You are on page 1of 18

Article

Mathematics and Mechanics of Solids


2015, Vol. 20(5) 522–539

Some remarks on metric and © The Author(s) 2013


Reprints and permissions:
sagepub.co.uk/journalsPermissions.nav
deformation DOI: 10.1177/1081286513506432
mms.sagepub.com

Salvatore Federico
Department of Mechanical and Manufacturing Engineering, The University of Calgary, Alberta, Canada

Received 17 August 2013; accepted 22 August 2013

Abstract
This work is aimed at emphasising the relationship between metric and deformation, under the light of a novel formalism
for the Polar Decomposition Theorem. All results are first presented in the classical formalism of Cauchy’s celebrated
theorem, and then in the proposed alternative formalism. Although the latter requires a little more work to be estab-
lished, it allows for directly defining all strain tensors as “covariant”, i.e. with both feet being covectors. Emphasis is also
placed on how, in the absence of the metric structure, the available mathematical tools are restricted to the deformation
gradient alone. Along with these main results, and in the didactical intention that permeates this work, several hints are
given, which could be useful in teaching Continuum Mechanics, e.g. the rigorous definition of the determinant of the
deformation gradient in Riemannian manifolds, and a caveat on the definition of the spatial Hencky logarithmic strain.
The setting is that of modern Continuum Mechanics, based on the description given by Differential Geometry in terms
of differentiable manifolds. However, passing to the simpler case of affine spaces takes almost no effort, paying attention
to keeping the distinction between vectors and covectors, and therefore allowing the matrices representing the metric
tensors to differ from the unit matrix.

Keywords
Metric, metric tensor, deformation, strain, generalised strain, polar decomposition

1. Introduction
In an oversimplified manner, deformation and strain can be defined as a “change of length” and a “relative
change of length”, respectively. From these elementary definitions, it follows that the notion of metric, intended
as the capability of measuring distances, is indispensable when introducing the concepts of deformation and
strain. In Geometry and Mechanics, normally the notion of metric is associated with that of metric tensor,
which is based on the idea of the familiar Euclidean distance.
When Cartesian coordinates are employed and an orthonormal basis is assigned, the metric tensor is repre-
sented by the unit matrix. Therefore, one can pretend that the metric tensor is “invisible”, can “forget” about
the distinction between a vector space and its dual space, i.e. between vectors and covectors, and can identify
all second-order tensors with the associated matrices in the given orthonormal basis. In this setting, the Green–
Lagrange and the Almansi strains are introduced starting from the right and left Cauchy–Green deformation
tensors, which are in turn obtained by pre- and post-multiplying the deformation gradient by its transpose,
respectively. Furthermore, one can introduce Hill’s generalised strains [1, 2] by employing Cauchy’s classical

Dedicated to the young Leo Federico, born on 14 May 2012.

Corresponding author:
Salvatore Federico, Department of Mechanical and Manufacturing Engineering, The University of Calgary, 2500 University Drive NW, Calgary, AB,
Canada T2N1N4.
Email: salvatore.federico@ucalgary.ca
Federico 523

Polar Decomposition Theorem, which states that any tensor can be decomposed into the product of an orthog-
onal tensor and a symmetric tensor or as the product of another symmetric tensor and the same orthogonal
tensor: the powers of the two possible symmetric tensors are used to define strains of every order, and the
Green–Lagrange and Almansi strains are retrieved as the strains of order two. This is the procedure used, for
example, in the classical text on non-linear elasticity by Ogden [2] and the introductory book by Bonet and
Wood [3].
In the more general case of curvilinear coordinates, or in the even more general case of differentiable Rie-
mannian manifolds, the metric must be explicitly employed in all passages described above. To be more precise,
one should speak of two metric tensors: one in the reference configuration (or the body manifold) and one in
physical space (or the space manifold).
This work is aimed at pointing out that there are two ways to define the right and left Cauchy–Green defor-
mation tensors (and therefore two ways to define the Green–Lagrange and Almansi strains), two ways to express
the polar decomposition Theorem, and consequentially two ways to define Hill’s generalised strains.
We shall show that this is closely connected with the distinction between the algebraic transpose and metric
transpose of a tensor [4], which correspond to the dual and simply the transpose, respectively, in the terminology
used by Marsden and Hughes [5].
The proposed alternative form of the Polar Decomposition Theorem allows us to directly relate its results
to the geometric definition of the fundamental deformation tensors as push-forwards/pull-backs of the metric
tensors.
In passing, we shall remark how, in pre-metric Continuum Mechanics [6], the only available tool for the
study of deformation is in fact the deformation gradient, the tangent map of the configuration. Moreover, we
shall give a rigorous definition of its determinant (the volume ratio) in Riemannian manifolds as well as in
manifolds without metric but with a non-vanishing volume form.
In this paper, the general notation on tensors and tensor spaces is based on that presented in a previous work
[4], which also contains a brief summary of the necessary concepts of linear algebra and tensor algebra.
Here, we refer to the general case in which the body B and the space S are differentiable manifolds, or
Riemannian manifolds, when the metric structure is present. If one wants to avoid invoking differentiable and
Riemannian manifolds, the physical space S can be regarded as an affine space modelled after R3 , and the body
and any arbitrary reference configuration as open subsets of S. Roughly speaking, an affine space A is a set,
called the point space, in which each couple (x, y) of points x, y ∈ A is mapped into an element u = y − x
of a vector space V, called the supporting or modelling space (see, e.g., [7]). In practice, one obtains a set in
which it is possible to attach a vector u = y − x at every point x, and the vector space of all vectors attached at
one point x is called the tangent space, Tx A, at that point. The dual of Tx A is called the cotangent space Tx A
and the collections of all tangent and cotangent spaces for all points x ∈ A are the tangent bundle TA and the
cotangent bundle T  A. When the point space and the supporting space are both R3 , one obtains the usual affine
space E3 of Classical Mechanics.

2. Pre-metric Continuum Kinematics


This section is an introduction to Continuum Kinematics based purely on the concepts of body and configura-
tion, with no reference to a metric structure. The reader will recognise that many familiar objects do not belong
to this picture, and can only be introduced when a metric is available, as it shall be shown in Section 3.
In the modern theory of Continuum Mechanics [5–9], the physical space S and a body B are differen-
tiable manifolds, and a configuration of the body is a diffeomorphism (i.e. an invertible differentiable map with
differentiable inverse)
φ : B → S, (1)

describing the deformation of the body. The image of φ, denoted φ(B), is also called a configuration of B,
whenever there is no danger of confusion with φ itself.
Although not mandatory, it is customary to choose one particular configuration φR , with image BR = φR (B),
and call it the reference configuration. In this way, one can study the deformation of the body B in terms of its
reference configuration BR , and define the map

χ = φ ◦ φR−1 : BR → S. (2)
524 Mathematics and Mechanics of Solids 20(5)

This shows that the treatment in terms of the body manifold or of an arbitrary reference configuration is exactly
the same, up to a diffeomorphism, which could even be viewed as a mere change of coordinates (see, e.g.,
Epstein and Elżanowski [10]).
A motion of the body is a configuration parameterised by time, i.e. a map
φ : B × R+
0 → S : (X , t) → x = φ(X , t), (3)
which in turn defines
φt : B → φ(B, t) ⊂ S : X → x = φt (X ) = φ(X , t), (4)
mapping B into the image Bt = φt (B) = φ(B, t), called current configuration at time t.
Once a point X and a time t have been fixed, the tangent map of the configuration is the linear map
F(X , t) = Tφ(X , t) : TX B → Tx S, (5)
such that its action on a tangent vector U ∈ TX B yields the directional derivative of the configuration φ with
respect to U:
F(X , t) U = [Tφ(X , t)] U = (∂U φ)(X , t) ∈ Tx S. (6)
In the traditional nomenclature of Continuum Mechanics, F takes the name of deformation gradient. However,
since the configuration φ is not a vector field, F is actually not a gradient [5] and, indeed, its component form
does not involve covariant differentiation, as shown below.
Technically, the directional derivative (6) is defined in terms of its representation in systems of coordinates
in B and S,
F a A (X , t) = φ a ,A (X , t), (7)
so that F(X , t) has tensor representation
F(X , t) = φ a ,A (X , t) ea (φ(X , t)) ⊗ EA (X ), (8)
where {EA }3A=1 and {ea }3a=1 the vector bases induced by the chosen systems of coordinates inB and S, respec-
tively,and {EA }3A=1 and {ea }3a=1 are the associated covector bases. Note that, when a system of coordinates is
chosen in S, each of the φ is a scalar field and its differential Tφ is indeed a gradient: in this view, F is the
a a

collection of three gradients, and this may be seen as a justification for naming it the “deformations gradient”.
As a two-point tensor, F(X , t) belongs to Tx S ⊗ TX B and, as a two-point tensor field,
F( · , t) : B → TS ⊗ T  B. (9)
By definition, the (algebraic) transpose of F is the two-point tensor field
FT ( · , t) : φ(B, t) → T  B ⊗ TS, (10)
such that, for every material vector field U (i.e. valued in TB) and every spatial covector field α (i.e. valued in
T  S),
U FT α = αFU. (11)
Notice that, in the definition (10), with respect to the definition of F (equation (9)), one switches TS and T  B
in the tensor product of spaces.
Given a material tensor field Z, valued in [TB]r s , its push-forward φ∗ [Z], valued in [TS]r s , is obtained by
contracting every vector foot (contravariant index) of Z with F and every covector foot (covariant index) with
F−T , i.e. in components,
(φ∗ [Z])a1 ...ar b1 ...bs = F a1 A1 . . . F ar Ar (F−1 )B1 b1 . . . (F−1 )Bs bs ZA1 ...Ar B1 ...Bs . (12)
Conversely, given a spatial tensor field Z, valued in [TS]r s , its pull-back φ ∗ [Z] is the tensor field, valued in
[TB]r s , obtained by contracting every vector foot (contravariant index) of Z with F−1 and every covector foot
(covariant index) with FT .
(φ ∗ [Z])A1 ...Ar B1 ...Bs = (F−1 )A1 a1 . . . (F−1 )Ar ar F b1 B1 . . . F bs Bs Za1 ...ar b1 ...bs . (13)
It is important to remark that no other measure of deformation can be introduced at this stage, as there is not
enough structure to do so. When the metric structure is not available, the deformation gradient is the only tool
one is equipped with. Even defining its determinant J is not trivial, although certainly possible, by invoking the
existence of a non-vanishing volume form in both B and S, as it shall be pointed out in Remark 5.2.
Federico 525

3. The introduction of the metric structure


If the differentiable manifolds body B and space S are equipped with metric tensor fields G and g, respectively,
they become Riemannian manifolds. This section is dedicated to elucidating some immediate consequences of
the introduction of the metric structure, that maybe have not been given enough emphasis in the literature. All
results are presented in the context of Continuum Mechanics, but can be adapted easily to generic tensors of
any type.
A metric tensor g on S is a symmetric, positive definite second-order tensor field valued in [TS]02 , i.e. it is a
bilinear form such that g(w, y) = g(y, w) for every w, y ∈ TS, and g(w, w) > 0 for every non-zero w. The metric
tensor defines the scalar product between two vectors as

w.y = w, y
g = g(w, y) = w g y = wa gab yb , (14)

The metric tensor also allows the contraction of two tensors of any order A and B in the case in which the last
foot of the first tensor and the first foot of the second tensor are both vectors; this type of contraction is denoted
by the same low dot “.” used for the scalar product of two vectors, resulting in A.B. Similarly, the inverse metric
tensor g−1 defines the scalar product between two covectors, and allows the contraction of two tensors in the
case in which the last foot of the first tensor and the first foot of the second tensor are both covectors. Full details
and examples on the contraction performed with the low dot are provided in a previous work [4]. Furthermore,
Section 4.2 reports an example in the case of second-order tensors, which is crucial to the purpose of this work.
The metric G on B is introduced in a perfectly analogous way.

3.1. Metric transpose


The metric transpose of a two-point tensor field such as the deformation gradient F is defined as the two-point
tensor field
Ft ( · , t) : φ(B, t) → TB ⊗ T  S, (15)
such that, for every spatial vector field w (i.e. valued in TS) and every material vector field Y (i.e. valued in
TB),
w, F Y
g = Ft w, Y
G , (16)
where w, y
g = w.y and W , Y
G = W .Y denote the scalar products induced by g and G, respectively. Note
that, in the definition (15), with respect to the definition of F (equation (9)), one switches and also takes the
duals of TS and T  B in the tensor product of spaces. From this, it follows that Ft has components [4, 5]

(Ft )A a = GAB (FT )B b gba = GAB F b B gba . (17)

Note that, by exploiting the symmetry of g and employing the special tensor products ⊗ and ⊗ introduced
by Curnier et al. [11], it is possible to obtain component-free expressions of Ft as a function of FT and F,
respectively [4]:
Ft = [G−1 ⊗ g] : FT = [G−1 ⊗ g] : F. (18)
Remark 3.1. Note that Marsden and Hughes [5] call the metric transpose Ft simply “transpose” of F and
denote it by FT , and the algebraic transpose the “dual” of F (because its domain and codomain are the dual
spaces of the codomain and domain, respectively, of F) and denote it by F (however, they often use the symbol
FT also for the dual, typically in push-forward/pull-back operations). The terminologies “algebraic transpose”
and “metric transpose” are taken from a previous work [4].
For a spatial tensor field h, valued in [TS]1 1 , i.e. a “mixed” tensor field, the metric transpose is defined
similarly to the case of two-point tensors, as the tensor field

ht ( · , t) : φ(B, t) → [TS]1 1 , (19)

such that, for all spatial vector fields w and y (i.e. valued in TS),

w, h y
g = ht w, y
g , (20)
526 Mathematics and Mechanics of Solids 20(5)

which, in components, reads


(ht )a b = gac (hT )c d gdb = gac hd c gdb . (21)
Similarly to the case of two-point tensors, the metric transpose can be expressed by
ht = [g−1 ⊗ g] : hT = [g−1 ⊗ g] : h. (22)
A tensor field valued in [TS]1 1 is called symmetric if ht = h and skew-symmetric if ht = −h. Indeed,
ht = ±h ⇒ g−1 hT g = ±h ⇒ hT g = ±gh (23)
and, for the symmetry of the metric tensor,
hT g = hT gT = (gh)T , (24)
and therefore
ht = ±h ⇒ (gh)T = ±gh, (25)
i.e. the symmetry of h in the sense of the metric transposition implies the symmetry of its “covariant” counter-
part gh = h , as in this example, or of its “contravariant” counterpart hg−1 = h . The converse is found by just
repeating the calculations in the inverse order.
Remark 3.2. When a metric is not defined, the concept of symmetry of a “mixed” tensor has no meaning
whatsoever, as the spaces TS and T  S are different, and practically one has a two-point tensor. The metric
induces an isomorphism between a vector space and its dual, and therefore enables us to define the symmetry
of “mixed” tensors via the symmetry of the associated “covariant” or “contravariant” counterparts of the tensor.
The case of “mixed” material tensors fields, i.e. tensor fields valued in [TB]1 1 , is treated in an analogous
way.

3.2. Orthogonal tensors


It is particularly interesting to study two-point orthogonal tensors, as the case of orthogonal tensors that, seen
as linear maps, have the same domain and codomain, is perfectly analogous.
A two-point tensor field R, valued in TS ⊗ T  B, is said to be orthogonal if it preserves the metric between
B and S, i.e. for all vector fields W and Y valued in TB, their scalar product in TB equals the scalar product of
their images in TS:
W , Y
G = RW , RY
g . (26)
For this reason, an orthogonal tensor such as R is also called an isometry, or isometric transformation, with
respect to the metric G in the domain and g in the codomain.
By definition of metric transpose (equation (15)), we can establish the chain of equivalences
W , Y
G = RW , RY
g = Rt RW , Y
G , (27)
from which, for the arbitrariness of W and Y , we deduce the fundamental property
Rt R = I, (28)
where I is the identity in TB, which implies that the inverse and the metric transpose of an orthogonal tensor
coincide:
Rt = R−1 . (29)
We could have produced the same result via RR = i, where i is the identity in TS, had we started from the
t

scalar product w, y
g of two spatial vector fields and the use of R−1 .
Remark 3.3. This is a good “excuse” to emphasise that the metric transpose and the inverse are tensors of the
same type, i.e. for a tensor field valued in TS ⊗ T  B, like R of this section or the deformation gradient F, the
metric transpose ( · )t and the inverse ( · )−1 are both valued in TB ⊗T  S, in contrast with the algebraic transpose
( · )T , which is valued in T  B ⊗TS. Note also that, for a second-order tensor field in [TS]1 1 or [TB]1 1 , the tensor
itself, its metric transpose, and its inverse belong to [TS]1 1 or [TB]1 1 , while its algebraic transpose belongs to
[TS]1 1 or [TB]1 1 (which are the dual spaces of [TS]1 1 or [TB]1 1 , respectively).
Federico 527

3.3. Determinants of second-order tensors of any type


The textbook definition of determinant is given for endomorphisms (i.e. linear maps of a vector space into
itself) or, equivalently, for their representing matrices in a given basis. In terms of tensors, an endomorphism
is equivalent to a “mixed” tensor. Indeed, if a is an endomorphism in the vector space V, it is a linear map
a : V → V, which, with the usual abuse of notation, we identify with the second-order tensor a : V  × V → R,
i.e. an element of V 1 1 [4, 12].
We start by briefly introducing r-forms and then, after recalling the standard definition of determinant for
a tensor a ∈ V 1 1 , which automatically induces the definition of the determinant of spatial second-order tensor
fields h valued in [TS]1 1 and material second-order tensor fields H valued in [TB]1 1 , we show the adjustments
needed to evaluate the determinant of second-order tensors fields of any type. The case of two-point second-
order tensors such as the deformation gradient F shall be treated in Section 4.3.
Let V be a vector space of dimension n. A multilinear form ω or order r ≤ n, i.e. a tensor in V 0r , is called
skew-symmetric if, and only if, it is invariant for even permutations of the arguments and changes sign for odd
permutations of the arguments, i.e. given a set of r vectors {v1 , . . . , vr } ⊂ V,

ω(vi1 , . . . , vir ) = i1 ...ir ω(v1 , . . . , vr ), (30)

where i1 ...ir is the Ricci/Levi-Civita permutation symbol, such that



+1 for {i1 , . . . , ir } even permutation of {1, . . . , r},
i1 ...ir = (31)
−1 for {i1 , . . . , ir } odd permutation of {1, . . . , r}.

Skew-symmetric multilinear forms are called multi-covectors of order r or also r-forms. Exhaustive intro-
ductions to r-forms and spaces of r-forms can be found, e.g. in the works by Epstein [7] and Segev
[6].
An n-form μ in a vector space V of dimension n has a single independent component in an assigned basis
{ei }ni=1 . Indeed, by definition of skew-symmetry (equation (30)),

μ(ei1 , . . . , ein ) = i1 ...in μ(e1 , . . . , en ), (32)

and, since h = μ(e1 , . . . , en ) is a well-defined scalar, once the basis {ei }ni=1 is assigned, it follows that h is the
only independent component of μ, and

μ(ei1 , . . . , ein ) = h i1 ...in . (33)

Now, let D be the unique n-form with independent component equal to 1 in the basis {ei }ni=1 , i.e. such that

D(e1 , . . . , en ) = 1. (34)

For every system of n vectors {v1 , . . . , vn } ⊂ V, the definition of multilinearity and equations (30) and (34)
imply
D(v1 , . . . , vn ) = D(vi11 ei1 , . . . , vinn ein ) = vi11 . . . vinn D(ei1 , . . . , ein ) = i1 ...in vi11 . . . vinn . (35)
The scalar (35) is called the determinant of the system of vectors {v1 , . . . , vn } with respect to the basis {ei }ni=1 .
If [[ai j ]] denotes the n × n matrix in which the jth column is given by the components vij of vector vj , i.e.
⎡ 1 ⎤
v1 · · · v1j · · · v1n
⎢. . .. . . .. ⎥
⎢ .. .. . . .⎥
⎢ i ⎥

[[a j ]] = ⎢v1 · · ·
i
vij · · · vin ⎥⎥. (36)
⎢ .. . . .. . . .. ⎥
⎣. . . . .⎦
vn1 · · · vnj · · · vnn

then
det[[ai j ]] ≡ D(v1 , . . . , vn ) = i1 ...in vi11 . . . vinn = i1 ...in ai1 1 . . . ain n (37)
528 Mathematics and Mechanics of Solids 20(5)

is the determinant of matrix [[ai j ]], with respect to the basis {ei }ni=1 , given in the classical textbook definition.
If a is an endomorphism of V, i.e. a linear map a : V → V, its matrix with respect to the basis {ei }ni=1 in the
domain and the codomain is given by considering y = a w and decomposing w and y in the same basis {ei }ni=1 .
Starting from
y = a w = yi ei = a(wj ej ) = wj a ej , (38)
we consider that the image a ej can be decomposed in the basis {ei }ni=1 and written as

a ej = [a ej ]i ei = ai j ei , (39)

where
ai j = [a ej ]i (40)
are called the components of a in the basis {ei }ni=1 in the domain and codomain. Note that, from equations (38)
and (40), we obtain the textbook result yi = ai j wj .
The determinant of the endomorphism a with respect to the basis {ei }ni=1 is identified with that of its
representing matrix [[ai j ]], i.e.
det a ≡ det[[ai j ]]. (41)
This automatically defines the determinant of a seen as a tensor in V 1 1 , or “mixed” tensor.
Let us now return to the realm of Riemannian manifolds, taking the example of the physical space S. The
case of the body manifold B is of course analogical.
At this point, the determinant of a “mixed” tensor field h valued in [TS]1 1 is established, and we need to
consider “contravariant” and “covariant” tensors fields valued in [TS]20 and [TS]02 , respectively. It is evident
that, since the definition of the determinant solely fits “mixed” tensors, the metric is indispensable to extend it.
We define the determinants of a tensor field b valued in [TS]20 and of a tensor field c valued in [TS]02
as the determinants of the associated “mixed” tensor fields b g and g−1 c, respectively, and then exploit the
identification of the determinant of a mixed tensor with that of its associated matrix, and the rule of the product
of matrices for determinants.
For the case of the tensor field b valued in [TS]20 , we have the chain of identities

det b ≡ det(b g) ≡ det[[bab gbc ]] = det([[bab ]] [[gbc ]]) = det[[bab ]] det[[gbc ]]. (42)

Analogously, for c valued in [TS]02 ,

det c ≡ det(g−1 c) ≡ det[[gab cbc ]] = det([[gab ]] [[cbc ]]) = det[[gab ]] det[[cbc ]]. (43)

Remark 3.4. It is important to note that this rigorous definition clarifies that the determinant of the metric
tensor g is identically one, whereas that of its representing matrix [[gab ]] is not necessarily so. Indeed, using, for
example, equation (43), we have

det g ≡ det(g−1 g) = det i ≡ det[[δ a b ]] = 1, (44)

and, by definition of inverse, naturally also det g−1 = 1. Therefore, it is extremely important to distinguish
between det g = det g−1 = 1 and det[[gab ]] = (det[[gab ]])−1 = 1, as the non-unitary coefficients det[[gab ]] and
det[[gab ]] appear in many calculations in Riemannian Geometry and, therefore, in Continuum Mechanics; as an
example, see equation (95) below, and many other cases described, e.g. by Marsden and Hughes [5].

4. Metric and deformation


The metric structure endows Continuum Mechanics with a virtually endless series of new objects associated
with deformation. Each of these objects is either a deformation or a strain, represented by a second-order tensor
that can be defined in two possible metric-related forms: “mixed”, i.e. with the first foot being a vector and the
second being a covector (equivalently, with the first index being contravariant and the second being covariant),
and “covariant” or “contravariant”, i.e. with both feet being covectors or vectors, respectively (equivalently,
with both indices being covariant or contravariant, respectively). This choice in the definitions gives rise to two
alternative forms of Cauchy’s Polar Decomposition Theorem.
Federico 529

This section is dedicated to the core result of this work: the two alternative forms of the Polar Decomposition
Theorem yield two forms of deformation and strain tensors, as well as two forms of all Hill’s generalised strains.
The tensors in these two formulations are connected by the appropriate metric tensor and, in practical terms, the
choice of either formulation is dictated purely by convenience.
First, we recall the classical form of the Polar Decomposition Theorem, following Marsden and Hughes [5].
Then, we propose an alternative form of the theorem, which is directly related to the geometric view in which
the standard deformation tensors are related to the metric tensors by push-forward or pull-back operations, as
appropriate. We subsequently define the natural deformation and strain tensors in the two formalisms, as well
as rigorously define the volume ratio (the determinant of the deformation gradient). Finally, we conclude with
Hill’s generalised strains in the two formalisms.
The Polar Decomposition Theorem states that any regular (i.e. with non-zero determinant) second-order
tensor admits a right decomposition into the product of an orthogonal tensor and a symmetric tensor, and a left
decomposition into the product of a symmetric tensor (which turns out to be the symmetric tensor of the right
decomposition, rotated by the orthogonal tensor) and the same orthogonal tensor.
Here, we present the enunciations of the two possible formulations of the theorem in the usual fashion
adapted to Continuum Mechanics, where the decomposition is applied to the deformation gradient tensor field,
F, which gives the orthogonal tensor the physical meaning of pure rotation, and gives the symmetric tensors the
physical meaning of pure deformations.

4.1. The classical Polar Decomposition Theorem


In the classical formulation, the Polar Decomposition Theorem is enunciated for matrices. In terms of tensors,
this corresponds to a situation in which all involved tensors are “mixed”, i.e. the first foot is a vector and the
second a covector (the first index is contravariant and the second covariant). This is the standard form reported,
e.g., by Marsden and Hughes [5],
F = R U = VR, (45)
where the rotation R is a proper (i.e. with determinant equal to +1; see Section 5 for details on the evaluation of
the determinant of a two-point tensor) orthogonal two-point tensor field that, like F, is valued in TS ⊗ T  B, and
the “mixed” right stretch U and left stretch V are symmetric tensor fields in the sense of metric transposition,
i.e. U t = U , and V t = V, valued in [TB]1 1 and [TS]1 1 , respectively. In components,
F a C = Ra A U A C = V a b Rb C . (46)
Left-multiplication of equation (45) by Rt = R−1 yields
U = R−1 V R, (47)
and, analogously, right-multiplication yields
V = R U R−1 . (48)
Equations (47) and (48) indicate that U and V are related by a purely rotational push-forward/pull-back relation.
Because of their symmetry, tensors U and V have real eigenvalues and orthogonal eigenvectors, resulting
from the eigenvalue problems
det(U − λ I) = 0, det(V − λ i) = 0. (49)
Moreover, the positive definiteness implies that their eigenvalues are all positive, and the fact that they are
related by equations (47) and (48) imply that they are similar, i.e. they have the same eigenvalues, denoted λ1 ,
λ2 , λ3 , and called principal stretches. The spectral decompositions are

3
U= λα N α ⊗ N α , (50)
α=1

where N α are the eigenvectors of U and N α = G−1 N α are the associated covectors, and

3
V= λα nα ⊗ nα , (51)
α=1
530 Mathematics and Mechanics of Solids 20(5)

where nα are the eigenvectors of V and nα = g−1 nα are the associated covectors, and equation (48) implies
nα = RN α . (52)

4.2. Alternative form of the Polar Decomposition Theorem


In the proposed alternative formulation, the Polar Decomposition Theorem reads
F = R.U = V.R, (53)
where the rotation R is the already seen proper orthogonal two-point tensor field valued in TS ⊗ T  B, the right
stretch U and the left stretch V are symmetric tensor fields in the sense of algebraic transposition, i.e. U T = U
and V T = V (which is equivalent to the symmetry of U and V in the sense of metric transposition) and valued
in [TB]02 and [TS]20 , respectively. In both cases, the low dot “.” indicates the presence of the appropriate metric
tensor, needed to contract the second foot of R with the first of U and the second foot of V with the first of R.
This is clarified by writing the metric tensors explicitly in equation (53), i.e.

F = R G−1 U = V g R, (54)
or by writing the component expression

F a C = Ra A GAB UBC = V ab gbc Rb C . (55)


Naturally, the the “covariant” U and the “contravariant” V are related to the “mixed” U and V of the standard
form of the theorem by the appropriate metric tensor:
U = U = G U, (56a)
 −1
V =V =Vg . (56b)
Substitution of equations (56) into the alternative form of the Polar Decomposition Theorem, equations (53)
and (54), yields the classical form, equation (45).
Left multiplication of equation (54) by G R−1 = G Rt yields

G Rt R G−1 U = G Rt V g R, (57)

from which, by using Rt R = R−1 R = I and then G−1 G = I on the left-hand side, and interposing i = g−1 g
between Rt and V in the right-hand side, we obtain
U = G Rt g−1 g V g R, (58)
which, using the relation between metric and algebraic transpose, G Rt g−1 = RT , and using the low dot to
indicate the metric tensor g, takes the final, familiar form
U = RT .V .R. (59)
Right multiplication of equation (54) by R−1 g−1 = Rt g−1 and analogous passages yield
V = R.U.RT . (60)
Equations (59) and (60) state that U is the purely rotational pull-back of V and, vice versa, V is the rotational
push-forward of U, and are analogical to equations (47) and (48), respectively. In both equations (59) and (60),
notice the presence of the appropriate metric tensors, indicated buy the low dots “.”: to be precise, indeed, one
should say that U is the purely rotational pull-back of V  = g V g (the “covariant” counterpart of V ) and, vice
versa, V is the purely rotational push-forward of U  = G−1 U G−1 (the “contravariant” counterpart of U).
Because of their symmetry, tensors U and V have real eigenvalues and orthogonal eigenvectors, which are
evaluated with respect to the material metric G and the inverse spatial metric g−1 , respectively, i.e. from the
eigenvalue problems
det(U − λ G) = 0, det(V − λ g−1 ) = 0, (61)
Federico 531

and, from the relation between the determinant of a “covariant” or “contravariant” tensor and the standard
determinant of a “mixed” tensor, they are identical to the eigenvalues λ1 , λ2 , λ3 of U and V. The spectral
decompositions of U and V are, therefore


3
U= λα N α ⊗ N α , (62)
α=1

and

3
V= λα nα ⊗ nα . (63)
α=1

Remark 4.1. The fact that the “mixed” U and V, the “covariant” U and the “contravariant” V have the same
eigenvalues implies that they have the same invariants I1 , I2 , I3 . This can also be seen considering the definitions
of the invariants in terms of trace and determinant. Indeed, since
I1 ( · ) = tr( · ), (64a)
I2 ( · ) = [(tr( · ) − tr(( · ) )],
1
2
2 2
(64b)
I3 ( · ) = det( · ), (64c)
it suffices to consider that the trace of a “covariant” or “contravariant” tensor is defined to be equal to that of the
associated “mixed” tensor, much like it has been done for determinants in Section 3.3. For example, the trace of
V is defined as tr(V ) ≡ tr(V g) = tr(V) = V a a = gac Vca . Note also that, similarly to the case of determinants,
the trace of all metric tensors is always equal to 3 (e.g. tr(g) ≡ tr(g−1 g) = tr(i) = δ a a = 3), although, naturally,
the trace of their matrix in a given basis may differ from 3 (i.e. tr([[gab ]]) = 3, in general).

4.3. Deformation tensors and strain tensors


The metric tensors G and g measure distances (or, to be more precise, the norms of tangent vectors) in B
and S, respectively. The deformation causes changes in the distance between the current positions in space
of any two points, with respect to their distance in the the body manifold (or the reference configuration),
and these changes are captured by the push-forward or pull-back of the metric tensors and their inverses, as
appropriate. The concept of strain is obtained by either comparing the pull-back of the spatial metric with the
body metric, or the spatial metric with the push-forward of the body metric. This view is directly related to the
Polar Decomposition Theorem by means of the alternative formalism presented here. Naturally, the relationship
can be established also in the classical formalism, but one has to fall back into what we here propose as the
alternative one.
In both cases, the connection between the deformation tensors and the Polar Decomposition Theorem comes
from the fact that the only way to evaluate the right and left stretch tensors is to directly evaluate their squares,
the right and left Cauchy–Green deformation tensors, and then extract the square root.
Following the classical form of the Polar Decomposition Theorem, as presented in Section 4.1, and recalling
that U and V are symmetric in the sense of metric transposition, we define the tensor fields

C = F t F = U t Rt R U = U U = U 2 , (65a)
b = F F = VR R V = V V = V ,
t t t 2
(65b)

valued in [TB]1 1 and [TS]1 1 , and called “mixed” right and left Cauchy–Green deformation tensors, respectively,
and with component form
C A C = GAB F a B gab F b C = U A B U B C , (66a)
ba c = F a A GAB F b B gbc = V a b V b c . (66b)
Tensors C and b are both symmetric and are coaxial with U and V, respectively, i.e. they have the same eigen-
vectors of U and V, respectively, and the same eigenvalues. Their spectral decompositions are analogous to
those of U and V, with eigenvalues λ2α .
532 Mathematics and Mechanics of Solids 20(5)

The inverses of C and b are tensor fields valued in the same spaces, and are called Piola deformation tensor
B and Finger deformation tensor c respectively,

B = C −1 = F−1 F−t = U −1 R−1 R−t U −t = U −1 U −1 = U −2 , (67a)


−1 −t −1 −t −t −1 −1 −1 −1 −2
c=b =F F =V R R V =V V =V . (67b)

Their component expressions are

B A C = (F−1 )A a gab (F−1 )B b GBC = (U −1 )A B (U −1 )B C , (68a)


−1 A −1 B −1 a −1 b
c a
c = g (F )
ab
b GAB (F ) c = (V ) b (V ) c, (68b)

and they are coaxial with U and V, with analogous spectral decompositions, and eigenvalues λ−2
α .
Comparison of the “mixed” right Cauchy–Green tensor C with the material identity tensor I yields the
“mixed” Green–Lagrange strain E, and comparison of the spatial identity tensor i with the “mixed” Finger
tensor c yields the “mixed” Almansi strain e,

E= 1
2
(C − I), (69a)
e= 1
2
(i − c), (69b)

where the coefficient 12 serves to guarantee that both strains reduce to the infinitesimal strain in the small-
displacement approximation. Note that E and e are related by push-forward/pull-back.
The definition of the four deformation tensors (65) and (67) and the two strain tensors (69) coming from the
classical Polar Decomposition Theorem is that presented as standard in the classical treatise by Marsden and
Hughes [5], but is somewhat artificial, as it does not immediately convey the physical meaning of each tensor
in terms of “measuring distances”. In order to recover the physical meaning of all tensors, Marsden and Hughes
[5] have to convert C, E, c and e into their “covariant” counterparts, and B and b into their “contravariant”
counterparts.
Following the alternative form of the Polar Decomposition Theorem proposed in Section 4.2, this becomes
completely natural.
Let us start from the standard geometrical result that the right Cauchy–Green deformation tensor is defined as
the pull-back of the spatial metric g and is a tensor field valued in [TB]02 , and the left Cauchy–Green deformation
tensor is defined as the push-forward of the inverse material metric G−1 and is a tensor field valued in [TS]20 ,
i.e.

C = φ ∗ [g] = FT g F = FT.F, (70a)


−1 −1
b = φ∗ [G ] = F G F = F.F . T T
(70b)

Now, using the alternative form (53) of the polar decomposition in equations (70), and the symmetry of U and
V with respect to the algebraic transposition, we obtain relations analogous to the (65), i.e.

C = (R.U)T .(R.U) = U G−1 RT g R G−1 U = U G−1 U = U.U = U 2 , (71a)


b = (V .R).(V .R)T = V g R G−1 RT g V = V gV = V .V = V 2 , (71b)

where we used G−1 RT g = Rt = R−1 and defined the squares U 2 = U.U = U G−1 U and V 2 = V .V = V g V .
In component form, C and b read

CAB = F a A gab F b B = UAC GCD UDB , (72a)


bac = F a A GAB F b B = V ac gcd V DB . (72b)

Tensors C and b are symmetric and coaxial with U and V , respectively, and the spectral decompositions are
analogous to those of U and V , with eigenvalues λ2α .
The inverse of C is the Piola deformation tensor B, a material tensor field valued in [TB]20 , pull-back of the
inverse spatial metric g−1 , and the inverse of b is the Finger deformation tensor c, a spatial tensor field valued
Federico 533

in [TS]02 , push-forward of the material metric G, and we directly report also their relation with the inverses of
U and V :

B = C −1 = φ ∗ [g−1 ] = F−1 g−1 F−T = F−1 .F−T = U −2 , (73a)


c = b−1 = φ∗ [G] = F−T G F−1 = F−T .F−1 = V −2 . (73b)

Their component expressions are

BAB = (F−1 )A a gab (F−1 )B b = (U −1 )AC GCD (U −1 )DB , (74a)


cab = (F−1 )A a GAB (F−1 )B b = (V −1 )ac gcd (V −1 )db , (74b)

and their spectral decompositions are


3
B= λ−2
α Nα ⊗ Nα, (75a)
α=1

3
c= λ−2 α α
α n ⊗n , (75b)
α=1

where we notice that they are coaxial with U and V in the sense of the associated eigenvectors and
eigencovectors.
Exactly as for the case of the two forms of the right and left stretch tensors (Remark 4.1), both forms of the
right and left Cauchy–Green tensors have the same eigenvalues and the same invariants.
Once C and c are defined as the pull-back of g and the push-forward of G, respectively, it is natural to define
the Green–Lagrange and the Almansi strain tensors via the expressions

E= 1
2
(C − G), (76a)
e= 1
2
(g − c), (76b)

which are still related by a push-forward/pull-back relation.


The definitions (70) (along with (71)), (73) and (76) are the most natural ones, as they allow to describe the
change of norm of a tangent vector. Indeed, if W ∈ TB and w ∈ TS are related by the pull-back W = φ ∗ [w] =
F−1 w and the push-forward w = φ∗ [W ] = F W , the squared norms of W and w, induced by the metric tensors
G and g, respectively, are

W 2 = W , W
G = W .W = W G W = (F−1 w) G (F−1 w) =
= w (F−T G F−1 ) w = w c w = w, w
c , (77a)
w = w, w
g = w.w = w g w = (FW ) g (FW ) =
2

= W (FT g F) W = W C W = W , W
C . (77b)

Therefore, the strain in the material direction W or the spatial direction w can be described by half of the
difference between the squared norms of the “current” vector w and the “referential” vector W :
1
2
(w2 − W 2 ) = 1
2
( w, w
g − W , W
G ). (78)

Now, using equations (77), we obtain, with some simple manipulation, the two expressions
1
2
(w2 − W 2 ) = W [ 12 (C − G)] W = W E W , (79a)
1
2
(w − W  ) = w
2 2
[ 12 (g − c)] w = w e w, (79b)

which give the Green–Lagrange and the Almansi tensors the physical meaning of strain in terms of pulled-back
metric C = φ ∗ [g] and pushed-forward metric c = φ∗ [G], respectively.
534 Mathematics and Mechanics of Solids 20(5)

Remark 4.2. This is naturally a well-known result, and a similar proof in Cartesian coordinates, without ref-
erence to the metric tensors, is reported by Bonet and Wood [3]. The novelty is in the direct relation with
the proposed alternative form of the Polar Decomposition Theorem. This direct relation plays an even more
important role in the definition of Hill’s generalised strains, as we shall show in Section 4.4.
As for the case of the right and left stretch tensors (equation (56)), the two forms of the deformation and
strain tensors are related by the appropriate metric tensor:
C = C  = G C, (80a)
 −1
b = b = bg , (80b)
 −1
B = B = BG , (80c)

c = c = g c, (80d)

E = E = G E, (80e)

e = e = g e. (80f)

4.4. Hill’s generalised strains


Hill [1] proposed a set of strains generalising the structure of the Green–Lagrange and Almansi strains (69) and
(76), written as a function of the right and left stretch tensors, rather than of the right Cauchy–Green and the
Finger tensor.
In the classical form of the Polar Decomposition Theorem, the Green–Lagrange and Almansi strains read

E = 12 (U 2 − I) = α 12 (λ2α − 1) N α ⊗ N α , (81a)
−2
1 −2 α
e = 2 (i − V ) = α 2 (1 − λα ) nα ⊗ n ,
1
(81b)
and, in the proposed alternative form,

α α
E= 1
2
(U 2 − G) = α 2 (λα − 1) N ⊗ N ,
1 2
(82a)

e= 1
2
(g − V −2 ) = α 12 (1 − λ−2 α α
α )n ⊗ n , (82b)

where U 2 = U G−1 U = U.U and V −2 = V −1 g−1 V −1 = V .V . Equations (81) and (82) show that the Green–
Lagrange and Almansi strain can be regarded as “strains of order two”, because of the power with which U and
V −1 feature in the definitions.
The generalised strains proposed by Hill [1] are obtained by replacing 2 by m ∈ N. In in the classical
formalism, they are all “mixed”, i.e. the material ones are valued in [TB]1 1 and the spatial ones are valued in
[TS]1 1 , and read

α
E (m) = m1 (U m − I) = α m1 (λm
α − 1) N α ⊗ N , (83a)
−m
−m α
e = m (i − V ) = α m (1 − λα ) nα ⊗ n .
(m) 1 1
(83b)
In the proposed alternative formalism, they are all “covariant”, i.e. the material ones are valued in [TB]02 and
the spatial ones are valued in [TS]02 , and read

α α
E(m) = m1 (U m − G) = α m1 (λmα − 1) N ⊗ N , (84a)
−m
1 −m α α
e = m (g − V ) = α m (1 − λα ) n ⊗ n .
(m) 1
(84b)
Again, the coefficient m1 guarantees that all strains reduce to the infinitesimal strain in the small-displacement
approximation.
Of all these strains, the only ones used for practical purposes are the strains of order two, as already seen,
and the strains of order one. The latter are both associated with the name of Biot. In the classical formalism,
these read

E (1) = U − I = α (λα − 1) N α ⊗ N α , (85a)


−1
−1 α
e = i − V = α (1 − λα ) nα ⊗ n ,
(1)
(85b)
Federico 535

and, in the proposed alternative formalism,


E(1) = U − G = α (λα − 1) N α ⊗ N α , (86a)


−1

(1)
e =g−V = α (1 − λ−1
α )n
α α
⊗n . (86b)
The Biot strains are important because they provide material and spatial large-displacement versions of the nom-
inal strain, i.e. two strains that correctly “filter” the rotation in their definition, also under large displacements.
The nominal strain of the infinitesimal theory, in contrast, fails in filtering the rotation under large displacements
(see the nice example in Chapter 1 of the textbook by Bonet and Wood [3]).
Another important couple of Hill’s generalised strains, not represented by equations (83) and (84), in which
m ∈ N, is obtained by considering m ∈ R and performing the limit m → 0; both strains are associated with the
name of Hencky.
The material Hencky strain is found recalling the fundamental limit
am − 1
lim = ln(a), (87)
m→0 m
and the spatial Hencky strain is found by exploiting the same limit, with the sign changed, and the substitution
a = b−1 , i.e.
1 − am 1 − b−m
lim = − ln(a) ⇒ lim = − ln(b−1 ). (88)
m→0 m m→0 m
In the classical formalism, the two Hencky strains read

E (0) = ln(U ) = α ln(λα ) N α ⊗ N α , (89a)


−1
−1 α
e = − ln(V ) = α − ln(λα ) nα ⊗ n ,
(0)
(89b)
and in the proposed alternative formalism,

E(0) = ln(U) = ln(λα ) N α ⊗ N α ,


α (90a)

e(0)
= − ln(V −1 ) = α − ln(λ−1 α α
α )n ⊗ n . (90b)
Finally, note that the “mixed” and the “covariant” forms of Hill’s generalised strains of any order are related
by expressions analogous to those in equations (80e) and (80f).
Remark 4.3. A quite subtle but common error often found in the literature is to write the spatial “covariant”
Hencky strain as ln(V ). This error is caused by an over-reliance on equation (88), the result of which can
naturally be written − ln(b−1 ) = ln(b), and on the usual tacit assumption of Cartesian coordinates. In the two
Hencky strain tensors, however, while the minus sign before the logarithm is purely algebraic, the −1 exponent
of the left stretch tensor is a tensor inversion! In the case of the spatial “mixed” Hencky strain, the tensor
inversion on V yields a tensor field valued in the same space [TS]1 1 in which V is valued, and therefore the
operation is safe, indeed

e(0) = − ln(V −1 ) = α − ln(λ−1 α ) nα ⊗ n


α

= α ln(λα ) nα ⊗ nα
= ln(V). (91)
In contrast, in the case of the spatial “covariant” Hencky strain,

e(0) = − ln(V −1 ) = α − ln(λ−1 α


α )n ⊗ n
α

= α ln(λα ) nα ⊗ nα
= ln(V  ) = ln(gV g) = ln(V ), (92)
because V −1 and V  are valued in the dual space [TS]02 of the space [TS]20 in which V is valued.
This warning is necessary because, in the literature on constitutive behaviour, reference is virtually always
made to the “covariant” Hencky strain e(0) = − ln(V −1 ) = ln(V ) and not to the “mixed” e(0) = − ln(V −1 ) =
ln(V).
As it happens in many similar cases in Algebra and Mechanics, the expression e(0) = ln(V ) is acceptable
only in Cartesian coordinates, when V and V  = gV g can be identified.
536 Mathematics and Mechanics of Solids 20(5)

5. The volume ratio


Here we give an explicit definition of the volume ratio, i.e. the determinant J of the deformation gradient F,
which can be immediately extended to any two-point tensor, for the case of Riemannian manifolds. The standard
definition of determinant of a “mixed” tensor, or endomorphism, given in equation (41) cannot be used here,
because a two-point tensor maps between two different spaces, i.e. it is not an endomorphism.
Nonetheless, in many works in the literature, the determinant of F is identified with the determinant of
its representative matrix [[F a A ]]. This is correct under the tacit assumption, made in the largest part of the
literature, of Cartesian coordinates. In fact, det F equals det[[F a A ]] exclusively in Cartesian coordinates, as it
has been shown by Marsden and Hughes [5]. However, the proof that Marsden and Hughes [5] provided for
the expression of det F was derived by starting from Cartesian coordinates and then transforming to curvilinear
coordinates. Here, we shall give a very simple definition, suited for general Riemannian manifolds B and S, in
which the Euclidean structure (and, therefore, Cartesian coordinates) is not available, taking the example of the
deformation gradient tensor field F.
The determinant of F is defined by the two equivalent expressions
√ √
det F = det C ≡ det C, (93a)
√ √
det F = det b ≡ det b, (93b)
where the equivalences det C ≡ det C and det b ≡ det b come from the definitions (43) and (42) of the
determinant of “covariant” and “contravariant” tensors. In terms of the representing matrices, we have

det F = det(F F) = det(G−1 FT g F) =
t

= det[[G (F )B gab F C ]] = det([[GAB ]] [[(FT )B a ]] [[gab ]][[F b C ]]),
AB T a b (94a)

det F = det(FFt ) = det(F G−1 FT g) =

= det[[F A G (F )B gbc ]] = det([[F a A ]] [[GAB ]] [[(FT )B b ]] [[gbc ]]).
a AB T b
(94b)

Since the determinant of the transpose of a matrix equals the determinant of the matrix, and the determinant of
the inverse of a matrix is the reciprocal of the determinant of the matrix, we obtain the final expression
1
J = det F = det[[gab ]] det[[F a A ]] , (95)
det[[GAB ]]
which is the same result reported by Marsden and Hughes [5], without the need to assume that the Euclidean
structure and Cartesian coordinates are available.
Remark 5.1. Note that equation (95) explicitly features the coefficients det[[GAB ]] = 1 and det[[gab ]] = 1
involving the representative matrices of G and g, but both metric tensors, seen as tensors, have always
determinants equal to one (equation (44)).
Remark 5.2. As briefly mentioned at the end of Section 2, it is possible to define the determinant J of F also
when no metric is defined on B and S, as long as a non-vanishing volume form is defined on both.
Volume forms on a manifold of dimension n are all non-vanishing n-forms, and have this alternative name
because they are the integrands in “volume integrals”. An introduction to the theory of integration based on
forms can be found in the works by Epstein [7] and Segev [6]. Here, we shall quickly go through the main
passages without any claim of extreme rigour.
From equation (33), it is easy to show that a volume form μ on S can be written as
μ = h a1 ...an ea1 ⊗ . . . ⊗ ean , (96)
where the scalar field h is the independent component of μ. Analogously, one can define a volume form on B
with independent component H as
M = H A1 ...An EA1 ⊗ . . . ⊗ EAn . (97)
Federico 537

The pull-back φ ∗ [μ] can be calculated explicitly, as


φ ∗ [μ] = h a1 ...an (FT ea1 ) ⊗ . . . ⊗ (FT ean )
= h a1 ...an F a1 A1 . . . F an An EA1 ⊗ . . . ⊗ EAn
= h det[[F a A ]] A1 ...An EA1 ⊗ . . . ⊗ EAn , (98)
which shows that φ ∗ [μ] is a volume form on B, with independent component h det[[F a A ]]. In terms of the
volume form M chosen in B in equation (97), φ ∗ [μ] reads
1
φ ∗ [μ] = h det[[F a A ]] H A1 ...An EA1 ⊗ . . . ⊗ EAn
H
1
= h det[[F a A ]] M
H
= J M, (99)
where
1
J = det F ≡ h det[[F a A ]] (100)
H
is defined as the determinant of F with respect to the volume form μ on S and the volume form M on B.
The meaning of this definition is clear if we look at the theorem of the change of variables in terms of volume
forms,

μ= φ [μ] = J M, (101)
φ(B) B B
where we immediately recognise the “coefficient” J = det F, needed to transform the integral from the domain
φ(B) to the domain B. In the formalism adopted in many textbooks, this reads
dυ = J dV , (102)
where dυ and dV are the “volume elements” corresponding to the volume forms μ and M.
Now, if metric tensors g and G are defined in S and B, they induce the standard volume forms [5]

θ = det[[gab ]] a1 ...an ea1 ⊗ ... ⊗ ean , (103a)

 = det[[GAB ]] A1 ...An EA1 ⊗ ... ⊗ EAn . (103b)
It is easy to see that the expression of J with respect to θ and  is exactly that in equation (95).
Finally, note that this definition reduces to the standard definition of determinant for the case of an endo-
morphism, i.e. a tensor in [TB]1 1 or [TS]1 1 : in that case, indeed, the two volume form components in equation
(100) coincide and cancel out.

6. Discussion
The metric structure is indispensable in order to give Continuum Kinematics something more than the con-
figuration map and its tangent map, the deformation gradient. Indeed, deformation and strain are defined with
respect to a given metric, in terms of push-forward or pull-back operations.
This work was aimed at remarking that all deformation and strain tensors can be defined in two forms.
This result had been reported by Marsden and Hughes [5], but its consequences have not been fully explored
yet. The first form descends from the classical form of Cauchy’s Polar Decomposition Theorem, which was
conceived with matrices in mind, and therefore is suited for “mixed” tensors (with the first foot being a vector
and the second a covector, i.e. the first index being contravariant and the second covariant). In the second form,
some of the deformation tensors are “contravariant” and some are “covariant” (with both feet being vectors
or covectors, respectively, and the indices being both contravariant or covariant, respectively), and the strains
are all “covariant”; the four fundamental deformation tensors (right and left Cauchy–Green, Piola, Finger) are
naturally defined by means of pull-back/push-forward operation on the spatial/material metric, as appropriate,
and can be conveniently related to the Polar Decomposition Theorem by means of the proposed formalism.
538 Mathematics and Mechanics of Solids 20(5)

The classical form of the Polar Decomposition Theorem is quite straightforward to establish, but it lacks a
direct connection with the idea that deformation is measured by means of pull-back/push-forward of the metric.
While this information can be recovered by contracting the “mixed” deformation tensors with the appropriate
metric to obtain the associated “contravariant” or “covariant” tensors, the link comes across as quite artificial.
On the other hand, while the proposed alternative formulation is more laborious to be established, it has the
advantage of directly introducing the right stretch U as a tensor with both feet being covectors (i.e. with both
indices being covariant) and the left stretch V as a tensor with both feet being vectors (i.e. with both indices
being contravariant). We regard the alternative formulation as the preferred one for two reasons. Firstly, it is
naturally connected to the definitions of the basic deformation tensors (right and left Cauchy–Green, Piola, and
Finger) seen as push-forwards/pull-backs of the metric tensors (see Section 4.3), which we see as an elegant
feature, from the geometric point of view. Secondly, it allows for directly defining all strains, including Hill’s
generalised tensors of order n ∈ N and Hencky’s logarithmic strains, as fully “covariant” tensors, which is the
standard use in constitutive theory (see Section 4.3).
Moreover, we have emphasised how the two forms of the four fundamental deformation tensors differ in the
use of the metric transpose in the classical form, and of the algebraic transpose in the proposed alternative form.
The clear distinction between the two transpositions avoids possible errors, confusion, and headaches. Among
the other things, we have also provided a general and rigorous proof (both in Riemannian manifolds and in
general differentiable manifolds with a non-vanishing volume form) of the expression of the volume ratio, the
determinant of the deformation gradient, and pointed out a subtle error that can be made in the definition of the
spatial logarithmic strain.
In conclusion, the proposed alternative form of the Polar Decomposition Theorem is more suited to the
purposes of Continuum Mechanics, as it preserves and clarifies the geometrical meaning of all quantities related
to the deformation. Future developments may include the application of this formalism to the definition of the
measures of strain in Cosserat continua (see, e.g., [13, 14] and references therein) and in higher-order gradient
theories [15], as well as in the multiplicative decomposition of the deformation gradient involved in the theories
of plasticity or growth (see, e.g., [16–19] and references therein).

Acknowledgements
The young Leo Federico is gratefully acknowledged for having let me spend time, which could have been converted, perhaps more
fruitfully, into playtime with him, to write this work. Prof Marcelo Epstein is gratefully acknowledged for his precious suggestions on
the expression of the determinant of the deformation gradient in terms of volume forms.

Funding
This work was supported in part by Alberta Innovates – Technology Futures (formerly Alberta Ingenuity Fund, Canada), through
the AITF New Faculty Programme, and by the Natural Sciences and Engineering Research Council of Canada, through the NSERC
Discovery Programme.

References
[1] Hill, R. J Mech Phys Solids 1968; 16: 229–242.
[2] Ogden, RW. Non-linear Elastic Deformations. New York: Dover, 1997.
[3] Bonet, J, and Wood, RD. Nonlinear Continuum Mechanics for Finite Element Analysis, second edition. Cambridge: Cambridge
University Press, 2008.
[4] Federico, S. Covariant formulation of the tensor algebra of non-linear elasticity. Int J Non-Linear Mech 2012; 47: 273–284.
[5] Marsden, JE, and Hughes, TJR. Mathematical Foundations of Elasticity. Englewood Cliff, NJ: Prentice-Hall, 1983.
[6] Segev, R. Notes on metric independent analysis of classical fields. Math Meth Appl Sci 2013; 36(5): 497–566.
[7] Epstein, M. The Geometrical Language of Continuum Mechanics. Cambridge: Cambridge University Press, 2010.
[8] Truesdell, C, and Toupin, RA. The classical field theories. In: Flügge S (ed) Encyclopedia of Physics, 3 Vols. Berlin: Springer-
Verlag, 1960, Vol 1, pp.226–793.
[9] Truesdell, C, and Noll, W. The Non-linear Field Theories of Mechanics, Vol 3, S. Flügge (ed) Encyclopedia of Physics, 3 Vols.
Berlin: Springer-Verlag, 1965.
[10] Epstein, M, and Elżanowski, M. Material Inhomogeneities and Their Evolution. Berlin: Springer-Verlag, 2007.
[11] Curnier A, He Q-C, and Zysset, P. Conewise linear elastic materials. J Elasticity 1995; 37: 1–38.
[12] Bishop, RL, and Goldberg, SI. Tensor Analysis on Manifolds. Englewood Cliff, NJ: Prentice-Hall, 1968.
Federico 539

[13] Neff, P. Existence of minimizers for a finite-strain micromorphic elastic solid. Proc Roy Soc Edinburgh A 2006; 136: 997–1012.
[14] Pietraszkiewiecz, W, and Eremeyev, VA. On natural strain measures of the non-linear micropolar continuum. Int J Solids Struct
2009; 46: 774–787.
[15] dell’Isola, F, Seppecher P, and Madeo, A. How contact interactions may depend on the shape of Cauchy cuts in n-th gradient
continua: Approach à la D’Alembert. Z Angew Math Phys 2012; 63: 1119–1141.
[16] Simo, JC. A framework for finite strain elastoplasticity based on maximum plastic dissipation and the multiplicative
decomposition. Part I: Continuum formulation. Comput Meth Appl Mech Eng 1988; 66: 199–219.
[17] Cuomo, M, and Contrafatto, L. Stress rate formulation for elastoplastic models with internal variables based on augmented
Lagrangian regularisation. Int J Solids Struct 2000; 37: 3935–3964.
[18] Epstein, M, and Maugin, GA. Thermomechanics of volumetric growth in uniform bodies. Int J Plasticity 2000; 16: 951–978.
[19] Grillo, A, Federico, S, and Wittum, G. Growth, mass transfer, and remodeling in fiber-reinforced, multi-constituent materials. Int
J Non-Linear Mech 2012; 47: 388–401.

You might also like