Académique Documents
Professionnel Documents
Culture Documents
des Grundbauinstitutes
der Technischen Universitt Berlin
Herausgegeben von S.A. Savidis
HEFT 44
Differential Geometry Applied to Continuum
Mechanics
N
1 (V)
(P )
(U)
X
X(U)
x(V)
X(P )
x((P ))
x X 1
Daniel Aubram
Berlin 2009
ISBN 978-3-8322-8154-0
ISSN 0342-3905
Shaker Verlag GmbH P.O. BOX 101818 D-52018 Aachen
Phone: 0049/2407/9596-0 Telefax: 0049/2407/9596-9
Internet: www.shaker.de e-mail: info@shaker.de
Disclaimer
This paper is not intended to serve as a monograph for specialists
about dierential geometry and continuum mechanics. Many interesting topics have been omitted and many of the presented key facts
and basic results are stated without proofs; they may be found in
the standard textbooks, e.g. [1, 2, 3, 4, 5, 6, 7, 8]. Comprehensive
treatises on linear geometry and linear algebra are, for example, [9]
and [10].
Daniel Aubram
Berlin, November 2008
Abstract
Dierential geometry provides the suitable background to present and discuss continuum mechanics with an integrative and mathematically precise terminology. By
starting with a review of linear geometry in ane point spaces, the paper introduces
modern dierential geometry on manifolds including the following topics: topology,
tensor algebra, bundles and tensor elds, exterior algebra, dierential and integral
calculi. The tools worked out are applied subsequently to basic topics of continuum
mechanics. In particular, kinematics of a material body and balance of mass are formulated by applying the geometric terminology, the principles of objectivity and material
frame indierence of constitutive equations are examined, and a clear distinction of the
Lagrangian formulation from the Eulerian formulation is drawn. Moreover, the paper
outlines a generalized Arbitrary Lagrangian-Eulerian (ALE) formulation of continuum
mechanics on dierentiable manifolds. As an essential part, the grid manifold introduced therein facilitates a consistent description of the relations between the material
body, the ambient space and the arbitrary reference domain of the ALE formulation.
Not least, the objective of the paper is to provide a compilation of important formulae
and basic results some of them with a full proof frequently used by the community.
If practical, point arguments and changes in points within equations will be clearly indicated, and component and direct (or absolute) tensor notation will be applied as
needed, avoiding a single-track approach to the subject.
Keywords: dierential geometry; continuum mechanics; large deformations; Arbitrary Lagrangian-Eulerian; manifold; tensor analysis
Zusammenfassung
Die Dierentialgeometrie bietet den geeigneten Hintergrund, um die Kontinuumsmechanik mit einer einheitlichen und mathematisch prazisen Terminologie darzulegen und
zu diskutieren. Ausgehend von einem R
uckblick auf die lineare Geometrie in anen
Punktraumen f
uhrt die Arbeit in die moderne Dierentialgeometrie auf Mannigfaltigkeiten unter Ber
ucksichtigung der folgenden Themen ein: Topologie, Tensoralgebra,
B
undel und Tensorfelder, Auere
Algebra sowie Dierential- und Integralkalk
ule. Die
erarbeiteten Werkzeuge werden anschlieend auf grundlegende Themen der Kontinuumsmechanik angewendet. Insbesondere wird die Kinematik eines materiellen Korpers
und die Massenbilanz vom geometrischen Standpunkt heraus formuliert, das Prinzip
der Objektivitat von Tensoren und von Materialgleichungen wird untersucht, und es
wird der Unterschied zwischen der Lagrangeschen und der Eulerschen Formulierung
auf klarende Weise dargestellt. Desweiteren skizziert die Arbeit eine verallgemeinerte Arbitrary Lagrangian-Eulerian (ALE) Formulierung der Kontinuumsmechanik auf
dierenzierbaren Mannigfaltigkeiten. Als wesentlicher Bestandteil ermoglicht dabei die
eingef
uhrte Gittermannigfaltigkeit eine konsistente Beschreibung der Beziehungen zwischen dem materiellen Korper, dem umgebenden Raum und dem beliebigen Referenzgebiet der ALE Formulierung. Nicht zuletzt besteht die Zielsetzung der Arbeit darin,
wichtige Formeln und grundlegende Ergebnisse auf den behandelten Gebieten teilweise
auch mit vollstandigem Beweis zusammenzustellen. Sofern es zweckmaig erscheint,
werden Punktargumente und der Wechsel der Bezugspunkte in den Gleichungen hervorgehoben. Auerdem wird je nach Bedarf sowohl die Komponentenschreibweise, als
auch die direkte oder absolute Schreibweise von Tensoren angewendet und dadurch ein
eingleisiges Vorgehen vermieden.
Schlagworte: Dierentialgeometrie; Kontinuumsmechanik; groe Verformungen; Arbitrary Lagrangian-Eulerian; Mannigfaltigkeit; Tensoranalysis
God is a geometer.
Plato
Contents
Preface by the Editor
iii
Abstract
Zusammenfassung
vii
Contents
xi
List of Figures
xiii
Notation
xv
1 Introduction
5
5
8
3 Dierential Geometry
3.1 Topology and Manifolds . . . . . . . . . . . .
3.2 The Tangent Space . . . . . . . . . . . . . . .
3.3 Tensor Algebra . . . . . . . . . . . . . . . . .
3.4 Bundles and Tensor Fields . . . . . . . . . . .
3.4.1 Sections of Fibre Bundles . . . . . . .
3.4.2 Action of Maps . . . . . . . . . . . . .
3.5 Exterior Algebra of Dierential Forms . . . .
3.6 Dierentiation on Manifolds . . . . . . . . . .
3.6.1 Covariant Derivative . . . . . . . . . .
3.6.2 Lie Derivative . . . . . . . . . . . . . .
3.6.3 Exterior Derivative . . . . . . . . . . .
3.7 Integration on Manifolds . . . . . . . . . . . .
3.7.1 Orientation and Determinants . . . . .
3.7.2 Stokes Theorem and Volume Measures
4 Application: Continuum Mechanics
xi
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
15
15
19
24
30
30
35
40
44
45
49
53
54
54
59
65
CONTENTS
xii
4.1
4.2
4.3
4.4
4.5
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
65
69
74
76
79
79
82
84
85
92
95
Bibliography
97
List of Figures
2.1 Position vectors of a point P with respect to dierent frames (O, gi ) and
(O , gi ). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
10
12
18
19
31
37
52
60
66
68
82
87
88
xiv
LIST OF FIGURES
Notation
General Conventions
Calligraphic (M, N , . . .)
Italic lightface (a, b, A, B, . . .)
,
f g = f (g)
U M
dim(Mndim ) ndim
A(v) Av
v = PQ
d(P, Q)
{g 1 , . . . , g ndim } {g i }
(O, gi )
xvi
,
SE(M)
SO(V)
(S, T )
int(S), cl(S), S
(U, x), x(U) Rndim
{xi }P = x(P )
A(M) = {(Ui , xi )}iI
c:RM
f :MR
TP M = {P } Vn
TP M
T M = P M TP M
T : TA TB
, dxi
xi
T S, T : S
a (v) a v
w[f ] = df w
T , T T
,
: E M, (E, , M)
: M TM
(T M)
(Tqp (M)) = Tpq (M)
( k T M) = k (M)
Alt : Tk0 (M) k T M
iu : k (M) k1 (M)
v
LuT = T
+ uT
t
d
1 ,...,n
dv n (M)
J
I, J, K, . . .
i, j, k, . . .
, , , . . .
NOTATION
pullback, pushforward by a map
special or proper Euclidian group of a point space M
special or proper orthogonal group of a vector space V
topological space; set S with topology T
interior, closure, and boundary of S
coordinate system on U; chart of U
coordinates of the point P U in a chart (U, x)
atlas of the manifold M = iI Ui
curve on M
real function or scalar eld on M
n-dimensional tangent space at P M
cotangent space; dual space
tangent bundle
tangent map; dierential of a map : A B
basis vectors, dual basis vectors in a tangent space
single contraction, double contraction of tensors
linear form applied to a vector v
directional derivative of a function f along w
adjoint, transpose of a tensor
index lowering operator, index raising operator
bre bundle over M
tangent bundle section; vector eld on the manifold M
set of all vector elds (bundle sections) on M
set of all pq -tensor elds on M
set of all k-forms on M
alternation mapping
interior product
covariant derivative along v
Lie derivative of a tensor eld T along u
exterior derivative of a k-form
Levi-Civita symbol; permutation symbol
Riemannian volume form on M; g-volume
Jacobian of a map
coordinate indices (Lagrangian formulation)
coordinate indices (Eulerian formulation)
coordinate indices (ALE formulation)
Chapter 1
Introduction
It is unquestioned for a long time that natural sciences can benet from dierential
geometry, since it makes a comprehensive theory of gravitation possible in general
relativity. The so-called geometric mechanics [1, 2, 11, 12, 13, 14, 15, 16], however, has
been implemented more recently and thus it is applied to a lesser extend, especially
in engineering sciences. Important examples are found in the theories of rods and
shells, in the Lagrange-Hamilton formalism, and relativistic elasticity. In the theory of
materials, the Lie derivative serves to obtain objective rates of stress measures, but it
is often used detached from the overall geometric context.
Dierential geometry [1, 3, 4, 5, 6, 7] provides the suitable background to present
and discuss the subjects of mechanics, and especially continuum mechanics, with an
integrative and mathematically precise terminology. It claries basic concepts and
opens up deep examination even of complex issues. For example, geometry reveals
that the determinant of the Jacobian matrix is not an invariant scalar, and that the
question, whether the rst Piola-Kirchho stress tensor is symmetric or not, does not
make sense. However, dierential geometry demands a large investment of eort and
persistency from its students.
Continuum mechanics [2, 8] is typically prepared for the Euclidian point space. This is
motivated by applications in engineering sciences, which prefer as simple spaces as possible. In some cases, e.g. for shells, it is reasonable to use local curvilinear coordinates
instead of the global cartesian coordinates. The former are not ane, that is, they do
not transform linearly as cartesian coordinates, and additional terminology metric,
covariant and contravariant basis vectors etc. enters. If the shell is understood as
embedded in the Euclidian space, the terms mentioned can refer to the ambient structure, namely if the position vectors of points are identied with those points. As a
result, the covariant vector basis, for example, arises from the partial derivatives of
the position vector with respect to the curvilinear coordinates of the related point on
the shell. The existence of a global position vector, however, already requires a global
linear structure of space. Without embedding the shell in the Euclidian space, a terminology built on position vectors, and hence the description of the shells geometry
and kinematics will collapse.
CHAPTER 1. INTRODUCTION
3
forming, free surface ow and impact processes. The Lagrangian and Eulerian formulations are generalized within the ALE formulation, as a time-dependent reference
domain uncoupled from the material body and its congurations in the ambient space
is used to describe the physical quantities under consideration. Due to the generalization, however, the governing equations become more complicated, especially when
compared to the Lagrangian formulation in solid mechanics.
In the references cited, the ALE formulation for Euclidian spaces is discussed in detail,
but to the knowledge of the author, no reference is available in which the framework
is extended to manifolds or Riemannian spaces, or which respond closer to the geometry of the ALE formulation. Therefore, another objective of the paper is to extend
the geometric methods of continuum mechanics by the Arbitrary Lagrangian-Eulerian
formulation on manifolds. It will be restricted to the geometry and kinematics of a
body and to the conservation of mass. Other balance equations as well as dynamical
and material theoretical aspects will not be considered.
The structure of the paper is as follows. Chapter 2 reviews some basic results of linear
geometry. In chapter 3, an introduction into the terminology of modern dierential
geometry will be given: topology, tensor algebra, bundles and tensor elds, exterior
algebra, dierential and integral calculi. After applying the terminology to continuum mechanics in the classic Lagrangian and Eulerian formulations in chapter 4, the
ALE formulation will be geometrically introduced and implemented into the overall
continuum mechanical context. The paper closes with some concluding remarks in
chapter 5.
CHAPTER 1. INTRODUCTION
Chapter 2
Review of Linear Geometry
The following chapter should motivate the construction of vector spaces and mappings
on general manifolds. It is devoted to the important topics of vector algebra and ane
point spaces used in continuum mechanics and the theory of materials, including the
concept of objectivity.
2.1
V
a+b=b+a
V
a = a , R ,
with existent unique neutral and inverse elements for both addition and scalar multiplication, is called vector space or linear space over the body R. The elements of V are
called vectors.
Denition 2.1.2. Let V be a vector space. If there exists a positive bilinear mapping
respectively an inner product
, : V V
(a, b)
a, b =
b, a ,
ai = 0, i = 1, 2, . . . , n
holds, where 0 is the zero vector, then the vectors g 1 , g 2 , . . . , g n are called linearly
independent, otherwise linearly dependent. If n vectors gi are linearly independent, but
(n + 1) vectors g i are always linearly dependent, then n ndim is the dimension of
Vndim and {g 1 , g2 , . . . , gndim } {gi } Vndim is called basis of Vndim . The latter can
then be expressed by
Vndim = U1 U2 . . . Undim ,
where denotes the direct sum (see denition below) of the subspaces Ui Vndim , and
g i Ui for all i {1, . . . , ndim }.
v = v g1 + v g2 + . . . + v gn =
n
vigi vigi .
i=1
By 2.1.3 and 2.1.5, the basis vectors can be arbitrarily chosen. They are, in general,
neither orthogonal nor normalized. Moreover, no origin has been used to dene the v i
or g i , that is, points do not exist in vector spaces.
Example 2.1.7. Every row resp. n-tuple {v 1 , v 2 , . . . , v n } Rn , as known from linear
algebra, is a vector, and if {e1 , e2 , . . . , en } is the canonical basis in Rn , then {v 1 , v 2 ,
. . . , v n } = v i ei is the local representative.
Denition 2.1.8. Let V, W be vector spaces. A map A : V W, with
1. A(v 1 + v 2 ) = Av 1 + Av 2 W, v 1 , v 2 V
and
2. A(v) = (Av) W, R ,
where Av A(v), is called linear transformation, linear map or homomorphism. If
A : V W is bijective such that A1 is its inverse and both V and W have the
same dimension, then A is called isomorphism. Linear transformations A : V V are
called endomorphisms, and isomorphisms with V = W are called automorphisms.
g i = B ii g i .
In the previous equation, the sum is over primed indices only. The matrix of I, written
(B ii ), is called the inverse matrix of the change of basis, that is, (B ii )1 = (B ii )
arranges the components of the direct change of basis, g i = B ii g i . Thus, g j =
B ij B ii g i = B ii B ij g i = ij g i and Ig j = ij g i , where ij is called the Kronecker
delta.
As studied later in the text, the passive transformation rule identies the vector with
a rank-one tensor. However, both interpretations are of fundamental importance in
physics, especially in continuum mechanics.
Aa, AbW =
a, bV
for all a, b V. An isometry A : V V is called an orthogonal map, having the
properties det A = 1 and A1 = AT , where AT is the transpose or adjoint of A. A
more general denition of an orthogonal map includes the isometries A : V W and
then stipulates AT A = I V and AAT = I W .
2.2
Denition 2.2.1. Let S = {A, B, C, . . .} be a set of points and V a vector space. The
pair (S, V) or simply S if the meaning is clear from the context together with the
map
SS
(A, B)
AB
v = AB.
2. If AB = CD, then AC = BD also holds (parallelogram axiom).
If V moreover has a metric according to 2.1.2, then (S, V) is called a Euclidian point
space.
AB + BC, BA = AB and AA = 0 denes the vector sum, the inverse element and
the neutral element, respectively.
Corollary 2.2.4. If v(A) = AB is a vector with base point A, and AB = CD, then
v(A) = v(C) .
Conclude that the parallelogram axiom of denition 2.2.1 renders ane point spaces
at this is the fundamental dierence between at spaces and manifolds.
space, and OP = x Vndim is called position vector of P with respect to O. The local
representative x = xi g i includes the ane coordinates {x1 , x2 , . . . , xndim } {xi } of P
with respect to the frame (O, gi ).
Note that in Euclidian point spaces, frame is used as a synonym for Euclidian
observer.
Denition 2.2.7. Given a frame (O, gi ) in the space S, one may construct coordinate
lines by varying one ane coordinate and keeping the other coordinate values xed,
i.e. by changing one coordinate (component) xi of the position vector. The family of
lines obtained is called ane coordinate system on S and is denoted by the pair (S, x).
If every coordinate line is orthogonal to each of the other coordinate lines, the ane
coordinate system is called cartesian.
Although ane coordinates are uniquely determined by the chosen frame, one should
be careful with the dierence between the frame and the coordinate system. Without
a frame it does not make sense to talk about coordinate systems!
One should also be careful with the terminology position vector, because it depends
Expanded in the primed basis, x becomes OP {gi } = Ix = B ii xi g i , where 2.1.12 has
been applied. Setting c = ci g i = O O, then in the primed frame the coordinates of
= x + c are
O P = O O + OP resp. x
xi = B ii xi + ci = i (x1 , . . . , xndim )
as desired.
10
= O P
x
P
x = OP
gi
Ix = x
O
O
g i
Figure 2.1: Position vectors of a point P with respect to dierent frames (O, gi ) and
(O , g i ).
Proposition 2.2.9. A vector of an ane point space (= position vector!) is coordinate-invariant (or frame-indierent) and it ts 2.1.13, whereas a position vector does
not.
Proof. Additional to the previously described situation, let Q S be another point
and let OQ = y ig i and O Q = y i g i denote its position vectors in the two frames.
Then by 2.2.1 and 2.2.2, v = Q P = OQ OP = O Q O P . Apply 2.2.8 to get
v = (y i xi )gi = (B ii y i + ci ) (B ii xi + ci ) g i
= (y i xi )B ii g i = (y i xi )g i
xi g i = OP is also a vector, but B ii xi = xi if O = O by 2.2.8, i.e. the position of P
with respect to (O, gi ) and (O , gi ) does not transform in terms of 2.1.13.
Corollary 2.2.10. Let OO = ai g i , O O = ci g i and g i = B ii g i be the translation
vectors and the change of basis of two frames (O, gi ) and (O , g i ), respectively. Since
OO = O O, one has
ai = ci B ii
and
ci = ai B ii .
i
= B ii .
j
x
Denition 2.2.12. Let (S, V) and (T , W) be two ane point spaces, O, P, Q S, and
A : V W a linear map. A map : S T is referred to as an ane transformation,
if for every P = O + OP ,
(P ) = (O) + A(OP )
T .
11
v = (Q) (P ) = (O) + Q(OQ) (O) + Q(OP ) = Q(OQ) Q(OP ). Since
Q is linear, Q(OQ OP ) = Q(Q P ) and the assertion follows.
components of the position vector O(P ) = x with respect to the g i are then given by
xi = Qii xi + ci ,
where ci are the components of the translation vector O(O) = c, and xi are the
12
Q(OP )
(P )
P
Qx
(B)
Qg i
(O)
O
B
gi
OP
Figure 2.2: Active objectivity of the vector x = OP in the Euclidian point space S
with respect to a superposed rigid motion of a subset B S (left), and with respect
to a change of framing (right).
but not proper orthogonal as for superposed rigid motions. By keeping this dierence
in mind, proposition 2.2.15 is equivalent to the following:
(O, gi ) =
Proposition 2.2.16. Let Q be orthogonal resp. an isometry, (O, gi )
(O , Qg i ) a change of framing in the Euclidian point space S and {gi } a basis dened
through Qg i = Qii g i . The components of the position vector OP = x with respect to
the gi are then given by
xi = Qii xi + ci ,
where ci are the components of the translation vector OO = c, and xi are the compo
nents of OP = x with respect to the basis {g i }.
Proof. Isometric Qs imply that P has the same coordinates with respect to the
frame (O , Qg i ) as P has with respect to (O, gi ) except for reordering of indices if
det Q = 1. Therefore,
O P = Q(OP ) = Qx = xi Qg i = Qii xi g i ,
and OP = OO + O P resp. x = Qx + c as before. Substitution of x = xi g i and
c = ci gi then gives the result.
The connection between the formula 2.2.15 resp. 2.2.16 and the formula 2.2.8 is similar
to that of a map A : V V and the identity map I in a vector space. Hence, it seems
likely to rewrite denition 2.1.14 for Euclidian point spaces.
13
(O) + Q(OP ) and keeping the Euclidian frame xed. In this case, v = Qv as in
2.2.14, where Q SO(V), and xi = Qii xi + ci are the ane coordinates of (P ) in
terms of the coordinates of P and (O) with respect to O. (Qii ) is the matrix of Q with
respect to a suitable basis, and ci are the components of O O. Another way to dene
active objectivity is to interpret as a change of Euclidian framing (O, g i ) (O , Qg i )
(cf. proposition 2.2.16 and gure 3.5).
On the other hand, if the passive interpretation of objectivity is applied, : (O, g i )
(O , g i ) is a change of framing on the xed Euclidian point space and which keeps
the vector xed, that is, v = Iv as in 2.2.9. In this case, xi = B ii xi + ci are the
ane coordinates of the same (xed) point in dierent frames, where (B ii ) is the inverse
matrix of the change of basis. Therefore, passive objectivity phrases the transformation
properties of v under a change of ane coordinates.
14
Chapter 3
Dierential Geometry
3.1
Manifolds are more general ane point spaces. Engineers nowadays accept the term
manifold as the mathematical expression for a continuum that has a dierentiable
structure, but the rich theory behind is often ignored. However, dierentiability as it
stands means that there is some continuity. Topology is the basic eld for the study of
continuity and for the creation of a general terminology of space a manifold is also
a topological space. Descriptively, a topology carries the relations or interconnections
between elements of a point set.
To non-mathematicians, managing books on topology and tensor analysis on manifolds
may become a dicult challenge. The following chapter should assemble the topics
needed by applying a notation consistent with the rest of the text.
Denition 3.1.1. Let S be a set. A topology T is a collection of subsets A, B S,
called open sets, which satisfy the following axioms:
1. T and S T .
iI
Ai T also holds.
The pair (S, T ) is referred to as the topological space, but write S instead of (S, T )
if the meaning is clear. A subset A S is called closed, if S\A T is open,
i.e. closed sets are the complements of open sets. From the denition, the trivial
topology T = {, S} of S is obvious. The elements of a topological space are called
points; these can be geometric points, material particles, thermodynamic states etc.
Corollary 3.1.2. Since S\S = and S\ = S, the sets S and are both open and
closed.
Denition 3.1.3. The interior or open cover int(M) of a topological space
M S is
the union of all open sets U which completely lie in M, i.e. int(M) = U {U | U T
and U M} . The closure cl(M) is the smallest closed set which completely includes
16
M, i.e. cl(M) = S U {U | U T and U (S\M)} . The dierence cl(M)
int(M) = M is called boundary of M.
Denition 3.1.5. Let S be a topological space with topology T . An (open) neighborhood of a point P S is an open set U T such that P U. A point P is called
isolated, if {P } is open. A basis for the topology of S is a sequence or collection B of
open sets such that every open set of S is a union of elements of B.
Intuitively, second countable means that there is at least one way of covering the space
with a nite number of sets. Note that every second countable space is also rst
countable, but not conversely.
Denition 3.1.9. A topological space S is referred to as a Hausdor space, if every
two points P, Q S, P = Q, can be separated by neighborhoods U(P ) S and
V(Q) S such that
U V = .
X Rn
x(P ) = {x1 , x2 , . . . , xn }P {xi }P , x1 ,
17
x x1
x(U U ) : x (U U ) x (U U )
is called chart transition or change of coordinates. is the composition operator. A
chart transition x x1 |x(U U ) is also called the relabelling of the subset U U .
18
x X 1
X(1 (V)U ) : X 1 (V) U x 1 (V) U .
The chart transition is also called the local representative or the localization of . If
xi denote the coordinate functions of (V, x), one abbreviates i = xi X 1 , so
xi = i (X). The map is called dierentiable at P 1 (V) U, if x X 1 is
dierentiable at P .
N
1 (V)
(P )
(U)
X
X(U)
x(V)
X(P )
x((P ))
xX
19
U
X
X Rndim
V
Y Rndim
3.2
In ane point spaces, a vector is identied with parallel translation (see section 2.2): a
specic vector can be attached to every point of the ane point space or, equivalently,
20
ane point spaces are at and any point could be an origin of that vector. For a
manifold, it is not possible to dene a parallel translation, because at this stage reached,
there is not even a connecting path between the points. Consequently, there are no
vectors in the traditional sense.
In order to geometrize manifolds further, this section gives a blue print of the tangent
space, that is, a vector space attached to each point of the manifold, by beginning with
curvilinear coordinates in ane point spaces.
Denition 3.2.1. Let S be an m-dimensional ane point space and y 1 , . . . , y m the
coordinate functions of an ane coordinate system (S, y). Let (A, x), with coordinate
functions x1 , . . . , xn , n m, be a coordinate system on an n-dimensional manifold
A. A curvilinear coordinate system on A embedded in S such that A S is a
submanifold is determined by m nonlinear functions f i involved in the map
Rn
(x1 , . . . , xn )
Rm
y i = f i (x1 , . . . , xn ) ,
i = 1, . . . , m ,
so that y i = f i on A in S.
Therefore, the simple rule for the change of ane coordinates 2.2.8, y i = i (y 1, . . . ,
y m) = B ii y i + ci , where the B ii and ci are constants and do not depend on the point,
changes to y i = (i f i)(x1 , . . . , xn ) if curvilinear coordinates are involved.
With curvilinear coordinates in Euclidian spaces, a vector basis of the tangent space
at point P A S can be constructed by taking the partial derivatives of the
Proposition 3.2.2. Under a transformation xi = i (x1 , . . . , xn ) of (curvilinear) coordinates, which is assumed to be invertible, the tangent basis vectors transform according
to
x
x i
i
g i = i = i
=
g,
x
x xi
xi i
that is, the inverse matrix of the change of basis coincides with the Jacobian matrix of
the change of coordinates.
i
i
vi =
xi i
v .
xi
21
This transformation rule identies the vector components with components of a 1-fold
contravariant tensor:
Denition 3.2.3. A set of coordinate-dependent functions ti which transform under
chart transitions (or changes of coordinates) xi xi according to the rule
xi i
t
t =
xi
i
xi
ti
xi
dsi
xi dsi
=
.
dt
xi dt
So, by 3.2.3, 1-fold contravariant tensors on dierentiable manifolds do exist.
Next, consider a real C 1 function resp. a C 1 scalar eld f : U R, where U M. In
a chart (U, x), f induces the eld f (x) = f x1 Rn as a function of the coordinates
xi . A change of coordinates xi xi transforms the derivations of f with respect to
the coordinates according to the scheme
f
xi f
=
.
xi
xi xi
Conclude that on dierentiable manifolds there are also 1-fold covariant tensors.
Denition 3.2.5. Let f : M R be C 1 , and let f (x) = f x1 be its picture in a
chart. One denes the directional derivative of f along a curve s(t) M at point s(0)
through
df
f si ds
f dsi
f
=
= i w i = Dwf = w[f ] .
i
i
dt
x s dt t=0
x dt t=0 x
t=0
22
i
The w =
dsi
dt
t=0
= w[xi ] are tangential to the curve s(t) through point s(0), and they
are called components of the tangent vector of the curve at s(0). The w i are well-dened
up to re-parametrizations
dt
dsi
w i =
= wi .
dt
dt
Proposition 3.2.6. The directional derivative of a real function is coordinate-invariant.
Proof. By using 3.2.4,
i
i
f i
x i
xi xi f i
f
x f
=
w =
w = i wi .
w
i
i
i
i
i
i
i
x
x x
x
x x x
x
is the local version of an ane frame of reference on M that has been introP, x
i
duced in 2.2.6.
w(P ) = w i (P )
(P ) TP M .
xi
Whenever one talks about a vector w in the manifold, the picture of the w i in the
tangent space is meant. However, to get a local representative of w, the rst step is to
choose a chart on M.
TP M
P M
23
,
xi xj
R
,
xi xj
= gij (P )
P
at every point P M. The gij = gji are called metric coecients. If M is also
torsion-free (k ji = i jk , see sec. 3.6.1), then M is called a Riemannian manifold.
Denition 3.2.12. A linear form is a map
a : V
v
R
a (v) a v .
As there are no traditional vectors on manifolds, but only tangent vectors, the definition of a linear form has to be revised. On manifolds, the linear forms are called
dierential 1-forms, or shortly 1-forms.
Denition 3.2.13. Let M be an n-dimensional dierentiable manifold. The cotangent space TP M is a vector space at point
P M that can be formally written
as TP M = {P } Vn . The union T M = P M TP M is referred to as the cotangent
bundle of M.
i
i
Proof. It has to be shown that the duality relation dxi x
j = j holds, where j
ndim
is the Kronecker delta on M and i, j {1, 2, . . . , ndim }. However, in R
contraction
reduces to ordinary multiplication such that for every P U,
xi
dx (P ) j (P ) = j (P ) = ij
x
x
i
dxi (P )
(P ) = ij .
j
x
24
i
j
= ai v j ij = ai v i ,
= ai v j dxi
a v = ai dx v
j
x
xj
where ai dxi is the local representative of a .
(ii) With 3.2.15, the derivative 3.2.5 of a C 1 real function f in the direction w TP M
can be written as
f
f
f
= df w .
w[f ] = i w i = i w j j i = i w j dxi
x
x
x
xj
Note that neither the terminology co- and contravariant basis vectors nor a metric
for raising or lowering the indices is involved to perform a v, as it is typically done
when using curvilinear coordinates in Euclidian spaces. This is an example of how
dierential geometry can oer advantages and clarication of the basic theory.
3.3
Tensor Algebra
Denition 3.3.1. A pq -tensor T (P ) at point P of a manifold M is a multilinear
mapping
T : TP M . . . TP M TP M . . . TP M R ,
qfold
pfold
qfold
TP M. . .TP M TP M. . .TP M
pfold
rfold
sfold
R,
where Q = (P ) and T is a function of P . Every pq sr -tensor is an element of a
(p q r s)-dimensional vector space. One refers to (p+q+r+s) as the rank of the
tensor.
25
Example 3.3.3. A dierential 1-form a is a 01 -tensor, also called a covariant rank
one tensor, and a vector v is a 10 -tensor by setting a (v) = v(a ).
Denition 3.3.4. The components of a tensor, that has been dened through 3.3.1,
are obtained by delivering the basis vectors of the cotangent space and of the tangent
space as arguments of the tensor:
1 ,2 ,...,p
1
2
p
dx , dx , . . . , dx ,
,
,...,
.
T
1 ,2 ,...,q = T
x1 x2
xq
Components of two-point tensors are unwrapped analogously.
x1 x2
xp x1 x2
xq 1 ,2 ,...,p
= 1 2 . . . p
.
.
.
T
1 ,2 ,...,q .
x x
x x1 x2
xq
For the sake of completeness, some generalization of the tensorial transformation criterion is given here.
Denition
3.3.6. (See also [5], ch. vii; [6], ch. 21; [25], ch. 5; and [4]) Let T be a kind
of pq -tensor whose components transform according to the rule
x
x 1
xp x1
xq 1 ,...,p
1 ,...,p
=
f
det
.
.
.
.
.
.
T
T
1 ,...,q ,
1 ,...,q
x
x1
xp x1
xq
x
where f [det( x
)] is a function of the determinant of the inverse Jacobian matrix.
x
x w
Then T is called an even relative tensor of weight w, if f [det( x
)] = det( x ) , and
x
x w
it is called an odd relative tensor of weight w, if f [det( x
)] = | det( x )| . Moreover,
x
x
T is a pseudotensor, if f [det( x
)] = sign[det( x )].
The even (odd) relative tensors of weight 1 are also called even (odd) tensor densities,
and the even (odd) relative scalars of weight 1 are also called even (odd) scalar densities.
x
An absolute tensor (or ordinary tensor) is obtained by setting f [det( x
)] = 1. Except
for special topics of integration theory, this paper solely deals with ordinary (two-point)
tensors.
Denition
3.3.7. Contraction
reduces the rank of tensors. For example,
contracting
1
1
1
i
a 1 -tensor T and a 0 -tensor resp. vector v = v xi yields the 0 -tensor
i
T v = T (v) = v T
, in components
T ij v j .
xi
26
The contraction of a
1
-tensor T and a 01 -tensor (1-form) a is
1
a T = a (T )
, in components
T ij ai .
The contraction two tensors T and S in the i-th covariant slot of T and the j-th contravariant slot of S are dened in a similar way, as if the covariant slot is a 1-form and
the contravariant slot is a vector, respectively. The contraction of two tensors T and S
in the i-th contravariant slot of T and the j-th covariant slot of S is straightforward.
Let T abcd and Sijkl be the components of two tensors T and S. Then, without a
specication of slots,
T S
with
T abcd Sijcd = T ab ij
with
the double contraction of T and S, respectively. A single tensor is contracted analogously, provided that the tensor has as well covariant as contravariant slots.
1
Denition 3.3.8. The contraction of a 1 -tensor is called the trace of the tensor:
tr T = T ii .
Denition 3.3.9. Let the tensors T and S have the same rank and be compatible in
terms of contraction, then
T , S denotes the contraction on all index slots resp. the
inner product. Taking the examples of T and S from denition 3.3.7, then
T , S = T ijkl Sijkl .
By the chosen denition of tensors, T is an operator acting on the vector slots and
1-form slots of some other tensor S, that is, T (S) = S(T ) in general!
Denition 3.3.10. Let T be a pq -tensor and S a rs -tensor. The tensor product
-tensor
T S yields an p+r
q+s
(T S)(P ) : TP M . . . TP M TP M . . . TP M
pfold
TP M
qfold
. . . TP M TP M
rfold
. . . TP M R
sfold
such that
(T S) a1 , . . . , ap , v1 , . . . , v q , b1 , . . . , br , w1 , . . . , ws
= T a1 , . . . , ap , v1 , . . . , v q S (b1 , . . . , br , w 1 , . . . , ws ) ,
27
1 ,...,p
(T S)
1 ,...,r
1 ,...,q
1 ,...,s
=T
1 ,...,p
1 ,...,q S
1 ,...,r
1 ,...,s
That is, the tensor product generates ordered tuples of basis vectors, and so for a pq
tensor T one may abbreviate T Tqp (M), where Tqp (M) = p T M q T M is the
p
-tensor bundle, and k denotes the k-th tensor power.
q
p
Corollary 3.3.11. By using the denitions 3.3.4, 3.3.7 and 3.3.10, a q -tensor has
the local representative
T (P ) = T 1 ,2 ,...,p1 ,2 ,...,q
.
.
.
x 1 x 2
xp
Denition 3.3.12. The denition 3.3.6 can be generalized further by applying the
tensor product. An absolute pseudotensor is obtained by tensor-multiplying an even
and an odd relative tensor of opposite weights, and a relative pseudotensor of weight
w = u + v is the results of the tensor product of an even relative tensor of weight u
(resp. v) and an odd relative tensor of weight v (resp. u).
The reader should carefully distinguish the tensors on manifolds from those in ordinary
vector spaces, as the latter do not carry point information. The following proposition
helps to clarify this aspect.
Proposition 3.3.13. Let V, W be vector spaces and {GI } V, {g i } W appropriate
bases, then every linear map A : V W corresponds with a 11 -tensor.
Proof. Note that from the denitions 2.1.10 and 3.3.2, AV = V I (AGI ) = AiI V I g i
W, where V = V I GI is a vector. On the other hand
AiI V I g i = AiJ V I JI g i = AiJ V I g i GJ GI = AiJ g i GJ V I GI ,
so A = AiI g i GI is the tensor of the linear map.
An alternative proof of the previous statement can be obtained by showing that the
AiI of the linear map transform as the components of a tensor.
To simplify notation, boldface italics are used for both, linear transformations and
tensors. The dierence will become clear when the object is applied to a vector v of
the tangent space: Av is the linear map A applied to the vector v, and A v is the
contraction of the tensor A and the vector v.
Denition 3.3.14. Let M be a manifold and P M. From the metric coecients
in 3.2.11, the metric tensor, or shortly metric g(P ) = gij dxi dxj can be dened, so
that the pair (M, g) is a metric space (just write M instead of (M, g) if there is no
1
k
l
ij
g g = gkl dx dx g
xi xj
= gkl g ij il dxk
= dxi
= IM
j
x
xi
28
is the second-rank unit tensor on M.
ik
ik
i
i
g gkj = g
,
=
dx
j
xk xj
xj
= g ik gkj dxj ,
xk
tensor indices can be raised by the inverse metric coecients, and lowered by the metric
coecients.
1
j
-tensor, then the associated tensors
Denition 3.3.16. Let T = T ij x
i dx be a
1
of T are
dxi = g ik
T = T ij
j = T ik g kj i j ,
i
x x
x x
where is the index lowering operator and is the index raising operator, respectively.
0
(M) is the tensor with all indices lowered, and S
If S Tqp (M), then S Tp+q
T0p+q (M) is the tensor with all indices raised.
Denition 3.3.17. Let P M and v TP M a vector, then dene the linear transformation g (P ) : TP M TP M by g (P )v =
v, P , and its inverse g (P ) : TP M
TP M, such that for T T11 (M), T = g T and T = T g , respectively.
and
I = ik g kj
= g 1 = g .
i
j
x
x
Example 3.3.19. Raising the indices of the dierential df of a scalar eld gives the
gradient f , which is a vector:
(df ) = f = g ij
f
.
xi xj
v, T (U )Q = T T(v), U P .
29
Proposition 3.3.21. If gij (Q) are the metric coecients on N and GIJ (P ) are the
coecients of the inverse metric on M, then (i) the components of T T are
(T T )I i (Q) = gij T jJ 1 GIJ 1 ,
and (ii) T T = G T g .
Proof. To proof (i), let U(P ) M, V(Q) N be neighborhoods with appropriate
charts(U, X), (V, x), respectively, in which X(U)
= {X I }U and x(V) = {xi }V . Then
TQ N is a basis at Q, so that
dX I TP M is a dual basis at P and x
i
i
I
i
T (P ) = T I (P ) xi dX is a local representative of T . Set U = U I X
I and v = v xi ,
then,
T (U ) = T iI U J i dX I
= T iI U I i
J
x
X
x
and
T (U ), vQ = v
T iI U I
,
i
x xj
= v j T iI U I gij .
T I i
i
On the other hand, T T(v) = (T T )I i v j X
I dx xj = (T ) i v X I , so
T
T I i
J
= U J (T T )I i v i GIJ .
,
U , T (v) P = U (T ) i v
I
J
X X
and
T 1 T (U ) = U
involve the inverse tensor T 1 . It is easy to verify that T 1 has the local representative
T 1 (P ) = (T 1 )I i (P )
dxi ,
X I
where (T 1 )I i are the components of the inverse of the matrix (T iI ). Dene the inverse
transposed tensor T T (Q) analogously from
T T (Q) T T (Q) = I N
I
yielding T T (Q) = (T T )j J (Q) x
i dX .
resp.
T T T T (v) = v ,
30
The local representatives of the transpose, the inverse and the inverse transpose of
(one-point) tensors can be obtained easily by choosing P = Q and N = M. It is
notable that T T calls for metrics on M and N , whereas T 1 does not.
Denition 3.3.23. A two-point tensor T : TP M T(P ) N is called orthogonal
provided that
and
T T T = IN .
T T T = IM
T is called proper orthogonal, if it is orthogonal and det T = +1.
Denition 3.3.24. Let N be a metric space. A 11 -tensor S : T N T N is called
symmetric, if S = S T .
Note that with two-point tensors it does not make sense to talk about symmetry!
j
be symmetric, Q N and gij the
Corollary 3.3.25. Let S(Q) = S ij (Q) x
i dx
metric coecients on N . Then by 3.3.21,
S ij
j
k li
j
i
j
dx
=
g
S
g
dx
=
S
dx
.
jk
l
j
xi
xi
xi
1
(tr S)2 tr(S 2 ) ,
2
3.4
3.4.1
The previous section was about tensors at single points of the manifold. To give an
precise denition of elds of vectors and tensors, the theory of bre bundles is briey
introduced. For further studies see, for example, [26] and [1].
/U
p
p
ppp
p
p
p pr1
px pp
1 (U) E
31
F
U M
1
Denition 3.4.3.
A collection (yi : (Ui ) Ui F )iI of bundle charts of (E, ,
M), such that iI Ui = M for I N, is called bundle atlas of (E, , M).
Denition 3.4.4. Let (E, , M) be a bre bundle. If all 1 (U), U M, are locally
trivializable and 1 (M) M F is a homeomorphism, then the bre bundle is
called globally trivializable. If the total space is the product topology E = M F ,
and (M, y = Id) is a global bundle chart such that = pr1 : M F M and
1 (P ) = {P } F , then (E, , M) is a trivial bundle.
Denition 3.4.5. If the bre space is an n-dimensional vector space Vn , a bre bundle
(E, , M) is called a vector bundle, and for E = M Vn it is called a trivial vector
bundle.
Denition 3.4.6. If the bre of a bre bundle over M is spanned by the tangent
vectors at each P M, i.e. 1 (P ) = TP M, then the total space is denoted by T M,
and = M : T M M respectively (T M, M , M) is referred to as the tangent
bundle over M (cf. 3.2.10). A local trivialization is then assumed to be dieomorphic,
not only homeomorphic.
32
Example 3.4.7. The simplest example of a tangent bundle is found in the ane point
space (Rn , Rn ). The set of all vectors v = {v 1 , . . . , v n } at all points P Rn , i.e. set of
all pairs (P, v), is just the cartesian product Rn Rn = T Rn . So T Rn is trivial, and
{P } Rn is a tangent space at P .
Example 3.4.8. A tube is a trivial bundle, in which the base space is the circular
cross section S1 , and S1 R1 is the total space. The bre 1 (P ), P S1 , is a line on
the tube parallel to the axis. The Mobius strip is a non-trivial bundle over S1 . Locally,
tube and Mobius strip are identical.
Proposition 3.4.9. Let M be an n-dimensional dierentiable manifold, then (i) T M
is a 2n-dimensional manifold, and (ii) M : T M M is a vector bundle.
Proof. (i) Let U M be a subset and (U, x) a chart, where x : U x(U) Rn .
Then, by recalling the denitions 3.2.8 and 3.2.10, a natural chart of the tangent bundle
would be (T U, T x) including the map
T x : T M|U
(P, v)
x(U) Rn
T x(P, v) = { 1(v(P )), 2(v(P )), . . .
. . . , n (v(P )), v 1(v), v 2 (v), . . . , v n (v)} .
Therein, i(v(P )) = xi (P ) are the local coordinates of P M, v i (v) = v[xi ] are the
components of v(P ) TP M, and T M|U is the restriction of T M to U.
The natural atlas of TM, then, is the collection T A = {(T Ui , T xi )}iI of natural
charts, where T M = iI T Ui and I N. Hence, T M is homeomorphic to R2n , so
dim(T M) = 2n. As M is Hausdoran and second countable by denition, and T x
is a dieomorphism by 3.4.6, T M is also Hausdoran and second countable, i.e. a
manifold.
(ii) By the denition of a local chart, x(U) is homeomorphic to U, and T M|U =
1
P U TP M = M (U), so
yT M :
1
M
(U)
v(P ) = (P, v)
U Rn
{P, v 1 , v 2, . . . , v n } ,
1
v i = v[xi ] being understood, is a local trivialization of T M and M
(P ) = TP M is
n
a bre. By 3.2.7 and the example 2.1.7, R is also a vector space, so that TP M
{P } Rn is an isomorphism. Therefore, T M has a vector bundle structure induced
by the dierentiable structure of M.
33
Vector elds are sections of vector bundles. Moreover, the tensor product of vector
spaces can be transferred to vector bundles to generate tensor bundles, whose sections
are then called tensor elds.
Denition 3.4.13. Let : E M be a bre bundle, then a map
:ME,
with ((P )) = P, P M, is called bundle section. If the bre bundle is a vector
bundle, then is called vector eld on M. The set of all sections of E is denoted by
(E).
In the physical and mechanical literature, instead of the correct (E) respectively
: M E it is common to write (P ) for the eld, indicating that depends on the
points P M.
Denition 3.4.14. Let : E M be a vector bundle with n-dimensional bre,
y : 1 (U) U Vn a bundle chart and {g i } a basis in Vn , then
i (P ) = y 1(P, g i ) ,
i = 1, . . . , n ,
denes the local basis sections or local basis vector elds 1 , . . . , n : U E|U for
every P U M.
The local basis sections are necessary to express vector elds and tensor elds in local
coordinates, so to get a local representative of tensor elds.
Denition 3.4.15. Let M : T M M be a tangent bundle and (T U, T x) a natural
chart of T M, with the chart map T x : T M|U x(U) Rn , as dened in 3.4.9(i)
resp. 3.4.10. If y Rn and {ei } : Rn T Rn are the canonical basis sections in Rn
so {ei } is the canonical basis at every y, then
1
1
(T x) (y, ei ) = x (y), i =
(P ) , i = 1, . . . , n ,
x
xi
denes the local basis sections of T M for all x1 (y) = P U M, that is,
: U T M|U .
,...,
xi
xn
If {ei } is the dual basis of {ei }, then (T x)1 ei = dxi .
34
Corollary 3.4.17. Let M : T11 (M) M be a 11 -tensor bundle, (U, x) a chart of
U M and T (P ) T11 (M) a tensor at P U. Then, by 3.4.2 and 3.4.9(ii),
1
(P )
yT M| 1 (P ) : M
{P } Rn
T (P )
(T ij )(P )
1+1
is a vector space isomorphism that gives the components of T for all P U, where
1+1
Rn = Rn Rn denotes the n n-matrices (see also [4], ch. 4).
The corollary can be applied in a simmilar form to arbitrary pq -tensor elds in order
to undress the tensor components at every point where the eld is dened.
Denition 3.4.18. Let (E1 , M, 1 ), ( E2 , M, 2 ) be vector bundles, 1 (E1 )
and 2 ( E2 ) sections, and : M N a continuous map. The section
T (P ) = ( 1 2 )(P ) (E1 E2 )
where P M, is called second-rank two-point tensor eld over on M, if ( E2 , M,
2 ) is the induced bundle of some (E2 , N , 2 ). With this,
2 : M T N
is called induced section or vector eld over . Note that if 2 (Q) (E2 ) and Q =
(P ), then
( 2 )(P ) = 2 ((P )) .
Example 3.4.19. Two-point tensors and induced sections play an important role in
continuum mechanics. A famous example of a two-point tensor eld is the deformation
gradient, as it acts on two dierent congurations of a material body. The Lagrangian
or particle velocity eld is an induced section V t : B T S, where B is the material
body and S is the ambient space (see section 4.2).
The algebraic operations on tensors dened in section 3.3 all carry over to tensor elds,
by applying the operation to each bre of the corresponding bundle.
3.4.2
35
Action of Maps
The following paragraphs should investigate the action of maps on tensor elds, by
starting with linear transformations and then concentrating on maps : M N
between manifolds.
Denition 3.4.20. Let A : V W be an isomorphism of vector spaces, T Tpq (V)
a pq -tensor eld, a1 , . . . , ap W and v 1 , . . . , v q W. Then
(A T ) a1 , . . . , ap , v 1 , . . . , v q
= T (A a1 ), . . . , (A ap ), (A1 v 1 ), . . . , (A1 v q ) .
is called the pushforward of T by A. The pullback by A is dened through A =
(A1 ) .
TM
(P, V )
TN
T (P, V ) = ((P ), D(P ) V )
i.e. D =
i
X I
TM
P,
(P )
X I
TN
i
(P ),
(P ) i ((P )) ,
X I
x
in coordinates.
Corollary 3.4.22. From the given coordinate expression the chain rule T ( ) =
T T can be easily veried.
I
I
V
(P
)
((P
))
=
V
.
T (V ) (P ) =
X I
xi
X I xi
=(T (V ))I
As the tangent map is linear, one may dene a two-point tensor F to obtain the same
result:
i
i
I
I
J
T (V ) = V
=V
dX
=F V .
I
i
J
i
X x
X x
X I
=F
Conclude that the tangent map acts on a vector like a linear transformation.
36
Denition 3.4.24. With the conventions of 3.4.21, the inverse tangent map over diffeomorphisms : N M is dened through
T (1 ) : T N
xi
TM
(1 )I
,
xi X I
dxi
T M
i
dX I ,
X I
Proposition 3.4.26. (Without proof; cf. [1], p. 165.) Let : M N be an immersion, then there is a neighborhood U(P ) M such that (U) N is a submanifold
(see denition 3.1.26) and U (U) is a dieomorphism.
The preceding proposition does not imply that (M) is a submanifold, and even if T
is injective, might not be. However, dene the following for being injective.
Denition 3.4.27. Let : M N be an injective immersion, then : M (M)
is a dieomorphism and (M) is called an immersed submanifold in N . An immersion
: M N is called embedding provided that it is a homeomorphism onto (M) with
the topology induced by N .
37
/ TN
y< O
y
y
y
yy
V
yyT (V ) V
y
yy
yy
/N
M
TM
O
Corollary 3.4.32. (i) If g = f arises from the pushforward of f , then the pullback
is the inverse operation:
( f )(P ) = f 1 (P ) = f (P ) .
(ii) For a composition of maps and , the chain rule gives
( ) =
and
( ) = .
38
Note that also marks induced sections of vector bundles, but the meaning is dierent!
Induced sections : M T N are not proper vector elds on M (or N ).
Denition 3.4.35. The specication of the pullback and pushforward operators for
elds of dierential 1-forms a : N T N and B : M T M, respectively, is
straightforward. They are being dened through their action on vector elds. Let
V : M T M, then the pullback of a is given by
(( a ) V )(P ) = (a ) (T V ) ,
for P M and (P ) N , clearly,
a = (a ) T .
The pushforward B : N T N of B is a 1-form on N dened by
B = (1 ) B = (B 1 ) T (1 ) .
This can be easily obtained from the denition of the pullback of a by setting B =
a .
The pushforward and the pullback of arbitrary tensor elds can be realized through the
application of the pushforward and the pullback of vector elds and elds of 1-forms,
respectively, to all index slots of the tensor.
Denition 3.4.36. Let : M N be regular, T Tpq (M) and t Tpq (N ), then
( T )(Q) a1 , . . . , ap , w1 , . . . , w q
= T (P ) ( a1 ), . . . , ( ap ), (w 1 ), . . . , ( w q ) ,
and
( t)(P ) B1 , . . . , Bp , V 1 , . . . , V q
= t(Q) ( B1 ), . . . , ( Bp ), ( V 1 ), . . . , ( V q ) ,
where P M and Q = (P ).
39
The local representative of 1-forms is dened analogously by adopting 3.4.35 and replacing by x.
Corollary 3.4.41. Let {ei } and {ei } be the canonical basis sections of T Rn and its
i
i
duals, respectively, then from 3.2.14 and 3.4.15, x (ei ) = x
i and x (e ) = dx , i =
1, . . . , n.
Proposition 3.4.42. Let (V, x), V N , be a chart and t Tpq (N ) a tensor eld,
then the local representative of t on V is
t|V = t1 ,...,p 1 ,...,q
. . . p dx1 . . . dxq .
1
x
x
1 ,...,p
1
p
.
,...,
t
1 ,...,q = t dx , . . . , dx ,
x1
xq
Setting y = x(P ) and applying 3.4.39, the local representative x t on the subset x(V)
Rndim would be
x (t|V ) = (t1 ,...,p 1 ,...,q x1 ) e1 . . . ep e1 . . . eq ,
Application of 3.4.38 and 3.4.41 at every point then proofs the statement.
40
3.5
Exterior algebra deals with the cross product in three-dimensional vector spaces and
with determinants. On manifolds, exterior algebra is a basic ingredient to detect symmetries and orientations, in order to establish measures in n-dimensional non-Euclidian
spaces and to generalize integrals, balance equations and the theorems of Gau and
Stokes. Fundamental objects of exterior algebra are totally skew-symmetric covariant
tensors, which are generally called alternating multilinear forms respectively dierential forms on manifolds.
In this section, lower case Greeks are used for coordinate indices, and lower case Latins
are used for labels.
Denition 3.5.1. A permutation of a set Sk = {1, . . . , k} N is a map : Sk Sk .
Note that the set k of all permutations {(1), . . . , (k)} on Sk has k! elements. Dene
the permutation of a k-tuple of any elements (a1 , . . . , ak ) by
(a1 , . . . , ak ) = (a(1) , . . . , a(k) ) .
A transposition is a permutation that interchanges only two elements. Every permutation is composed of even or odd numbers of transpositions. An even (odd)
permutation is obtained from an even (odd) number of transpositions. Dene the signature of the permutation by sgn : k {1, +1}, then set sgn = +1 for even
permutations, and sgn = 1 for odd permutations, so that sgn = (1) .
R.
T M, where
k
41
is called k-th exterior
1
(sgn ) T (v (1) , . . . , v (k) ) ,
k!
k
It should be clear that k T M isa subspace of Tk0 (M). If n is the dimension of M,
then the dimension of k T M is nk .
Denition 3.5.5. A (dierential) k-form (P ) k TP M at P M is a k0 -tensor
such that Alt = .
Corollary 3.5.6. For T Tk0 (M), Alt T k T M, thus Alt(Alt T ) = Alt T by
3.5.5.
Proposition 3.5.7. Let (P ) k TP M be a k-form and v 1 , . . . , v k TP M, then
(v (1) , . . . , v (k) ) = (sgn ) (v 1 , . . . , v k ) .
Proof. First,note that (sgn )2 = 1 and the number of all permutations on {1, . . . , k}
is k!, that is, k a = k! a. Then from Alt = ,
1
(sgn ) (v (1) , . . . , v (k) ) = (v 1 , . . . , v k )
k!
k
1
(sgn )2 (v 1 , . . . , vk ) .
=
k!
k
Denition 3.5.8. Let T (P ) Tk0 (M) and S(P ) Tl0 (M) be tensors, then the exte
rior product or wedge product (T S) k+l TP M is a (k + l)-form dened by
T S =
(k + l)!
Alt (T S) ,
k! l!
TP M and (P )
TP M is dened in the same manner.
forms (P )
(1 + 1)! 1
(a b b a ) = a b b a
1!1! 2!
= (b a a b ) = b a .
42
Proposition 3.5.10. Let (P ) k TP M be a k-form, (P ) l TP M an l-form
and v 1 ,. . ., v k+l TP M vectors, then
( )(P )(v 1 , . . . , v k+l ) =
(sgn ) (v (1) , . . . , v (k) )(v (k+1) , . . . , v (k+l) ) ,
where
k,l
k,l
Proof. By 3.5.7, alternation of (or ) reduces to the sum with a single summand:
1
Alt (v 1 , . . . , v k ) =
(sgn ) (v (1) , . . . , v (k) )
k!
k
=
(v (1) , . . . , v (k) ) .
(1)<...<(k)
Therefore, the sum in Alt ( ) = Alt (Alt Alt ) is only over the shue permutations that satisfy (1) < . . . < (k) and (k + 1) < . . . < (k + l). From 3.5.3, the
, so
number of all shue permutations is (k+l)!
k! l!
k! l!
(sgn ) (v (1) , . . . , v (k) )(v (k+1) , . . . , v (k+l) ) ,
(k + l)!
k,l
where the denition 3.3.10 of the tensor product has been used. Substitution into 3.5.8
then gives the result.
k
Proposition 3.5.11.
TP M,
(Without proof.) For a R and forms
l
TP M and m TP M, the exterior product has the properties
(i) a( ) = a = a (bilinearity),
(ii) ( ) = ( ) (associativity), and
(iii) = (1)kl (supercommutativity).
Corollary 3.5.12. For a1-form TP M, = 0 by 3.5.11(iii) (see also corollary
3.5.9). However, if k TP M, then = 0 in general.
Corollary 3.5.13. Recall 3.3.4 for the components of a tensor, and let (U, x), with
U M a subset, be a chart with coordinate functions {x }.
(i) If T T20 (M) is a 02 -tensor, the components of Alt T are
(Alt T ) = Alt T
,
x x
1
1
,
,
T
= (T T ) .
=
T
2
x x
x x
2
(ii) Let 2 TP M be a 2-form, then 3.5.7 yields
,
,
=
= .
=
x x
x x
k
TP M and
( )1 ...k+l
where
k,l
43
l
k
(ii) {dx1 . . . dxk }, with
TP M, which
n1 1n!< . . . < k n, is a basis of
therefore has the dimension k = k!(nk)! .
Proof. (i) Through k T M Tk0 (M), every k-form is a k0 -tensor that has a local
representative
= 1 ...k dx1 . . . dxk
in a chart (U, x), U M, by 3.3.11. Here 1 ...k = x1 , . . . , xk , and {dx1
. . . dxk } is the tensor basis of Tk0 (M) at P U. Alternation and 3.5.8 then gives
= Alt = 1 ...k Alt (dx1 . . . dxk ) =
1
1 ...k dx1 . . . dxk .
k!
Note that the Einstein summation convention is still in force and the sum runs over
all choices of indices 1 , . . . , k {1, . . . , n}, including those where not all indices are
distinct. However, in that latter case, 1 ...k = 0 from 3.5.12. For
cases where
the other
1
1 , . . . , k are distinct, applying 3.5.7 for both x1 , . . . , xk and dx . . . dxk
yields
1 ...k dx1 . . . dxk = (1 )...(k ) dx(1 ) . . . dx(k )
for any permutation k and by noting that (sgn )2 = 1. Now, since rearranging the indices does not change 1 ...k dx1 . . . dxk , the sum needs to be
only
over one of the k! index permutations. Choose 1 < . . . < k to nally get
= 1 <...<k 1 ...k dx1 . . . dxk as desired.
(ii) Linear independency of the dx1 . . .dxk can be shown by
using the complemenk+1
n
1
n
, . . . , dx and the condition (dx . . . dx ) x1 , . . . , xn = 1 (see,
tary set dx
for example, [1], p. 332). As the number of ways that k coordinate dierentials can be
chosen from n coordinate
dierentials regardless of order (equivalently, with one xed
n
order) is just k , the assertion follows.
Corollary 3.5.15. Let be a k-form and an l-form, then in a coordinate system
{x }, the local representative of = (Alt ) (Alt ) is
=
1
... ... dx1 . . . dxk dxk+1 . . . dxk+l .
k! l! 1 k k+1 k+l
44
k
Note
k that in the literature, the term k-form is generally reserved for the sections of
T M, and not for the totally skew-symmetric covariant tensors at single points of
a manifold. This is because of their predominance in dierential and integral
calculus.
By abuse of language, k-form will be adopted for both elements of k T M and
k (M) in the paper, as long as the meaning will be clear from the context.
Proposition 3.5.17. Let : M N be a map, k (N ) and l (N ), then
( ) = .
Proof. Every dierential form constitutes a covariant tensor, so 3.4.38 can be applied,
and alternation commutes with the pullback.
Denition 3.5.18. For k (M), dene the interior product or degree-1 derivation
(= inner product!) by contracting a vector eld u (T M) with the rst index of
at every P M:
iu : k (M)
(v 1 , . . . , v k )
k1 (M)
iu(v 2 , . . . , vk ) = (u, v 1 , . . . , v k ) .
In local coordinates {x },
iu =
1
u 2 ...k dx2 . . . dxk .
(k 1)!
3.6
Dierentiation on Manifolds
So far only the algebra of tensor elds on manifolds has been analyzed. However, in
continuum mechanics, tensor elds generally depend on time and location. Due to
the various geometric facilities on manifolds, beyond partial derivatives there are three
possible types of dierential calculi: covariant dierentiation, Lie dierentiation and
exterior dierentiation. Each is appropriate for special problems, but all types have to
satisfy Leibniz rule
D(AB) = (DA)B + A(DB) .
45
D is the dierential operator under consideration and A, B are arbitrary tensor elds.
The so-called exterior derivative d satises the general Leibniz rule
d( ) = d + (1)k d ,
where k (M) and is an arbitrary form. First, the covariant derivative is being
briey introduced.
3.6.1
Covariant Derivative
(t, s)
TM
dc
T c(t, s) = c(t), (t) s ,
dt
and T c(, 1) : I T M.
Corollary 3.6.2. Let c, w be as in 3.6.1. Moreover, let {xi }P = x(P ) be the local
i
i
coordinates of P = c(t) U M in a chart (U, x), w = wi x
i and c = x c. Then
w i(c(t)) = dxi w(c(t)) = dxi (T c(t, 1)) =
dci
(t) = ci (t) ,
dt
Denition 3.6.4. Manifolds with connection according to 3.6.3 are called anely
connected.
In general, parallelism is not global this holds only for Euclidian point spaces. A
locally parallel transported tangent vector of a curve remains tangent to the curve.
Indeed, the main dierence between Euclidian and non-Euclidian geometry rests on
the fact that for the latter any geometric consideration is local, whereas the former
implies the existence of a global parallelism.
The abstract denition 3.6.3 should now be motivated step by step. The total dierential dv of a vector eld does not transform as a tensor, hence avoiding invariant results.
This drawback is eliminated by dening the absolute dierential.
46
Dv i(P ) = dv i + v k dxj k ij
are referred to as the absolute dierentials of the components of v at P U. If
a = ai dxi (T M) is a 1-form, then dene
Dai(P ) = dai + aj dxk k ji
in the same chart. The k ji are called the connection coecients.
In Riemannian spaces (M, g), the k ji are called Christoel symbols of the second kind.
As they depend on the metric coecients gij and g ij in a chart (U, x) of M via
1 kl gjl gil gij
k
i j = g
+ j
,
2
xi
x
xl
which is not shown here, one has k ji = i jk . For that reason Riemannian spaces are
called torsion-free, and the connection on (M, g) is referred to as the Levi-Civita
connection.
k ji
+
.
k i
xk
xj
xi
xm
xk xi
i v j (P ) =
(v j + Dv j ) v j
Dv j
v j
(P
)
=
=
+ v k k ji
i
i
i
x
x
x
Denition 3.6.9. Since the i v j are components of a 11 -tensor eld, dene the covariant derivative of v pointwisely through
v(P ) = i v j dxi
xj
wv = v(w) = j v w
=
xi
i
T11 (M) .
v i j
k j
i
w + v w k j
j
x
xi
is called the covariant derivative of v along w. Note that the latter gives a condition
for the i jk :
= k ij i .
j
= j
k
k
x x
x x
x
47
Descriptively, the covariant derivative is the deviation of the locally parallel transported
vector from the original vector in the vicinity of a point.
Proposition 3.6.10. For covariant-constant vector elds the absolute dierentials of
their components vanish, that is, in a coordinate system {xi }, Dv i = 0 and
v i
= v k k ij .
xj
Proof. In 3.6.9, set w = c and use the condition 3.6.3, then
i
v
j
k
i
+ v k j
= 0.
c v = c
j
x
xi
As c is arbitrary and the
xi
a (P ) =
ak i j dxi dxj
xj
in a chart (U, x), one has
i T jk (P ) =
T jk
+ T lk l ji T jl k l i ,
xi
and
wT (P ) = k T ij w k i dxj
i
x
T j k
=
w + T lj w k l ik T il w k j lk
dxj .
k
i
x
x
Proposition 3.6.12. Let (M, g), g = gij dxi dxj , be a metric space with connection
, then
g = 0 .
Proof. From 3.6.11 and 3.6.5,
i gjk
gjk
gjk
=
2gjl k li =
i
x
xi
gij gkj
gki
+
k
i
x
x
xj
= 0.
The covariant derivative can also be applied to vector and tensor elds over maps, from
which the following important result can be obtained.
Theorem 3.6.13 (Induced Connection). (See also [3], sec. 5.7) Let N be a manifold with connection and v (T N ). A regular map : M N induces a unique
connection on M such that for t TP M, with P M, one has
t (v ) = T (t) v
( T N ) ,
48
Proof. Let P U M and Q V N , and let (U, X), (V, x), respectively, be
i
i
appropriate charts such that t(P ) = tI X
I and v(Q) = v xi . Moreover, let F I (P ) =
i
1
x X
(P ) be the components of the tangent map T : T M T N , and j ik (Q) the
X I
coecients of . By 3.6.9 and 3.4.23, locally there is
i
v
I k
j i
T (t) v = t F I
+ v j k
k
x
xi
i
j
I i
v
k I
=
F
t
+
v
F kI
I
j k
k
x
xi
I i
(v i ) I j
=
t
+
v
= t (v ) ,
j
I
X I
xi
where j iI (P ) = (j ik ) F kI are the coecients of the induced connection on
M.
Corollary 3.6.14. Additional to the previous denitions, let I JK (P ) be the
connection coecients at every P M, and T (T11 (M) T11 (N )) a 11 11 -two-point
tensor eld over : M N . Use 3.6.11 and 3.6.13 to get
(i) At P M, the components of the covariant derivative of T are
I T JKij =
T JKij
+ T LKij L JI T JL ij K LI + T JK kj k il F lI T JK ik j kl F lI ,
X I
W T = I T JKij W I
K
dX
dxj
J
i
X
x
T JKij I
W + T LKij W I L JI T JL ij W I K LI + T JK kj W I k il F lI
I
X
J i
I k
l
T K k W j l F I
dX K i dxj .
J
X
x
Here the symbol should cover the covariant derivative of the TP M- and TP M-slots,
S JKi j
xk
W
(S ) (P )
+ S LKi j L Jk S JLi j K Lk
+ S JK l j l ik S JKi l j lk
dX K i dxj
J
X
x
= T (W ) S .
p
q
- and
49
p r
-tensor elds is straightforward.
q s
v i
+ v j j ii .
xi
det gkl
gij
= (det gkl )g ij ,
gmn
det gjk gmn
det gjk
=
= (det gkl ) g mn
.
i
i
x
gmn x
xi
g
gij
il
Substitution into i kj = 12 g kl xjli + g
yields
xj
xl
j ii
det gik
det gik
1 ikl gik
1
1
= g
=
=
,
2
xj
2 det gik xj
xj
det gik
so nally,
v i
1
det gik
1
j
i
div v = i + v
=
det gkl v .
x
xj
det gik
det gkl xi
3.6.2
Lie Derivative
50
Denition 3.6.18. In a dynamical system, let the states of the system at time t be
represented by points P (t) M, in which M is then called the state space or phase
space. At starting time s, P (s) = Ps are the starting points that constitute the initial
state. After some time t s has elapsed, the state changes to
t,s (Ps ) = P (t)
at time t. The operator t,s is referred to as the time-dependent ow on M provided
that
t,s s,r = t,r
and
t,t = IdM .
and
tt = 0 = IdM .
By noting that t+s = s+t = s t also holds, t fullls the group properties.
i
i
i
i
i
where u = ui x
i and c = x c. Moreover, c(0) = P becomes c (0) = x c(0) = x (P )
in the chart, so there is a system of n ordinary dierential equations with given initial
conditions. As u resp. the ui are C k , by Picard-Lindelofs theorem they have a unique
solution satisfying the initial conditions: the C k+1 functions ci (t).
51
Denition 3.6.22. Let u(P ), P M, be the (time-independent) vector eld generating the ow t , then the Lie derivative of a possibly time-dependent tensor eld
T Tpq (M) is dened by
t (T (t (P ))) T (P )
t T 0 T
d
LuT (P ) = lim
.
= lim
= t T
t0
t0
t
t
dt
t=0
t is the pullback concerning t .
The Lie derivative approximately answers the question how a tensor eld T changes
under some ow.
Denition 3.6.22 is only for time-independent vector elds u. In general, however, the
generating vector eld is time-dependent, i.e. u : M I T M, where I R. What
is the ow that will be generated? Clearly, it is asked for the solution of the dierential
equation
c(t)
= u(c(t), t) and c(s) = Ps ,
where s is the starting time and Ps = c(s) is the initial condition. However, this is
again an ordinary dierential equation which has the unique solution c(t) = t,s (Ps ),
where t,s is a time-dependent ow on M note this is a more general version of
proposition 3.6.20. Again, a proof can be performed with the aid of Picard-Lindelofs
theorem, and uniqueness of the solution gives the properties 3.6.18 of the ow (see also
[1], ch. 4; and [4], ch. 5).
Denition 3.6.23. A mapping t,s : M M, t, s I R, is referred to as the
time-dependent ow of u : M I T M, if for each s and P M
c(t) = t,s (Ps ) = (Ps , s, t)
is the unique integral curve of u starting at Ps at time t = s, i.e.
c(t)
=
d
t,s (Ps ) = u(t,s (Ps ), t) and c(s) = s,s (Ps ) = Ps .
dt
t,s
Tt Ts
t,s
T t s,s
Ts
d
LuT (Ps , t) = lim
= lim
= t,s T t
,
t0
t0
t
t
dt
t=s
with t = t s and the pullback operator t,s
concerning the ow t,s . Fixing t in
T t = T (, t) gives the autonomous Lie derivative
t,s
T s s,s
Ts
T
d
lim
= t,s T s
= LuT
= uT .
t0
t
dt
t
t=s
52
u
Luv t
v(Ps ) = (s,s
v)(Ps )
v(t,s (Ps ))
P (t) = t,s (Ps )
(t,s
v)(Ps )
Ps = s,s (Ps )
d
Tt ,
dt t,s
that is, the Lie derivative of an arbitrary tensor is obtained by pulling it back from t to
s, performing the time derivative, and then pushing it forward to t again.
Proof. Let t = s and f : M R be a dierentiable function on M, then
d
d
r,t fr
=
fr r,t t,s
d
d
=
ft
(r,t t,s ) fr
= t,s
dr
dt
r=t
by the property 3.4.32(ii) of pullbacks. Since this result holds for f replaced by arbitrary, and possibly time-dependent tensor elds T (see [1], sec. 5.4 for a detailed
discussion), the assertion follows.
Example 3.6.26. Let f : M I R be a dierentiable time-dependent scalar eld,
where ft (P ) = f (P, t) and P M, and let u(P, t) be a time-dependent vector eld
the time-dependency of P being understood. By 3.4.30, (t,s
ft )(Ps ) = ft t,s =
f (t,s (Ps ), t). Therefore
d
d
Luf (P, t) = t,s ft
= f (t,s (Ps ), t)
dt
dt
t=s
t=s
i
t,s
f
f
f
f
=
(P, t) + i (P, t)
(P ) =
(P, t) + i (P, t) ui(P, t) ,
t
x
t
t
x
that is, Luf =
f
t
+ u[f ].
53
vi j
u
xj
ui j
v
xj
xi
v
+ [u, v] ,
t
3.6.3
Exterior Derivative
The third type of a calculus of dierentiation on manifolds involves the exterior derivative, which is restricted to elds of dierential forms.
Denition 3.6.31. Let k (M) be a k-form on M, then its exterior derivative is
the (k + 1)-form
d = d =
=
1
d1 ...k dx1 . . . dxk
k!
1 1 ...k
dx dx1 . . . dxk
k! x
k+1 (M) .
Corollary 3.6.33. (i) Since d d = 0 from Schwarz theorem, every exact form is
closed, but the converse need not hold. (ii) By 3.5.12 and 3.5.14, every (k = ndim )form is closed.
54
Proposition 3.6.34. (Without proof.) (i) (d) = d( ), i.e. the exterior derivative commutates with the pullback. (ii) Let u be a vector eld, then u = iud+diu
(Cartans formula).
Summarizing the preceding sections, the covariant derivative needs an additional structure on manifolds, whereas the Lie derivative restricts the direction of dierentiation.
The exterior derivative is exclusively for dierential forms. All three types lead to
proper tensors, whereas the usual partial derivative does not. The coordinate-invariant
or covariant formulation of physical equations, therefore, is ensured when applying
the covariant, Lie, and exterior calculus.
3.7
Integration on Manifolds
3.7.1
Denition 3.7.1. Let n here and in the remainder of this section denote the
dimension of the manifold M. An n-form n (M) such that (P ) = 0 for all
P M is called volume form, and the set of all volume forms on M is the volume
bundle. If there exists a volume bundle on M, then M is called orientable and the
pair (M, ) is referred to as a volume manifold.
55
Example 3.7.4. The Moebius strip is a non-orientable manifold, because the equivalence classes [] and [] can only be dened locally.
Proposition 3.7.5. Volume forms (and n-forms) have a single component, that is,
dim(n (M)) = 1.
Proof. This follows directly from 3.5.14 by setting k = n.
Corollary 3.7.6.
(i) 3.5.14(i) states that the order of the coordinate dierentials in
the basis of k TP M is arbitrary. So carrying this over to each bre of k (M) and
using 3.7.5, one obtains a local representative of the volume form at P M in a
coordinate system {x }U on U M:
(P ) = (P ) dx1 . . . dxn .
(ii) By 3.7.5, any other volume form = f is a linear combination of the basis
{dx1 . . . dxn } n (M), by choosing suitable f : M R. If f > 0, then
[], and if f < 0, then [].
U
and
x
>
0
(resp.
<
0)
for
all
P
U,
then
the
basis
is
TP M a basis. If (P ) x
,
.
.
.
,
1
xn
called positively (resp. negatively) oriented relative to the volume form .
Denition 3.7.8. For every chart (U, x) on orientable M with orientation [], where
x : U Rn , dene a map x through
x |U : n (U) n (x(U)) ,
and set x (dx1 . . . dxn ) = f dx1 . . . dxn . The chart (U, x) is then called
positively oriented, if x (|U ) is equivalent to
dx1 . . . dxn n (x(U)) ,
the standard volume form on Rn .
The reader should notice that the map x has been already dened in a similar form to
obtain the local representative of tensor elds. However, for n-forms one sets x (dxi
i
i
. . . dxn ) = dx1 . . . dxn instead of taking
n x dx = e (see 3.4.41). It can be shown
that x is an isomorphism on each bre of
T U , so the inverse map x indeed exists.
Corollary 3.7.9. For M being orientable, there is an atlas A(M) = {(Ui , xi )}iI ,
with I N, of which all charts are positively oriented, that is, the determinant of the
Jacobian matrix of every two chart transitions xj (Uj Uk ) xk (Uj Uk ), j = k I,
is positive.
Theorem 3.7.10 (Transformation of n-Forms). Let ai = a i x TP M, where
1 i n, be vectors (the subscript labels are for convenience), and let n (M),
then at P M,
(a1 , . . . , an ) = det (a i )
,
,...,
x1
xn
where (a i ) is the matrix whose columns are the components of ai .
56
Proof. It is convenient to prepare the proof in two steps. First, it will be shown that
1
a . . . an
,...,
1
x
xn
i
= det a
,
x
where a1 , . . . , an are 1-forms. For the purpose of this paper it is sucient to consider
the case n = 2; a generalization can be found in [1], sec. 6.2. Dene ai = ai dx
TP M, then, by 3.5.10,
a a
,
1
x x2
=
a
(sgn ) a
x(2)
1,1
= a11 a22 a12 a21 = det ai .
x(1)
Here 1,1 denotes the sum over the two (1, 1)-shues on {1, 2}, and (ai ) =
i
a x is the matrix whose rows are the components of ai , as desired.
Nowdene = a1 . . . an . As a special case of the previous result is dx1 . . .
= 1, one has
dxn x
1 , . . . , xn
dx . . . dx
1
,...,
1
x
xn
= det (a i )
,...,
1
x
xn
a (ai ) =
ai ai dx
= ai ai = ai a i = 1
for all i {1, . . . , n}. From this and by using dened above, dx1 . . . dxn
. . . , xn = (a1 , . . . , an ), which proofs the theorem.
,
x1
x X 1
X
( dx . . . dx ) = det
( ) dX 1 . . . dX n ,
x X 1
where
is the Jacobian matrix of with respect to x and X.
X
dX 1 . . . dX n ,
57
both sides with X 1 , . . . , X n , then the right hand side becomes just det X
. On
the left hand side, however,
1
n
,...,
(dx . . . dx )
X 1
X n
1
n
, . . . ,
= (dx . . . dx )
X 1
X n
1
n
= (dx . . . dx )
,...,
X 1 x
X n x
1
n
,...,
(dx . . . dx )
= det
X
x1
xn
= det
X
by 3.4.35 and theorem 3.7.10.
= det
= det
det
det
.
det
X
x X X
x
X
X
Thus det( X
) is not a proper scalar function in general, i.e. it is not coordinateinvariant.
Denition
3.7.14. Let A : V W be an isomorphism of vector spaces and
n
W an n-form with respect to the vector space W, then, by 3.7.11 and 3.4.20, a
coordinate-free denition of the determinant can be obtained:
A = (det A) .
Note that det A = det (Ai ), where (Ai ) is the matrix of the linear transformation
(see 2.1.10).
58
(P ) : TP M . . . TP M
n-fold
a1 , . . . , an
TP M
a1 . . . an = 1 ...n a1 . . . an ,
1 ...n
+1 if 1 , . . . , n is an even permutation of 1, . . . , n
1 if 1 , . . . , n is an odd permutation of 1, . . . , n
=
0 if i = j for some i = j
n
n
1
. . . dxn from 3.3.5. Substitution into the
. . . dxn = 1 ...n x1 . . . x
dx
xn
x 1
preceding then yields
1 ...n
as desired.
x
= det
x
x1
xn
...
...
xn 1 n
x1
Note that as a tensor has nn components, but only n! are non-zero. Hence, the even
relative -tensor of weight 1 boils down to the even relative scalar dx1 . . . dxn =
(dx1 , . . . , dxn ) of weight 1 when applied to the n-tuple (dx1 , . . . , dxn ).
3.7.2
59
The n-forms on n-dimensional manifolds have a single component, thus they can be
integrated over open sets by prescribing the following:
Denition 3.7.17. Let X Rn be open and f dx1 . . . dxn n (X ), f =
f (x1 . . . xn ) being understood, then dene
$
$
$
1
n
1
n
f dx . . . dx =
f dx . . . dx
f,
where
%
X
This fundamental theorem well-known from the analysis of real functions leads to the
answer of how n-forms are to be integrated over manifolds.
Proposition 3.7.19. Let (U, x), U M, be a positively oriented chart and
n (M) an n-form, then
$
$
=
x1 .
U
x(U )
Proof. Without loss of generality assume that (P ) = (P ) dx1 . . . dxn for every
P M, then
x (|U ) = ( x1 ) dx1 . . . dxn
by 3.7.8 together with 3.4.42, and
$
$
x =
x(U )
x1
x(U )
In other words, n-forms get access to the calculus of integration of real functions by
dening their integral in the chart as the ordinary Riemann integral in Rn (cf. [1],
sec. 7.1). Clearly, dx1 . . . dxn is replaced by its Lebesgue measure, and the 1-form
valued coordinate dierentials dxi on U are read as the usual coordinate dierentials
dxi of the chart x(U) Rn note that this approach is comparable to the denition
of a dierentiable structure on manifolds presented in section 3.1.
If the atlas A(M) = {(Ui , xi )}iI of M = iI Ui contains of more than one chart,
then the integral of over M is dened via the sum of integrals over the subsets Ui
by assuming that there is a so-called partition of unity subordinate to the atlas.
60
M
U
x
xn = 0
xn
x(U)
Rn1
nx = (0, . . . , 0, 1)
On Riemannian manifolds, i.e. metric spaces, volume forms enable volume measurement. The volume measure should always be non-zero and positive to be consistent
with the specication in ordinary R3 .
From linear algebra, the volume of the parallelepiped spanned
by the three independent
vectors w1 , w2 , w3 R3 is given by Vol(w 1 , w 2 , w3 ) = det
w i , wj , if w1 , w2 and
w3 are positively oriented, and by
&
Vol(w 1 , w2 , w3 ) = |det
wi , w j | ,
if w 1 , w2 and w 3 are negatively oriented. det
w i , wj denotes the determinant of the
matrix (Wij ), whose elements are given by the inner products Wij =
w i , wj . To
61
carry this over to manifolds, one needs to dene quantities that involve the absolute
value of a determinant.
Denition 3.7.22. A multilinear mapping
:
V .
. . V R
n-fold
over a vector space V is called an -density, if for every v 1 , . . . , vn V and every
endomorphism A : V V,
(Av 1 , . . . , Av n ) = |det A| (v 1 , . . . , v n ) .
The set of all -densities over V is denoted by |
are also called odd or twisted n-forms.
n
-densities can be constructed
from n-forms as follows: If n V is an n-form, then
(x x1 )
=
det
,...,
x1 , . . . , xn ,
x1
xn
x
|dv| (w 1 , . . . , wn ) = |det
w , w | 2 .
62
TP M be
Corollary 3.7.25. In a regular chart (U, x) on (M, g), let x
1 , . . . , xn
&
1
oriented charts and let = x X denote the coordinate functions of x concerning with respect to {X }P = X(P ), for every P 1 (V) U M. Let dV , dv be
the Riemannian volume forms on M and N , respectively, then
dv = J dV ,
(det g )
where J (P ) = det X
, called the Jacobian of , is a proper scalar.
det G
Proof. By 3.7.11,
dv =
det g
=
det g (dx1 . . . dxn )
dX 1 . . . dX n .
det
X
1
det g x
x
Therefore,
x
x
det(g )
= det
det(G )
det(g ) det(G )
det(g )
= det
X
det(g ) det(G ) det(G )
det(g )
x
X
det
det
= det
X
x
X
det(G )
det(g )
= det
= J ,
X
det(G )
J
X
x
x
, see 3.7.13 indeed is a scalar.
63
Proposition 3.7.27. In the light of 3.3.12, Riemannian volume forms are absolute
pseudoscalars such that the formula of 3.7.26 boils down to
dv = J dV .
Proof. For the moment, drop the condition that both (V, x) and (V , x ) previously
dened are positively oriented, then
'
x x
x
.
det
=
det g = det g
det
g
x x
x
Let dv = det g dx1 . . . dxn = det g dx1 . . . dxn , then use 3.7.12(i) to
obtain
x
1
n
det g dx . . . dx = det
det g dx1 . . . dxn
x
x
det x
det g dx1 . . . dxn
=
det x
x
= sign det
det g dx1 . . . dxn ,
x
as desired. Now since dv is a scalar, one arrives at dv = dv by applying denition
3.4.30.
Corollary 3.7.28. det g is an odd relative scalar of weight 1, i.e. an odd scalar
density, and dv = det g 1 ...n dx1 . . . dxn = det g (dx1 , . . . , dxn ).
Denition 3.7.29. = det g is called the Levi-Civita tensor.
Corollary 3.7.30. By 3.3.12, 3.7.16 and 3.7.28 the Levi-Civita tensor is an absolute
pseudotensor.
Proposition 3.7.31. (Without proof.) If dv is a Riemannian volume form, then
udv = d(iudv) = (div u) dv.
Note this characterization of divergence does not require any metric or ane connection.
Proposition 3.7.32. Let M be a manifold with boundary M, w (T M) a vector
eld and n (T M) the unit normals on M. Then, on M,
iwdv = (n w) da ,
where da, dened through n da = dv, is called area form.
64
Proof. Note that da is eectively the dv for M, thus it also has a single component.
Now let {x } be a coordinate system
for M and choose M to be the plane where
1
x = 0 and n = dx , then da = det gab dx2 . . . dxn (a, b = 2, . . . , n). By 3.5.18,
3.5.19, and noting that 0! = 1,
=w
1
det gab
Evaluated
on M, i.e. at x1 = 0 = const., the second term vanishes and one is left
(n w) da .
Proof. The proof can be obtained by using Stokes Theorem 3.7.21 and applying the
propositions 3.7.31 and 3.7.32.
Chapter 4
Application: Continuum Mechanics
The geometric concepts introduced in the previous chapter should be applied to continuum mechanics in the following. It is being restricted here to the equations of kinematics with important strain measures, and the conservation of mass as example of balance
equations. The derivations are mainly based on the contributions [2, 11, 12, 13]. The
concept of an abstract material manifold follows Noll [27, 28]. In the last section,
a framework is outlined so as to extend the geometric continuum mechanics by an
Arbitrary Lagrangian-Eulerian (ALE) formulation [e.g. 22] of kinematics on manifolds.
4.1
66
B
t
P
B
t (B)
t
s
t,s = t s1
s (B)
Qs
T
B
dX
P is its dual in the cotangent space. Since S is Riemannian, GIJ (P ) =
,
are the metric coecients on B. The current conguration contains the
X I X J P
current places Q = t (P ) S of the particles of the material body. Charts of the
neighborhoods V(Q) S are denoted by the pairs (V, x), with {xi }Q = x(Q) R3 ,
i
that
space. The denition of a local base
is, x are the coordinates of the ambient
i
T
S
and
the
dual
base
{dx
}
T
i
t (P ) S is straightforward. The gij (Q) =
x
t (P )
,
are the metric coecients on S.
xi xj Q
As congurations are understood as embeddings, chart transitions from X to x and
vice versa are compatible. Herein, regularity of the local coordinate systems is demanded everywhere in order to exclude breaks and cracks of the material respectively
the ambient space.
67
The dierence between Q and t (P ) is important: the spatial points Q and also
S are independent of the motion of the body. By contrast, t (P ) denotes the point
occupied by the particle P at time t in the ambient space, and t (B) only considers
those spatial points which are occupied by the body during its motion.
As a common notation, upper case Latins are used for coordinates (I, J, K, . . .
{1, 2, 3}), vectors and tensors of the reference conguration, or are related to the
Lagrangian formulation dened later on. Lower case Latins are related to the current conguration, the ambient space, or to the Eulerian formulation. This is only a
general convention, inconsistency might be useful. Moreover, in order to reduce the
amount of necessary symbols within the text and in formulas, the picture X(P ) in
the chart is often identied with the particle P and a function f (X) = f X(P ), for
example, is written f (P ). However, it should be emphasised that there is a fundamental dierence between the object in the manifold and its picture or localization in the
chart.
The embedding of the body in space enables the observation of the body, and the measurement of physical properties or deformations, as B benets from the inner product
in S. In the material manifold M it does not exist any metric because of the arbitrary
choice of the map [27, 28]. Lengths and angles change permanently during a deformation process, hence it does not make any sense to dene a distance and suchlike
measures. The material manifold is merely a continuous assembly of particles with
certain physical properties. It is indeed a useful construction in fracture mechanics
and in material sciences, for example. Therein, it is used to establish microstructure of the material or forces on cracks, inclusions and dislocations of crystal lattices
(cf. conguration mechanics, Eshelby mechanics).
It is assumed throughout this chapter that both the ambient space and the body are
three-dimensional. However, for some concepts it is illuminative to include the case
where dim(B) < dim(S). For example, if B is a shell, note that dim(B) = dim(B) =
2, because is an embedding. Since t is also an embedding and t (B) S is a
submanifold, dim(t (B)) = 2 and charts of t (B) and S are compatible in the sense of
3.1.26. Note that even if B and S have dierent dimensions, t can be invertible, i.e. a
regular map.
Denition 4.1.4. Let t : B S be the motion of a material body, and let U B,
V S and 1
. Additionally, let (U, X), (V, x) be appropriate charts,
t (V) U =
then, analogous to denition 3.1.21,
x t X 1
X (1 (V)U ) :
t
X 1
x 1
t (V) U
t (V) U .
denes the chart transition concerning t or the localization of the motion. If {xi }Q =
x(Q) for Q V, then abbreviate it = xi t X 1 , so xi t = it are the coordinates
on t (B).
68
vt
T t
/ TS
pp7 O
p
p
pp
ppp
p
v = v =V 1
p
pp V t =T t (vt ) t t t t t
p
p
pp
ppp
/S
T OB
it
Vt =
= Vti (P ) i (Q)
i
t P x
x
describes the particle velocity eld at time t, called the Lagrangian velocity eld over
t . Clearly: V t : B T S, with V t (P ) = V (P, t) at xed t. The corresponding spatial
or Eulerian velocity eld,
i
1
1
v t = V t t =
t
= vti(Q) i
(T S) ,
i
t P
x
x
where Q = (P, t), is obtained by switching the point arguments (gure 4.2).
t .
J I
vt J = vIt (P ) I (Q) .
V t = T t (vt ) =
I
X
t
t
The I (Q, t) are time-dependent basis vectors that are scribed on B and follow its
If B and S have dierent dimensions, as this is the case for shells, the Eulerian velocity
eld v t is tangent to S, but it is not necessarily tangent to t (B). Thus, dening
the material velocity through v t = t vt has no physical meaning. Note that in the
literature, v is often called the convective velocity. This, however, conicts the term
used for a fundamental velocity eld in the ALE formulation.
69
1
t,s
,
t, s R .
Proof. Note that before reaching Q = t (P ) at time t, the particles pass some points
Qs = s (P ) at t = s (gure 4.1). For the components of the spatial velocity v at Qs ,
vti (Qs )
it
i 1
i
t s (Qs ) = t,s
=
1
(Qs ) .
s =
t P
t
t
Proposition 4.1.9. The Lie derivative of arbitrary t Tpq (S) along the spatial velocity
is
d
d
d
Lv t = (t 1
(t 1
(s t t) = t t t .
s )
s ) t = t s
dt
dt
dt
Proof. This follows from the previous proposition, by recalling the chain rule 3.4.32,
using 3.6.25, and by noting that (s )1 = (1
s ) = s .
4.2
X I
T(P ) S
i
=
T
= F
,
I
I
i
X
X x
X J
I
i
dX
=
F
(P
)
dX I
I
X I xi
xi
( T S T B) .
The expression T S denotes the induced bundle of T S over the map according to
3.4.12, i.e. x
i attached to (P ) being understood. If t is a motion of B in S, then F
I
depends on both P B and t R, that is, F (P, t) = F iI (P, t) x
i dX .
70
Corollary 4.2.2. (i) By 3.3.20 and 3.3.21, the transpose of the deformation gradient
is
F T : T S T ((B))
xi
TB
FT
j
T I
IJ
=
(F
)
=
g
F
G
ij
i
J
xi
X I
X I
F (Q) = gij
IJ
j
1
1
G
dxj
J
J
X
X
i
= F I 1 dX I
.
xi
(ii) Recall 3.3.22 and 3.4.24, then conclude that the inverse deformation gradient
F 1 (P ) : T(P ) S TP B corresponds to the inverse tangent map with switched base
points, i.e. F 1 = T (1 ) . Therefore,
F
(P ) =
where (1 )I = X I 1 x1 and
(1 )I
xi
(1 )I
(Q)
xi
dxi ,
I
X
(1 )I
(P ).
xi
(iii) From 3.3.22, also obtain the inverse transpose of the deformation gradient:
F
(Q) = g
)
1
dX I
G
IJ
j
i
x
x
i
1 I
1
dx
= (F ) i
.
X I
ij (
1 J
Applying the relations to 3.4.36 proofs the following corollary concerning the pushforward and pullback of tensor elds on B and S, respectively.
Corollary 4.2.3. (i) Let T Tpq (B), a1 , . . . , ap (T S) and w1 , . . . , w q (T S),
and let : B S be a regular conguration. Then, by dropping the point argument,
T a1 , . . . , ap , w1 , . . . , w q
= T (a1 F ), . . . , (ap F ), (F 1 w 1 ), . . . , (F 1 wq ) .
(ii) On the other hand, if t Tpq (S), B1 , . . . , Bp (T B) and W 1 , . . . , W q (T B),
then
t B1 , . . . , Bp , W 1 , . . . , W q
= t (B1 F 1 ), . . . , (Bp F 1 ), (F W 1 ), . . . , (F W q ) .
71
Example 4.2.4. Let t = tij dxi dxj T02 (S) and W k = WkI X
I (T B), k = 1, 2,
then
t (W 1 , W 2 ) = t ((F W 1 ), (F W 2 ))
i
K
j
L
dX
dX
dx
dx
xk
X I
xl
X J
I
J k l
i
j
dx
= tij W1 W2 F I F J dx
xk
xl
= tij W1I W2J F iI F jJ .
ik
= tij g
, F () dxj F ()
k
x
j
T
ik
,
()
dx
F
()
= tij g
F
xk
= tij F T dxi () dxj F ()
= (F T t F ) (, )
by 3.3.16 and 3.3.20. It is important to keep in mind that for the proposition to be
valid, a metric structure on S is required.
Corollary 4.2.6. Operate analogously to show that if (i) s T20 (S), then s =
F 1 s F T . (ii) Let S T20 (B), then S = F S F T , and (iii) T = F T T F 1
for T T02 (B).
Denition 4.2.7. In numerical applications of continuum mechanics it is the tangent
map T = F of the conguration : B S that is usually given, and not itself.
Therefore, it might be convenient to borrow the notation of denition 3.4.20 and to
replace by F , and by F . For example, if T = R : TP B TQ S is orthogonal
and t T02 (S), then R t = RT t R denotes the pullback of t over pure rotational ,
also called the R-pullback of t.
The deformation gradient includes stretches as well as rigid rotations of the body. There
are dierent choices to split o the stretches and, therefore, dierent strain measures
can be dened. Their applicability depends on the problem and is controversially
discussed. However, all strain measures are symmetric tensors by denition.
72
dX J .
I
X
Proposition 4.2.9. Let g = gij dxi dxj T02 (S) be the local representative of the
spatial metric tensor, then
C = g .
Proof. By 4.2.5, 4.2.2(i) and 3.3.16,
g(P ) = (F T ) (g ) F = (gij )F iI F jJ dX I dX J = C (P ) .
2E (P ) = (C I B )(P ) = C IJ IJ (P )
dX J ,
X I
where I B is the second-rank unit tensor and IJ is the Kronecker delta on B, respectively. Note that E = 12 (C G), where G = GIJ dX I dX J is the metric on
B.
dxj .
xi
In contrast to the right Cauchy-Green tensor, the left Cauchy-Green tensor requires
: B S to be regular (i.e. invertible).
Denition 4.2.12. Let the conguration map be regular, then dene the EulerAlmansi strain tensor or spatial strain tensor e : T S T S through
2e = iS c = ij cij
dxj ,
xi
in which c = b1 is called the Finger tensor, iS is the unit tensor and ij is the
Kronecker delta on S, respectively.
73
2D(P, t) = C(P, t) .
t
Note that an equivalent denition is D = E/t.
i
V j i
gik j k i
j V
V
F
F
+
g
F
+
g
F
.
ij
ij
K
J
K
xj
X K J
X J
V i
X I
to obtain
gik j k i
V F KF J
2GIK D IJ =
j
x
1 j m i gmi
gji
gjm
k
i
+gik (K V )F J V F K F J
+ m
2
xj
x
xi
1 j i k gik gjk gji
i
k
+
k
+gik (J V )F K V F J F K
2
xj
xi
x
(
)
gik
= gik (K V k )F iJ + (J V i )F kK + V j F iJ F kK j
x
1 j i k gki 1 j i k gik
V F JF K j V F JF K j
2
2
x)
( x
k
i
i
k
= gik (K V )F J + (J V )F K ,
g
gij and xijk evaluated at P being understood. Multiplying both sides with GKL and
relabelling the indices then gives the result.
74
2d = g .
t
d
Now setting t
= dt
t makes sense, because the pullback describes a curve of tensors
with parameter t. The result then follows from proposition 4.1.9. (Consult [2], p. 98,
for an alternative proof.)
4.3
The following section should give a precise denition of the Lagrangian and Eulerian
formulations of classic resp. traditional continuum mechanics, and it points to the
dierent use of the term convective.
Denition 4.3.1. A pq -tensor-valued physical eld f on the ambient space S, also
called a spatial eld, is a function
f : S I Tqp (S) ,
where I R. Dependent on the order of p and q, the physical eld thus appears as
a tensor-valued function of (Q, t) (S I), i.e. a time-dependent scalar, vector or
tensor eld f on S. The eld is presumed to be measurable on a subset V S lled
with matter; however, f would be meaningless in an empty space.
75
Corollary 4.3.3. The pq -tensor-valued Lagrangian eld F is understood as a map
F : B I Tqp (S). The dierence between F and f dened vanishes for xed t,
that is,
Ft = ft t : B Tqp (S) ,
so Ft = ft t is a
p
-tensor eld over t .
q
It is notable that the distinction between the Lagrangian formulation and the Eulerian
formulation of the same physical eld is only valid if there are ane point spaces
resp. general manifolds together with a point map. In ordinary vector spaces, the
equation Ft (P ) = ft t does not make sense, as the related tensors have no base
points. However, for simplicity, the dierence between f, F and their localizations
f x1 and F X 1 , respectively, is dropped throughout.
Denition 4.3.4. Another way to formulate the spatial eld f in terms of
ence conguration B is to pull it back by the motion t , such that (t ft )
which is then called the convective representative of f at time t [2, 11, 12].
one better uses the term material rather than convective, because in
community and so within this paper, convective has a dierent meaning.
+ v f
= (Q, t) ,
f (Q, t) =
t
Q
in which (Q, t) denotes a source term, e.g. a constitutive equation. The convective
term v f is the covariant derivative of f along the spatial velocity eld v, and
denotes the connection on S.
It should be clear that v f is due to Q = (P, t), thus legitimating the term material.
Corollary 4.3.6. Let f : S R R be a scalar eld, then
f
f
f
i
+ v[f ] = Lv f .
v
=
f =
t
Q xi
t
Q
Corollary 4.3.7. Let Ft (P ) = ft t be the Lagrangian representative of a spatial eld
f (Q, t). By noting that
F
resp.
ft =
(ft t ) 1
f((P, t), t) =
t
t
t
through the chain rule, an important link between the Lagrangian formulation and the
Eulerian formulation is
Ft
f
1
+ v f
= (Q, t) .
t =
t
t
Q
76
A(P, t) = V (P, t)
t
is called the particle acceleration or Lagrangian acceleration. The spatial or Eulerian
acceleration, dened by at = At 1
v i
Vj
xj
v j
v
v
k j i
+ v v = v
+
v
+
v
v
=
a(Q, t) =
k j
t
Q
xj
xi
t
Q
as desired.
4.4
Proposition 4.4.1. Let the material manifold be a measure space (M, m) with a nonnegative scalar measure m, called the distributed mass, and a mass m(P) R should
be assigned to every part P B of the material body B M. Furthermore, let the
measure be handed down to every orientable
% placement V = t (P) S in the ambient
space at time t I R, and let dv and V dv denote the Riemannian volume form
and the volume measure (Lebesgue measure) of V, respectively, then there is a function
: V I R such that
$
m(P) =
dv .
V
Proof. Note that m(P) = m(V), because P and t (P) are assumed to have the same
mass measure. The rest of the proof can be done with the aid of Radon-Nikodyms
theorem, but it should be omitted here.
Denition 4.4.2. (Q, t) is called spatial or Eulerian mass density at Q V and time
t, and dm = dv is called mass form on V.
77
+ div( v) = 0.
(ii)
t Q
%
Proof. (i) By 0 (U) U, conservation of mass can be written t (U ) (Q, t) dv =
%
(P ) dV . Application of the change of variables theorem 3.7.18 and the formula
U ref
t dv = J dV then gives
$
$
(Q, t) J (P, t) dV =
ref (P ) dV .
U
( J ) = J +
= J + ( div v)J = 0 ,
t
t
where is the material time derivative of (Q, t). Multiplying both sides with 1/J
i
and noting that div( v) = x
i v + div v then proofs the assertion.
Denition 4.4.6. One refers to 4.4.5(ii) as the continuity equation, and to
as its conservative form.
( J )
t
=0
78
J t ((T S)) (T B) ,
Proposition 4.4.9. (Without proof; a proof can be found in [2], p. 117.) Let dV , dv
be the Riemannian volume forms on B and S, respectively, then W is the Piola transform of w, if and only if for a motion t : B S,
t (iwdv) = iW dV .
Theorem 4.4.10 (Piola Identity). Let DIV, div denote the divergence operators on
B and S, respectively, and let W be the Piola transform of w with respect to the motion
t : B S, then
DIV W = (div w t )J .
Proof. Note that by 3.6.34, d commutes with pullback, so by propositions 3.7.31,
4.4.9, and the formula t dv = J dV ,
(div w t )J dV = t ((div w) dv) = t d(iwdv) = dt (iwdv)
= diW dV = (DIV W ) dV ,
that is, DIV W = (div w t )J as desired.
Proposition 4.4.11 (Nansons Formula). Let dA, da be the area forms on positively oriented B and t (B), respectively, then
1
da (Q, t) = (J dA F 1
t ) t .
Proof. dA and da are dened through N dA = dV and n da = dv, respectively, where N and n are the related unit normals (also consider the remark after
3.7.32). By using the property 3.5.11(i), J dV = N (J dA). However, setting
N = t n and applying 3.5.17 yields
J dV = t (n da) = N t da ,
that is, t da = J dA. Note that t da is evaluated on B and has a single covariant
component, so pushing forward and recalling 4.2.6(iii) nally gives (point arguments
are suppressed)
da = t (t da) = t (J dA) = J dA F 1
t .
4.5
4.5.1
79
The notion of objectivity has been introduced in section 2.2 for vectors in at spaces,
in particular Euclidian point spaces, and it should be generalized to arbitrary tensor
elds on dierentiable manifolds. By applying the pushforward and pullback operators
for tensor elds derived in section 3.4.2, objectivity is dened as follows:
Denition 4.5.1. Let S, S be dierentiable manifolds, t Tpq (S) a tensor eld on S
and : S S a dieomorphism, then
t = t
is called the objective transformation of t under the map . t is called invariant under
the transformation , if t = t, where it is being understood that both sides are to
be evaluated at the same point.
80
Denition 4.5.4. Let Q, Q S and R : TQ S TQ S an orthogonal spatial twopoint tensor, then dene the invertible change of framings (Q, v) (Q , R(Q)v) for
every spatial vector v, and (Q, a ) (Q , a R1 (Q)) for every spatial one-form a ,
respectively, by assuming the inverse R1 to exist.
TS
(Q, v)
TS
T (Q, v) = ((Q), D(Q) v) ,
S.
T(Q)
Corollary 4.5.5 phrases that changes of framings dened point by point through 4.5.4
are compatible with superposed rigid motions , that is, they can be interchanged
with the dierential resp. tangent T . Moreover, changes of framings are invertible for
every Q, Q S by denition, even if R is only orthogonal, but not proper orthogonal.
Therefore, changes of framings even if S is Riemannian (curved) and not Euclidian
(at) can be applied to elds of vectors and one-forms on S as pushforwards and
pullbacks (see section 3.4.2). Finally, the pushforward t under a change of framing
is well-dened by 3.4.36:
Denition 4.5.6. Let t Tpq (S), and t = t its (active) objective transformation
under a change of framing that is compatible with a superposed rigid motion by
means of 4.5.5, then t is called Euclidian frame-indierent (EFI). To make a notational
dierentiation, write t = EFI t for the EFI interpretation and t = IRBM t for the IRBM
interpretation of objectivity, respectively.
Denition 4.5.7. A tensor eld t on the ambient space S which transforms objective
under superposed rigid motions : S S, i.e. under spatial isometries, is just called
objective (the same applies to viewed as a change of framing). If t transforms objectively under arbitrary spatial dieomorphisms, then it is called spatially covariant.
Objective resp. covariant tensors become manifest in the mathematical formulation of
physical theories, e.g. the theory of materials and also in general relativity. However,
the principle of objectivity applied to the tensor elds of the theory should not be
confused with that of invariance of the governing equations of the theory. While the
former demands the use only of those tensors which are objective with respect to
some group of transformations, i.e. which are geometric objects, the latter requires the
equations set up to be form-invariant (or generally covariant), i.e. to have the same
form before and after a transformation belonging to the group of transformations.
81
Denition 4.5.8. Let a physical theory be described by some spatial tensor elds
s, t, . . . Tpq (S), and the governing equations of the theory have the form f (s, t, . . .) =
0. The equations are called generally covariant (also called form-invariant or dieomorphism-invariant), if for any dieomorphism : S S,
(f (s, t, . . .)) = f ( s, t, . . .) ,
so that f remains functionally unchanged under the map . A theory is covariant if all
its governing equations are covariant. Let = IdS , then generally covariance requires
the equations to be form-invariant under arbitrary coordinate transformation, which
is then referred to as their passive dieomorphism invariance.
82
V
x
/ V
x(V)
/ x
x
(V) = x(V)
If {xi }, {xi } denote the local coordinate functions of x and x , respectively, and i =
xi x1 is the coordinate transition concerning , then i = xi (x1 x) x1
i . Therefore, the components of the tangent map T are equal to the partial derivai
tives of the relabelling of points x i , and thus for a local representative v(P ) = v i x
i
of the vector eld,
i
v i 1
v = T v 1 =
i
x
xi
by 3.4.33. Because the pushforward by : S S of arbitrary pq -tensor elds on S is
realized in a similar way setting i i , the assertion follows.
The gauge freedom underlying the principle of general covariance is an important
ingredient to restrict or reduce the governing equations of a physical theory considerably
which is basically the main goal of the endeavor. However, if the particular equation
f (s, t, . . .) = 0 is physically substantial or not depends on the physical theory under
consideration.
4.5.2
Material Frame-Indierence
In continuum mechanics, a typical application of the principles of objectivity and general covariance can be found in the material frame-indierence of constitutive equations
of materials. The remainder of the section will be devoted to this topic, but only a few
key facts will be outlined using the example of simple hyperelastic material.
Denition 4.5.10. A hyperelastic or Green elastic material is an ideally elastic material that relates the conguration resp. deformation : B S of the material body
B at every particle P B (time-dependency will be again omitted) to the stress by
the stored energy at P .
83
Proposition 4.5.11. (Without proof; cf. [2]) A hyperelastic constitutive equation can
locally be written in the form
P = P(P, F ) ,
in which F is the deformation gradient at P B, and P : TP B T(P ) S is called the
rst Piola-Kirchho stress tensor.
Note that P is a two-point tensor having the one leg at P and a spatial leg at
Q = (P ) S.
Normally, the theory of materials concerns the elds of stress and deformation rather
than their point values. The hyperelastic constitutive equation then becomes P =
P(F ), in which P : B T S T B now is the rst Piola-Kirchho stress eld, and
the constitutive function is a map
P: E F,
where E is the bundle over B whose bre at P B is T(P ) S TP B, and F = T S
T B.
Denition 4.5.12. Let : S S be an arbitrary spatial dieomorphism resp. a
superposed motion taking spatial points Q to Q (equivalently, can be understood as
a change of framing), then the hyperelastic constitutive equation P = P(F ) is called
spatially covariant provided that
P = P(F ) ,
where F (P ) : TP B TQ S and Q = (Q) = ((P )) S. Spatial covariance means
that the constitutive equation P itself is form-invariant under spatial dieomorphisms.
If the spatial dieomorphism is replaced by a superposed rigid body motion, i.e. a
spatial isometry such that TQ : TQ S TQ S is proper orthogonal and P = P(F ) still
holds, then the constitutive equation is called material frame-indierent (MFI).
Material frame-indierence in Euclidian point spaces has often been the matter in
dispute in the last decades, and it is usually stated as an independent assumption
or axiom. However, it can be derived from the more general principles of Euclidian
frame-indierence and form-invariance:
Proposition 4.5.13. (See also [29, 30].)
(i) Both Euclidian frame-indierence (EFI) and form-invariance (FI) of constitutive
equations together demonstrate material frame-indierence (MFI); conceptually, EF I+
F I = MF I.
(ii) Indierence of constitutive equations with respect to superposed rigid body motions
(IRBM) is equivalent to material frame-indierence; conceptually, IRBM = MF I.
Proof. (i) F = F = T (F ) 1, so F transforms objectively (clearly, the spatial
leg of F ), and since P is a two-point tensor of the same class, it also does. Now
84
P = EFI P demands (P(F )) = EFI (P(F )) from the constitutive equation. On
the other hand, form-invariance forces EFI (P(F )) = P(EFI F ) = P(F ), therefore,
P = P(F ) as desired.
(ii) Indierence of the constitutive equation P with respect to superposed rigid body
motions means that IRBM P = P(IRBM F ). As shown in (i), however, P and F
transform objectively in the same manner, so IRBM and MFI are equivalent.
Basically, proposition 4.5.13(i) derives material frame-indierence from the relative
motion of dierent Euclidian observers, whereas 4.5.13(ii) involves two motions of a
body the original motion and the motion overlayed by a rigid one with respect
to the same observer. However, both viewpoints are equivalent, except for the fact
that the rst requires the linear transformation of both frames to be orthogonal, while
the latter includes motions with proper orthogonal tangents only, since rigid motions
necessarily are orientation-preserving.
In their encyclopedia ([8], appendix 19A), Truesdell and Noll summarize that the form
4.5.13(i) is historically connected with the names Zaremba and Jaumann, and 4.5.13(ii)
is a form that has essentially been proposed by Hooke, Poisson and Cauchy. Truesdell
and Noll [8] side themselves with Zaremba and Jaumann, as Noll [31] did already
before; he called material frame-indierence at that time the principle of objectivity
of material properties. Therefore, the following notational dierentiation would be
convenient.
Denition 4.5.14. Refer to the fact 4.5.13(i) as the Zaremba-Jaumann-Noll form,
and to 4.5.13(ii) as the Hooke-Poisson-Cauchy form of material frame-indierence.
The reader probably knows that material frame-indierence is a useful concept to
reduce the form of constitutive equations, but not all materials comply with it. For
example, kinetic gas does not fulll invariance with respect to superposed rigid body
motions and thus it is not material frame-indierent [29, 30].
4.6
Large deformation initial boundary value problems in continuum mechanics are usually
solved by applying the nite element method. In this context, however, the classical
Lagrangian and Eulerian formulations have some shortcomings. In the Lagrangian
nite element formulation, excessive element distortions may occur that may lead to
unstable and inaccurate numerical analyses, or even terminate the calculation. From
the Eulerian viewpoint, the discretized domain is xed in space and, therefore, following
free surfaces and moving material interfaces becomes a cumbersome task.
The Arbitrary Lagrangian-Eulerian (ALE) formulation succeeds in combining the advantages of both classic Lagrangian and Eulerian viewpoints by choosing the nite
element mesh as a time-dependent reference domain dierent from the material (Lagrangian) and spatial (Eulerian) congurations [20, 21, 22]. For the last two decades the
85
ALE framework has been developed to a powerful analysis tool for large deformation
problems, especially metal forming processes, free surface ows and uid-deformable
structure interaction.
Consistent with the geometrical overall context of the paper, the reference domain
underlying the ALE formulation is introduced as a grid manifold that is embedded in
the ambient space, and which is independent of the material and its congurations.
The ane connection and the inner product of the ambient space are used to establish a
covariant derivative on the reference domain and to enable measurement of deformation
of the grid, respectively. It has to be noted that this additional functionality is not
intrinsic, whether to the underlying material manifold, nor to the grid manifold, but
it is induced to the latters congurations in space.
The important convective velocity eld, which is well-known in the ALE community,
will be dened subsequently as a section of the tangent bundle of the ambient space.
Dierential geometry, then, reveals three important facts resulting from the relative
motion between the body and the reference domain. First, the pullback of the convective velocity concerning the relative motion establishes the so-called referential velocity
eld, which is essential for the denition of a material time derivative on the reference
domain. Second, the fundamental ALE operator takes advantage of an induced connection on the reference domain given by the Levi-Civita connection of the ambient
space and, third, mesh regularization or optimization of the mesh, in context of the
ALE nite element method, can be interpreted as the numerical representative of a
time-dependent ow of the reference domain in the ambient space.
4.6.1
The required homeomorphism ensures the one-to-one correspondence of grid nodes and
particles of the body. Hence, the grid manifold G is also a continuum and the reference
grid R inherits the topology of the body. The grid cells are innitely small and every
grid node has a neighborhood containing other grid nodes. Descriptively, the grid is a
prototype of a manifold: it has no further properties besides its topology.
Due to the arbitrary choice of the reference grid, it does not make sense to dene an
inner product or an ane connection on the grid manifold. But in order to specify the
state of the grid, that is, stretches of the grid cells and distances of the grid nodes, R
is embedded in the ambient space in the same way as the material body (gure 4.4):
Denition 4.6.2. Considering that the ALE formulation has its origin in computational mechanics, one may refer to M, dened by the map : G R M S and
being dierentiable, as the computational representative or the model of B, that is, M
86
Q
coordinate indices related to the reference domain will be denoted by lower case Greeks,
with , , . . . {1, 2, 3}. If appropriate, vectors, tensors etc. will be marked with a
hat.
Denition 4.6.4. Let the spatial coordinates xi be given as functions of the C 1 grid
motion t , that is, it (M) = xi t 1 , then
it
= ti (M) i (Q)
t (M) =
i
t M x
x
denes the grid velocity over t . The corresponding spatial grid velocity eld t =
t 1
and t (Q) = (Q, t) at xed t. Then t and t are called the relative map and the
referential map of R = t (M), respectively, provided that
t = t t1 ,
that is, the physical motion t is a superposed motion of the body relative to the
reference domain. Both t and t are in general explicitly time-dependent.
The next theorem states how to specify the ow of the material particles via the ow
of the reference points.
87
material body
t
S
reference
conguration
t (B)
model of B
R = t (M)
current
conguration
reference domain
(ALE)
t
R
reference grid
Figure 4.4: General Arbitrary Lagrangian-Eulerian formulation: reference grid, congurations and related mappings.
88
t (B)
t (M)
s (M)
t,s = t s1
Figure 4.5: Flows associated with the motion t of the body B and the motion t of
its model M in the ambient space.
Theorem 4.6.6 (Compatibility of Flows). Let t,s be the time-dependent ow of
the spatial grid velocity , and t , t the relative map and referential map of R =
of the regular physical motion t :
t (M), respectively, then the ow t,s = t 1
s
B S can be obtained from
t,s = t t,s 1
s .
Proof. Since t (B) and t (M) evolve synchronously, compatibility of the maps requires (see gure 4.5)
1
t t 1
s = t s s ,
t, s R .
t, s R .
Substitution into the previous equation and composing both sides with 1
s from the
right then gives the result.
Example 4.6.7. In a numerical implementation when using the nite element method, the discretization of the domain in its current conguration, i.e. the deformed
nite element mesh at time t, is a numerical representative of the reference domain
R = t (M). A mesh regularization or an r-adaption of the mesh that preserves the
mesh topology, then, represents the discretization of the time-dependent ow t,s :
s (M) t (M).
The localizations of the relative map t : R S and the inverse referential map
t1 : B R should be discussed next.
89
x 1
(t )x = x t 1
(1 (V)W ) : 1
t (V) W
t (V) W
t
and
(t1) X = t1 X 1
X(t (W )U ) :
X (t (W ) U) (t (W ) U) ,
respectively for those who want to make a dierence between t and (t1 )1 : the
)
latter is meant here. If {xi }Q = x(Q) are the coordinates of Q V and { }Q = x(Q
= t1(P ) W , then abbreviate
the coordinates of Q
it = xi t 1
and
(t1) = t1 X 1 ,
= i (Q,
t) and (t1 ) (P ) = ( 1 ) (P, t), respectively.
where it (Q)
resp.
(t1 )
= t
t
= t (Q)
(T R) ,
P
and
it
i
1
=
w
(Q)
= wt
t
t
t Q
xi
xi
(T S)
The relative velocity w t might be confused with the grid velocity t , but notice there
is the following relation:
90
(it t )
it k
i
=
w
+
.
t
t
t
t
kt t
M
Corollary 4.6.12. By 4.1.8, the time-dependent ow of wt is t 1
s , where t, s R.
Denition 4.6.13. The tangent of the map t is
T t : T R T (t (R)) T S
it
,
xi
and its inverse is given by
T (1
t ) : T S T (t (R)) T R
(1
t )
,
i
i
x
x
1
1
1
in which (1
t ) = t x , i.e. (t ) (x(Q)) = t .
Again, the case should be mentioned where dim(R) < dim(S). Since t is an embedding and t (R) S is a submanifold, T t (T R) = T (t (R)) T S is a subspace of
i
R (cf. denitions 3.1.22 and
T S, so that the Jacobian matrix t is invertible at Q
3.1.26, and section 4.1).
Denition 4.6.14. The two-point tensor elds F t (t T B T R) and F t
(t T S T R) associated with the tangent maps T t and T t , respectively, are
dened by
I
t) = t d
F (Q,
X I
and
i
t) = t d ,
F (Q,
xi
where F 1
t (Q, t) =
(t1 )
X I
Theorem 4.6.16. Let v be the spatial velocity of the motion of a material body, w
the spatial velocity of the relative map (i.e. the relative velocity) and the referential
velocity (see also 4.6.10 for denitions), then
v t wt = t t .
91
Proof. Substituting 4.6.9 for the components of the spatial velocity v t (Eulerian
velocity eld) gives in detailed expression
it
(xi t (1 ) t1 )
1
1
=
t =
t
t
P
t
P
i
1
t
(t )
1
1
t
(Q) +
=
t
(Q)
t (Q)
(Q)
t
Q
t
P
i
t
i
1
= wt (Q) +
t
(Q) .
t
vti(Q)
Recalling 3.4.33, it should be clear that the second term on the right hand side includes
the components of the pushforward of the referential velocity eld. More general, for
vector elds (T R) one has
1
t = T t 1
t = (F t ) t
(T S) ,
The convective velocity c should not be confused with the Finger tensor (see denition
4.2.12). However, the meaning will be clear from the context.
Note that the convective velocity is a proper vector eld on S, and it provides a
fundamental link between the body, its congurations and the reference domain. It
denotes the relative velocity between the particles P = 1
t (Q) and the reference points
1
Lv t = t + w+ t = t + wt + ( ( t)) ,
t
t
92
where the time index is omitted for notational convenience. By 3.6.24, 4.6.10, and
the corollaries 3.4.32 and 4.6.12,
t + wt + ( ( t))
t
d
1
= (t s ) tt
+ ( ( t))
dt
t=s
d
+ ( ( t))
= t (1
s ) (t (t tt ))
dt
t=s
d
= s (t tt )
+ ( ( t))
dt
t=s
d
=
( tt )
+ ( t) = (L ( t)) .
dt t
t=s
Lv t =
4.6.2
resp.
Ft = ft t1
p
at xed t. The map f : R I Tqp (S), where ft 1
t = ft Tq (S), is then called the
Arbitrary Lagrangian-Eulerian (ALE) formulation of the physical eld f . In an ALE
R serve as the independent point variables.
formulation, the reference points Q
material time derivative of the Eulerian eld f (Q, t) has been dened
f
through f = t Q + v f , in which the rst term on the right hand side represents the
local time derivative at xed Q, and the second term (the convective term) denotes
the covariant derivative of f along the spatial resp. Eulerian velocity eld v. In what
t) should be established.
follows, the material time derivative of the ALE eld f(Q,
Proposition 4.6.20. Let f (Q, t) = ft (Q) be (the Eulerian formulation of ) a physical
eld, and t at least C 1 , then the material time derivative of the ALE eld ft = ft t
on the reference domain R is
t) ,
f =
= (Q,
+ f
t
Q
93
t) is a source
in which is the referential velocity eld from denition 4.6.10 and (Q,
term accounting for some corresponding response function (e.g. a constitutive equation).
f is the covariant derivative of f along in terms of the t -induced connection
on R.
Proof. Proving the local time derivative is trivial, but the convective term bares
some intricacy. Recall that the Eulerian eld ft Tpq (S) has been dened in 4.3.1 as
an honest tensor eld on S. Therefore, the corresponding ALE eld ft = ft t is a
tensor eld over t , that is, an induced section of the tensor bundle Tqp (S) (see denition
3.4.18). As proposed by theorem 3.6.13 in section 3.6, the covariant derivative of an
induced section calls for a connection on R that is induced by the connection
on S via the map t .
Without loss of generality, for the rest of the proof it will be considered the case where
ft (T S) is a vector eld. In a chart (V, x) on S, ft has the local representative
(t T S) ,
(fti t )
j
i
k
+ (ft t ) j k t (F t ) d i .
x
by theorem
Therein, j ik (Q) are the coecients of on S, and (j ik (F t )k )(Q),
3.6.13, are the coecients of on R induced by t . Moreover, note that (fti t )
of the reference domain
above functionally depends on the coordinates { }Q = (Q)
R.
ft =
vector eld must also be a vector eld over t for the proposed material time derivative
for f to make sense. This, however, can only be realized by a honest vector eld on R.
For example, taking the referential velocity eld t = t (T R) gives
f (Q)
k
(f i t )
i
t
j
+ (f t ) j k t
xi
(t T S) .
On the other hand, for t = IdS , the relative map t : R S just becomes a coordinate
transformation or relabelling of points, and the proposed material time derivative for
f must reduce to the material time derivative for the Eulerian eld f dened in 4.3.5.
However, in this case one also has t = IdS t1 by 4.6.5, and thus v = (IdS ) by
4.6.10. Therefore, the referential velocity is the proper vector eld to accomplish the
convective term f, and to proof the assertion.
94
F
f
t) .
= (Q,
=
+ (c) f
t
t
Q
1
t .
resp.
f ( (P, t), t) =
ft =
ft t
t
t
In addition, from theorem 3.6.13, and noting that ct t = T t ( t ) by 4.6.17,
f = (c) f .
Use both relations to replace the respective terms in 4.6.20, and nally obtain the
desired result.
Finally, the substitution of the ALE operator for material time derivatives in the balance equations of continuum mechanics enables the solution of initial boundary value
problems with respect to an arbitrary time-dependent reference domain.
Proposition 4.6.22. Let the material body B, the reference domain R and the ambient
t) and (Q, t) be the mass densities on
space S have the same dimensions, and let (Q,
R and S, respectively, then conservation of mass in ALE setting reads
i
+
c + div v t = 0 .
t
Q
xi
Proof. The mass densities and are scalar elds, hence the covariant derivative is
i
t
i
=
.
( )(Q)
= i =
c t = ((c) )(Q)
x
xi
Also note that from proposition 4.4.5(ii) and by using the relation ft = ft t1 1
t ,
continuity reads
t t1 1
t = div v ,
By 4.6.21, one is allowed to write
t1 1
+ i ci + div v = 0 ,
t
t Q
x
where the left hand side is evaluated at spatial points Q S. However, since each
term is a proper scalar, the assertion follows by switching the point arguments.
Chapter 5
Summary and Conclusions
After an introduction to modern dierential geometry, some geometric concepts in
continuums mechanics have been reviewed. In particular, the formulation of kinematics
of a body and balance equations benet from the precise geometric terminology. On
the one hand, a clear and precise distinction of the Lagrangian formulation from the
Eulerian formulation can be drawn. On the other hand, by using a general term
of space, those issues can be revealed, for which the introduction of a metric, an
ane connection or other geometric structures physically make sense, or merely lead
to simplications.
Subsequent to the classic formulations according to Lagrange and Euler, an implementation of the geometric concepts in the context of the Arbitrary Lagrangian-Eulerian
formulation has been presented. As an essential component, the introduced grid manifold facilitates a consistent description of the relations between the material, the ambient space and the reference domain. This also concerns the numerical implementation
by using the nite element method. The grid manifold establishes the topology of
the reference domain respectively the nite element mesh, whereas the time-dependent
placement of the grid in the ambient space represents the conguration of the mesh,
owning the physical properties and the state of the material body. The continuous mesh
regularization in the course of an ALE nite element simulation has been formulated
geometrically as a ow of the grid nodes in the ambient space.
By dening mappings between the introduced manifolds namely the material body,
its congurations and the reference domain, pushforward and pullback operators have
been obtained that provide the mapping of arbitrary spatial and material tensor elds
onto the reference domain. As a fundamental result, the convective velocity eld on the
current conguration of the body, which is well-known in the ALE community, emerges
from the pushforward of the referential velocity eld over the relative motion between
body and grid. Finally, the paper uncovered that the fundamental ALE operator
involves the notion of an induced connection, so that ALE computation can be done
without dening an extra ane connection on the reference domain.
96
Bibliography
[1] R. Abraham, J. E. Marsden, and T. Ratiu. Manifolds, Tensor Analysis, and
Applications. Addison-Wesley, Reading, MA, 1983.
[2] J. E. Marsden and T. J. R. Hughes. Mathematical Foundations of Elasticity. Dover
Publications, New York, 1994. (Originally published: Prentice-Hall, 1983).
[3] R. L. Bishop and S. I. Goldberg. Tensor Analysis on Manifolds. The Macmillan
Company, New York, 1968.
[4] M. Spivak. A Comprehensive Introduction to Dierential Geometry, volume 1.
Publish or Perish, Inc., Houston, Texas, 3rd edition, 1999.
[5] J. L. Synge and A. Schild. Tensor Calculus. Dover Publications, New York, 1978.
(Originally published: University of Toronto Press, 1949).
[6] C. W. Misner, K. S. Thorne, and J. A. Wheeler. Gravitation. W. H. Freeman and
Company, San Francisco, 1973.
[7] P. K. Raschewski. Riemannsche Geometrie und Tensoranalysis. Verlag Harri
Deutsch, GmbH, 2nd edition, 1995. (in German).
[8] C. Truesdell and W. Noll. The Non-Linear Field Theories of Mechanics. SpringerVerlag Berlin Heidelberg New York, 3rd edition, 2004.
[9] R. A. Sharipov. Course of Linear Algebra and Multidimensional Geometry. Publications of Bashkir State University, Ufa, Russia, 1996. URL
http://arxiv.org/abs/math.HO/0405323.
[10] D. Lau. Algebra und Diskrete Mathematik 1. Springer-Verlag Berlin Heidelberg
New York, 2nd edition, 2007. (in German).
[11] J. C. Simo and J. E. Marsden. On the rotated stress tensor and the material version
of the doyle-ericksen formula. Archive for Rational Mechanics and Analysis, 86
(3):213231, 1984.
[12] J. C. Simo, J. E. Marsden, and P. S. Krishnaprasad. The hamiltonian structure
of nonlinear elasticity: The convective representation of solids, rods, and plates.
Archive for Rational Mechanics and Analysis, 104(2):125183, 1988.
[13] A. Yavari, J. E. Marsden, and M. Ortiz. On spatial and material covariant balance
laws in elasticity. Journal of Mathematical Physics, 47(042903), 2006.
97
98
BIBLIOGRAPHY
[14] M. Epstein and R. Segev. Dierentiable manifolds and the principle of virtual work
in continuum mechanics. Journal of Mathematical Physics, 21(5):12431245, 1980.
[15] R. Segev. Forces and the existence of stresses in invariant continuum mechanics.
Journal of Mathematical Physics, 27(1):163170, 1986.
[16] R. Segev. The geometry of cauchys uxes. Archive for Rational Mechanics and
Analysis, 154(3):183198, 2000.
[17] B. Riemann. On the hypotheses which lie at the bases of geometry. Nature, 8
(183-184):1417, 36, 37, 1873. (Reprint, Translated by W.K. Cliord).
[18] J. Earman and J. D. Norton. What price spacetime substantivalism? the hole
story. British Journal for the Philosophy of Science, 38:515525, 1987.
[19] J. D. Norton. General covariance and the foundations of general relativity: Eight
decades of dispute. Reports on Progress in Physics, 56(7):791858, 1993.
[20] T. J. R. Hughes, W. K. Liu, and T. K. Zimmermann. Lagrangian-eulerian nite element formulation for incompressible viscous ows. Computer Methods in
Applied Mechanics and Engineering, 29:329349, 1981.
[21] J. Huetink, P. T. Vreede, and J. van der Lugt. Progress in mixed eulerianlagrangian nite element simulation of forming processes. International Journal
for Numerical Methods in Engineering, 30:14411457, 1990.
[22] J. Donea, A. Huerta, J.-Ph. Ponthot, and A. Rodrguez-Ferran. Arbitrary
Lagrangian-Eulerian Methods, volume 1 of Encyclopedia of Computational Mechanics, chapter 14. John Wiley & Sons, Ltd., 2004.
[23] T. Belytschko, W. K. Liu, and D. Moran. Finite Element for Nonlinear Continua
and Structures. John Wiley & Sons, Ltd., 2000.
[24] A. Einstein. Die formale Grundlage der allgemeinen Relativit
atstheorie. Sitzungsberichte der Preuischen Akademie der Wissenschaften zu Berlin, pages 1030
1085, 1914. (in German).
[25] L. Florack. Image Structure. Kluwer Academic Publishers Dordrecht Boston
London, 1997.
[26] N. Steenrod. The Topology of Fibre Bundles. Princeton University Press, 1951.
[27] W. Noll. Materially uniform simple bodies with inhomogeneities. Archive for
Rational Mechanics and Analysis, 27(1):132, 1967.
[28] W. Noll. A new mathematical theory of simple materials. Archive for Rational
Mechanics and Analysis, 48(1):150, 1972.
[29] B. Svendsen and A. Bertram. On frame-indierence and form-invariance in constitutive theory. Acta Mechanica, 132:195207, 1999.
[30] A. Bertram and B. Svendsen. On material objectivity and reduced constitutive
equations. Archives of Mechanics, 53(6):653675, 2001.
BIBLIOGRAPHY
99
Erschienene Hefte:
Herausgegeben von H. Mller-Kirchenbauer
[1] Aufbau und Arbeitsgebiete des Instituts 1977, 166 S.
ISBN 3 7983 578 1
(vergriffen)
3,00
4,50
3,50
4,50
4,00
4,00
2,50
4,00
8,00
12,00
8,00
10,50
10,00
11,50
10,50
10,50
10,50
10,50
10,50
10,50
15,50
15,50
[24] Selbstheilungsvermgen toniger Erdstoffe in Dichtungssystemen von Deponien bei Schdigung durch
mechanische Risse und Austrocknung
Mallwitz, K., 1996, 266 S.
ISBN 3 7983 1693 7
18,00
15,50
15,50
[27] Untersuchungen zum Filtrationsverhalten von Feinstbindemittelsuspensionen bei der Injektion in Sande
Mittag, J., 2000, 208 S.
ISBN 3 7983 1855 7
18,00
[28] Numerische Verfahren zur Berechnung von Baugrund-BauwerkInteraktionen im Zeitbereich mittels GREENscher Funktionen
fr den Halbraum
Bode, C., 2000, 168 S.
ISBN 3 7983 1856 5
16,00
31,00
15,50
18,00
18,00
19,00
19,00
25,00
16,50
18,00
45,00
45,00
45,00
45,00
49,80
48,80
15,00
Grundbauseminar
Vorstellung der neuen EAB mit dem Teilsicherheitskonzept, 2007, 236 S
25,00
25,00