
SnR over ∂1 " or u over ∂2 " (or "), from the relation
1 1 1
uo ⊗ SnR + u ⊗ so + u ⊗ bo ∈ Sym (12.24)
∂1 " ∂2 " "

(Exercises 2, 6, and 7). Notice that, for both Mo (", f ) and Mo (", i) to be sym-
metric, the information needed is, roughly speaking, that the reference and the
deformed shape have the same overall symmetries; since an assumption on the
deformed shape is involved, that information has an implicit, but indubitable, con-
stitutive nature.

EXERCISES

1. Let the Kirchhoff stress measure be defined by


4
T := (det F)T. (12.25)
STRESS 45

T = T = TR = 3
Show that, to within O(|∇u|)-terms, 4 TR .
2. In Figure 10, let a = α c, α > 0, |c| = 1. Show that (12.17) implies that
- .
1# $
u(0) − u(−λ) + u(λ) × c = 0 (12.26)1
2
for the free–free beam of Figure 10a; and that (12.24) implies
u(0) × c = 0 (12.26)2
for the hinged–hinged beam in Figure 10b.

(a)

(b)
Figure 10.

Figure 11.

3. Use the polar decomposition theorem (Chapter 1, footnote## on p. 9) to find


all rotations that solve equation (12.19) for the hinged-free beam whose reference
shape " is visible in Figure 11.
4. Let A ∈ Lin, Q ∈ Orth. Show that
QA ∈ Sym ⇔ AQ ∈ Sym. (12.27)
Use this result to prove that, if a rotation R solves (12.19) (and thus the system of
loads (so , bo ) is balanced in the rotated shape determined by R), then the rotated
system of loads (RT so , RT bo ) is balanced in the reference shape ".
46 P. PODIO-GUIDUGLI

5. A system of loads (so , bo ) for " is said to be in astatic equilibrium if the


system of loads (Qso , Qbo ) is balanced in the reference shape for all Q ∈ Rot.
Under the assumption that Mo (", i) ∈ Sym, prove that a system of loads is in
astatic equilibrium if and only if r(") = 0 and Mo (", i) = 0.#
6. Observe that (12.24) reduces to
1
u ⊗ bo ∈ Sym, (12.28)
"

both when uo ≡ 0 over ∂1 " = ∂" and when uo ≡ 0 over ∂1 ", so ≡ 0 over ∂2 ".
In application of (12.29), for the clamped-free, 2plate-like, homogeneous body in
Figure 12, loaded by its own weight, show that ( " u) × c = 0.

Figure 12.

7. Consider again the clamped-free plate-like body in Figure 12, this time sub-
ject to null body forces and null applied tractions on the free walls, and denote by
5l , 5r the clamped walls (so that ∂l " = 5l ∪ 5r ), by 6u , 6l the upper and lower
faces. Prove the following implications of (12.24):
-1 .
(i) Sn × a = 0
5r

for so ≡ 0 over 6u and 6l , uo ≡ 0 on 5l and uo = a on 5r ;


-1 .
(ii) u×a =0
6u

for uo ≡ 0 over the clamped walls, so ≡ a over the upper face and so ≡ 0 over the
lower face.

# As an appropriate preliminary to the proof, show that the assumption that M (", i) ∈ Sym does
o
not entail any loss of generality.
47

CHAPTER III
Constitutive Assumptions

13. Linearly Elastic Materials


The stress-strain law of a linearly elastic material is the assignment of a linear
transformation C of Lin into itself: for

S = C[H], (13.1)

S is interpreted as the stress accompanying H, the gradient of the displacement


from a reference shape chosen once and for all. The choice of C is customarily
restricted by three symmetry requirements that we now motivate.
(i) We know from our local analysis of deformation in Chapter I that a de-
formation of small gradient H may be regarded as consisting of a simple
deformation measured by E = sym H and a rotation measured by W =
skw H; the linearity of C implies that

C[H] = C[E] + C[W]. (13.2)

Now, translations induce no stress, since the gradient of a translation is the


null tensor and C[0] = 0. If we wish to guarantee that rotations induce no
stress as well, we are led to stipulate that

C[skw H] = 0, H ∈ Lin. (13.3)

(ii) We have seen in Chapter II that the balance law of angular momentum re-
quires that the stress field be symmetric-valued everywhere. To fulfill this
requirement it is sufficient to assume that

skw C[H] = 0, H ∈ Lin. (13.4)

(iii) The stored energy of a linearly elastic material described by C is a smooth


mapping

σ : Sym → IR, σ = σ (E), (13.5)


48 P. PODIO-GUIDUGLI

such that
∂E σ (E) = C[E], E ∈ Sym, (13.6)
and that the following normalization condition is satisfied:
σ (0) = 0. (13.7)
The value of σ at E is interpreted as the elastic energy stored per unit volume
at a point of the reference shape when the infinitesimal strain tensor takes
the value E at that point. Given C, (13.6) may be regarded as a differential
equation for σ , with the initial condition (13.7). It can be shown (cf. [11,
Section 24]) that a solvability condition for (13.6), (13.7), i.e., a condition for
the existence of a stored-energy mapping σ , is that C satisfies
A · C[B] = B · C[A], A, B ∈ Lin; (13.8)
if (13.8) holds, it turns out that
1
σ (E) = E · C[E]. (13.9)
2
As is usual in linear elasticity, we here take (13.3), (13.4) into account by
thinking of C as a linear transformation of Sym into itself (cf. Exercise 1), and
accordingly require that (13.8) prevail over Sym; we call such a mapping C an
elasticity tensor. The elasticity tensor determines the mechanical response of a
linearly elastic material in that the stress
S = C[E] (13.10)
depends linearly, and the stored energy quadratically, on the strain E via C; in
particular, then, both stress and energy are null in the reference shape, as well as
in any other shape reached from it through an infinitesimal rigid displacement of
type.
We shall also make the additional assumption that C is positive definite, i.e.,
A · C[A] > 0, A ∈ Sym\{0}; (13.11)
if so, then the stored-energy mapping has nonnegative values, and, in addition, C
is invertible. We shall denote the inverse of C by C−1 , and call it the compliance
tensor. With the compliance tensor, we can express the strain as a linear function
of stress:
E = C−1 [S]; (13.12)
C−1 shares the symmetry and positivity properties of C.
We shall write Lin for the linear space of all fourth-order tensors, i.e., the linear
transformations of Lin into itself. The elasticity and the compliance tensor may
be considered as elements of Lin; we shall find it expedient to introduce other
CONSTITUTIVE ASSUMPTIONS 49

elements in Section 15. Given an orthonormal basis {c1 , c2 , c3 }, it is sometimes


useful to compute the components of a fourth-order tensor A with respect to this
basis:
Aij hk = ci ⊗ cj · A[ch ⊗ ck ]. (13.13)
In terms of components, formulae (13.10) and (13.9) can be written as
1
Sij = Cij hk Ehk , σ = Cij hk Eij Ehk , (13.14)
2
while (13.3), (13.4), and (13.8), take, respectively, the forms
Cij hk = Cij kh , Cij hk = Cj ihk , and Cij hk = Chkij (13.15)
(cf. Exercise 1). The relations (13.15)1,2 express the so-called minor symmetries of
an elasticity tensor, whereas (13.15)3 expresses its major symmetry. As to the last
property, we note that the transpose AT of an element A ∈ Lin is defined by
A · A[B] = B · AT [A], A, B ∈ Lin, (13.16)
and thus,
ATij hk = Ahkij . (13.17)

A is symmetric whenever A = AT , or rather Aij hk = ATij hk ; in particular, in view of


(13.8) and (13.16) or (13.15)3 , all elasticity tensors are symmetric.
Equations (13.15) together allow us to conclude that the 81 elasticities Cij hk are
determined by at most 21 of those; said differently, that the most general linearly
elastic material is specified by 21 independent moduli (as we shall see, such consti-
tutive prescriptions as response symmetries and internal constraints further reduce
the number of independent elasticities).

REMARKS. 1. Our present description of the linearly elastic response originated


in two papers by G. Green dated 1839 and 1841 (cf. [9, Section 24; 33, Section 301
et seq.], Cauchy held different views, based on his construction of a molecular
model for linear elasticity. As we shall see in Section 25, Cauchy’s elasticities
satisfy six linear symmetry restrictions additional to (13.15); hence, there are at
most 15 independent moduli. Cauchy’s elasticity has been called by Pearson the
rari-constant theory, to distinguish it from the standard multi-constant theory of
Green (cf. [32]).
2. We have left tacit a possible dependence of the mechanical response on the
material point. If instead we were to stress such dependence, we would write, e.g.,
C = C(p), p ∈ ", and
σ : " × Sym → IR, σ = σ (p, E), (13.18)
in place of (13.5).
50 P. PODIO-GUIDUGLI

3. In the remark that ends Section 11 we have briefly considered the stored-
energy mapping
σ : Lin+ → IR,
4 σ =4
4 σ (F), (13.19)

which furnishes the Piola stress


4
TR (F) = ∂F4
σ (F) (13.20)

accompanying the deformation gradient F in a general elastic material. It is stan-


dard to require that 4
σ be invariant under observer changes, i.e., that

σ (F) = 4
4 σ (QF), F ∈ Lin+ , Q ∈ Rot;# (13.21)
other commonly accepted, convenient assumptions are

σ (I) = 0 and
4 σ (I) = 4
∂F4 TR (I) = 0, (13.22)

so that both energy and stress are null in the reference shape.
It is natural to ask how the mapping 4
σ described by (13.19), (13.21), and (13.22),
relates to the stored-energy mapping σ in (13.5)–(13.7), which characterizes the
linear elastic response. Now, as to the normalization conditions (13.22), σ does
satisfy the corresponding conditions

σ (0) = 0 and ∂E σ (0) = C[0] = 0; (13.23)

but the choice of σ is not restricted by any invariance requirement. Yet, it can be
shown that
1 # $
σ (F) = E · ∂F(2)4
4 σ (I)[E] + O |F − I|3 , E = sym(F − I), (13.24)
2
a relation that compares directly with (13.9), where the fourth-order tensor ∂F(2)4
σ (I),
just as C, satisfies (13.3) and (13.4) as a consequence of (13.21) and (13.22)2
(Exercises 4 and 5).
This result allows one to regard σ as an approximation of 4 σ , to within terms
of the third order in the smallness parameter |F − I|. There is no need to do so,
however: as a mathematical theory, the constitutive theory of linear elasticity –
just as linear elasticity in the whole – stands on its own feet, without any need to
be deduced as an approximation of a more encompassing theory, be it nonlinear
elasticity or another theory.

# Invariance under observer changes formalizes the physical expectation that the energy stored
due to whatever a deformation (of gradient F) be measured the same when the deformed shape is
observed after whatever a rigid motion (of gradient Q).
CONSTITUTIVE ASSUMPTIONS 51

EXERCISES

1. Show that conditions (13.3) and (13.4) have the alternative formulations
(13.15)1 , and (13.15)2 , respectively, and that they are together equivalent to
C[W] = 0, W ∈ Skw; C[E] ∈ Sym, E ∈ Sym. (13.25)
2. Show that, if σ has the form (13.9), then it satisfies (13.6).
3. Show that C is invertible if it is positive definite.
4. In the manner of Section 4, define the mapping
σ̌ (ε) := 4
σ (I + εH), H ∈ Lin, (13.26)
and, granted the necessary smoothness, show that
1 # $
σ̌ (ε) = σ̌ (0) + ε σ̌ 1 (0) + ε 2 σ̌ 11 (0) + O ε 3 , (13.27)1
2
σ̌ (0) = 4
σ (I), σ̌ 1 (0) = ∂F4
σ (I) · H,
(13.27)2
σ̌ 11 (0) = H · ∂F(2)4
σ (I)[H];

moreover, with (13.27) and (13.22), show that


1 # $
σ (F) = [F − I] · ∂F(2)4
4 σ (I)[F − I] + O |F − I|3 . (13.28)
2
5. (i) Derive the following differential consequences of condition (13.21) of
invariance under observer changes for the stored-energy mapping:
Q∂F4 σ (F) = ∂F4
σ (QF), F ∈ Lin+ , Q ∈ Rot; (13.29)1
# $
σ (F) FT ∈ Sym,
∂F4 F ∈ Lin+ ; (13.29)2

σ (F) = ∂F(2)4
W∂F4 σ (F)[WF], F ∈ Lin+ , W ∈ Skw; (13.29)3
## (2) $ # $ $
∂F 4 σ (F) HT ∈ Sym. F ∈ Lin+ , H ∈ Lin. (13.29)4
σ (F)[H] FT + ∂F4

(ii) Show that, if (13.22)2 holds, (13.29)3 and (13.29)4 imply, respectively, that
the fourth-order tensor ∂F(2)4
σ (I) in (13.28) satisfies both relations (13.25).

14. Material Symmetry


It is common experience that certain rigid transformations of the reference shape
are mechanically undetectable, in that two otherwise identical deformational ex-
periments, the one performed before and the other after any one of those rigid
52 P. PODIO-GUIDUGLI

transformations, give identical results in terms of induced stresses and stored en-
ergies. For example, consider a square piece of a square-mesh net of elastic fibers
(a window grate, say, or a handkerchief, or else, for a slightly different example, a
reinforced concrete slab): we do expect the same force to be needed to pull apart
either pair of opposite sides of the same amount. Similarly, if we cut two identical
radial test-probes out of one slice of a timber log (Figure 13) and then apply equal
tensile loads to both probes, we expect to get substantially equal elongations.#

Figure 13.

Thus, one could not tell whether the grate had been rotated in its plane of any
integer multiple of π/2, or the log about its axis of any angle, before performing
the mechanical experiments described above. An important issue in the constitutive
theory is to construct a method of describing such symmetries, and of efficiently
exploiting them in sharpening our specification of the mechanical response of a
given material. We here take up such issue for linearly elastic materials.
Let u(y) be the displacement field in a neighborhood of a typical point p ∈ "
of a body having elasticity tensor C (Figure 14).

Figure 14.

# But not so if we were applying the same tensile load to another geometrically identical probe
cut parallel to the log axis. Thus, in general rigid rotations from a reference placements are indeed
revealed by appropriate mechanical experiments.
CONSTITUTIVE ASSUMPTIONS 53

Consider a rotation of " about p:


y + (y) = p + Q(y − p), Q ∈ Rot, (14.1)
and define
# $
u+ y + = Qu(y) (14.2)
whenever y + and y are related by (14.1). The purpose of this definition is to pic-
ture a situation when, given an experiment inducing the displacement u(y) at the
point y of ", the same experiment, if performed after a rotation of ", induces the
displacement u+ (y + ) at the point y + where the rotation has taken the point y.
One may ask which gradient the displacement field u+ has at p. Let H and H+
denote, respectively, the gradients of the mappings y $→ u(y) and y + $→ u+ (y + ).
Differentiation of the left and right sides of (14.2) gives, with the help of (14.1),
# $ # $
∇ u+ ◦ y + = H+ Q, ∇ Qu(y) = QH. (14.3)
It follows that
H+ (p) = QH(p)QT , (14.4)
i.e., H+ (p) is the orthogonal conjugate of H(p) with respect to Q; in particular, it
is not difficult to show that
E+ (p) = QE(p)QT (14.5)
(Exercise 1).
For the oriented material plane through p of normal n, the stress vector at p
corresponding to the displacement field u is given by
! "
s(p, n) = S(p)n, S(p) = C(p) E(p) . (14.6)
The material plane oriented by n is transformed by (14.1) into the plane through p
of normal
n+ = Qn, (14.7)
and the stress vector at p corresponding to the displacement field u+ is given by
# $ ! "
s+ p, n+ = S+ (p)n+ , S+ (p) = C(p) E+ (p) . (14.8)
We call Q a material symmetry transformation at p for the linearly elastic material
in question if Q is such that s+ is just the rotated version of s, i.e.,
# $
s+ p, n+ = Qs(p, n) (14.9)
for all pairs of displacement fields u, u+ as in (14.2) and normals n, n+ as in (14.7).
It follows from this definition that, just as the strains E and E+ , the stresses S and
S+ are orthogonal conjugates with respect to Q:
S+ (p) = QS(p)QT . (14.10)
54 P. PODIO-GUIDUGLI

It also follows, with the use of (14.6)–(14.10), that Q is a material symmetry


transformation if and only if
! "
QC[E]QT = C QEQT , E ∈ Sym (14.11)

(here and henceforth the dependence on p is again left tacit). Given C, the collec-
tion GC of all rotations Q satisfying (14.11) is a subgroup of Rot that is called the
material symmetry group of C.#

EXERCISES

1. Show that the algebraic operations of orthogonal conjugation and symmetriza-


tion commute.
2. Establish (14.10) and (14.11).
3. Define 3GC to be the group of all orthogonal tensors that obey (14.11) (this is
the definition usually found in textbooks, cf. [11]). Show that, just as Orth can be
represented as the direct product of Rot and the two-element group {−I, I}, 3 GC is
the direct product of GC and {−I, I}.

15. Fourth-Order Tensors


Various fourth-order tensors are encountered in the study of the linearly elastic
response; in this section we collect some related algebraic material to be used in
the sequel.
Recall from Section 13 that the fourth-order tensors are the linear transforma-
tions on Lin, and that we write Lin for their collection. In addition to the elasticity
tensor C and the compliance tensors C−1 , we have already encountered various ele-
ments of Lin, e.g., the transposer TA := AT and the operators constructed in terms
of it and the identity I, namely, sym and skw (Exercise 1).## For a given Q ∈ Rot,
we have also had occasion to consider the linear transformation A $→ QAQT deliv-
ering the orthogonal conjugate of A with respect to Q. We now wish to show that
this mapping is a special case of one of two product rules that generate elements of
Lin starting from two arbitrary elements of Lin.
Given A, B ∈ Lin, the conjugation product of the ordered pair (A, B) is the
fourth-order tensor

A # B[C] := ACBT , C ∈ Lin; (15.1)


# We shall prove that G is a group in Section 16.
C
## Lin is closed under composition, e.g., C ◦ skw and skw ◦ C are elements of Lin (and (13.3),
(13.4) may be read as the requirement that these fourth-order tensors both map all of Lin into its null
element).
CONSTITUTIVE ASSUMPTIONS 55

the dyadic product is


A ⊗ B[C] := (B · C)A, C ∈ Lin. (15.2)

It follows from the definition of transpose (13.16) and, respectively, (15.1) and
(15.2) that
(A # B)T = AT # BT , (A ⊗ B)T = B ⊗ A; (15.3)

thus, conjugation products are not symmetric (unless the factors A, B are in Sym),
nor are dyadic products (unless the factors are parallel). The following composition
rules hold true:

(A # B)(C # D) = (AC) # (BD); (15.4)


# $
(A ⊗ B)A = A ⊗ AT B , A(A ⊗ B) = (AA) ⊗ B, (15.5)

for all A, B, C, D ∈ Lin and A ∈ Lin.


Choosing A = B = I in (15.1) and (15.2) we recover other useful fourth-order
tensors (some of which we already encountered in Section 5), namely,
I ⊗ I = 3 sph, I # I = I. (15.6)

We note that, since |I|2 = dim(V) = 3, we may write (15.6)1 as


I I
sph = ⊗ , (15.7)
|I| |I|
and recognize sph and dev as the complementary orthogonal projectors associated
with the decomposition

ISym = sph + dev (15.8)

of the identity of Sym. When we choose A = B = Q ∈ Rot in (15.1) we write


Q := Q # Q. (15.9)

Just as a rotation Q of Lin preserves scalar products of elements in V, its associated


orthogonal conjugator Q preserves scalar products in Lin:

QA · QB = QAQT · QBQT = A · QT QBQT Q = A · B; (15.10)


we then have reason to say that Q is a rotation of Lin. It is not difficult to see that,
for each Q ∈ Rot, Q commutes with sph and dev:

Q sph = sph Q = sph, Q dev = dev Q. (15.11)


56 P. PODIO-GUIDUGLI

EXERCISES

1. Show that

T = sym − skw, I = sym + skw. (15.12)


Moreover, use the definition (13.13) to show that

Tij hk = δik δj h ,
2(sym)ij hk = δih δj k + δik δj h , (15.13)
2(skw)ij hk = δih δj k − δik δj h
(here δij is the Kronecker symbol, whose value is 1 if i = j, 0 if i )= j ).
2. (i) Establish (15.3), (15.4) and (15.5). (ii) Show that
# T $
(A ⊗ B)(C # D) = A # ⊗ C #$D [B] ,
T
(15.14)
(A # B)(C ⊗ D) = A # B[C] ⊗ D.
3. Establish (15.11).
4. Prove that, if p ∈ U and P = p ⊗ p, then
P # P = P ⊗ P. (15.15)

16. Problems of Classification and Representation


With the use of (15.9) we may write (14.11) as
QC = CQ, (16.1)

and say that a rotation Q is in the symmetry group GC of C if Q and C commute


(cf. [19]). For linearly elastic materials, such commutativity is the algebraic sub-
stance of the physical property that certain rotations of the reference shape are not
detectable by means of experiments measuring stress or energy (Exercise 2).
Taking advantage of (16.1) we can promptly verify that GC is a group. Indeed,
I ∈ GC for all C. Moreover, if Q ∈ GC , then Q−1 = QT is associated with QT , and,
as QT Q = QQT = I, (16.1) imply that

CQT = QT QCQT = QT CQQT = QT C, (16.2)


so that QT ∈ GC . Finally, if Q1 ,Q2 ∈ GC , as Q1 Q2 is associated with Q1 Q2 by
(14.4), we have that

Q1 Q2 C = Q1 CQ2 = CQ1 Q2 , (16.3)


and thus Q1 Q2 ∈ GC .
CONSTITUTIVE ASSUMPTIONS 57

Given C, the classification problem consists in finding GC , and then classifying


the corresponding linearly elastic material according to the richness of GC itself.
The classification ranges from isotropic materials, those for which GC = Rot, to
aelotropic materials, for which GC = {I}. In between these extreme cases, a fistful
of discrete groups and one continuous group suffice to describe all symmetries
of interest in the constitutive theory of anisotropic linearly elastic materials.# The
material class with a continuous symmetry group is the transversely isotropic class,
for which GC = Rot(a), the collection of all rotations about a fixed axis a.##
Another typical problem in the constitutive theory of linearly elastic materials
is the representation problem, which consists in finding, given a subgroup G ⊂
Rot, a representation formula for all elasticity tensors C such that GC ⊃ G, i.e.,
in describing the class of all elasticity tensors sharing a given symmetry group.‡
A well-known example of solution of the representation problem is Lamé’s formula
for the isotropic case, when G = Rot and
Ciso = 2µISym + λI ⊗ I. (16.4)
We shall now delineate a technique to solve the representation problem in general,
and apply it to derive not only (16.4), but also a representation formula for the
transversely isotropic case.
To begin with, recall that for an elasticity tensor C, when seen as a linear and
symmetric transformation of the vector space Sym (a space of dimension 6), we
may formulate the so-called spectral problem, i.e., the problem of finding all pairs
(γ , C), with γ ∈ IR and C a subspace of Sym, such that
C[C] = γ C, |C| = 1, C ∈ C. (16.5)
There are at most six proper pairs (γ , C) solving (16.5) such that the real numbers
γi are all distinct (cf., e.g., [13]) if (γi , Ci ) and (γj , Cj ) are proper pairs with
γi )= γj , then Ci · Cj = 0. Moreover, the following spectral representation holds
for C:
*6
C= γ i Ci ⊗ Ci , (16.6)
i=1
where it is understood that proper numbers are repeated as many times as their
algebraic multiplicity.‡‡ For example, if γ has multiplicity 2, there are solutions
# Cf. [9, 10; 11, Section 21]. Among the classes having discrete symmetry groups are the mon-
oclinic, the orthotropic–rhombic and the orthotropic–tetragonal classes, in the order of increasing
richness in material symmetry; for the last class, e.g., GC consists of all rotations of an integer
multiple of π/2 about a fixed axis.
## Hence, transversely isotropic materials have the symmetry group of a circle in the plane.
‡ Problems of classification and representation are at the technical, if not the conceptual, core
of the constitutive theory for whatever material class one may wish to consider. Aside from linear
elasticity, it is only for finite elasticity that those problems have received a rather extensive, although
still incomplete, treatment.
‡‡ The algebraic multiplicity of a proper number γ is the dimension of the associated proper
space C.
58 P. PODIO-GUIDUGLI

C1 , C2 of (16.5) such that C = span(C1 , C2 ); moreover, (16.6) may be given the


form
6
*
C = γ (C1 ⊗ C1 + C2 ⊗ C2 ) + γ i Ci ⊗ Ci , (16.7)
i=3

where the first two dyadic products yield a decomposition of the identity of C:
IC = C1 ⊗ C1 + C2 ⊗ C2 (16.8)
(IC can also be seen as the orthogonal projector of Sym on C). Whatever the mul-
tiplicity of proper numbers, it is always possible to choose an orthonormal basis
{Ci , i = 1, . . . , 6} of proper vectors for Sym such that
6
*
ISym = Ci ⊗ Ci . (16.9)
i=1

In view of (16.1) and (16.6) we have that, given G ⊂ Rot, all elasticity tensors
C having GC ⊃ G must be such to satisfy
6
*
CQ = QC = γi (QCi ) ⊗ Ci (16.10)
i=1

for each Q ∈ G, and therefore,


C(QCk ) = γk QCk , (16.11)
for each index k, here unsummed. In other words, C must be such that each of its
proper spaces C satisfies
QC = C, Q ∈ G, (16.12)
i e., is left invariant under the action of each Q associated with a rotation belonging
to the given group G. Thus, the representation problem is solved when all nontrivial
subspaces of Sym obeying (16.12) are found.#
In the isotropic case it is not difficult to show that the only subspaces of Sym,
other than {0} and Sym itself, that fulfill (16.12) for G = Rot are C1 = Dev and
C2 = Sph (cf. (15.11)). This, in view of (5.5) and (16.7) is tantamount to (16.4).
The class of all isotropic materials is parametrized by two material moduli; we
may choose the Lamé constants, as in (16.4), or the proper numbers γ1 = 2µ and
γ2 = 3λ + 2µ (with γ1 of multiplicity 5, and γ2 simple):
Ciso = γ1 dev + γ2 sph. (16.13)
# Solving the representation problem by looking for invariant subspaces is somewhat easied by
the fact that, if C obeys (16.12), then C ⊥ , its orthogonal complement in Sym, obeys it as well
(Exercise 4).
CONSTITUTIVE ASSUMPTIONS 59

In the transversely isotropic case, when G ⊃ Rot(a), the solution of the repre-
sentation problem is slightly more complicated. If we let a ≡ c3 , it can be shown
(vid. [27, 29]) that each C must have two 2-dimensional proper spaces,

√1 = span(C1 , C2 ),
C
(16.14)
2Cα = cα ⊗ c3 + c3 ⊗ cα , α = 1, 2,
and

√2 = span(C3 , C4 ),
C √ (16.15)
2C3 = c1 ⊗ c2 + c2 ⊗ c1 , 2C4 = c1 ⊗ c1 − c2 ⊗ c2 ;
and two 1-dimensional proper spaces that together compose the orthogonal com-
plement K of C1 ⊕ C2 in Sym. If we choose the following orthonormal basis for
Sym:
& √ '
Ci , i = 1, . . . , 4; D1 = c3 ⊗ c3 , 2D2 = c1 ⊗ c1 + c2 ⊗ c2 , (16.16)
we arrive at the following representation formula for C:
C = γ1 (C1 ⊗ C1 + C2 ⊗ C2 ) + γ2 (C3 ⊗ C3 + C4 ⊗ C4 ) + C, (16.171 )

C = δ1 (D1 ⊗ D1 ) + δ2 (D1 ⊗ D2 + D2 ⊗ D1 ) + δ3 (D2 ⊗ D2 ).# (16.172 )

We see that the class of all transversely isotropic materials is parametrized by


five material moduli; all members of the class share the proper spaces C1 and C2 ,
but the two one-dimensional proper spaces spanning K may vary from one mem-
ber to another (Exercise 6).

EXERCISES

1. Show that the elasticity and the compliance tensor have the same symmetry
group: GC = GC−1 .
2. Show that, for all Q ∈ GC ,
# $
σ (E) = σ QEQT , E ∈ Sym, (16.18)
i.e., σ = σ ◦ Q.
3. Prove that the only nontrivial subspaces of Sym that satisfy (16.12) are Dev
and Sph.##
4. The orthogonal complement of a subspace C of Sym is defined to be
C ⊥ := {E ∈ Sym | E · C = 0, C ∈ C}.
# Note that all of the pairs (γ , C ), . . . , (γ , C ) solve the spectral problem (16.5), whereas none
1 1 2 4
of (δi , Dα ) are proper pairs. Representations alternative to (16.17) are given in [30]; vid. also [5, 27].
## A proof may be found in [28].
60 P. PODIO-GUIDUGLI

Show that, if Q ∈ Orth is such that (16.12) holds, then QC ⊥ = C ⊥ (and con-
versely).
5. Prove that necessary and sufficient conditions for an elasticity tensor to be
positive definite are that all its proper numbers are positive. In particular, for the
elasticity tensor of an isotropic material, show that the positivity conditions are

µ > 0, 3λ + 2µ > 0. (16.19)


6. For C as in (16.17),
(i) determine the proper pairs of C;
(ii) prove that necessary and sufficient conditions for the elasticity tensor of a
transversely isotropic material to be positive definite are that

γ1 > 0, γ2 > 0, δ1 > 0, δ1 δ3 − (δ2 )2 > 0; (16.20)

(iii) show that GC ⊃ Rot(c3 ), i.e., that, for each Q ∈ Rot(c3 ), Q and C commute.
7. Show that the isotropic elasticity tensor

Ciso = 2µ dev + (2µ + 3λ) sph (16.21)


obtains from (16.17) for

γ1 = γ2 = 2µ, δ1 = λ + 2µ, δ2 = 2 λ, δ3 = 2(λ + µ). (16.22)

17. Internal Constraints


In continuum mechanics the standard notion of an internal constraint, i.e., of a
restriction on possible strains, is modelled on the notion of a bilateral, perfect and
frictionless positional constraint in mass–point mechanics; we begin by recapitu-
lating the latter.

17.1. POSITIONAL CONSTRAINTS IN MASS – POINT MECHANICS

Let M be a surface in E, represented as the locus of zeroes of a smooth scalar map-


ping x $→ µ(x) on E. We say that a mass–point X, of mass m > 0, is constrained
to move on M, or that M is a (bilateral, positional) constraint manifold for X, if
all possible trajectories τ $→ x(τ ) in E of X must lie in M:
# $
µ x(τ ) = 0 (17.1)

(Figure 15).
Differentiating (17.1) with respect to τ we have
# $
∇µ x(τ ) · x. (τ ) = 0, (17.2)
CONSTITUTIVE ASSUMPTIONS 61

Figure 15.

where x := x − o denotes the position vector of a point x ∈ M with respect to


a fixed origin o; thus, as a direct kinematical consequence of a bilateral positional
constraint, we see that, at a point x(τ ) ∈ M, the possible velocities x. (τ ) are
orthogonal to the gradient ∇µ evaluated at x(τ ), and therefore lie in the tangent
plane TM (x(τ )) to M at x(τ ).
In addition to such a kinematical implication, the imposition that a mass–point
move on a prescribed surface is usually accompanied by the following dynamical
stipulations:
(i) the total force f acting on X splits into a reactive and an active part:

f = f (R) + f (A); (17.3)

(ii) the reactive part (briefly, the reaction) has the representation

f (R) = ϕ (R) n, with ϕ (R) (x) ∈ IR,


∇µ(x) (17.4)
n(x) := for x ∈ M,
|∇µ(x)|

the scalar multiplier ϕ (R) being indeterminate, in the sense that it is not the
object of a specific constitutive prescription;
(iii) the active part at a point (x, τ, v) ∈ M × IR × TM (x) is given by a vector-
valued mapping

(x, τ, v) $→ f (A) (x, τ, v). (17.5)

REMARK. The above constraint is termed bilateral, because the reaction mul-
tiplier ϕ (R) may have any sign; perfect, because the magnitude of ϕ (R) is inde-
terminate; and frictionless, because the reactive force expends no power in any
admissible motion:

f (R) (x) · v = 0, for all x ∈ M and v ∈ TM (x). (17.6)


62 P. PODIO-GUIDUGLI

In that it restricts the set of possible motions and specifies to some extent the
forces that may accompany them, the assignment of a constraint is constitutive
by nature. The associated evolution problem has peculiarities that make it dif-
ferent from the unconstrained problem (indeed, strictly speaking, the latter may
be regarded as an important, but special, case of the former). From the motion
equation
f (R) + f (A) = (mx. ). (17.7)
and the initial conditions, one seeks to determine the trajectory of X, a curve on M,
and the reactions when X travels along its trajectory. The two problems are solved
in series: first, the “pure” (reaction-free) motion equation
Pf (A) = P(mx. ). , P(x) := I − n(x) ⊗ n(x), (17.8)
obtained by projecting (17.7) onto the current tengent plane, is used to find the
trajectory; with this, (17.7) yield the reaction:
! " & ! "'
f (R) = (I − P) (mx. ). − f (A) = n · (mx. ). − f (A) n.# (17.9)

17.2. INTERNAL CONSTRAINTS IN LINEAR ELASTICITY

An especially simple instance in constrained mass–point mechanics is encountered


when the constraint manifold is flat, and therefore may be globally identified with
its tangent plane. Constraint manifolds are also flat in linear elasticity, where an
internal constraint is the prescription of a subspace of Sym to be interpreted as the
collection of all possible strains: typically, as we have seen in Section 6, at each
fixed point of a body a constraint tensor V ∈ Sym is assigned, and the associated
constraint space M is defined to be
M = {E ∈ Sym | V · E = 0}. (17.10)
Thus, we can write
Sym = M ⊕ M⊥ , M⊥ := span(V). (17.11)
The accompanying dynamical stipulations are that the stress splits into a reactive
and an active part:
S = S(R) + S(A) , (17.12)
# In (17.9) the constitutive prescription (17.5) is used to determine the active force at each point
x(τ ) of the trajectory:
.
f (A) = f (A) (x(τ ), τ, x (τ )).

Thus, in particular,
. . .
ϕ (R) (x(τ )) = n(x(τ )) · [(mx (τ )) − f (A) (x(τ ), τ, x (τ ))].
CONSTITUTIVE ASSUMPTIONS 63

with
S(R) = ψ (R) V, ψ (R) ∈ IR, (17.13)

and with the active stress specified by a linear transformation of M into itself:

C: M → M, S(A) = C[E].# (17.14)


In particular, then, reactive stresses expend no power in any admissible motion, in
the sense that
. .
S(R) · E = 0, E ∈M (17.15)
(cf. (17.6) and Remark 1 in Section 12; the flat set M coincides with its own tangent
space).
In the presence of an internal constraint the evolution equation of linear elastic-
ity has the form

Div S(R) + Div S(A) + bo = ρo v. . (17.16)

The solution technique consists once again in looking first for a “pure” conse-
quence of (17.16) that determines the trajectory, and then computing the reactions.
The mathematical issue is to find an operator that annihilates the term Div S(R) in
(17.16), just as the projection operator P annihilates f(R) in (17.7). We note that, in
general,
# $ # $
Div S(R) = Div ψ (R) V = V ∇ψ (R) + ψ (R) Div V, (17.17)

and then consider some examples.

EXAMPLE 1 (Preservation of volume). Choose the constraint tensor field V(p) =


I, p ∈ ". Then,

I · E(u) = Div u = 0; (17.18)


in view of (4.12), the material in question is everywhere incompressible. Moreover,
the reactive stress is spherical: S(R) = ψ (R) I; and, Div S(R) = ∇ψ (R) . Thus, the
reaction annihilator is the differential operator Curl; the pure motion equation is
! "
Curl Div S(A) + bo − ρo v. = 0. (17.19)
# It follows from (17.13) that S(R) is an arbitrary element of M ⊥ that, by (17.14), is orthogonal
to S(A) . The latter property is a consequence of the “normalization” of the active stress that has been
implicity performed when M has been chosen as the range of C. This normalization is admissible
because one exploits the indeterminacy of the scalar multiplier in the representation (17.13), and
formally absorbs into the reactive stress a possible component in M⊥ of the active stress. We might
have manufactured an even more stringent analogy than the one we are resting upon here by requiring
that the reactive and active forces on a constrained mass–point were mutually orthogonal in all
possible motions; however, this requirement is artificial in mass–point mechanics.
64 P. PODIO-GUIDUGLI

EXAMPLE 2 (Preservation of length). Let now V(p) = e(p) ⊗ e(p), p ∈ ",


e(p) ∈ U, so that

e(p) ⊗ e(p) · E(p) = E(p)e(p) · e(p) = 0, (17.20)

and the material at p is inextensible in the direction of the material fiber along e(p).
Then,
# $
S(R) = ψ (R) e ⊗ e, Div S(R) = ∂e ψ (R) e + ψ (R) Div(e ⊗ e). (17.21)

In the easy case when the constraint is uniform: e(p) ≡ e, we have that Div S(R) =
(∂e ψ (R) )e, and the annihilator is the projection (I−e⊗e) onto the plane orthogonal
to e; otherwise, the annihilator cannot be an algebraic operator (Exercise 1).

EXAMPLE 3 (Preservation of orthogonality). In this case we choose the con-


straint tensor field to be, at each p ∈ ", V(p) = e(p) ⊗ f(p) + f(p) ⊗ e(p),
with e(p), f(p) ∈ U, and e(p) · f(p) = 0, so that the material at p is unshear-
able with respect to the pair of orthogonal directions e(p), f(p). This time S(R) =
ψ (R) (e ⊗ f + f ⊗ e) and
# $ # $
Div S(R) = ∂f ψ (R) e + ∂e ψ (R) f + ψ (R) Div(e ⊗ f + f ⊗ e) (17.22)

(Exercise 2). Thus, if the constraint is uniform, the reaction is annihilated by the
projection operator (I − e ⊗ e − f ⊗ f) (Exercise 3).

The assignment of n linearly independent constraint tensors Vi leads to a con-


straint space whose dimension is (6 − n); for n > 1, easy modifications of the
above developments are in order, e.g., the reactive stress is written as
n
*
S(R) = ψi(R) Vi . (17.23)
i=1

A constraint space M is proper if dim M < 6; rigidity corresponds to M = {0}


(Exercises 4 and 5).#

REMARK. The examples we considered have special importance in applications:


real materials such as vulcanized rubbers, crude or synthetic, which are found
reluctant to volume changes no matter the deformations imposed and the loads
applied, are modelled as incompressible; fiber-reinforced materials, be the fibers
included by the matrix in an orderly or in a random manner, are modelled as inex-
tensible in the direction of fibers; and thin bodies, like plates or rods, whose typical
deformations involve modest changes of the right angle of their cross-section to
# A mechanical theory of linearly elastic materials with internal constraints has been given by
Pipkin in [23].
CONSTITUTIVE ASSUMPTIONS 65

their axis, are modelled as unshearable with respect to each pair formed by a
cross-sectional direction and the axial direction.

EXERCISES

1. Show that, if the general inextensibility constraint discussed in Example 2


prevails, then

(I − e ⊗ e)Div S(R) = ψ (R) (∇e)e = ψ (R) ∂e e. (17.24)

2. For e, f, g three mutually orthogonal vector fields, prove that

g · Div(e ⊗ f + f ⊗ e) = −∇g · (e ⊗ f + f ⊗ e). (17.25)

3. Let the constraint of orthogonality preservation prevail in a plate-like region,


with e = εα (p1 , p2 , p3 )cα and f = ϕα (p1 , p2 , p3 )cα . Use (17.25) to show that the
orthogonal projection c3 ⊗ c3 annihilates all reactive terms in the motion equation.
4. It may happen that more than a constraint tensor is needed to prescribe a
constraint space. As an example, for c1 , c2 , c3 three mutually orthogonal vectors,
consider the space

M2 = {E ∈ Sym | Ec3 · cα = 0, α = 1, 2} (17.26)

that obtains when preservation of orthogonality of fibers along both the pairs of
directions (c3 , c1 ) and (c3 , c2 ) is requested. Show that
(i) if E ∈ M2 , then orthogonality of all pairs of fibers of type (c3 , c = γα cα ) is
preserved (and conversely; cf. relation (5.12));
(ii) dim M2 = 4.
5. (The rigidity constraint). Show that

E(u) ≡ 0 in " ⇔ u(p) = u(po ) + Wo (p − po ), Wo ∈ Skw, p ∈ ".

18. Constraints and Material Symmetry


In the presence of a constraint the domain of the elasticity tensor C is the constraint
space M. In view of (14.11), if we wish to list some Q ∈ Rot among the symme-
try transformations of the material described by C, we must require that, for all
E ∈ M, QE belong to M as well. Therefore, with each constraint space M we
associate a group GM consisting of all rotations that leave M invariant:

GM := {Q ∈ Rot | QM = M}. (18.1)


66 P. PODIO-GUIDUGLI

The constraint group GM specifies the maximal material symmetry compatible with
the constraint M, in the sense that for an elasticity tensor C and a constraint M to
be compatible we must verify that
GM ⊃ GC . (18.2)
Let us consider two constraint spaces that are frequently encountered in appli-
cations. If material fibers are inextensible in a given direction, c3 , say, the relative
constraint space is
M1 = {E ∈ Sym | Ec3 · c3 = 0} (18.3)
(cf. (17.10) and (17.20)); this constraint is basic to the theory of fiber-reinforced
materials. If, in addition to being inextensible in the direction c3 , the material is
incapable of shearing with respect to all pairs of directions (c3 , c), with c3 · c = 0
(Exercise 17.4, where the constraint space M2 is introduced, and Exercise 2), it is
not difficult to see that the appropriate constraint space is
M3 = {E ∈ Sym | Ec3 = 0};# (18.4)
this constraint plays a central role in plate and shell theories. It can be shown that
the constraint group of M3 (as well as, a fortiori, the constraint groups of M1 and
M2 ), contains the subgroup Rot(c3 ) of Rot of all rotations about c3 , but does not
contain Rot itself (Exercise 3). Thus, the constraint of inextensibility, both by itself
and combined with unshearability, is compatible with the response symmetry of
transversely isotropic materials, for which GC ⊃ Rot(c3 ), but incompatible with
isotropic materials, for which GC = Rot.
The question arises as how to formulate and solve the representation problem
in the presence of internal constraints.
As to a formulation, we lay down the following. For M a subspace of Sym, and
G a subgroup of Rot, let Ela(M, G) be the class of all elasticity tensors whose
response symmetry is determined by G in a manner compatible with the con-
straint M, i.e., the class of all linear symmetric transformations C from M into
itself such that GM ⊃ GC ⊃ G.## The constrained representation problem consists
in finding (a basis for) Ela(M, G).
To gain insight for solution, suppose that for a given unconstrained material
class we have the partial spectral representation
C = γ C ⊗ C + C, C[C] = 0; (18.5)
# Note that dim M = 5, since one constraint tensor, c ⊗ c , specifies the constraint space; on the
1 3 3
other hand, dim M3 = 3, since three linearly independent constraint tensors are needed to specify
M3 , namely, V1 = c1 ⊗ c3 + c3 ⊗ c1 , V2 = c2 ⊗ c3 + c3 ⊗ c2 , and V3 = c3 ⊗ c3 . Note also that,
when deformations are constrained as (18.4) specifies, an application of (17.23) gives the following
representation for the reaction stress:
(R) (R)
S(R) = Sα3 (cα ⊗ c3 + c3 ⊗ cα ) + S33 c3 ⊗ c3 .
## Thus, G ⊂ G is a necessary condition for Ela(M, G) to be not empty.
M
CONSTITUTIVE ASSUMPTIONS 67

and suppose, in addition, that we seek a representation formula for the elastic-
ity tensor 3
C of a material with the same response symmetry, constrained by the
assignment of
M = {E ∈ Sym | C · E = 0} (18.6)
for the space of possible deformation tensors. First, we observe that we have no
problems in satisfying (18.2), since M is orthogonal to the proper space span(C)
of C, and since (16.12) implies that, for each proper space C of C, QC ⊥ = C ⊥ for
all Q ∈ GC . Secondly, we define the symmetry group of 3 C to be
& '
G3 3 3
C := Q ∈ GM | QC = CQ|M . (18.7)
Finally, as

C = GC
G3 (18.8)
(Exercise 4), we conclude that the desired constrained material has elasticity tensor
3
C = C|M = C|M . (18.9)

REMARK 1. Structural mechanics offers us here an interpretative key. Indeed,


in that theory a typical heuristic introduction to the concept of a positional con-
straint consists in (i) considering a deformable simple support, say, a linearly elastic
support offering the reaction
r = γ εc + γ̄ ε̄c̄, γ , γ̄ > 0,
to the displacement (ε, ε̄) (Figure 16), and (ii) performing a thought experiment
where the support rigidity γ is made to grow to infinity while, for the vertical reac-
tion to stay finite, the deformation ε has to tend to null: in the limit, the constraint
ε = 0 emerges and, consequently, the support’s response becomes
r = ρ (R) c + γ̄ ε̄c̄,

Figure 16.
68 P. PODIO-GUIDUGLI

with an indeterminate vertical reaction ρ (R) . Similarly, we have from (18.5) that
S = C[E] = γ (C · E)C + C[E],
and, if the stress has to remain finite when γ → +∞, then necessarily the con-
straint (18.6) emerges; the resulting reaction has the direction of C and indetermi-
nate magnitude, since we only require that γ (C · E) stays bounded.

Guided by these developments, whenever we are given a constraint space M, a


subgroup G of Rot, and a representation formula for all elasticity tensors in the un-
constrained material class Ela(Sym, G), we see that we may construct Ela(M, G)
as the collection of restrictions to M of all those elements of Ela(Sym, G) that have
(0, M⊥ ) as a proper pair. Accordingly, for the typical element of the constrained
material class under examination, the constitutive relation is
S = S(R) + S(A) , S(R) ∈ M⊥ , S(A) = 3
C[E], 3
C ∈ Ela(M, G). (18.10)
It may happen that O, the null fourth-order tensor, is the only element of
Ela(Sym, G) having (0, M⊥ ) as a proper pair; we then say that the data M and
G are incompatibile (vid. Exercises 5 and 6). Otherwise, the solution of the con-
strained representation problem is deducible from the solution of the corresponding
unconstrained problem. We exemplify the deduction procedure in the cases of a
transversely isotropic material and the constraints M1 , M2 , and M3 , in the order.
Formula (16.17) offers an expedient point of departure. From (18.3) and (16.16)
we see that
(M1 )⊥ = span(D1 ), (18.11)
so that
C[D1 ] = δ1 D1 + δ2 D2 . (18.12)
Thus, C[D1 ] = 0 if and only if δ1 = δ2 = 0, and we conclude that, for transversely
isotropic materials inextensible in the direction c3 , the elasticity tensor has the
following representation:
3
C = γ1 (C1 ⊗ C1 + C2 ⊗ C2 ) + γ2 (C3 ⊗ C3 + C4 ⊗ C4 )
+ δ3 (D2 ⊗ D2 ), (18.13)
where 3C is a linear transformation of C1 ⊕ C2 ⊕ C3 , C3 := span(D2 ), and (δ3 , C3 )
is now a proper pair.
Assume now that the constraint M2 prevails. Then, (17.25) and (16.14) show
that C1 = (M2 )⊥ , so that this time 3
C maps C2 ⊕ K into itself and is given by
3
C = γ2 (C3 ⊗ C3 + C4 ⊗ C4 ) + C, (18.14)1

C = δ1 (D1 ⊗ D1 ) + δ2 (D1 ⊗ D2 + D2 ⊗ D1 ) + δ3 (D2 ⊗ D2 ). (18.14)2


CONSTITUTIVE ASSUMPTIONS 69

Finally, if the constraint space is M3 , it is sufficient to take into account the


statement in Exercise 2 to conclude that the elasticity tensor must be
3
C = γ2 (C3 ⊗ C3 + C4 ⊗ C4 ) + δ3 (D2 ⊗ D2 ), (18.15)

and have domain D := C2 ⊕ C3 .

REMARK 2. Since, in the light of (15.6)2 , we have that

I # I|Sym = ISym ,

(16.4) can be written as

Ciso = 2µI # I + λI ⊗ I, (18.16)

with (18.16) regarded as a transformation of Sym. Relation (18.15) can be given


an equivalent form reminiscent of formula (18.16), namely,

C = 2µ̃(I − c3 ⊗ c3 ) # (I − c3 ⊗ c3 )
3
+ λ̃(I − c3 ⊗ c3 ) ⊗ (I − c3 ⊗ c3 ), (18.17)

with (18.17) regarded as a transformation of M3 . Note that the restrictions to {c3 }⊥


of (I − c3 ⊗ c3 ) and I coincide, and that
# $%
(I − c3 ⊗ c3 ) # (I − c3 ⊗ c3 ) %M = IM3 . (18.18)
3

EXERCISES

1. Let M be a subspace of Sym, and let M⊥ be its orthogonal complement in


Sym; moreover, let Q be the fourth-order tensor associated by (15.9) with a given
Q ∈ Rot. Show that QM = M ⇔ QM⊥ = M⊥ ⇔ QT M = M.
2. Show that M3 = M1 ∩ M2 .
3. Prove that Rot(c3 ) ⊂ GM3 , and that Rot ⊂
/ GM3 .
4. Recall that, given the elasticity tensor C, the associated symmetry group can
be represented as

GC = {Q ∈ Rot | QC = CQ}. (18.19)

Prove (18.8) for C, 3


C and GC specified by (18.5), (18.9) and (18.7), respectively.
5. Show that the only constraints compatible with isotropy are
(i) rigidity;
70 P. PODIO-GUIDUGLI

(ii) preservation of volume:


M = {E ∈ Sym | I · E = 0} (18.20)
(cf. (17.18)); and
(iii) preservation of shape:
M = {E ∈ Sym | E = εI, ε ∈ IR}. (18.21)
6. Given a constraint space M = {E ∈ Sym | V · E = 0} and an elasticity
tensor C, show that M and C are incompatible if and only if the constraint tensor
V has nonvanishing projection on each of the proper spaces of C.
7. Observe that C2 = span(c1 ⊗ c1 , c2 ⊗ c2 ), {c3 }⊥ = span(c1 , c2 ), and PM3 =
I|M3 = C3 ⊗ C3 + C4 ⊗ C4 + D2 ⊗ D2 , with PM3 the orthogonal projector onto
M3 . For 3C as in (18.17), prove that
(i) for Ciso such that λ = λ̃ and µ = µ̃,
3
C = PM3 Ciso |M3 ,# (18.22)
so that an unconstrained isotropic material and a material which is trans-
versely isotropic with respect to the direction of c3 , inextensible in that di-
rection, and unshearable along planes orthogonal to c3 , have the same stress
response to strains E belonging to M3 ;
(ii) the “Lamé” constants λ̃, µ̃ in (18.17) are expressible as follows in terms of
the proper numbers γ2 , δ3 in (18.15):
1 1
λ̃ = (δ3 − γ2 ), µ̃ = γ2 ; (18.23)
2 2
(iii) 3
C is positive definite if and only if
µ̃ > 0, λ̃ + µ̃ > 0. (18.24)
8. Show that the elasticity tensor of an incompressible transversely isotropic
material has the form
C = γ1 (C1 ⊗ C1 + C2 ⊗ C2 ) + γ2 (C3 ⊗ C3 + C4 ⊗ C4 ) + C, (18.25)1
- .
1 1
C = δ1 D1 ⊗ D1 − √ (D1 ⊗ D2 + D2 ⊗ D1 ) + D2 ⊗ D2 . (18.25)2
2 2

19. Interpretation of Material Moduli


The purpose of this section is to illustrate the physical meaning of the material
moduli introduced so far.
# C, then, is the composition of the linear mappings C |
iso M3 and PM3 .
CONSTITUTIVE ASSUMPTIONS 71

We begin with isotropic materials. From (16.13) we quickly deduce that the
stress response to a dilatation E = εI is a uniform traction S = ε(3λ + 2µ)I,
to a shearing deformation E = τ (s ⊗ t + t ⊗ s), s · t = 0, a simple shear S =
τ 2µ(s ⊗ t + t ⊗ s); the related technical material moduli are, respectively, the bulk
and the shear modulus β and G:
S·I t · Ss
3β := = 3λ + 2µ, 2G = = = 2µ. (19.1)
E·I t · Es
Two other moduli are extensively used in technical applications in the place of
the Lamé moduli λ and µ; to introduce them, we note preliminary that, for positive
definite materials, (16.13) and (16.19) imply that the compliance tensor has the
form
- .
1 1 1 λ
C−1 = dev + sph = I− I ⊗ I .# (19.2)
2µ 3λ + 2µ 2µ 3λ + 2µ
If the stress state is uniaxial in the direction s:

S = ψs ⊗ s, (19.3)

then the strain state is


- .
−1 1 λ
E = C [S] = ψ s⊗s− I . (19.4)
2µ 2µ(3λ + 2µ)
It follows from (19.4) that Young’s modulus E, the ratio of the axial stress compo-
nent and the corresponding axial strain, is
s · Ss µ(3λ + 2µ)
E := = ; (19.5)
s · Es λ+µ
moreover, Poisson’s modulus,## the opposite of the inverse ratio of the axial strain
to the transversal strain in any direction t such that t · s = 0, is
t · Et λ
ν := − = . (19.6)
s · Es 2(λ + µ)

REMARK. The positivity inequalities (16.19) are equivalent to either pair of


simultaneous inequalities

β > 0, G>0 (19.7)

and

E > 0, −1 < ν < 1/2. (19.8)


# As this formula suggests, for C to be invertible it is sufficient that both µ )= 0 and 3λ + 2µ )= 0.
## Oftentimes called also the modulus of lateral contraction.
72 P. PODIO-GUIDUGLI

Moreover, λ is positive if and only if ν is, and it is only for ν > 0 that, in a uni-
axial state of stress such as (19.2), fibers orthogonal to the load direction shorten
(lengthen) when parallel fibers lengthen (shorten): in other words, the opposite
behaviour is not excluded by the restriction that the stored energy be positive
definite.

We now direct our attention to transversely isotropic materials. In general, an


unconstrained transversely isotropic material has one bulk modulus, but more than
one shear modulus, Young’s modulus, or Poisson’s modulus (Exercise 1). For the
constrained materials characterized by the constraint space M3 and the elasticity
tensor (18.17) the technical moduli corresponding to β and G are
β̃ = λ̃ + µ̃ 3 = µ̃;
and G (19.9)
their definitions parallel (19.1), and refer to situations when, respectively, E =
ε(I − c3 ⊗ c3 ) or E = τ (s ⊗ t + t ⊗ s), with the unit vectors s, t, and c3 , mutually
orthogonal. If the positivity restrictions (18.24) prevail, the compliance tensor is a
well-defined linear transformation of D, namely,
1
C−1 =
3 (I − c3 ⊗ c3 ) # (I − c3 ⊗ c3 )
2µ̃
λ̃
− (I − c3 ⊗ c3 ) ⊗ (I − c3 ⊗ c3 ) (19.10)
4µ̃(λ̃ + µ̃)
(cf. (18.17)); if in (19.3) we choose s · c3 = 0, we get Young’s modulus

3 = 4µ̃(λ̃ + µ̃) ,
E (19.11)
λ̃ + 2µ̃
whereas, for s as above and t such that t·c3 = s·t = 0, the corresponding Poisson’s
modulus is
λ̃
ν̃ = . (19.12)
λ̃ + 2µ̃
We deduce from (18.24) that β̃, G 3 and E 3 are strictly positive; moreover, if λ̃ > 0,
then ν̃ ∈ ]0, 1[, so that, in a stress state S = ψs ⊗ s, fibers orthogonal to both s and
c3 can only shorten (lengthen) when fibers parallel to s lengthen (shorten).

EXERCISES

1. Consider the uniaxial stress states S(3) = ψc3 ⊗ c3 and S(1) = ψc1 ⊗ c1 ,
and the associated strains E(3), E(1) . Show that, for an unconstrained transversely
isotropic material,
c3 · S(3)c3 c1 · S(1)c1
E(3) := )= E(1) := , (19.13)
c3 · E(3)c3 c1 · E(1)c1
CONSTITUTIVE ASSUMPTIONS 73

Figure 17.

c1 · Ec1 c3 · Ec3
ν(1,3) := − )= ν(3,1) := − . (19.14)
c3 · Ec3 c1 · Ec1

Note, in particular, that E(3) = (C−1 −1 −1 −1


3333 ) , E(1) = (C1111 ) .

2. Construct an interpretation of the coefficient γ1 (γ2 ) in (18.13) as the modulus


relative to a shearing deformation accompanied by a strain proportional to Cα , α =
1, 2 (C3 ) (Figure 17).
75

CHAPTER IV
Equilibrium

20. Classical, Strong, and Weak Formulations


Boundary value problems of interest in applications admit various formulations of
different generality and scope. A classical formulation of the equilibrium problem
in linear elasticity consists of an assignment of data, a list of unknowns, and a set
of equations that the unknowns should satisfy for given data.
The data are:
– a regular region " of E, whose boundary ∂" has a well-defined outer normal
n almost everywhere and is composed of two complementary and disjoint
portions ∂1 " and ∂2 ":
∂1 " ∪ ∂2 " = ∂", ∂1 " ∩ ∂2 " = ∅; (20.1)
– a displacement field uo over ∂1 ";
– a system of loads (so , bo ), with the volume loads bo defined over " and the
surface loads so defined over ∂2 ";
– an elasticity tensor C describing the stress response to deformation of the
material comprising the region " (with C depending in general on the place
in ").
It is required to find an elastic state (u, E, S), i.e., a triplet consisting of one vec-
tor and two symmetric-valued tensor fields over " that correspond to equilibrium,
in the sense that they satisfy the following equations:
– (kinematical conditions)
sym(∇u) = E in ", u = uo in ∂1 "; (20.2)
– (statical conditions)
Div S + bo = 0 in ", Sn = so in ∂2 "; (20.3)
– (constitutive condition)
S = C[E] in ". (20.4)
76 P. PODIO-GUIDUGLI

Within this setting one distinguishes three basic equilibrium problems, accord-
ing to the type of boundary data uo and so : the displacement problem, when ∂1 " ≡
∂" and only displacement boundary data are assigned; the force problem, when
∂2 " ≡ ∂" and surface forces are assigned over the entire boundary (cf. Sec-
tion 12); and the mixed problem, when the surface measure of ∂1 ". is not null
(or, at least, ∂1 " contains three noncoplanar points, so that rigid motions are ruled
out anyway).#

REMARK 1. The displacement and load data model two types of interactions of
the environment with the body: confinement to a specific region in space, through
uo ; and exertion of external forces, both at a distance (bo ) and of contact (so )
(Section 7). Classically, neither the displacement data nor the load data depend
functionally on the solution, but there are instances of practical importance where
more far ranging possibilities have to be considered.##

The classic formulation has the merit of making clear the distinction between
kinematical, statical and constitutive conditions on elastic states; however, it is
defective, because the function space setting is left unspecified, and therefore, the
well-posedness issue cannot be dealt with. In the manner of Hadamard, a linear
problem is well-posed whenever one can show that a solution exists, is unique,
and depends continuously on the data: just to start talking with some rigor about
existence of solutions one has to specify in what function space they are sought,
and to be explicit about the smoothness of the data.
We stipulate that both the displacement and the force data uo and (so , bo ) be
continuous fields over ∂1 ", ∂2 " and ", respectively; moreover, we let the depen-
dence of C on place, if any, be smooth. We then introduce the following collections
of fields over ":
– the space of strong solutions
& '
U := u ∈ C 2 (") ∩ C 1 (") | u = uo in ∂1 " ;

– the space of variations


& '
V := v ∈ C 1 (") ∩ C 0 (") | v = 0 in ∂1 " ;

– the space of statically admissible stresses


&
S := S ∈ C 1 (") ∩ C 0 (") | S ∈ Sym in ",
'
Div S + bo = 0 in ", Sn = so in ∂2 " .
# Although this set of problems may seem quite inclusive, this is not so; for example, in the theory
of structures such as rods or plates, it may happen that the boundary portions where displacement
and forces are assigned are not disjoint. We shall discuss more general assignments of boundary data
in Section 24.
## Cf. footnote# , Section 11, p. 37.
EQUILIBRIUM 77

A strong formulation of the equilibrium problem is given as follows. First, the


kinematical condition (20.2)1 is inserted into the constitutive condition (20.4) to
obtain
! "
S(u) = C sym(∇u) ; (20.5)

secondly, (20.5) is inserted into the static conditions (20.3) to arrive at a differential
equation for the displacement field u:
! "
Div C sym(∇u) + bo = 0 in ", (20.6)

subject to the boundary condition

C[sym(∇u)]n = so in ∂2 "; (20.7)

thirdly, it is required to find a displacement field u ∈ U that satisfies (20.6) and


(20.7); such a displacement field is called a strong solution of the equilibrium
problem.
Now, let S be an arbitrarily chosen, statically admissible stress field. Then, for
each v ∈ V , with the use of the divergence identity (6.6) we have that
1 1 1 1
# # T $ $
0= Div S · v + bo · v = Div S v − S · ∇v + bo · v,
" " " "

from which, by (6.5) and the definition of S, the so-called virtual work equation
follows:
1 1 1
S · ∇v = bo · v + so · v, (S, v) ∈ S × V .# (20.8)
" " ∂2 "

The derivation of this last relation suggests that the statical conditions (20.3) can
be given a weak form defined by the requirement that (20.8) hold for all admissible
variations:
1 1 1
S · ∇v = bo · v + so · v, v ∈ V .## (20.9)
" " ∂2 "

In particular, (20.9) is satisfied by the stress field S(u), with S(u) as in (20.5),
whenever u is a strong solution of the equilibrium problem; but it also makes
sense for displacement fields of lesser smoothness. To bring the latter remarkable
property into light we introduce
– the space of weak solutions
& '
U := u ∈ C 1 (") ∩ C 0 (") | u = uo in ∂1 " ,
# Notice that the symmetry of the stress field plays no role in the derivation of (20.8), just as in
the characterization (11.2) of force balance. The qualifier weak is used here in the sense of integral,
or global, as opposed to differential, or local.
## Note that S need not be differentiable for (20.9) to hold.
78 P. PODIO-GUIDUGLI

and give the following weak formulation of the equilibrium problem:


to find a displacement field u ∈ U such that
1 1 1
S(u) · ∇v = bo · v + so · v, v ∈ V . (20.10)
" " ∂2 "

REMARK 2. Displacement fields in U have the same smoothness as we stipulated


for deformations when we developed a local theory of strain in Chapter I. One may
ask whether a sufficiently smooth weak solution of (20.10) is a strong solution of
the equilibrium problem. The answer is in the affirmative, as it is enough first to
use the divergence identity “backwards” to get
1 1
# $
Div S(u) + bo · v − (Sn − so ) · v = 0, v ∈ V ,
" ∂2 "

and then to appeal to the arbitrariness in the choice of v.

EXERCISES

1. Let A ∈ Sym, a ∈ V be given. A well-known condition for existence and


uniqueness of the solution to the problem of finding u ∈ V such that Au = a is
that det A )= 0. Prove the following a priori estimate that establishes the continuous
dependence of the solution on the data when A is positive definite:
|a|
|u| $ , (20.11)
αmin (A)
where αmin (A) is the smallest proper number of A.
2. In structural mechanics a classical formulation of the mixed equilibrium
problem for the axial deformation of a rod subject to axial loads (σo , bo ) is to
find a triplet (u, ε, σ ) consisting of axial displacement u, strain ε, and normal
force σ , such as to satisfy all of the following kinematical, statical, and constitutive
conditions:
u1 = ε in ]0, λ[, u(0) = uo , (20.12)

σ 1 + bo = 0 in ]0, λ[, σ (λ) = so , (20.13)

σ = ρε in ]0, λ[, ρ(p) > 0 (20.14)

(here λ, ρ are the rod’s length and extensional rigidity, respectively, and a prime
denotes differentiation with respect to the axial coordinate p). Give strong and
weak formulations of problem (20.12)–(20.14), and observe that they both provide
EQUILIBRIUM 79

an image stripped to the bone of the equilibrium problem of linear elasticity, in its
versions (20.6), (20.7) and (20.10), respectively.
3. With the notation and the mechanical interpretation of the preceding exercise,
consider solving the ordinary differential equation
# 1 $1
ρu + bo = 0 (20.15)
in the space
& # $ # $ '
U = u ∈ C 2 ]0, λ[ ∩ C 1 [0, λ] | u(0) = 0 .#
(i) Use the Cauchy–Schwarz inequality
%1 λ % -1 .1/2 - 1 .1/2
% % λ λ
% (uv)%% $ u2 v2 (20.16)
%
0 0 0

to establish the following a priori pointwise


2 bound for the solution in terms
1 λ 12
of the total stored energy 5(u) = 2 0 ρu :

25
u2 (p) $ p, p ∈ [0, λ], ρmin := min ρ(p). (20.17)
ρmin p∈]0,λ[

(ii) Prove that each function


& # $ # $ '
u ∈ U = u ∈ C 1 ]0, λ[ ∩ C 0 [0, λ] | u(0) = 0

satisfies the following Poincaré-type inequality:


1 λ 1 λ
u $λ
2 2
u12 . (20.18)
0 0

(iii) Use (20.18) to establish continuous-dependence for problem (20.15), in the


following form:
8u8 $ α1 8bo 8 + α2 |so |,
#√ $−1
α1 = (2ρmin )−1 λ2 , α2 = 2ρmin λ3/2, (20.19)
- 1 λ .1/2
8u8 := u2
0
(cf. (20.11)).
# The equilibrium shape of a string, fixed at its ends and subject to a transverse load b (p) per unit
o
length, is the solution of the following particular case of problem (20.14):

ρu11 + bo = 0, u ∈ U, u(λ) = 0,

with u the string’s deflection and ρ the constant tensile stress in the string.
80 P. PODIO-GUIDUGLI

21. Variational Formulation. The Principle of Minimum Potential Energy


Consider
– the space of kinematically
& # 1admissible # (displacement,
$$ strain) pairs
0 0
K := (u, E) ∈ C (") ∩ C " × C (") | E ∈ Sym in ",
'
sym(∇u) = E in ", u = uo in ∂1 " .
In the light of (20.2) and (20.3), this notion of a kinematically admissible (dis-
placement, strain) pair is complementary to the notion of a statically admissible
stress with respect to the notion of state suggested by the classical formulation of
the equilibrium problem (Exercise 1); moreover, if u is a strong solution of the
equilibrium problem, and if we set E(u) = sym(∇u) and S(u) = C[E(u)], then
(u, E(u), S(u)) is an elastic state, with (u, E(u)) ∈ K and S(u) ∈ S.
Clearly, if (u, E) ∈ K, then u ∈ U ; conversely, for u ∈ U , (u, E(u)) ∈ K
provided E(u) ∈ C 0 ("). Given the functional over K:
1 1 1
1
3
8{u, E} := σ (E) − bo · u − so · u, σ (E) = E · C[E], (21.1)
" " ∂2 " 2

one is led to consider the potential-energy functional, i.e., the functional over U
defined by
1 1 1
# $
8{u} := σ E(u) − bo · u − so · u; (21.2)
" " ∂2 "

a straightforward consequence of definitions (21.1) and (21.2) is that


& ' # $
8{u} = 83 u, E(u) , u ∈ U , E(u) ∈ C 0 " . (21.3)

The stored-energy functional


1
$
5{u} = σ (E(u) ,
"

a quadratic functional, is interpreted as the elastic energy stored in the body when
the displacement u from the reference placement occurs; as C is positive definite,
the stored energy density σ is never negative over U , so that the stored-energy
functional vanishes at u ∈ U if and only if u is a rigid displacement. The linear
functional
-1 1 .
T{u} = − bo · u + so · u
" ∂2 "

is the load potential, and accounts for the energy of the system of applied loads.
Accordingly,

8{u} = 5{u} + T{u} (21.4)


EQUILIBRIUM 81

is often referred to as the total potential energy associated with the displacement
field u.#
With the use of the potential-energy functional we can demonstrate an interest-
ing variational characterization of weak solutions to the equilibrium problem of
linear elasticity. Let the first variation of 8 be defined as
d
δ8{u}[v] := 8{u + εv}|ε=0 , v ∈ V. (21.5)

Observe that (21.1)2 yields, for each ε ∈ IR and for w = u + εv,
# $ # $ ! " # $
σ E(w) = σ E(u) + εE(v) · C E(u) + ε 2 σ E(v) . (21.6)

Thus,
-1 1 1 .
8{u + εv} = 8{u} + ε S(u) · E(v) − bo · v − so · v
" " ∂2 "
1
+ ε 2 σ (E(v) (21.7)
"

and
1 1 1
δ8{u}[v] = S(u) · E(v) − bo · v − so · v. (21.8)
" " ∂2 "

Recalling (20.10), we conclude that u ∈ U is a weak solution of the equilibrium


problem if and only if the potential-energy functional is stationary at u, i.e.,

δ8{u}[v] = 0, v ∈ V. (21.9)

Accordingly, a variational formulation of the equilibrium problem consists in find-


ing the stationary set of the potential-energy functional; further description of this
set is achieved when strong equilibrium solutions are considered.
For u a strong solution of the equilibrium problem, we pick any u ∈ U , and
write the algebraic identity (21.6) for ε = −1 and w = u − u:
# $ # ! " # $
σ E(w) = σ E(u)) − E(u) · C E(u) + σ E(u) . (21.10)

In view of (13.8) and (21.1)2 , we see that


# It can be shown (vid. [24] and the literature cited therein) that the second law of thermody-
namics takes the following form for linearly thermoelastic materials undergoing strongly isentropic
processes
1 1
.
bo · u + . d
so · u − 5(u) ! 0,
" ∂2 " dt

with the left side of the inequality interpreted as the dissipation, namely, the working of the applied
loads minus the time rate of change of the stored energy.
82 P. PODIO-GUIDUGLI
! "
E(u) · C E(u) = S(u) · E(u)
# $
= S(u) · E(w) + S(u) · E(u) = S(u) · E(w) + 2σ E(u) . (21.11)
Thus, (21.10) and (21.11) imply that
# $ # $ # $
σ E(w) = σ E(u) − σ E(u) − S(u) · E(w). (21.12)

We now note that w is an element of V , whereas S(u) ∈ S, so that, by the virtual


work equation (20.8),
1 1 1
S(u) · ∇w = bo · w + so · w; (21.13)
" " ∂2 "

with (21.13), integration of (21.12) over " yields


1 1 1
# $ # $ # $$
σ E(w) = σ E(u) − σ E(u
" " "
1 1
# $ # $
− bo · u − u − so · u − u
" ∂2 "
= 8{u} − 8{u}. (21.14)
As the stored energy is never negative, we then have that
8{u} ! 8{u}, u ∈ U ;# (21.15)
moreover, as the stored-energy functional only vanishes for a rigid displacement,
we also have that in (21.15) equality holds only if u and u differ at most by a
rigid displacement field over ". We have just proved the following principle of
minimum potential energy: if u is a strong solution of the equilibrium problem,
then the potential-energy functional attains a global minimum at u; moreover, to
within at most a rigid displacement, such minimum point is unique.

REMARKS.
1. A key hypotheses to prove the above result is that C be positive definite
(Section 13). A number of interesting well-posedness results hold true under less
stringent, or different, requirements on C (such as strong ellipticity, homogeneity,
isotropy, etc.), accompanied by more specific assumptions on " (e.g., that " be
star-shaped) and/or consideration of either the displacement or the traction problem
alone.##
# Note that this result implies the following inequality:

3 E} ! 8{u},
8{u, (u, E) ∈ K.
## The existence issue is masterfully treated in [7]; as to uniqueness, vid. [11, Section 32; 14].
C is strongly elliptic (at a point p ∈ ") when it is positive definite over the collection of nonnull
dyads A = a ⊗ b, a, b ∈ V, homogeneous when C(p) ≡ const. " is star-shaped if there is a point
po ∈ " such that the line segment from po to any point p ∈ ∂" intersects ∂" only at p itself.
EQUILIBRIUM 83

2. One can prove (21.15) also for u a weak solution of the equilibrium problem.
Indeed, the only change needed to extend the given proof would be to establish
(21.13) by appealing to (20.10), instead of (20.8). Note that, for u and u two weak
solutions of the equilibrium problem, (21.15) implies that 8(u) = 8( u ). Thus,
the potential-energy functional has constant and minimum value over its stationary
set; the latter, when it is not empty, consists at most of one significant point, plus
all its rigid equivalents.
3. The validity of the principle of minimum potential energy does not pre-
suppose that an equilibrium solution exists, be it strong or weak. Similarly, even
without proving that the potential-energy functional has indeed a minimum, it can
be shown that a converse to this principle holds, namely, that if u ∈ U happens to
minimize 8 and be sufficiently smooth, then u provides us with a strong solution
of the equilibrium problem (vid. [11, Section 36]).
4. The existence of a solution to the minimum problem for 8 in a suitable
Hilbert space follows from fairly general and by now completely standard tech-
niques of functional analysis (vid. [21, Chapter 5]); equally standard regularization
methods, both at interior points of " and at its boundary, allow us to interpret such
solutions as equilibrium solutions (vid. [18, Vol. I]).

EXERCISE

1. The basic duality of the spaces of kinematically admissible (displacement,


strain) pairs K and of statically admissible stresses S is expressed by
1 1 1 1
S·E = bo · u + so · u + Sn · uo ,
" " ∂2 " ∂1 "
# $
(u, E), S ∈ K × S, (21.16)

with the equation in relation (21.16) regarded as an identity over K × S (cf. the
virtual work relation (20.8), where the equation is regarded as an identity over
S × V ).#
(i) Show that, if the equation in (21.16) is supposed to hold for all (u, E) ∈ K,
then S ∈ S; and that, conversely, if that equation holds for all S ∈ S, then
(u, E) ∈ K. Thus, depending on the choice of the quantifier, the identical
fulfillment of the equation in (21.16) can be seen as a characterization either
of those stress fields that balance the given loads (Section 12 and (20.3)) or of
those (displacement, strain) fields that are mutually compatible in the sense
of (20.2).
# This identity supports the view, promoted by E. Sternberg (cf. [11]), that a notion of state plays
a central role in linear elasticity, and suggests that in fact that notion should be somewhat more
structured than simply a triplet (u, E, S), as writing ((u, E), S) implies.
84 P. PODIO-GUIDUGLI

(ii) Show that (20.9) obtains from (21.16) when the latter is written for a fixed
S ∈ S and for (u, E)α ∈ K (α = 1, 2).
(iii) Show that, when (21.16) is written for a fixed (u, E) ∈ K and for Sα ∈ S (α =
1, 2), then
1 1
T·E= Tn · uo , T ∈ T, (21.17)
" ∂1 "

where T , the space of stress variations, is defined to be


& # $
T := T ∈ C 1 (") ∩ C 0 " | T ∈ Sym in ",
'
Div T = 0 in ", Tn = 0 in ∂2 " .

(iv) Confirm that relation (21.17) can be used to give the following weak formula-
tion in terms of stress of the equilibrium problem: to find a stress field S ∈ S
such that
1 1
T · E(S) = Tn · uo , T ∈ T , (21.18)
" ∂1 "

where E(S) = C−1 [S]. # This formulation should be compared with the weak
formulation in terms of displacement given in Section 20.

22. Minimum Complementary Energy. Variational Principles


In this section we first establish the other classical minimum principle of linear
elastostatics, the principle of minimum complementary energy, and then discuss
two among the many variational principles that have been proposed, the Hellinger–
Prange–Reissner and Hu–Washizu principles.

22.1. THE PRINCIPLE OF MINIMUM COMPLEMENTARY ENERGY

Consider the following functional over the space S of statically admissible stresses:
1 1
1
9{S} := σ̌ (S) − Sn · uo , σ̌ (S) := S · C−1 [S]. (22.1)
" ∂1 " 2
Clearly, by (13.9),
σ̌ (S) = σ (E) for S = C[E]. (22.2)
# The crux is to prove that a strain field E = C−1 [S] that solves (21.18) is kinematically admis-
sible, in the sense that there is a displacement field u such that (u, E) ∈ K. The analytical tool is an
orthogonal-decomposition theorem for second-order tensor fields (vid. [31]) applied to the collection
of kinematically admissible strains and the space of stress variations (vid. also [25]).
EQUILIBRIUM 85

We wish to motivate why (22.1) is called the complementary-energy functional by


showing that, for u a strong solution of the equilibrium problem, the potential-
energy functional and the complementary-energy functionals sum up to zero:
& '
8{u} + 8 S(u) = 0. (22.3)
To this end we establish a preparatory result due to Lamé.
In the identity (6.22), interpret the vector field v as a displacement field u,
and the tensor field A as a stress field S that satisfies the equilibrium differential
equation (20.3)1 in the region ". Then,
1 1 1
Sn · u + bo · u = S · E(u). (22.4)
∂" " "

Assume now, in addition, that the material comprising " is linearly elastic, with
stored energy density σ as in (21.1)2 , and that the field S in (22.4) is such that
! "
S = S(u) = C E(u) . (22.5)
Then, (22.4) yields Lamé’s result:
1 1 1
# $
S(u)n · u + bo · u = 2 σ E(u) . (22.6)
∂" " "

With definitions (21.2) and (22.1), (22.2) and (22.6), we obtain the announced
result:
& '
8{u} + 9 S(u)
1 1 1 1 1
# $ # $
= σ E(u) − bo · u − so · u + σ̌ S(u) − S(u)n · uo
" " ∂2 " " ∂1 "
1 1 1
# $
= 2 σ E(u) − bo · u − S(u)n · u = 0.# (22.7)
" " ∂"

Now, let S = C[E(u)] be the stress field corresponding to a strong solution u


of the equilibrium problem; and, for any chosen S ∈ S, let W = S − S. Then,
following steps like those that led us to (21.12), we arrive at the analogous relation
σ̌ (W) = σ̌ (S) − σ̌ (S) − W · C−1 [S]
= σ̌ (S) − σ̌ (S) − W · E(u). (22.8)
At this point, we note that W satisfies the homogeneous version of (20.3), namely,
Div W = 0 in ", Wn = 0 in ∂2 "; (22.9)
and we call upon (6.22) to obtain
1 1
W · E(u) = Wn · uo . (22.10)
" ∂1 "
# Here we have used (20.5) and (20.7) to obtain that s = S(u)n in ∂ ".
o 2
86 P. PODIO-GUIDUGLI

Relations (22.8) and (22.10) imply


1 1 1 1
σ̌ (W) = σ̌ (S) − σ̌ (S) − (Sn − Sn) · uo
" " " ∂1 "
= 9(S) − 9(S). (22.11)

Thus, since C−1 is positive definite, we have established the following principle
of minimum complementary energy: if u is a strong solution of the equilibrium
problem, then the complementary-energy functional attains a global minimum at
the stress field S engendered by u:

9{S} ! 9{S}, S ∈ S; (22.12)

moreover, the inequality in (22.12) is strict unless S ≡ S.

22.2. THE STATIONARITY PRINCIPLE OF HELLINGER – PRANGE – REISSNER

The variational principles of linear elasticity may be given a unifying formulation


that proves especially suitable for those which are not extremum principles. This
formulation is in terms of a notion of state; it asserts that the solution set of the
equilibrium problem can be identified with the stationary set of a functional defined
over a collection of admissible states. In our present context, a state is a triplet
(u, E, S) of a vector field u and two symmetric-valued tensor fields E, S over "; a
state is admissible whenever:
(i) the displacement field u is of class C 2 ("), with both u and sym(∇u) of class
C 0 (");
(ii) the strain field E is of class C 0 ( " );
(iii) the stress field S is of class C 1 ( " ) ∩ C 0 ( " ), with Div S ∈ C 0 (") (cf. [11,
Section 34]). We shall denote by A the space of all admissible states, and
by AK the subset of all kinematically admissible states in A, namely, those
obeying the kinematical condition (20.2)1 :

sym(∇ u) = E. (22.13)

The Hellinger–Prange–Reissner principle asserts that u is a strong solution of


the equilibrium problem if and only if u renders stationary the following functional
over AK :
1 1 1 1 1
:{u, E, S} := σ̌ (S) − S · E + bo · u + Sn · (u − uo ) + so · u.
" " " ∂1 " ∂2 "
(22.14)

To prove this statement we first differentiate at ε = 0 the mapping

ε $→ :{u + εv, E + εV, S + εW}, sym(∇v) = V, (22.15)


EQUILIBRIUM 87

to evaluate the first variation of the functional (22.14) at (u, E, S):


1 1 1
# −1 $
δ:{u, E, S}[v, V, W] = C [S] − E · W − S · V + bo · v
"
1 1" "
1
+ (u − uo ) · Wn + Sn · v + so · v.
∂1 " ∂1 " ∂2 "
(22.16)
We then note that, by identity (6.22) and condition (22.15)2 ,
1 1 1
S · V = − v · Div S + Sn · v. (22.17)
" " ∂"

Thus,
1 1
# $
δ:{u, E, S}[v, V, W] = C−1 [S] − E · W + (Div S + bo ) · v
" "
1 1
+ (u − uo ) · Wn − (Sn − so ) · v. (22.18)
∂1 " ∂2 "

It is clear from (22.18) that, if u solves the equilibrium problem in its strong
formulation, then
& '
δ: u, E(u), S(u) [v, V, W] = 0, (v, V, W) ∈ AK , (22.19)
i.e.,
& # $ # $'
δ: u, E u , S u = 0. (22.20)
Conversely, if (22.19) holds, careful use of the arbitrariness in the choice of
(v, V, W) yields the conclusion that u is indeed a strong solution of the equilibrium
problem.#

22.3. THE STATIONARITY PRINCIPLE OF HU – WASHIZU

Positivity of C is not a requirement for the Hellinger–Prange–Reissner principle


to hold, in contrast to minimum principles; invertibility of C is enough. Neither
invertibility of the stress-strain relation nor requiring that the admissible states meet
the kinematical condition (22.13) is needed for the Hu–Washizu principle. This
principle asserts that u is a strong solution of the equilibrium problem if and only
if u renders the following functional over A stationary:
1 1 1
;{u, E, S} := σ (E) − S · E − (Div S + bo ) · u
" " "
1 1
+ Sn · uo + (Sn − so ) · u. (22.21)
∂1 " ∂2 "
# Extensions of the fundamental localization lemma of calculus of variations are needed,
expounded in [11, Section 35].
88 P. PODIO-GUIDUGLI

The first variation of this functional is

δ;{u, E, S}[v, V, W]
1 1 1
# $
= C[E] − S · V − (Div S + bo ) · v + (Sn − so ) · v
" " ∂2 "
1 1 1 1
− W · E − Div W · u + Wn · u + Wn · uo . (22.22)
" " ∂2 " ∂1 "

But (6.22) applied to W and u furnishes


1 1 1
W · sym(∇u) = − Div W · u + Wn · u, (22.23)
" " ∂"

and thus (22.22) can be given the form

δ;{u, E, S}[v, V, W]
1 1
# $
= C[E] − S · V − (Div S + bo ) · v
" "
1 1 1
! "
− W · E − sym(∇ u) − Wn · (u − uo ) + (Sn − so ) · v.
" ∂1 " ∂2 "
(22.24)

Clearly, a strong solution u of the equilibrium problem satisfies


& '
δ; u, E(u), S(u) [v, V, W] = 0, (v, V, W) ∈ A. (22.25)

Once again the final, subtler part of the proof consists in drawing the desired con-
clusion from the assumption that the first variation vanishes identically (vid. [11,
Section 38]).

REMARKS.
1. There are some interesting connections between the minimum and the vari-
ational principles we have presented: the Hu–Washizu functional ;{u, E, S} re-
duces to the potential-energy functional 8{u} whenever the (displacement, strain)
pair (u, E) in the triplet (u, E, S) is kinematically admissible; and the Hellinger–
Prange–Reissner functional :{u, E, S} reduces to the complementary-energy func-
tional 9{S} whenever the stress S in the triplet (u, E, S) is statically admissible.
Similar connections are found in Exercise 2.
2. The variational approach to equilibrium problems we have discussed has
a partial counterpart in finite elasticity. A variational characterization of weak
solutions is achievable in finite elasticity, but a variational principle involving com-
plementary energy would be in general vacuous, because it would presume invert-
ibility of the constitutive relation, a feature generally undesirable in finite elasticity
(vid., e.g., [10; 32, Section 88]). Nevertheless, a stationarity principle of Hellinger–
Prange–Reissner type may be established for the special class of materials named
EQUILIBRIUM 89

after St. Venant and Kirchhoff (Exercise 3); it has been used as a basis for derivation
of nonlinear plate equations by the use of scaling methods.#

EXERCISES

1. Suppose that the equilibrium problem has a strong solution. Show, then, that
3{u, E} + 9{S} ! 0
8 (22.26)
for each kinematically admissible (displacement, strain) pair (u, E) and each stati-
cally admissible stress S.
2. Use the divergence identity (6.22) to show that
(i) ;{u, E, S} = 8{u} for all states (u, E, S) such that (u, E) ∈ K;
(ii) :{u, E, S} = 9{S} for all states (u, E, S) such that S ∈ S.
Under the assumption that C is invertible, show that
(iii) :{u, E, S} = −8{u} for all states (u, E, S) such that (u, E) ∈ K and S =
C[E];
(iv) ;{u, E, S} = −9{S} for all states (u, E, S) such that S ∈ S and S = C[E].
3. A St.Venant–Kirchhoff material is an elastic material such that the Cosserat
stress measure 3TR introduced in (10.13) depends linearly on the nonlinear strain
measure D defined by (3.2)1 :
3TR = C[D]. (22.27)
Take a kinematically admissible state to be a triplet (u, D, 3 TR ) of smooth fields
such that
1
sym(∇ u) + (∇ u)T (∇ u) = D (22.28)
2
(cf. (3.4)1 ); assume that C in (22.27) is invertible;## and consider the functional
1 1 1
# $
3 3
:{u, D, TR } := 3 3
σ̌ TR − TR · D + bo · u
" "
1 1"
+ TR nR · (u − uo ) + so · u, (22.29)
∂1 " ∂2 "
TR = F3
TR , F = I + ∇u,
defined over the set of kinematically admissible states. Show that the first varia-
tion of the functional (22.29) vanishes at a state (u, D, 3
TR ) that solves the exact
equilibrium problem
# $ ! "
Div F3TR + bo = 0, C−1 3 TR = D in "; (22.30)1
# These methods are discussed in [6, Vol. II].
## E.g., in [6, Vol II], it is assumed that C has the form (16.4), and that positivity restrictions
stronger than (16.19) hold, namely, λ > 0 and µ > 0.
90 P. PODIO-GUIDUGLI
# $
u = uo in ∂1 ", F3
TR nR = so in ∂2 " (22.30)2

(cf. (10.16); of course, the fields D and F in (22.30) depend on the displacement
field u as specified by (22.28) and (22.29)3 , respectively).
4. The functional : 3 (u, D, 3
TR ) is a nonlinear version of the Hellinger–Prange–
Reissner functional; the former can be used only for St.Venant–Kirchhoff mate-
rials, the latter for all linearly elastic materials with invertible elasticity tensor. Is
there a nonlinear version of the Hu–Washizu functional for St.Venant–Kirchhoff
materials (or perhaps other elastic materials with nonlinear response)?

23. Compatible Field and Boundary Operators


Given a field operator, the choice of an accompanying boundary operator can have
crucial importance to the well-posedness of the resulting boundary-value problem.
As a rule, an assignment of the restriction to the boundary of the unknown never
leads to inconsistencies. Remarkably, other types of compatible boundary operators
are suggested by the variational format, when the problem under study admits it,
as is always the case for elasticity.
To substantiate this statement, we first consider a simple model problem. For "
a regular domain in the plane p3 = 0, the field operator ruling the equilibrium of
a (linearly elastic, homogeneous, isotropic) membrane taut over " is the harmonic
operator
L[u] := −γ 6u, (23.1)
with γ a positive constant, and u(p1 , p2 ) the deflection. The associated stored-
energy functional is the functional
1
1
5{u} := γ |∇u|2 , (23.2)
2 "
whose first variation at u ∈ C 2 (") ∩ C 1 (") is
1 1
# $
δ5{u}[v] = (−γ 6u)v + (γ ∂n u)v, v ∈ C 1 (") ∩ C 0 " . (23.3)
" ∂"

This suggests that we formulate inhomogeneous (i.e., with nonnull boundary data)
boundary-value problems describing the equilibrium of membranes by associating
with the field equation
L[u] = bo in ", (23.4)
the boundary equations
B0 [u] := u = uo in ∂1 ", (23.5)

B1 [u] := γ ∂n u = so in ∂2 ". (23.6)


EQUILIBRIUM 91

In these equations bo is the assigned load per unit area of ", and uo , so are, re-
spectively, the assigned displacement and load per unit length on disjoint boundary
portions ∂1 " and ∂2 " whose union exhausts the boundary itself. The prescription
of the boundary deflection constituted by the operator B0 is usually incorporated
into the definitions of both the domain of the stored-energy functional and the space
of variations (cf. the definitions of the spaces U and V in Section 20); when this is
done, the requirement that the first variation vanish identically yields the boundary
operator B1 , that prescribes the boundary traction in terms of the normal derivative
of the deflection on the boundary. A different way of assigning the gradient of u at
the boundary would generally lead to an ill-posed problem,# for which there might
be no solution.

REMARK 1. In terms more or less standard in the calculus of variations, condition


(23.5)2 (more generally, all conditions bearing directly on the unknown and its
derivatives) is a geometric boundary condition, whereas condition (23.6)2 is a static
condition, because it has the primary meaning of a prescription of force, rather
than displacement, at the boundary. Conditions (23.6)2 et sim. are also often called
natural boundary conditions, because they result from the variational automatisms
in ways like that leading to (23.6)2 itself. Other qualifiers that for historical rea-
sons are frequently used for the operators B0 , B1 and the corresponding boundary
conditions are “of Dirichlet type” and “of Neumann type”, respectively.

Just as was done for the membrane problem, the field operator of three-dimen-
sional linear elasticity
! "
L[u] := −Div C sym(∇u) (23.7)
is associated with the geometric boundary operator
B0 [u] := u (23.8)
and with the static boundary operator
! "
B1 [u] := C sym(∇u) n. (23.9)

To see this, it is enough to perform the first variation at u ∈ C 2 (") ∩ C 1 (") of the
quadratic functional
1
1 ! "
5{u} = C sym(∇u) · ∇u (23.10)
2 "
in the direction of any v ∈ C 1 (") ∩ C 0 ("): the variation is
1 1
# ! "$ ! "
δ5{u}[v] = − Div C sym(∇ u) · v + C sym(∇ u) n · v. (23.11)
" ∂"
# Hadamard’s notion of a well-posed linear problem has been recalled in Section 20.
92 P. PODIO-GUIDUGLI

In all of the formulations of the equilibrium problem that we have given so far
we have systematically embodied the geometric condition
B0 [u] = uo in ∂1 " (23.12)
into our definitions of the spaces of solutions and variations; and we have stipulated
the statical condition
B1 [u] = so in ∂2 ". (23.13)
However, in three-dimensional elasticity there is much more freedom in the choice
of the boundary operators than in the membrane case: in the latter, either the deflec-
tion u or the traction γ ∂n u in the combination u(γ ∂n u) must be assigned at a point
of the boundary; in the former, the corresponding combination u · S(u)n allows
for an assortment of partial, complementary specifications of the displacement
vector u and the traction vector S(u)n at the boundary. We shall study the various
possibilities in the next section.

REMARK 2. A distinguished feature of both the force and the mixed problems
of elasticity – and a source of characteristic difficulties in adapting to genuine
3-dimensional problems the analytical techniques that work well in lower dimen-
sional cases – is that the operator B1 [u] involves not only the normal derivatives of
the boundary displacement, but the tangential ones as well (Exercise 2).

EXERCISES

1. Consider the strong formulation of the membrane problem:


γ 6u + bo = 0 in ", u ≡ 0 on ∂". (23.14)
With the use of Poincaré inequality
1 -1 .1/2
8u82 $ (diam ")2 |∇u|2 , 8u8 := u2 , (23.15)
" "

prove that the solution of problem (23.14) depends continuously on the data, in the
sense that
8u8 $ γ −1 (diam ")2 8bo 8 (23.16)
(cf. Exercises 1 and 3, Section 20).
2. Show that, for an isotropic material, the boundary operator (23.10) may be
written as
B1 [u] = 2µ∂n u − µ(Curl u)n + λ(Div u)n. (23.17)
EQUILIBRIUM 93

(cf. the definition (6.9) of the Curl operator).


3. Consider the model equation
# $ # $
u11 + po = 0, u ∈ C 2 ]0, 1[ ∩ C 1 [0, 1] (23.18)
(cf. Exercises 2 and 3, Section 20). Show that, for a given function po , the gen-
eral solution of (23.18) is parametrized by two out of the four boundary data
u(0), u1 (0), u(1), and u1 (1), whose values are restricted by the conditions
1 1
u1 (1) − u1 (0) + po = 0, (23.19)1
0
1 11 ζ
u(1) − u(0) − u1 (0) + po = 0. (23.19)2
0 0

When the field equation (23.18) is accompanied by a prescription of the pair (u(0),
u1 (0)) (or, equivalently, of the pair (u(1), u1 (1)), an initial-value problem obtains,
whereas a prescription of any one of the remaining pairs yields a boundary-value
problem; all these problems have a unique solution except the one which ob-
tains when a pair (u1 (0), u1 (1)) is chosen consistently with (23.19)1 , for which
the solution is unique to within an arbitrary translation.
4. Show that, for an isotropic material, the boundary operator (23.13) may be
written as
B1 [u] = 2µ∂n u − µ(Curl u)n + λ(Div u)n (23.20)
(cf. the definition (6.9) of the Curl operator).
24. Generalized Boundary Conditions
The boundary conditions that most frequently occur in the theory of structures are
of neither the geometric nor the static type discussed so far. Consider the situations
of Figure 18.

Figure 18.
94 P. PODIO-GUIDUGLI

In case (a) (respectively, (b)) at the point po ∈ ∂" we must prescribe the normal
(tangential) component of the displacement and the tangential (normal) component
of the traction. Thus, the portions ∂1 " and ∂2 " of the boundary where geomet-
rical and statical conditions are assigned are no longer disjoint, as they are in
the formulations of the mixed boundary value problem; instead, a complementing
combination of components of both the displacement and the traction vector is
specified at each point of ∂".
To model such situations we introduce a tensor field Po over ∂" whose values
are orthogonal projections;# precisely, we stipulate that
& '
Po (p) ∈ I, 0; n(p) ⊗ n(p), I − n(p) ⊗ n(p) , p ∈ ∂". (24.1)

At a point of ∂" we then write the combination u · Sn of the displacement and the
traction vectors as follows:

u · Sn = Po u · Sn + (I − Po )u · Sn
= Po u · Sn + u · (I − Po )Sn, (24.2)

and accordingly replace (23.9) and (23.10) by

B0 [u] := Po u (24.3)

and

B1 [u] := (I − Po )S(u)n, (24.4)

respectively. In this fashion the classical assignment of boundary data is replaced


by the assignment of a triplet (uo , so , Po ) of fields over ∂", such as to satisfy the
mutual consistency requirements

(I − Po )uo = 0, Po so = 0 (24.5)

following from

B0 [u] = uo , B1 [u] = so . (24.6)

It is easy to check that, at a point p ∈ ∂", Po (p) = I and Po (p) = 0 yield,


respectively, the usual geometrical and statical conditions. The other two comple-
mentary choices, Po (p) = n(p) ⊗ n(p) and Po (p) = I − n(p) ⊗ n(p), yield the
contact conditions
#
(u − uo ) · n = 0, (I − Po ) S(u)n − so ) = 0, (24.7)

# Recall from Exercise 1, Section 2, that P ∈ Sym is an orthogonal projection if P2 = P, and that
an orthogonal projection other than I and 0 has one of the following complementary representations:
u ⊗ u, I − u ⊗ u for some u ∈ U.
EQUILIBRIUM 95

and, respectively,
# $
(I − n ⊗ n)(u − uo ) = 0, S(u)n − so · n = 0, (24.8)

exemplified in Figure 18.


One may ask whether the equilibrium problem with the generalized boundary
conditions (24.3)–(24.6) continues to admit a variational formulation and, if so,
whether a principle of minimum potential energy holds, etc. An affirmative answer
to such question is obtained on defining the spaces of weak solutions, variations,
and statically admissible stresses as follows:
& # $ '
U o := u ∈ C 1 (") ∩ C 0 " | Po u = uo in ∂" ,
& # '
Vo := v ∈ C 1 (") ∩ C 0 " ) | Po v = 0 in ∂" ,
# $
So := {S ∈ C 1 (") ∩ C 0 " | S ∈ Sym in ",
'
Div S + bo = 0 in ", (I − Po )Sn = so in ∂" .

Then, the virtual work equation becomes


1 1 1
S·∇v = bo · v + so · v, (S, v) ∈ So × Vo , (24.9)
" " ∂"

and the weak formulation of the equilibrium problem is to find a displacement field
u ∈ U o such that
1 1 1
S(u) · ∇ v = bo · v + so · v, v ∈ Vo , (24.10)
" " ∂"

(cf. (20.8) and (20.10), respectively).# Moreover, introducing the generalized poten-
tial-energy functional over the space U
1 1 1
# $
8o {u} := 4 σ E(u) − bo · u − (I − Po )so · u, (24.11)
" " ∂"

we can construct a notion of variational solution, discuss the role of strong and
weak solutions as minimizers, and so on (vid. [11, Section 40]).

# Note that, due to (24.5)


1 1
so · v = (I − Po )so · v.
∂" ∂"
96 P. PODIO-GUIDUGLI

25. Elastic Equilibrium with the Cauchy Relations


Remarkably, not all of the elasticities Cij hk enter in the general expression of the
principal part of the field operator (23.8).# Indeed, given sufficient smoothness that
the order of spatial differentiation is immaterial, then
Cij hk uh,kj = 4
Cij hk uh,kj , (25.1)
for
4
C = C − Č, 2Čij hk = Cij hk − Cikhj (25.2)
(Exercise 1).
In particular, when C(p) ≡ const., as is almost invariably the case in classical
elasticity, the field operator (23.8) can be written as
! "
L[u] = −Div 4 C sym(∇ u) . (25.3)
We conclude that in a displacement problem (Section 20), where the field equation
ruled by (25.3) is associated with the boundary condition (23.12) over all of ∂",
the complex Č of material moduli has no effect. This conclusion is, however, false
in the case of force, or mixed problems, because
Čij hk uh,k nj )= 0 (25.4)
in general at a boundary point, and hence,
! " ! "
B1 [u] = C sym(∇ u) n )= 4 C sym(∇ u) n. (25.5)
Can we set
Č = O, (25.6)
as Cauchy proposed to do on the basis of his molecular theory## of elastic response?
The answer, Cauchy’s authority notwithstanding, must be negative. If we did,
given that (25.6), due to (25.2)2 , amounts to six linear conditions on C, the Cauchy
relations,‡ we would be left with an elasticity tensor 4C having only 15 independent
# The principal part of a differential operator is the part of it which contains all the derivatives of
highest order. The principal part is so called because only the coefficients of the highest derivatives
determine the classification of the related differential problem. In the case of elasticity, the definition
yields
+ + ,,
p.p. − Div C[∇ u] := −Cij hk uh,kj ,
i

and the local classification of the related elasticity problem depends on C(p) only (cf. footnote## in
Section 21, p. 82).
## A molecular model had been proposed by Navier as early as 1823 (cf. [35, Section 301]).
‡ The component form of (25.6) is

Cij hk = Cikhj .
EQUILIBRIUM 97

components (instead of the canonical 21). Such a 4 C always has the major symmetry
(13.16)3 , but in general (that is, for arbitrary material symmetry), has neither of the
minor symmetries (13.16)1,2 . Consequently, we could not exclude that infinitesimal
rigid deformations generate nonnull stress, or guarantee that the action of 4 C at a
given infinitesimal strain delivers a symmetric stress (see, however, Exercises 2
and 3).
The significance of Č in linear elasticity is made clear when we split the stored
energy density into the parts determined, respectively, by 4 C and Č:
σ (E) = 4
σ (E) + σ̌ (E), σ (E) = E · 4
24 C[E], 2σ̌ (E) = E · Č[E] (25.7)
(cf. (13.9)). One then recognizes σ̌ (E) as a null Lagrangian, that is, the diver-
genceless density of a functional whose representation as a volume integral can be
replaced by a representation as a surface integral: in this case,
1 1
∇ u · Č[∇ u] = ∇ u · Č[u ⊗ n].# (25.8)
" ∂"

REMARK. Addition of a null Lagrangian to the energy functional of a variational


problem changes only the boundary part of the associated Euler–Lagrange oper-
ator. Such change is, as a rule, crucial: for an example taken from the theory of
structures, the energy term missing in each of S. Germain’s various proposals and
later supplied by Lagrange (cf. [33]), the term that makes the Germain–Lagrange
theory of plate bending successful (Exercise 4), is a null Lagrangian.

EXERCISES

1. Show that it follows from (25.2) that


24
Cij hk = Cij hk + Cikhj . (25.9)
2. Use (16.4) and (15.13)2 to obtain the following component representation of
the Lamé elasticity tensor, regarded as a linear transformation of Lin:
Cij hk = µ(δih δj k + δik δj h ) + λδij δhk . (25.10)
Show that, consequently,
# $
24
C[A] = 2µA + (λ + µ) (tr A)I + AT , (25.11)1
# $
2Č[A] = (λ − µ) (tr A)I − AT , (25.11)2
so that the following are equivalent:
# Null Lagrangians in linear elasticity are treated in [17], from which much of the material in this
section is taken and where many relevant references are listed; other references are found, and more
general issues are treated, in [26].
98 P. PODIO-GUIDUGLI

(i) Č = O;
(ii) λ = µ;
(iii) 4
C[A] = µ(A + AT + (tr A)I).
3. Note that (iii) in Exercise 2 implies that
4
C ◦ skw = skw ◦ 4
C=O (25.12)
(cf. (13.3) and (13.4)). Construct 4
C and Č for a transversely isotropic material of
type (16.17). Does (25.12) hold in this case?
4. The standard bending-energy functional for a plate of cross-section P is
1
1 ! "
5(w) = D (6w)2 − d(w,11 w,22 − (w,12 )2 , (25.13)
2 P

where w(p1 , p2 ) is the transverse deflection, and D, d are given physical con-
stants. Check which part of the integrand determines the biharmonic field oper-
ator L[w] = D66w, and find out how the remaining part enters into the static
boundary operator B1 [w].
5. Let n denote the unit normal to the deformed shape
q = pα cα + w(p1 , p2 )c3
of the cross-section P of the plate considered in the preceding exercise.
(i) Show that
# $−1/2
n = 1 + |∇w|2 (c3 − ∇w).
(ii) Define the linearized curvature tensor K to be
K := −∇ (2) w ∼
= −∇n.
Show that
tr K = −6w, det K = w,11 w,22 −(w,12 )2
(cf. (25.13)).

26. Elastic Equilibrium in the Presence of Internal Constraints


As a part in the formulation of a problem of elastic equilibrium, the stipulation
of an internal constraint implies restrictions on the accompanying choices of both
the boundary data and the material response: e.g., in a displacement equilibrium
problem for an incompressible body the boundary displacements must be assigned
in such a way as to conserve the body’s volume (cf. the remark that ends Sec-
tion 6); and inextensibility is incompatible with the response symmetry of isotropic
materials, as discussed in the beginning of Section 18. These examples tell us that
EQUILIBRIUM 99

constrained equilibrium problems must be posed with care. We now extend the
weak formulation of the mixed problem given in Section 20 so as to account for
internal constraints.
For simplicity, we consider the case when one constraint-tensor field V ∈ C 0 (")
is assigned, with V(p) ∈ Sym, and we let
& '
U c := u ∈ U | V(p) · ∇ u(p) = 0 in "

be the associated space of constrained weak solutions. Moreover, for


& '
M(p) = E ∈ Sym | V(p) · E = 0 (26.1)

the constraint space at point p, and for


# $
M⊥ (p) := span V(p) (26.2)

(cf. definitions (17.10) and (17.11)2 ), we let the stress be split in the manner of
(17.12) into a reactive stress

S(R) = ψ (R) V, ψ (R) ∈ IR, (26.3)

and an active stress

S(A) = C[E], C: M → M, (26.4)

with the elasticity tensor C positive definite:

A · C[A] > 0, A ∈ M\{0} (26.5)

(Section 13).
We then give the following weak formulation of the mixed problem of elastic
equilibrium in the presence of an internal constraint:
• to find a displacement field u ∈ U c and a reactive stress field S(R) = ψ (R) V ∈
C 0 (") such that, for
! "
S = C E(u) + S(R), (26.6)1
1 1 1
S·∇v= bo · v + so · v, v ∈ V . (26.6)2
" " ∂2 "
(R)
Interestingly, a solution (u, ψ ) to problem (26.6) does not always exist; if
there is one, then the displacement field u is uniquely determined (cf. [23]).
To prove uniqueness of the equilibrium displacement, a straightforward adap-
tation of the classical argument by Kirchhoff is enough. Calculating upon a con-
tradiction, assume that there are two solutions, and let w denote their difference.
Then, w ∈ V , and (26.6) yields
1
! "
C E(w) · E(w) = 0; (26.7)
"
100 P. PODIO-GUIDUGLI

with this, the positivity of C, (17.27), and the fact that meas (∂1 ") )= 0, imply that
w ≡ 0 in ".
With a view to showing that nonexistence is to be expected for arbitrarily given
boundary displacements, note that the equilibrium active stress S(A) (u) = C[E(u)]
is unique, because such is the equilibrium displacement u. Note also that, if there
are two reactive stress fields associated with u at equilibrium, then their difference
T = τ V must satisfy
1
T · ∇ v = 0, v ∈ V . (26.8)
"

Suppose now that


(i) there is a solution (u, ψ (R) ) to problem (26.6);
(ii) equation (26.7) has a nontnvial, smooth solution T = τ V.
With the use of the standard identity (6.6), assumption (ii) implies that
# $
Div τ V = 0 in ", τ Vn = 0 on ∂2 ". (26.9)
From (i) and (26.9) we deduce that
1 1 1
0= u · Div(τ V) = − τ V · ∇ u + τ Vn · uo ,
" " ∂1 "

whence the following existence condition on the boundary displacement:


1
τ Vn · uo = 0. (26.10)
∂1 "

If, for a given field uo over ∂1 ", there is one (nontrivial, smooth) solution of
(26.7) that falsifies (26.10), then the assumption that problem (26.6) has a solution
cannot hold true.

REMARKS
1. We have already observed that the unconstrained problem can be regarded as
a special case of the general constrained problem. In particular, as we know also
from our discussion in Section 21 of the stationary points for the potential-energy
functional, for an unconstrained body uniqueness of the equilibrium displacement
is guaranteed; moreover, the existence condition (26.10) is empty.
2. Once the equilibrium displacement u has been found, the strong form of the
equations for the reaction multiplier ψ (R) is:
# $
Div ψ (R) V = −Div S(A) (u) − bo in ", (26.11)1

ψ (R) Vn = −S(A)(u)n + so on ∂2 " (26.11)2

(Section 17). Now, a constraint tensor V may have maximal rank (as is the case
for incompressibility) or not (inextensibility has rank 1, angle preservation has
EQUILIBRIUM 101

rank 2); and it may or may not be uniform. Maximality and uniformity together
reduce system (26.11) to its easiest form (Exercise 1). Even if the uniformity as-
sumption is dropped, maximality still keeps the mathematical complexity of the
mixed equilibrium problem down to a reasonable level (cf. [1, 14, 31]); unfortu-
nately, incompressibility seems to be the one example of maximal constraint of
interest in applications. To our knowledge, a full treatment of the submaximal case
(Exercises 2 and 3) has not been given to date.
3. The weak formulation of the mixed problem in terms of stress given in Ex-
ercise 1, Section 21, can be adapted so to hold when an internal constraint is in
force. Formally, equation (21.17) is replaced by the requirement to find a stress
field S ∈ S such that
1 1
# (A)$
T·E S = Tn · uo , T ∈ T , (26.12)
" ∂1 "

where, at each given point p ∈ ", S(A) = S − (S · V)V is the orthogonal projection
of S onto the constraint space M;# where E(S(A)) = C−1 [S(A)]; and where, we
recall from Section 21, the space of stress variations T is defined to be
& # $
T := T ∈ C 1 (") ∩ C 0 " | T ∈ Sym in ",
'
Div T = 0 in ", Tn = 0 in ∂2 "
(Exercise 4).

EXERCISES

1. Show that, for a maximal uniform constraint, system (26.10) takes the form
# $
∇ψ (R) = −V−1 Div S(A)(u) + bo in ", (26.13)1
# $
ψ (R) = −V−1 S(A)(u)n − so · n on ∂2 ". (26.13)2

2. A material surface S of normal n is characteristic for the constraint field V if


Vn = 0 on S, (26.14)

Show that:
(i) for T = τ V a solution of equation (26.7), τ ≡ 0 on the part of ∂2 " which is
not characteristic for the given constraint;
(ii) on a characteristic surface, the constraint condition is expressed in terms of
tangential derivatives only:

0 = V · ∇t u, ∇t u := ∇u(I − n ⊗ n); (26.15)

# Here we tacitly assume that the constraint tensor is normalized so as to have |V| = 1.
102 P. PODIO-GUIDUGLI

(iii) a constraint of maximal rank has no characteristic surface, while surfaces of


normal n = e1 × e2 are characteristic for a rank-2 constraint with proper
directions e1 and e2 , and surfaces orthogonal to e are characteristic for the
rank-1 constraint V = e ⊗ e.
3. Let " be a right cylinder, of cross-section P and length l, made of a fiber-
reinforced material with fibers in the direction e of the cylinder’s axis. Show that:
(i) the assignment of displacement data on the lateral surface ∂P ×]0, l[ is unre-
stricted, whereas the condition
1
# + $
e· uo − u− o =0 (26.16)
P

must be satisfied in case data u±


o are assigned on the upper and lower ends;
(ii) when uniform tractions s± o = ±σo e are assigned over the ends P ± , then
+ −
ψ ≡ σo over P ∪P ; moreover, the equilibrium displacement must satisfy
(R)

1 l
# $
0=e· Div S(A)(u) + bo . (26.17)
0

4. Show that the difference of two stress fields S(α) ∈ S such that
S(1) − (S(1) · V)V = S(2) − (S(2) · V)V

is a tensor field T ∈ T such that T(p) ∈ M⊥ (p) and that


1
0= Tn · uo (26.18)
∂1 "

(cf. (26.9)).
EQUILIBRIUM 103

References

1. D.N. Arnold and R.S. Falk, Well-posedness of the fundamental boundary value problems for
constrained anisotropic elastic materials. Arch. Rational Mech. Anal. 98 (1987) 143–165.
2. E. Benvenuto, La Scienza delle Costruzioni e il suo Sviluppo Storico. Sansoni, Firenze (1981).
3. E. Benvenuto, An Introduction to the History of Structural Mechanics. Part I: Statics and
Resistance of Solids. Part II: Vaulted Structures and Elastic Systems. Springer, Berlin (1991).
4. P. Bisegna and P. Podio-Guidugli, Mohr’s arbelos. Meccanica 30 (1995) 417–424.
5. A. Campanella, P. Podio-Guidugli, and E.G. Virga, Restrizioni costitutive per materiali elastici
lineari trasversalmente isotropi. In: Atti VIII Congr. AIMETA, Torino, Vol. II (1986) 465–470.
6. P.G. Ciarlet, Mathematical Elasticity, Vol. I: Three-Dimensional Elasticity. North-Holland,
Amsterdam (1988). Vol. II: Plates. North-Holland, Amsterdam (1997).
7. G. Fichera, Existence theorems in elasticity. In: Handbuch der Physik VIa/2. Springer, Berlin
(1972).
8. G. Fichera, Existence theorems in linear and semi-linear elasticity. Z. Angew. Math. Phys. 54
(1974) T24–T36.
9. S. Forte and M. Vianello, Symmetry classes for elasticity tensors. J. Elasticity 43 (1996) 81–
108.
10. M. François, G. Geymonat, and Y. Berthaud, Determination of the symmetries of an experimen-
tally determined stiffness tensor: application to acoustic measurements. Int. J. Solids Structures
35 (1998) 4091–4106.
11. M.E. Gurtin, The linear theory of elasticity. In: Handbuch der Physik VIa/2, Springer, Berlin
(1972).
12. M.E. Gurtin, An Introduction to Continuum Mechanics. Academic Press, New York (1981).
13. P.R. Halmos, Finite-Dimensional Vector Spaces. Van Nostrand (1958).
14. M.J. Horta Dantas, On the boundary value problems of linear elasticity with constraints.
J. Elasticity 54 (1999) 93–111.
15. O.D. Kellogg, Foundations of Potential Theory. Springer, Berlin (1929), reprinted by Dover
(1953).
16. R.J. Knops and L.E. Payne, Uniqueness Theorems in Linear Elasticity, B.D. Coleman (ed.),
Springer Tracts in Natural Philosophy, Vol. 19. Springer, Berlin (1971).
17. M.R. Lancia, P. Podio-Guidugli, and G. Vergara Caffarelli, Null Lagrangians in linear elasticity.
Math. Models Methods Appl. Sci. 5 (1995) 415–427.
18. J.L. Lions and E. Magenes, Problèmes aux Limites non Homogènes et Applications. Dunod,
Paris (1968).
19. L.C. Martins and P. Podio-Guidugli, A new proof of the representation theorem for isotropic,
linear constitutive relations. J. Elasticity 8 (1978) 319–322.
20. L.C. Martins and P. Podio-Guidugli, A variational approach to the polar decomposition
theorem. Rend. Accad. Naz. Lincei Classe Sci. Fis. Mat. Nat. 66 (6) (1979) 487–493.
21. S.G. Mikhlin, Mathematical Physics, an Advanced Course. North-Holland, Amsterdam (1970).
22. W. Noll, A general framework for problems in the statics of finite elasticity. In: G.M. de la
Penha and L.A. Medeiros (eds), Contemporary Developments in Continuum Mechanics and
Partial Differential Equations. North-Holland, Amsterdam (1978).
23. A.C. Pipkin, Constraints in linearly elastic materials. J. Elasticity 6 (1976), 179–193.
24. P. Podio-Guidugli, Inertia and invariance. Annali Mat. Pura Appl. (IV) CLXXII (1997) 103–
124.
25. P. Podio-Guidugli, The compatibility constraint in linear elasticity. J. Elasticity 59 (2000) in
press.
26. P. Podio-Guidugli and G. Vergara Caffarelli, Surface interaction potentials in elasticity. Arch.
Rational Mech. Anal. 109 (1990) 343–383.
104 P. PODIO-GUIDUGLI

27. P. Podio-Guidugli and M. Vianello, The representation problem of constrained linear elasticity.
J. Elasticity 28 (1992) 271–276.
28. P. Podio-Guidugli and M. Vianello, Constraint manifolds for isotropic solids. Arch. Rational
Mech. Anal. 105 (1989) 105–121.
29. P. Podio-Guidugli and M. Vianello, Internal constraints and linear constitutive relations for
transversely isotropic materials. Rend. Mat. Acad. Lincei s. 9, 2 (1991) 241–248.
30. P. Podio-Guidugli and E.G. Virga, Transversely isotropic elasticity tensors. Proc. Roy. Soc.
London A 411 (1987) 85–93.
31. R. Rostamian, Internal constraints in linear elasticity. J. Elasticity 11 (1981) 11–31.
32. I. Todhunter and K. Pearson, A History of the Theory of Elasticity and of the Strength of
Materials, Vol. 1. Cambridge Univ. Press, Cambridge (1886).
33. C. Truesdell, Sophie Germain: Fame earned by stubborn error. Boll. Storia Sci. Mat. XI (1991)
3–24.
34. C. Truesdell and W. Noll, The non-linear field theories of mechanics. In: Handbuch der Physik
III/3. Springer, Berlin (1965).
35. C. Truesdell and R.A. Toupin, The classical field theories. In: Handbuch der Physik, III/1.
Springer, Berlin (1960).

View publication stats

You might also like