DRAF

December 7, 2008 10:51 pm

What is deformation?

Rebec

ca Br
ann

on

Kinematics:
The mathematics of deformation
Continuum mechanics deals with the mathematics, mechanics, kinematics and physics
of deformable bodies. The word kinematics refers to the mathematics of motion. No
physics principles are invoked in kinematics. Velocity, acceleration, rotation, etc. are all
kinematics concepts. The notions of force and torque are mechanics concepts. Physics
deals with the relationships between loads and motion observed in the physical world. For
example, the famous particle equation F = ma states that the force applied to a particle is
proportional to the particles acceleration, with mass being the proportionality factor.
More correctly, force equals the rate of linear momentum. A different, but analogous,
equation applies when to deformable bodies.

1. What is deformation?
Let o denote the initial configuration of a physical body (e.g., an airplane part, a
blob of putty, etc.). As illustrated in Fig. 1.1, let X denote the initial position of a point in

the body. Disallowing fracture (where particles break into two separated particles), each
point in a body can have only one initial (pre-deformation) location. Consequently, the initial position vector, X , can be regarded as a unique name of any particular particle of

interest. When we say particle X we are referring to the particle that was initially

located at position X . Let denote the configuration of the body after deformation. The

goal of this chapter is to define the standard methods and lexicon used to quantify the relationship between and o . Because each material particle X moves to a new location,

x and because no two distinct particles are allowed to deform into the same location, the

key relationship between and o is the mapping function, x = ( X ) that relates

the initial and deformed position vectors. Often, a mapping function does not need to be
found per se. Instead, kinematics is founded on the notion that a mapping function exists.
Figure 1.1 shows a pseudo 2D deformation in which a square grid has been conceptually painted onto the material. We have filled in a few of the material elements (grid
cells) with solid color assist with some upcoming discussions. Of course, since the out-ofplane thickness has changed (by about 50% from the looks of it), this deformation is actually three-dimensional. A pseudo 2D deformation has out-of-plane thickness changes, but
no out-of-plane shearing; motion is characterized principally with respect to the two inplane coordinates. Most of the concepts in this document will be illustrated in the context
of pseudo 2D deformations, but they apply equally well to general deformations in which
material elements can shear and rotate in arbitrary directions.

1
Copyright is reserved. Individual copies may be made for personal use. No part of this document may be reproduced for profit.

T
F
A
DR
ann
ca Br
Rebec

December 7, 2008 10:51 pm

What is deformation?

on

Lo
L

Figure 1.1.
A general 2D deformation. A slice of putty is deformed into a new configuration in
pseudo two-dimensional manner. The cross-section and the out-of plane length change arbitrarily, but no
out-of-plane shearing is allowed. In this sketch, a grid has been conceptually painted onto the material
to facilitate visualizing the deformation.

Deformation
Since each particle occupies its own unique
location in space in both the initial and
deformed configurations, we may assert that
there exists a one-to-one function (called a
mapping function) such that

Material
interpenetration?
Nonphysical!

x = (X)
(1.1)

Examples of mapping functions are given in

Figure 1.2.
A locally admissible, but
Section 5. Naturally, no single material element
globally inadmissible deformation.
(i.e., no grid cell in Fig. 1.1) should be permitted to invert. This constraint is called local admissibility of the deformation, and it
is ensured if the matrix [ x i X j ] has a positive determinant. In finite element simulations, a technique called hourglass control is typically used to ensure local admissibility.
However, local admissibility is not sufficient to guarantee global admissibility,
which requires that the mapping function in Eq. (1.1) be invertible and therefore prohibits
material inter-penetration. Figure 1.2, for example, shows a globally inadmissible deformation that is everywhere locally admissible. In computational mechanics, contact algorithms must be used to avoid the globally inadmissible phenomenon of material
interpenetration. We will soon discuss in great detail the deformation gradient tensor, F ,

2
Copyright is reserved. Individual copies may be made for personal use. No part of this document may be reproduced for profit.

DRAF

December 7, 2008 10:51 pm

The deformation gradient

Rebec

ca Br
ann

on

whose ij components are given by [ x i X j ] . Occasionally, problems in mechanics

specify only how the F tensor varies in space without making explicit reference to the
mapping function, x = ( X ) . In these problems, another admissibility condition (called

the compatibility condition)

ensures that the mapping function exists even though it is
not being used explicitly compatibility ensures that the F tensor field is integrable

even if we choose not to integrate it.

Displacement
The displacement of a particle is a vector extending from a particles initial location
to its deformed location (see Fig. 1.1):
uxX

(1.2)

The initial and final position vectors, x and X , are origin-dependent (they depend on

where you place the origin). The displacement

vector
u is a free vector it wont change
displacement can be generalized
if you move the origin. The definitions of positions and
to permit different origins to be used in the initial and deformed configurations. If the
deformed origin is located by extending a vector c from the initial origin, then x in the

above expressions must be replaced by c + x . The origin-to-origin vector c does not vary

with position it is spatially constant.

2. The deformation gradient

The deformation gradient tensor (graphical introduction)
GOAL: Provide a physical understanding of the deformation gradient tensor prior to defining
it mathematically.

The sketch of deformation (Fig. 1.1) shows a square grid conceptually painted onto
the undeformed body, and some of the grid cells have been filled in with solid color to
help us better visualize the deformation. Note that each little material square deforms to a
shape that is close to a parallelogram in shape. This approximation (squares-to-parallelograms) becomes exact in the limit of an infinitely refined grid. In three dimensions, cubes
deform into parallelepipeds. On surfaces embedded in 3D space, infinitesimal squares
deform into infinitesimal parallelograms. The deformation gradient is a tensor that quantifies both the 3D and 2D shape change as well as overall material rotation, making it superior to strain as an all-encompassing measure of deformation of material elements. If you
have the deformation gradient tensor, then you can compute the strain; the converse is not
true. The purpose of this section is to describe how to graphically approximate the deformation gradient tensor by simply looking at the deformation. We will postpone giving an
exact mathematical definition of the deformation gradient tensor.

3
Copyright is reserved. Individual copies may be made for personal use. No part of this document may be reproduced for profit.

T
F
A
DR
ann
ca Br
Rebec

December 7, 2008 10:51 pm

The deformation gradient

on

Any cube can be characterized by the three orthonormal vectors that form its edges.
Likewise, the deformed parallelepiped can be characterized by the three vectors that form
its edges. The deformation gradient tensor F quantifies the changes in these edge vectors
into columns of a 3 3 matrix of compoby assembling the three deformed edge vectors
nents. For rectangular Cartesian coordinate systems, the columns of the deformation gradients component matrix contain the deformed parallelepiped edge vectors, expressed
relative to the undeformed edge vectors that defined the initial cube. By relative, we mean
that all length changes are expressed as multiples of the initial cubes edge lengths, and all
directions are expressed relative to the initial cubes edge directions. Consequently (by
appropriately defining our unit of length) we may pretend that the initial infinitesimal
cube is a unit cube whose three edge vectors are aligned with the orthonormal laboratory basis, { E1, E2, E3 } . Upon deformation, these edge vectors deform into a new set of

vectors, { g 1, g 2, g 3 } , that define the edges of the deformed parallelepiped. If you find the
of
the i th deformed edge vector and assemble those components into the
lab components
i th column of F , then the result will be the component matrix for F with respect to the
basis. Mathematically, this means that

{ E 1, E 2, E 3 } (lab)

gk = F Ek

(2.1)

or, equivalently,
g k = F jk E j ,

(2.2)

where F jk are the jk components of F with respect to the lab basis.* The k th column of
[ F jk ] contains the lab components of g k . These equations provide us with a method for

approximating the local deformation gradient

at a point whenever we have accurate pictures of the deformed cube (or, for 2D, the deformed square).
The deformed g k edge vectors are not generally orthogonal or of unit length. They are

called material vectors

because they travel along with the same set of material points
(rotating with and stretching in proportion to the grid lines). In general, if M is a material
vector in the initial configuration (not necessarily of unit length), then it will flow with the
material to become a new (stretched and rotated) vector m in the deformed configuration,

given by
m = F M , and the fiber stretch is defined

= m M

(2.3)

* Comparing Eqs. (2.1) and (2.2), you might (wrongly) think that we should have written Fkj instead
of F jk in Eq. (2.2). To see why Eq. Eq. (2.2) is correct, note that any vector v may be expressed in
terms of the lab basis as v = v k E k , where v k = v E k . Stated differently, any vector v may be writ

ten v = ( v Ek )E k , or, since

the dot product is commutative,
v = ( E k v )E k . This identity applies

for any vector v . Applying it to one of the stretched edge vectors, g k , gives g k = ( E i g k )E i , or,

substituting Eq. (2.1), g k = Fik E i , where F ik = E i F E k .

4
Copyright is reserved. Individual copies may be made for personal use. No part of this document may be reproduced for profit.

DRAF

December 7, 2008 10:51 pm

The deformation gradient

Rebec

ca Br
ann

on

Thus, if the F tensor is constructed from knowledge of how the unit lab base vectors
material at a point, then you can immediately compute how any material
would flow with
vector (not necessarily aligned with the lab basis and not necessarily of unit length) would
flow with the material. This is the fundamental characteristic of a linear transformation.
We will explain later that F ij = x i X j . The mapping function x = ( X ) is generally

nonlinear, but its derivative at a point is linear with respect to small changes in position.
This statement is analogous to what a freshman calculus student encounters when first
introduced to the concept of differentiation. A function y = f ( x ) might be nonlinear, but
one can always (for smooth functions) define a straight (i.e., linear) line that is tangent to
the curve at any point and the slope of the line is given by dy dx . The deformation gradient tensor, F ij = x i X j , plays a role in three dimensions that is analogous to the tangent slope dy dx in one dimension. The difficulty lies in visualization. In one dimension,
we can always plot the function y = f ( x ) and then show the tangent line on the graph. In
three dimensions, we must understand the mapping function x = ( X ) in a more abstract

manner by talking about what it does (locally) to little cubesof material.

Our perspective
here would be analogous to describing the 1D mapping y = f ( x ) by stating that a little
line segment dx deforms into a new line segment dy that differs in length by a ratio
s = dy dx so that dy = sdx . For generalization into 3D mapping, this last equation
becomes dx = F dX .

Before refining this loose mathematical discussion, lets spend some time constructing
some deformation gradient tensors graphically. What we will be doing here is analogous
to taking out a ruler and estimating the local slope of a 1D curve y = f ( x ) by measuring
local rise over run.

EXAMPLES
Consider the top material element sketched in Fig. 2.1. Zooming in on the undeformed
and deformed shapes of these elements, and introducing a ruler for which the initial
cube has unit length gives the following drawing:

g = 1.78E 1 + 1.27E 2

E2

Deformed

Undeformed

g 1 = 1.3E 1 0.05E 2

E1

Figure 2.1.
A graphical determination of the deformation gradient This sketch shows a zoomed in
picture of the small material element at the top of Figure 1.1 on page 2. A ruler has been drawn on this
figure so that the undeformed square is a unit square.

5
Copyright is reserved. Individual copies may be made for personal use. No part of this document may be reproduced for profit.

T
F
A
DR
ann
ca Br
Rebec

December 7, 2008 10:51 pm

The deformation gradient

on

Using the rulers as a guide, we can read off the components of each of the deformed
edge vectors to obtain:
1.3
{ g1 } =

0.05

and

1.78
{ g2 } =

1.27

(2.4)

Assembling these components into columns of a 2 2 matrix gives

[ F ] = 1.3 1.78

0.05 1.27

(2.5)

The fiber stretches are

1 = g 1 E 1 = 1.3
and 2 = g 2 E 2 = 2.2
(2.6)

which mean that an infinitesimal fiber originally aligned with E 1 will, after deformation,
have a length 1.3 times as large. A fiber originally aligned with E 2 will change length by

a factor of 2.2.
The deformation in Fig. 1.1 is actually three dimensional the out-of-plane thickness
has increased by about 50%, which corresponds to an out-of-plane stretch ( L Lo ) equal
to 1.5 . Consequently, the E 3 base vector has deformed to become g 3 = 1.5E 3 . Conse

quently, the pseudo 2D deformation

gradient in Eq. (2.5) can be upgraded
to 3D as
1.3 1.78 0
[ F ] = 0.05 1.27 0

0
0 1.5

(2.7)

6
Copyright is reserved. Individual copies may be made for personal use. No part of this document may be reproduced for profit.

DRAF

December 7, 2008 10:51 pm

The deformation gradient

Rebec

ca Br
ann

on

If we zoom in on the middle element in our original sketch and graphically measure the
deformed material element edge vectors, then the deformation gradient is found as follows:

g = 1.5E 2

2
E2

Deformed
Undeformed

g = E1

E1

Component arrays:
1
{ g1 } =

0
{ g2 } =

1.5

[F] = 1 0

0 1.5

Figure 2.2.
A graphical approximation of the local deformation gradient This figure
shows an enlarged picture of the middle material element shown in Figure 1.1 on page 2. This
material element happens to be in a state of uniaxial strain in the 2-direction.

Finally, the material element in the lower left of Figure 1.1 on page 2 is deformed
according to a deformation gradient approximated graphically as follows:

g = 0.4E 1 + E 2

E2

Undeformed

Deformed

g = E1

E1

Component arrays:
1
{ g1 } =

0
Figure 2.3.

0.4
{ g2 } =

[ F ] = 1 0.4

0 1

Graphical determination of the deformation gradient for the bottom element in Fig. 1.1

This material element coincidentally appears to be in a state of pure shear.

7
Copyright is reserved. Individual copies may be made for personal use. No part of this document may be reproduced for profit.

T
F
A
DR
ann
ca Br
Rebec

December 7, 2008 10:51 pm

The deformation gradient

on

Physically interpreting a deformation gradient tensor

Consider the inverse problem: suppose you know the deformation gradient (e.g., as
output from a structural mechanics finite element code), and you wish to visualize the
nature of the deformation by sketching the material element. The process is simple. Just
interpret the i th column of [ F ] to contain the lab components of the deformed edge vec
tors, and then draw the deformed
element accordingly. Suppose, for example, that you are
given
1
--- 2
[F] = 2

1
1 --4

(2.8)

The first column contains g 1 and the second column contains g 2 . Specifically,

1 2
2
and { g 2 } =
{ g1 } =

1 4

E2

(2.9)

g = 2E 1 1--4- E 2

rmed
Defo

Undeformed

E1

g =0.5E 1 E 2

Figure 2.4.
Sketching the deformation when the deformation gradient is known This deformation corresponds to the deformation gradient in Eq. (2.8).

Being only a collection of four numbers, the deformation gradient matrix in Eq. (2.9)
does not have much obvious physical meaning by itself. But, by drawing the associated
deformation, we can now see that this deformation is essentially a movement of the material consisting of stretching it in the 2-direction, then shearing it in the 1-direction, and
finally rotating it clockwise.

8
Copyright is reserved. Individual copies may be made for personal use. No part of this document may be reproduced for profit.

DRAF

December 7, 2008 10:51 pm

The deformation gradient

Rebec

ca Br
ann

on

Sequentially applied deformations

Suppose that some total deformation F is achieved via two separate deformation phases,

the first being defined by one deformation

gradient tensor F ( 1 ) and the next defined by
(
2
)
F . At the end of the first phase, a material vector M willstretch and rotate into a new

vector
m intermediate = F ( 1 ) M

At the end of the second phase, this vector will become

m = F ( 2 ) m intermediate = F ( 2 ) F ( 1 ) M

Thus, the overall final deformation gradient tensor is

F = F(2) F(1)

(2.10)

(2.11)

(2.12)

Note that the tensors are multiplied in reverse order of application. Changing the order of
application will result in a different material deformation (See Fig. 2.5).

stretch
horizontally

then rotate
45 degrees

then stretch
horizontally

rotate
45 degrees

Figure 2.5.

Illustration that order of application matters!

9
Copyright is reserved. Individual copies may be made for personal use. No part of this document may be reproduced for profit.

T
F
A
DR
ann
ca Br
Rebec

December 7, 2008 10:51 pm

The deformation gradient

on

Tracking volume changes

A volume element in the undeformed configuration can be defined by the parallelepiped formed by of any three (infinitesimal) material edge vectors, { M 1, M 2, M 3 } . The
the triple
scalar
undeformed volume of this infinitesimal material element is given by
product:
V o = [ M 1, M 2, M 3 ]
(2.13)

The initial material vectors each stretch and rotate, becoming new deformed vectors,
{ m 1, m 2, m 3 } , that define the deformed shape of the material element. The deformed vol isgiven
by
ume
V = [ m 1, m 2, m 3 ]
(2.14)

or, noting that m k = F M k ,

V = [ F M 1, F M 2, F M 3 ]
(2.15)

Expanding this out in indicial notation and using the definition of a determinant reveals
that
V = ( detF ) [ M 1, M 2, M 3 ] = detF V o ,

(2.16)

V----= detF
Vo

(2.17)

or

where J is called the Jacobian of the deformation, defined by

J detF
(2.18)

In words, the determinant of the deformation gradient tensor equals the ratio of the
deformed to the undeformed infinitesimal volume elements. For realistic deformations,
both volume elements must be positive. Therefore, we may assert that
J>0

(2.19)

Therefore, the deformation gradient tensor must be invertible. This is the mathematical
statement of local admissibility, discussed on page 2.

Tracking surfaces
The gk vectors move with the material. They always lie along the same set of points

(more correctly,
theyre always tangent to and stretch with the grid lines). In three dimensions, the triad { g 1, g 2, g 3 } forms the edges of a parallelepiped that defines the deformed
element.

shape of the material

Upon deformation, a face of the parallelepiped is defined
by the two material edge vectors that form that face. If, for example, we consider the face
formed by g 2 and g 3 , then the normal to this face is proportional to g 2 g 3

10
Copyright is reserved. Individual copies may be made for personal use. No part of this document may be reproduced for profit.

DRAF

December 7, 2008 10:51 pm

The deformation gradient

Rebec

ca Br
ann

on

Even though the edge vectors move with the material, the outward normals to planes
do not move with the material. Consider the deformation shown in Fig. 2.4, for which the
deformation gradient is
1
--2

2
[F] =

1
1 ---

(2.20)

To extend this example to three dimensions (so that we can talk about planes), lets suppose that the deformation is actually that of a thick plate (no stretching occurs in the outof-plane direction), so the 3D deformation gradient is
1
--2

2 0

[F] =
1
1 --- 0

(2.21)

0 0 1
We have already explained that g k comes from the k th column of F , which means

gk = F Ek
(2.22)

Now we introduce complementary vectors k defined by

i = F T E i
(2.23)

Note that
i g k = ik
(2.24)

Incidentally, because the vector i obeys Eq. (2.24), it is called the dual or contravari in publications that employ convected coordinates, the
ant vector associated with g k and,
g i (which is distinguished from g because it uses a supervector i is typically denoted
i
will continue to use the symbol
for the dual vector.

script rather
than a subscript). We
i

Suppose, for example, 1 is perpendicular to both g 2 and g 3 ; hence, it must be per as the normal goes, the

pendicular to the material surface

formed by g 2 and g3 . As far
for computational convenience.
magnitude of 1 is inconsequential and may be chosen

Specifically, our definition in Eq. (2.23) corresponds to

( g2 g3 )
1 = --------------------
- , where g o g 1 ( g 2 g 3 )
go

(2.25)

In three dimensions
[ FT ]

1 4 0
2
= ------ 8 2 0
15
0 01

(2.26)

11
Copyright is reserved. Individual copies may be made for personal use. No part of this document may be reproduced for profit.

T
F
A
DR
ann
ca Br
Rebec

December 7, 2008 10:51 pm

The deformation gradient

on

The columns of this matrix contain the lab components of the v k vectors, sketched in Fig.

2.6. This picture shows quite clearly that normals to material planes
are not material vectors the dont flow with the material.

N2

E2

Undeformed

g
1

E1

rmed
Defo

N1

g
2

1
Figure 2.6.
Tracking material surfaces This deformation corresponds to the deformation gradient in
Eq. (2.8). The deformed edges are coincident with g i = F E i outward normals to the material surfaces

are parallel to i = F T Ei .

In general, if a material plane (i.e., one that moves with the material) has a normal N (not
the
required to be of unit length) in the undeformed configuration, then after deformation,
vector
= FT N
(2.27)

(again not generally of unit length) will be normal to the plane in the deformed configuration. If the unit normal is desired, then this result can simply be normalized.

12
Copyright is reserved. Individual copies may be made for personal use. No part of this document may be reproduced for profit.

DRAF

December 7, 2008 10:51 pm

The deformation gradient

Rebec

ca Br
ann

on

Area vectors
Let M 1 and M 2 be any two material vectors in the undeformed configuration. The
parallelogram

area of the
formed by these two vectors is given by
Ao = M1 M2

(2.28)

If the material vectors are infinitesimal, then this area vector will be infinitesimal. Upon
deformation, the M k vectors deform into new stretched and rotated vectors, m 1 and m 2 ,
vector

that define a new area

A = m1 m2

(2.29)

Substituting Eq. (2.3) into this expression gives

A = ( F M 1 ) ( F M2 )

Using an identity from tensor analysis, this result can be written as
A = F c ( M1 M 2 )

(2.30)

(2.31)

or
A = F c Ao

(2.32)

where F c is the cofactor of F . Since the deformation gradient is invertible, the cofactor
may be alternatively written
F c = ( detF )F T

(2.33)

Recalling that J = detF and also recalling that M 1 M 2 is just the undeformed area vec

tor, A o , Eq. (2.32) maybe written

A = JF T A o

(2.34)

This famous result is called Nansons relation.

13
Copyright is reserved. Individual copies may be made for personal use. No part of this document may be reproduced for profit.

T
F
A
DR
ann
ca Br
Rebec

December 7, 2008 10:51 pm

The deformation gradient

on

Deformation gradient (precise mathematical definition)

So far, we have described a graphical method of approximating the deformation gradient. Namely, you can paint a small square in the material, and then construct the deformation gradient by assembling the deformed material vectors into columns of [ F ] . Now

we will provide the quantitative definition of the deformation gradient. The deformation
gradient is defined such that an infinitesimal material line segment dX deforms into a new

infinitesimal line segment dx so that

dx = F dX

(2.35)

or, by the chain rule,

dx
F = ------dX

(2.36)

This conclusion holds because (recall) the deformation can be described through a mapping:
x = ( X)
(2.37)

In a rectangular Cartesian coordinate system, the components of the deformation gradient

tensor are
x i
F ij = -------X j

(2.38)

In matrix form,
x x x
--------1- --------1- --------1X1 X2 X 3
x x x
[ F ] = --------2- --------2- --------2
X1 X2 X 3

(2.39)

x x x
--------3- --------3- --------3X1 X2 X 3

Examples may be found starting on page 44.

Homogenous deformations
We have described the deformation gradient tensor in the context of inhomogenous
deformations. Referring to the definition in Eq. (2.38), note that the deformation gradient
tensor itself generally varies with position. Consequently, the deformation of one infinitesimal material cubes generally will be different from that of another cube located elsewhere
in the body.

14
Copyright is reserved. Individual copies may be made for personal use. No part of this document may be reproduced for profit.

DRAF

December 7, 2008 10:51 pm

The deformation gradient

Rebec

ca Br
ann

on

A homogenous deformation is a very special kind of deformation in which the

deformation gradient has the same value everywhere in the body. In this case, Eq. (2.36)
can be integrated to give the mapping function,
x = F X + c,
(2.40)

where c is a constant of integration representing material translation. For homogenous

deformations,
straight lines deform to straight lines, planes deform to planes, cubes (no
matter how large) deform to parallelepipeds, and spheres deform to ellipsoids. Contrast
this with an inhomogenous deformation such as the one illustrated in Fig. 1.1, where originally straight grid lines become curved and only infinitesimal cubes deform to parallelepipeds. These distinctions are illustrated in Fig. 2.7.

Pure
Rotation

Initial Configuration

General inhomogenous
Deformation

Homogeneous
Deformation

Figure 2.7.
Distinctions between pure rotation, homogenous deformation and inhomogenous deformation.

15
Copyright is reserved. Individual copies may be made for personal use. No part of this document may be reproduced for profit.

T
F
A
DR
ann
ca Br
Rebec

December 7, 2008 10:51 pm

The deformation gradient

on

Displacements for Homogenous deformations. Recall that, for a homogenous deformation,

x = F X + c,
(2.41)

where F is spatially uniform and c (which is also spatially uniform) represents the dis of the particle originally located at the origin in the reference frame.
placement
The displacement field, u = x X , is therefore

u = H X + c , where H F I

Of course, Eq. (2.41) can be alternatively written

(2.42)

X = F 1 x C , where C F 1 c
(2.43)

Since neither F nor c vary spatially, the new alternative translation vector C is also spa
tially constant. The vector
C is the displacement of the particle currently at the origin.

Using Eq. (2.43), the displacement,

u = x X , can be written

1
u = h X + C , where h I F
(2.44)

Both Eq. (2.42) and Eq. (2.44) will give the same result for the displacement field.
EXAMPLE: Consider purely circular motion about the 3-direction. Since the particle
at the origin does not move, we have c = C = 0 . For circular motion about the x 3 axis,

the deformation gradient is given by

cos sin 0
[ F ] = sin cos 0

0
0 1

(2.45)

Therefore
x 1 = X 1 cos X2 sin
x 2 = X 1 sin + X 2 cos
x3 = X3

(2.46)

We can introduce cylindrical coordinate systems such that

X 1 = R cos , X 2 = R sin

(2.47)

x 1 = r cos , x 2 = r sin

(2.48)

Then Eq. (2.46) becomes

r cos = R cos cos R sin sin
r sin = R cos sin + R sin cos ,

(2.49)

or, using trig identities on the right hand sides,

r cos = R cos ( + )
r sin = R sin ( + ) ,

(2.50)

16
Copyright is reserved. Individual copies may be made for personal use. No part of this document may be reproduced for profit.

DRAF

December 7, 2008 10:51 pm

The deformation gradient

Rebec

ca Br
ann

on

from which we obtain the intuitively obvious result

r = R

and

= +

(2.51)

In other words, under circular motion, the radial coordinate remains unchanged while the
angular coordinate increases by the rotation angle.
Now lets look at the far less obvious results for the displacement vector. It would be
easy to fall into the trap of thinking that the radial component of displacement must be
zero. However, even though the radius does not change, the radial displacement is nonzero. To prove this methodically, lets go back to the RCS result of Eq. (2.46). The displacements are found by simply subtracting the initial Cartesian coordinate X k to give
u 1 = X 1 ( cos 1 ) X 2 sin
u 2 = X 1 sin + X 2 ( cos 1 )
u3 = 0

(2.52)

From here, we can introduce a cylindrical coordinate system. The base vectors for the system are different in the reference and current systems. Namely, the reference cylindrical
basis is
E R = cos E 1 + sin E 2

E = sin E 1 + cos E 2

and therefore, with respect to this system,

(2.53)

u R = u E R = u 1 cos + u 2 sin = cos 1

u = sin

(2.54)

Rebecca: add more details proving this result

A similar analysis using the spatial cylindrical system gives
u r = u e r = u 1 cos + u 2 sin = 1 cos

u = sin

(2.55)

Constructing the deformation gradient using dyads

In the previous example, we considered only the deformation of material fibers that
were initially aligned with the lab basis. This is the easiest way to construct the deformation gradient tensor. If however, you start with three linearly independent material fibers
{ M 1, M 2, M 3 } (this time not necessarily of unit length) in the undeformed direction, then
for constructing the deformation gradient must be modified. These concepts
the procedure
are useful in experimental diagnostics. For example, application to strain rosettes is discussed at the end of this section.
Method 1: You can assemble the lab components of the { M 1, M 2, M 3 } vectors into col
umns of a matrix f so that
o

17
Copyright is reserved. Individual copies may be made for personal use. No part of this document may be reproduced for profit.

T
F
A
DR
ann
ca Br
Rebec

December 7, 2008 10:51 pm

The deformation gradient

on

Mk = f Ek .
(2.56)

o
If these vectors become { m 1, m 2, m 3 } in the deformed configuration, then you can
into a second matrix f so that
likewise assemble their lab components

mk = f Ek .
(2.57)

We seek the deformation gradient tensor that connects the initial and final configuration
directly. In other words, we seek a tensor F such that

mk = F Mk .
(2.58)

Solving Eq. (2.56) for E k gives E k = f 1 M k . Substituting this result into Eq. (2.57)

o
gives
m k = f f 1 M k

o
Comparing Eq. (2.58) with Eq. (2.59) tells us that
F = f f 1

(2.59)

(2.60)

Method 2: Another way to achieve the same result is to employ dyads and dual bases.
Again, let { M 1, M 2, M 3 } denote three linearly independent infinitesimal material vectors
3D configuration. Let { , , } be the associated dual (or contrain the undeformed
1 2 3

variant) vectors. That is,
M2 M3
M3 M 1
M1 M2

,
1 = -------------------- 2 = -------------------- 3 = --------------------Mo
Mo
Mo

(2.61)
where M o ( M 1 M 2 ) M 3

If the material line segments { M 1, M 2, M 3 } deform to become new line segments

tensor may be constructed by
{ m 1, m 2, m 3 } , then the deformationgradient

3

F =

m k k

(2.62)

k=1

Here, the symbol denotes dyadic multiplication. For any two vectors a and b , the

dyad, a b , is a second order tensor whose ij component equals a i b j .

Eq. (2.60) and Eq. (2.62) will both give the same result.
Example: Suppose that a mounting bracket for the landing gear on an airplane was photographed during the most recent inspection of the plane. Following an unusually rough
landing, a new photograph revealed that the bracket had undergone some severe deformation as shown in Fig. 2.8. We will use the lines formed by the ACME landers company
logo to determine the deformation gradient for the bracket. For the purpose of this exam-

18
Copyright is reserved. Individual copies may be made for personal use. No part of this document may be reproduced for profit.

DRAF

December 7, 2008 10:51 pm

The deformation gradient

Rebec

ca Br
ann

on

ple, we will presume that the bracket underwent homogenous deformation. Of course, in
real life, the deformation would likely be inhomogenous and the technique illustrated
below would have to be applied using smaller line segments (such pairs of intersecting
microscopic scratches) to find the deformation at various points in the bracket.

Before

AC
M

Figure 2.8.

After

ers
d
n
a
l

ACME

d
lan

er

Airplane landing gear mounting bracket, before and after deformation

Introducing rulers as shown, the straight lines forming the company logo can be
regarded as vectors. The line below the word ACME will be taken to be M 1 , pointing

up and to the left; the line below landers will be M 2 , pointing up and to the right.
Before

deformation, the rulers show these vectors to have lab components as follows:
M 1 = 0.875E 1 + 0.7E 2

M 2 = 0.45E 1 + 0.125E 2

After deformation, the logo vectors deformed into

(2.63)

m 1 = 1.1E 1 + 0.1E 2

m 2 = 0.325E 1 + 0.175E 2

We will presume that the bracket did not deform at all out of plane so that

(2.65)

(2.64)

(2.66)

M 3 = E 3 and m 3 = E 3
(2.67)

Using method #1 for finding the deformation gradient tensor, the f is constructed by
o
assembling the lab components of the undeformed vectors into columns:
[f ] =
o

0.875 0.45 0
0.7 0.125 0
0
0 1

(2.68)

Similarly assembling the deformed vectors gives the f tensor as

19
Copyright is reserved. Individual copies may be made for personal use. No part of this document may be reproduced for profit.

T
F
A
DR
ann
ca Br
Rebec

[f] =

December 7, 2008 10:51 pm

The deformation gradient

on

1.1 0.325 0
0.1 0.175 0
0
0 1

(2.69)

Applying Eq. (2.60) gives

[ F ] = [ f ] [ f ] 1

1.1 0.325 0 0.875 0.45 0

= 0.1 0.175 0
0.7 0.125 0
0
0 1
0
0 1

1.1 0.325 0 0.2945 1.060 0

0.860 0.496 0
= 0.1 0.175 0 1.6495 2.062 0 = 0.259 0.467 0
0
0 1
0
0 1
0
0
1

(2.70)

Solving the same problem using method #2, we must first construct the dual vectors associated with { M 1, M 2, M 3 } . To start, note that

M 1 M 2 = 0.4244E 3
(2.71)

(2.72)
M 2 M 3 = 0.125E 1 0.45E 2

(2.73)
M 3 M 1 = 0.7E 1 0.875E 2

Therefore, referring to Eq. (2.61),

M o ( M 1 M 2 ) M 3 = 0.4244

1 = 0.29455 E 1 + 1.0604E 2

2 = 1.6495E 1 + 2.06186E 2

3 = E3 ,

Recall that the deformed vectors were measured to be

(2.74)

(2.75)

m 1 = 1.1E 1 + 0.1E 2
(2.76)

(2.77)
m 2 = 0.325E 1 + 0.175E 2

Referring to Eq. (2.62), we need to compute the dyads, m 1 1 , m 2 2 , and m 3 3 :

1.1
0.324005 1.16644 0
m 1 1 = 0.1 0.29455 1.0604 0 = 0.029455 0.10604 0

0
0
0
0

(2.78)

0.325
0.5360875 0.6701045 0
m 2 2 = 0.175 1.6495 2.06186 0 = 0.2886625 0.3608255 0

0
0
0
0

(2.79)

20
Copyright is reserved. Individual copies may be made for personal use. No part of this document may be reproduced for profit.

DRAF

December 7, 2008 10:51 pm

The deformation gradient

Rebec

0
0 0 0
[ m2 2 ] = 0 0 0 1 = 0 0 0

1
0 0 1

ca Br
ann

on

(2.80)

Now, applying Eq. (2.62) adding up the above tensors gives

0.860 0.496 0
[ F ] = 0.259 0.467 0

0
0
1

(2.81)

which agrees with the result obtained using method #1. Obviously, for this particular problem, method #1 is superior in its computational simplicity. The fundamental concepts
associated with method #2 are nonetheless important to understand and often easier to
work with in theoretical discussions.
Strain rosette example: a strain rosette orients three strain gages at equal ( 120 ) angles.
To ensure measurement at a single point, the gages are centered at the same location; however, to analyze the rosette, you can imagine that the gages are actually placed along edges
of an equilateral triangle and then presume that the material in the triangle deforms homogeneously by the same amount as the original central point. Strain gages return information about length changes but not orientation changes. In principle, if you could accurately
measure the orientation changes of the first two gages, then the third one would not be
necessary. The redundant third strain gage directly measures the relative orientation
change between the first two gages. Thus, strain rosettes really measure only stretch
they do not characterize rotation. The full deformation gradient (which includes rotation)
can always be constructed by first analyzing the rosette with the fiducial assumption
that the first gage does not rotate. This will produce a fiducial deformation gradient and
the actual deformation gradient can be constructed by then superimposing the actual rotation of the first gage onto the fiducial deformation. In three dimensions, constructing a
deformation gradient is more complicated because the reorientations of two gages must be
tracked.

21
Copyright is reserved. Individual copies may be made for personal use. No part of this document may be reproduced for profit.

T
F
A
DR
ann
ca Br
Rebec

December 7, 2008 10:51 pm

Polar decomposition

on

3. Polar decomposition
A rotation is a special kind of deformation in which material vectors permissibly
change orientation, but they dont change length. In this case, it can be shown [see gobag
rotation document] that the associated deformation gradient will be orthogonal (its inverse
will equal its transpose). Furthermore, since element inversions are prohibited, rotations
will be proper orthogonal (determinant will equal +1 ) Rotations are discussed in great
detail in the gobag Rotation document.
A stretch is a completely different special kind of deformation in which there exist
three material vector orientations in the 3D initial configuration that will change in length,
but not in orientation. In this case, the deformation gradient tensor will be both symmetric
and positive definite. The proviso about positive definiteness is important. If a deformation gradient tensor is symmetric, that does not necessarily mean it is a stretch. For example, a pure rotation of 180 will result in a symmetric deformation gradient, but its not a
stretch because it possesses a negative eigenvalue. Being symmetric, a stretch is diagonal
in its principal basis. The principal values, called principal stretches, equal the ratio of
deformed to undeformed lengths of the three non-rotating material fibers. Material fibers
that are not aligned with the principal directions of a stretch will change orientation, but
for every fiber rotating one way, there will be another rotating in the opposite direction,
making the net rotation of material fibers zero for a pure stretch deformation.
The polar decomposition theorem is a mathematical statement that the (local) deformation of a material element may be regarded as a combination of a pure stretch and a
pure rotation. You can stretch a material and then rotate it or vice versa. Recall that order
of application of sequentially applied deformations matters. Therefore, if you want to end
up with the same deformation, then you must conclude that the stretch used in the stretchthen-rotate scenario would have to be different from the stretch used in the rotate-thenstretch scenario. You might also suspect that the rotations would also have to be different
in the two scenarios. While it is true that the stretches are different, it can be proved that
the rotation is the same in both cases. The polar decomposition states that there exists a
rotation tensor R , a stretch-first tensor U and a stretch last tensor V such that

Stretch then rotate:

Rotate then stretch:

F = RU

F = VR

Note:

Note:

U is called the right stretch

V is called the left stretch

(3.1)

(3.2)

Note that the Jacobian is J = detF = detU = detV . The following page summarizes

key polar decomposition formulas (proof omitted).

22
Copyright is reserved. Individual copies may be made for personal use. No part of this document may be reproduced for profit.

DRAF

December 7, 2008 10:51 pm

Polar decomposition

Rebec

ca Br
ann

on

POLAR DECOMPOSITION
F = RU

F = VR

First STRETCH
then ROTATE

First ROTATE
then STRETCH

Same starting
configuration
[F] = 1 0

01

Stretch & compress

along red lines
(40% strain)

Rotate by 150
degrees

Different
intermediate
configurations

Rotate by 150
degrees

Stretch & compress

along red lines
(40% strain)

Same ending
configuration
[ F ] = 0.795 0.925

0.154 1.074

This figure shows a deformation for which principal Hencky (logarithmic) strains
are 40% and the material rotation is 150 . This deformation can be accomplished via
two stages: stretch and rotation. The polar decomposition theorem says that you will get
the same final result regardless of whether you rotate first or rotate last. The key governing equations are:
U = ( FT F ) 1 / 2
V = (F FT)1 / 2 = R U RT

1
R = FU
R = V F

For 2D deformations, the rotation tensor can be directly computed by

F + F 22 F 12 F 21
1
[ R ] = ----------------------------------------------------------------------- 11

( F 11 + F 22 ) 2 + ( F 12 F 21 ) 2 F 21 F 12 F 22 + F 11
Then U = R T F and V = F R T

23
Copyright is reserved. Individual copies may be made for personal use. No part of this document may be reproduced for profit.

T
F
A
DR
ann
ca Br
Rebec

December 7, 2008 10:51 pm

Polar decomposition

on

Recall that the deformation of an infinitesimal line element dX is given by

dx = F dX

In terms of the polar rotation and stretch, this becomes

(3.3)

dx = R U dX
(3.4)

Broken down in this manner, we see that U dX quantifies the stretching of the material
that results strictly from the material dis
fiber dX , as well as the part of the fiber rotation
tortion (i.e., its shape change only). Under the pure stretch, U dX , fibers that are aligned
with the principal direc
with the principal directions of U wont rotate. Fibers not aligned

tions will rotate, but for any fiber that rotates one way, there will exist a different fiber
rotating in the opposite direction. Consequently, the overall material rotation under a pure
stretch will be zero. The total deformation is obtained after the pure stretch by the application of the rotation tensor R , which rotates all fibers about the same axis by the same
angle of rotation. Hence, R is really is an excellent measure of overall material rotation.

EXAMPLE: Consider the deformation in Eq. (2.81):

0.860 0.496 0
[ F ] = 0.259 0.467 0

0
0
1

(3.5)

This is a two-dimensional deformation, so the rotation tensor is found by

F + F22 F12 F 21
1
[ R ] = ----------------------------------------------------------------------- 11

( F 11 + F 22 ) 2 + ( F 12 F 21 ) 2 F 21 F 12 F 22 + F 11
1
= ------------------------------------------------------- 1.327 0.755
( 1.327 ) 2 + ( 0.755 ) 2 0.755 1.327
= 0.8692 0.4945
0.4945 0.8692

(3.6)

Since, for 2D deformations, R 11 = cos and R21 = sin , where is the rotation
angle, we note that = 30 .
The reference stretch is
[ U ] = [ R ] T [ F ] = 0.8692 0.4945 0.860 0.496 = 0.876 0.200

0.4945 0.8692 0.259 0.467

0.200 0.651

(3.7)

The spatial stretch is

[ V ] = [ F ] [ R ] T = 0.860 0.496 0.8692 0.4945 = 0.993 0.006

0.259 0.467 0.4945 0.8692
0.006 0.534

(3.8)

24
Copyright is reserved. Individual copies may be made for personal use. No part of this document may be reproduced for profit.

DRAF

December 7, 2008 10:51 pm

Polar decomposition

Rebec

ca Br
ann

on

Note that the off-diagonals ( V 12 and V 21 ) are nearly equal to zero, which means the principal directions of the spatial stretch just happened to turn out to coincide with the lab
basis for this particular example. Also note that the 11 component of the stretch is nearly
equal to 1. For this particular problem, if it werent for computational round-off, V 12 and
V 21 have turned out to be identically zero and V 11 would have been identically equal to 1.
We are able to make this assertion because we know how we created Fig. 2.8 (which corresponds to this deformation) in the first place. That figure was created by using the computers drawing tools to first rotate the undeformed configuration by an angle = 30
and then to reduce the height of the rotated drawing by approximately 50% (note that V 22
is nearly equal to 50% of unity). Since the drawing was obtained by first rotating and then
stretching, we know that the rotation angle computed from the polar decomposition had to
come out equal to our initial rotation angle, and the spatial stretch had to come out to be a
50% vertical compression, as it has! We could have alternatively constructed the drawing
by using a stretch then rotate sequence involving a 50% compression of the initial configuration along a line oriented at 30 followed by a rotation of +30 . In general, the principal directions of stretch are neither coincident with the lab directions nor related in any
way to the rotation angle.
Incidentally, upgrading the above stretches and rotations to 3D gives
0.8692 0.4945 0
[ R ] = 0.4945 0.8692 0

0
0
1

(3.9)

0.876 0.200 0
[ U ] = 0.200 0.651 0

0
0
1

(3.10)

0.993 0.006 0
[ V ] = 0.006 0.534 0

0
0
1

(3.11)

25
Copyright is reserved. Individual copies may be made for personal use. No part of this document may be reproduced for profit.

T
F
A
DR
ann
ca Br
Rebec

December 7, 2008 10:51 pm

Strain

on

4. Strain
A infinite number of strain definitions
The deformation gradient is truly the best measure of deformation. It is especially
attractive because it contains information about both stretch and material rotation. For
one-dimensional homogenous uniaxial deformations, the axial strain is often defined as
L Lo
= --------------- = 1 ,
Lo

where

L
= ----Lo

(4.1)

The parameter is called the axial stretch.

Axial strain is often defined alternatively as
L Lo
= --------------- = [ 1 1 ]
L

(4.2)

Many advanced materials models define the strain as

= ln
(4.3)
Note that all you really need to describe the amount of deformation is the stretch . All of
these strains definitions are of the general Seth-Hill form,*
1
= --- [ k 1 ]
k

(4.4)

The parameter k characterizes the strain

definition. By appropriate choices for k ,

any of the above strain definitions can be
achieved. For example, Eq. (4.1) is
obtained by choosing k=1 , while
k=2
Eq. (4.2) corresponds to k= 1 . The log
strain in Eq. (4.3), also known as Hencky
strain, is obtained in the limit as k 0 .
All of these various strain definitions are
equivalent to first order when length
k0
changes are small (i.e., when is close to
Figure 4.1.
1). Thats because a two-term Taylor
sures.
series expansion of Eq. (4.4) centered at
=1 gives 1 , regardless of the value of k .

k=2
k0
k= 2

k= 2

The Seth-Hill family of strain mea-

* Referred to as Erikson-Doyle in the rheology literature.

26
Copyright is reserved. Individual copies may be made for personal use. No part of this document may be reproduced for profit.

DRAF

December 7, 2008 10:51 pm

Strain

Rebec

ca Br
ann

on

The variation of strain with stretch is shown in Fig. 4.1 for various choices of the SethHill parameter k . All of the strain measures are negative in compression ( < 1 ) and positive in tension ( > 1 ) . For positive choices of k , extreme compression toward the limit
as 0 results in the strain measure approaching a finite limit. Physically, one would
expect that an infinite load would be required to compress a material down to nothing.
Consequently, if a choice k > 0 were to be used in a linear stress-strain relationship,
= E , this behavior would imply that complete compression (a physical impossibility)
could be obtained via a finite load. Such a result is nonphysical and therefore a nonlinear
constitutive model must be used when using k > 0 in a highly compressed material. Otherwise numerical solutions will exhibit compression instabilities. When choosing a negative
Seth-Hill parameter ( k < 0 ) , the opposite happens: positive choices for k become unstable if a linear stress-strain relation is used in a highly tensile (large stretch) application.*
The logarithmic strain ( k 0 ) is the only strain choice that avoids both of these (compression and tension) pitfalls. This does not mean that a materials stress-strain relationship will necessarily be linear with respect to a logarithmic strain measure; it only means
that a first-order Taylor series expansion of the actual (generally nonlinear) stress-strain
function is likely to be more accurate than other strain measures. Other strain measures
can be used, and the same level of accuracy can be obtained if more terms in a Taylor
series expansion are employed.
Recall that a stretch is the ratio of deformed to undeformed lengths. For general deformations, we have already shown that the Jacobian ( J = detF = detU = detV ) equals

analog of
the linthe ratio of deformed to undeformed volumes. Thus, J is the volumetric
ear stretch and we may define volume strains in a manner consistent with the linear
strain definition given in Eq. (4.4). Namely, the consistent volumetric strain is defined
1
v = --- [ J k 1 ]
k

(4.5)

This volumetric strain is well posed for any 3D deformation. Soon, we will define strain
tensors and then prove that the consistent volumetric strain is not equal to the trace of the
strain tensor except when k=0 (i.e., log strain).

* This is less likely to be troublesome in practice because usually some other inelastic failure mechanism (e.g. fracture) usually initiates well before the unstable tensile regime is reached.

27
Copyright is reserved. Individual copies may be made for personal use. No part of this document may be reproduced for profit.

T
F
A
DR
ann
ca Br
Rebec

December 7, 2008 10:51 pm

Strain

on

Strain tensors
For 3D deformations, linear strain measures can still be defined in terms of the ratio of
the deformed to undeformed lengths of an individual material fiber. However, for 3D, differently oriented fibers will experience different amounts of stretch. Furthermore, two
material fibers can change their relative orientations under 3D deformations, so we need to
quantify that phenomenon as well. We have already emphasized that the deformation gradient tensor already does an excellent job in this respect. The deformation gradient is, by
far, the superior measure of material deformation. Nonetheless, a lot of people are more
comfortable with using strain instead of the deformation gradient, so we will describe now
how to compute strain from a deformation gradient tensor. Keep in mind that strain tensors
lose information (about rotation). You can construct any of the strain tensors if you have
the deformation gradient, but you cannot do the reverse.
The 1D definition of strain given in Eq. (4.3) was phrased in terms of the ratio of
deformed to undeformed lengths. We can extend this concept to 3D by using the polar
decomposition theorem. The stretch tensor from the polar decomposition is positive definite and symmetric and positive definite. Consequently, it is diagonal in its principal basis
and all of the eigenvalues are positive. In the absence of material rotation (i.e., if R = I )

any material fiber oriented in one the i th principal stretch directions will change length
by
a factor equal to the eigenvalue, i , but it will not change orientation. This suggests that
we can use the Seth-Hill generalized strain definition in each of these directions, and then
assemble the resulting three principal strains into a tensor that is diagonal with respect to
the stretch directions. When there is material rotation (i.e., if R I ) then two strain defini one
that uses the spatial
tions are possible, one that uses the reference stretch U and
stretch V . Hence, to define a 3D generalized strain tensor, we can upgrade the Seth-Hill

strain measure
in Eq. (4.4) to either a reference strain,
1
= --- [ U k I ]
k

(4.6)

or a spatial strain,
1
= --- [ V k I ]
k

(4.7)

Of course, if k=0 is used, these expressions become, in the limit,

= ln ( U )

= ln ( V )

and

(4.8)

28
Copyright is reserved. Individual copies may be made for personal use. No part of this document may be reproduced for profit.

DRAF

December 7, 2008 10:51 pm

Strain

Rebec

ca Br
ann

on

In general, for non-even powers of k ,* computing U k [or ln ( U ) ] requires expressing U

in its principal basis, raising each principal stretch to the k th power

[or taking logs of the
principal stretches if log strain is desired], and then transforming back to the lab basis.
Specifically if [ U ] is the matrix of lab components of U and if a matrix [ Q ] is con eigenvectors of U , then
structed to contain the lab components of the orthonormalized

[ U ] = [ Q ]T [ ] [ Q ]
(4.9)
where [ ] is a diagonal matrix containing the principal stretches associated with the principal basis. The lab components of the general Seth-Hill strain are found by
1
[ ] = --- ( [ Q ] [ k ] [ Q ] T [ I ] )
k

(4.10)

Similarly, components of the logarithmic strain with respect to the lab basis are found by
[ ] = [ Q ] [ ln ] [ Q ] T
(4.11)

where [ ln ] is a diagonal matrix containing the natural logs of the principal stretches.
For the logarithmic strain, note that
tr = ln ( 1 ) + ln ( 2 ) + ln ( 3 ) = ln ( 1 2 3 ) = ln J = v
(4.12)

The logarithmic strain is the only large-deformation strain measure whose trace equals the
consistent volumetric strain. Some sketches of pure stretch deformations and their associated logarithmic principal strains are given on the following page.

* We will explain later why an eigenvalue decomposition is not necessary for even Seth-Hill parameters. Thats the key reason why even powers are used at all -- they are simpler computationally.

29
Copyright is reserved. Individual copies may be made for personal use. No part of this document may be reproduced for profit.

T
F
A
DR
ann
ca Br
Rebec

December 7, 2008 10:51 pm

Strain

on

PURE STRETCH

Hencky strain: 0%

Undeformed configuration

[ F ] = 0.766 0.264

0.264 1.396

Hencky strain: 40%

Hencky strain: 10%

[ ] = 0.306 0.257
0.257 0.306

Hencky strain: 30%

Hencky strain: 20%

30
Copyright is reserved. Individual copies may be made for personal use. No part of this document may be reproduced for profit.

DRAF

December 7, 2008 10:51 pm

Strain

Rebec

ca Br
ann

on

PURE STRETCH FOLLOWED BY ROTATIONS

Rotation: 30 deg

Undeformed
configuration
Hencky strains: 40%
Rotation: zero

Rotation: 60 deg

Rotation: 90 deg
Rotation: 120 deg

Rotation: 180 deg

Rotation:
210 deg
Rotation: 150 deg

[ F ] = 0.795 0.925 [ R ] = 0.866 0.5

0.5 0.866
0.154 1.074
[ U ] = 0.766 0.264

0.264 1.396

[ V] =

[ ] = 0.306 0.257
0.257 0.306

[] =

Rotation:
240 deg

Hencky strains: 40%

Rotation: 270 deg

Rotation: 300 deg

31
Copyright is reserved. Individual copies may be made for personal use. No part of this document may be reproduced for profit.

T
F
A
DR
ann
ca Br
Rebec

December 7, 2008 10:51 pm

Strain

on

For pure stretch deformations, there is no difference between and defined in

spatial
and referEqs. (4.6) and (4.7). However, whenever there is material rotation, the
ence strains will differ from each other by
= R RT

(4.13)

where R is the rotation from the polar decomposition theorem. This distinction (illus the previous page) becomes of paramount importance when constructing matetrated on
rial constitutive laws for anisotropic materials. Modeling an anisotropic material requires
specification of the material orientation. If a constitutive model is constructed using the
spatial strain, then that model must transform the initial material orientation into the new
(rotated) orientation prior to applying the model. Additionally, the principle of material
frame indifference demands that models that use spatial strain must introduce special
objective rates when ever the constitutive model is applied in rate form. When using the
reference strain, however, neither rotation of the material orientation nor special rates are
required. The price paid for using reference strain measures is the spatial strain (which is
whats usually available from structural mechanics host codes) has to be transformed back
into the unrotated configuration prior to application of the constitutive model, and then the
output of the model must be transformed back to the spatial configuration. Computationally, the reference strain option is more efficient and guaranteed to satisfy the principle of
material frame indifference.
We already mentioned that an advantage of the logarithmic strain is that its trace
equals the volumetric strain. Furthermore, the log strain tends to be the choice that makes
the stress-strain function most linear in uniaxial tests. However, the logarithmic strain
requires a polar decomposition in order to obtain a stretch tensor, and the polar decomposition is labor-intensive because it requires an eigenvalue analysis to take the square root
of a tensor. Moreover, computing rates of the logarithmic strain is extraordinarily complicated because, in addition to the principal stretches being a function of time, the principal
directions also change in time. For many people, these disadvantages far outweigh the
advantages.
Recall (from the polar decomposition summary sheet) that U = ( F T F ) 1 / 2 . Conse

be written
quently, the generalized Seth-Hill reference strain in Eq. (4.6) may
1
= --- [ ( F T F ) k / 2 I ]
(4.14)
k

All of the disadvantages we cited for the logarithmic strain will apply whenever the SethHill parameter not an even (and nonzero) integer. Even values for the parameter k are
computationally attractive because there is no need to compute the square root of a tensor.
In particular, the choice k=2 corresponds to the what is called the Lagrangean strain
tensor. With this choice, the square root in Eq. (4.14) goes away, giving
1
E = --- [ F T F I ]
2

(4.15)

32
Copyright is reserved. Individual copies may be made for personal use. No part of this document may be reproduced for profit.

DRAF

December 7, 2008 10:51 pm

Strain

Rebec

ca Br
ann

on

The Lagrange strain is a reference strain. Consequently, it may be regarded as a strain that
results from the pure reference (right) stretch prior to the rotation part of the deformation.
Superimposing even more rotation will have no effect on the Lagrange strain. Anisotropic
constitutive models that use the Lagrange strain in conjunction with a reference stress
measure do not require rotation of the material orientation into the current state. These
models will automatically satisfy the principle of material frame indifference without
needing to introduce any special objective rates.
The Lagrange strain can be readily computed directly from the deformation gradient
tensor (no polar decomposition is required), and its rate is also easily found by

1
E = --- [ F T F + F T F ]
2

(4.16)

A fundamental identity from continuum mechanics states that the rate of the deformation
gradient is related to the materials spatial velocity gradient, L v x , by

F = LF
(4.17)

Therefore, Eq. (4.16) can be written

E = FT D F ,

where

1
D --- [ L + L T ]
2

(4.18)

The tensor D is often called the rate of deformation, which is an unfortunate and misleading name because, despite having some rate-like properties, it is not the proper rate of
any tensor. By this, we mean that it is possible to construct closed deformation paths* for
which the time integral of D is not zero. Frequently in mechanics, it is useful to introduce

a general unrotation operation,

denoted by an overbar, and defined for any spatial second-order tensor Y by

T
YR YR
(4.19)


Using the overbar notation, Fig. 4.18 can be written

E = VDV

(4.20)

where
D RT D R

(4.21)

V = FT R = RT V R = U

(4.22)

and

* i.e., deformations that begin at one configuration, move through other configurations, and then
return to the starting configuration.

33
Copyright is reserved. Individual copies may be made for personal use. No part of this document may be reproduced for profit.

T
F
A
DR
ann
ca Br
Rebec

December 7, 2008 10:51 pm

Strain

on

Reference displacement gradients

The displacement gradient tensor is just the derivative of the displacement with
respect to the reference position vector:
du
H = ------dX

u i
or, in RCS, H ij = -------X j

Recalling that u = x X , note that

du
d(x X)
dx dX
H = ------- = ------------------- - = ------- ------- = F I
dX
dX
dX dX

Solving for F in Eq. (4.24) and substituting the result into Eq. (4.15) gives

1
E = --- [ H + H T + H T H ] This is eq. (3.19) in the book.
2

(4.23)

(4.24)

(4.25)

This is the formula one typically finds in an elasticity textbook, whereas continuum
mechanics textbooks typically favor the more elegant and computationally simple
Eq. (4.15).

A wrong-headed rotation measure

Many elasticity books define a tensor by

1--T
T
= [H H H H]
2

(4.26)

or
= HE
(4.27)

This tensor is well defined, but it is not a sensible or useful measure of material rotation,
as claimed (by some) in the elasticity community. We will call this tensor the wrongheaded rotation (WHR) tensor. In terms of the deformation gradient, the WHR tensor
may be written
1
= F I --- [ F T F I ]
2

In terms of the polar stretch and rotation tensors,

(4.28)

1
= R U I --- [ U 2 I ]
(4.29)
2

In the complete absence of material stretch (i.e., when U I ) the tensor + I does
finite stretches,

for
such
indeed approximately equal the polar rotation R . However,
an

interpretation does not apply and the WHR tensor is essentially useless. Under a pure
stretch (i.e., a non-rotational deformation where F = U ), the WHR tensor becomes

34
Copyright is reserved. Individual copies may be made for personal use. No part of this document may be reproduced for profit.

DRAF

December 7, 2008 10:51 pm

Strain

Rebec

ca Br
ann

on

= e E , where e is the engineering ( k = 1 ) strain measure. Thus, since engineering

equal to Lagrange strain, the WHR tensor will not be zero or the identity even
not

strain
is
under pure stretches. It is a lousy measure of material rotation. For finite stretch problems,
forget about the WHR, and use the polar rotation tensor instead.
Incidentally, noting that the last term in Eq. (4.28) is the Lagrange strain, the WHR
tensor may be written
= FIE

(4.30)

Spatial strain measures

In the previous sections, we discussed the Lagrange strain tensor E , defined by

1
E = --- [ U 2 I ]
(4.31)
2

This strain measure is a reference strain measure because it is computed using the reference stretch tensor. It will not change if you superimpose extra rotation on the deformation. The Lagrange strain has a spatial counterpart, called the Signorini strain, defined
1
E = --- [ V 2 I ]
2

(4.32)

The Signorini strain is related to the Lagrange strain by

E = R E RT
(4.33)

Both the Lagrange and Signorini strains are of Seth-Hill type with a Seth-Hill parameter
k = 2 . Recall that the general form for a Seth-Hill spatial strain uses the spatial (left)
stretch:
1
= --- [ V k I ]
k

Recalling that V = ( F F T ) 1 / 2 , the Seth-Hill spatial strain can be written

1
= --- [ ( F F T ) k / 2 I ]
k

(4.34)

(4.35)

Therefore, once again, choosing a Seth-Hill parameter that is an even integer has the computational advantage that no polar decomposition is required.
The Eulerian strain e is a spatial strain measure corresponding to a Seth-Hill

parameter k = 2 :
1
1
= --- [ I ( F F T ) 1 ] = --- [ I ( F T F 1 ) ]
2
2

(4.36)

35
Copyright is reserved. Individual copies may be made for personal use. No part of this document may be reproduced for profit.

T
F
A
DR
ann
ca Br
Rebec

December 7, 2008 10:51 pm

Strain

on

SPATIAL displacement gradients

The spatial displacement gradient tensor is just the derivative of the displacement with
respect to the spatial (deformed) position vector:
du
h = -----dx

u
or, in RCS, h ij = -------i
x j

(4.37)

Recalling that u = x X . Differentiating this with respect to the spatial position vector

gives
du
d(x X)
dx dX
h = ------ = ------------------- - = ------ ------- = I F 1
dx
dx
dx dx

Solving for F 1 in Eq. (4.38) and substituting the result into Eq. (4.36) gives

1
= --- [ h + h T h T h ]
2

(4.38)

(4.39)

Small displacement gradients

Recall from Eq. (4.1) that all Seth-Hill strain measures become equivalent if the linear
stretch = L Lo is close to 1. Similarly, if the stretch tensor is close to the identity (i.e.,
if all of its eigenvalues are close to 1), then all reference strain measures are approximately
equivalent, so we may define a small-stretch* reference strain tensor as
SS = U I

Likewise, we can define

(4.40)

SS = V I
(4.41)

Here SS means small stretch (i.e., each principal stretch is nearly equal to 1). For small
stretch deformations, any identity that was true only for a particular Seth-Hill parameter
becomes true (approximately) for the small stretch tensor as well. For example, the volumetric strain is in general equal to the trace of the strain tensor only for logarithmic strain
definitions. However, for small stretches, this statement holds approximately regardless of
the strain definition.
Even if stretches are small, however, there might still be large rotations, so
SS SS

(4.42)

* The phrase small stretch does not mean the principal stretches are close to zero; it means they are
close to 1.

36
Copyright is reserved. Individual copies may be made for personal use. No part of this document may be reproduced for profit.

DRAF

December 7, 2008 10:51 pm

Strain

Rebec

ca Br
ann

on

In order to assert that all strain definitions are approximately equivalent, we must add the
proviso that the rotation is small as well. This means that the deformation gradient tensor
is approximately equal to the identity and therefore the displacement gradient tensor is
approximately zero.
If the displacement gradients are small, then the last term in Eq. (4.25) is negligible,
and Eq. (4.25) becomes
1
SDG = --- ( H + H T )
2

(4.43)

Here, SDG stands for small displacement gradients. For small displacement gradients,
not only is the stretch small, but the rotation is small too. Therefore
SDG = U I ,

(4.44)

or
1
U = --- ( H + H T ) + I
(4.45)
2

For small displacement gradients, the infinitesimal rotation tensor is approximately

1
R SDG = --- ( H H T )
2

(4.46)

This approximation must be handled carefully since this expression does not correspond
to an orthogonal tensor.

Volume strain
In the section on the polar decomposition, we noted that the Jacobian J equals the determinant of stretch, which is simply the product of the principal stretches:
J = 1 2 3

(4.47)

Furthermore, in the section on the deformation gradient, we showed that the ratio of
deformed to initial volumes equals the Jacobian
V
J = -----Vo

(4.48)

Thus, the Jacobian is the volumetric analog of the linear stretch, = L L o . When a
strain is defined using a particular choice k for the Seth-Hill strain parameter, then
1
= --- ( U k I )
k

(4.49)

37
Copyright is reserved. Individual copies may be made for personal use. No part of this document may be reproduced for profit.

T
F
A
DR
ann
ca Br
Rebec

December 7, 2008 10:51 pm

Strain

on

The consistent definition of volumetric strain uses the same value for the Seth-Hill parameter so that
1
v = --- ( J k 1 )
k

(4.50)

In this section, we will show how the consistent volumetric strain is related to invariants of
the strain tensor we will prove that it is not equal to the trace of the strain tensor except
under conditions of small strain or when k 0 .
From Eq. (4.49), we know
U k = I + k

Therefore, taking the determinant,

(4.51)

J k = det ( I + k )
(4.52)

Using a fundamental theorem from 3D matrix analysis, the determinant on the right hand
side can be expanded to give
J k = det ( I ) + I C : ( k ) + I : ( k ) C + det ( k )
(4.53)

where the superscript C denotes the cofactor. The cofactor of the identity is just the
identity itself, so the second term on the right hand side simplifies to kI 1 , where I 1 tr .

Also, ( k ) C = k 2 ( ) C , so the third term becomes k 2 I 2 , where I 2 is the second invariant

of strain, given by I 2 = tr ( ) C . Finally, the last term is just the third invariant I 3 times
k 3 . Thus, the above equation may be written
J k = 1 + kI1 + k 2 I 2 + k 3 I 3

(4.54)

and the consistent volumetric strain becomes

v = I 1 + kI 2 + k 2 I 3

(4.55)

For arbitrary deformations, note that the volume strain will be identically equal to the trace
of the strain only if k=0 . In other words, the logarithmic (Hencky) strain measure is the
only one whose consistent volume strain measure is given by the trace of the strain tensor.
Of course, for small deformations, the strain measures become indistinguishable because
I 3 I 2 I 1 , so approximating the volumetric strain by the trace of the strain is acceptable
in that case.

38
Copyright is reserved. Individual copies may be made for personal use. No part of this document may be reproduced for profit.

DRAF

December 7, 2008 10:51 pm

Strain

Rebec

ca Br
ann

on

Why isnt the rate of deformation a true rate?

Recall that the velocity gradient is given by

L = F F 1

and the rate of deformation is

(4.56)

D --- ( L + L T ) = sym ( F F 1 )
2

The following two deformation gradients represent quite different motions:

[F] =

cosh sinh 0
sinh cosh 0
0
0 1

and

1 2 0
[F] = 0 1 0

0 0 1

(4.57)

(4.58)

These deformations have different stretches and different velocity gradients but they
both have the same rate of deformation tensor. This alone should tell you that theres
something strange about the rate of deformation tensor. It fails to comprise the same
information as the true rate of a stretch tensor.

Comparison
of reference
stretches
In this section, we will demonstrate that D is not a true rate. We will likewise demonstrate that the unrotated rate of deformation,
D = RT D R

is not a true rate either. (Here, R is the polar rotation tensor).

Note that
1

D = --- ( F F 1 + F T F T )
2

(4.59)

(4.60)

39
Copyright is reserved. Individual copies may be made for personal use. No part of this document may be reproduced for profit.

T
F
A
DR
ann
ca Br
Rebec

December 7, 2008 10:51 pm

Strain

on

The right hand side is linear with respect to F , so it follows that there must exist a fourth

order tensor Y , which depends on F , but is independent

of F , such that

( D ij = Y ijmn F mn ) (4.61)
D = Y :F

Specifically,
1
1 + F 1 )
Y ijmn = --- ( im F jn
jm in
2

(4.62)

Lets suppose (for the purpose of contradiction) that there does exist some strain tensor

that is a function of F such that D = . If so, then the chain rule would require that

dE

= ------- :F
(4.63)
dF

If D = , then comparing with Eq. (4.61), it would have to be true that

d
------- = Y
(this will be contradicted)
(4.64)
dF

or, using Eq. (4.62),

ij
1
1 + F 1 )
-----------= --- ( im F jn
jm in
2
Fmn

(this will be contradicted)

(4.65)

On the surface, this looks like a fairly straightforward set of partial differential equations
(one equation for each of the ways to select values for the free indices). You might throw
caution to the wind and simply launch into trying to integrate this system (if you do, keep
1 equals ( F 1 ) , not 1 F ). However, you will find that no matter how
in mind that F rs
rs
rs
hard you try, you cant find a solution. This system has no solution. There are two ways to
prove it. The first way looks at an integrability condition. At the root, we need to prove
that

Y :F is not an exact differential

If it were, then Eq. (4.64) would be valid, which would imply that
2 E ij
2 Eij
Y ijpq
Y ijmn
--------------- = ------------------------ = ------------------------ = ------------F pq
F mn F pq
F pq F mn
Fmn

(this will be contradicted)

(4.66)

(4.67)

Conversely, if this condition were true, then Eq. (4.64) would be integrable. Hence, we
need to show that Eq. (4.67) is false. First we need a lemma:
1
Frs
1
------------ = F rj1 F ks
F jk

(4.68)

Then differentiating Eq. (4.65) with respect to F gives

40
Copyright is reserved. Individual copies may be made for personal use. No part of this document may be reproduced for profit.

DRAF

December 7, 2008 10:51 pm

Strain

Rebec

Y ijmn
1 F 1 + F 1 F 1 )
--------------- = -----1- ( im F jp
qn
jm ip qn
2
F pq

ca Br
ann

on

(4.69)

Swapping roles of mn with pq gives

Y ijpq
1 F 1 + F 1 F 1 )
-------------- = -----1- ( ip F jm
nq
jp im nq
F mn
2
The integrability condition requires that

Y ijmn
Y ijpq
---------------- = -------------Fpq
Fmn

(4.70)

for all F , or

1 F 1 + F 1 F 1 = F 1 F 1 + F 1 F 1
im F jp
qn
jm ip qn
ip jm nq
jp im nq

(this will be contradicted) (4.71)

This relationship must hold for all F for Eq.(4.65) to be integrable. This relationship does

not hold for all F , but its not real obvious

why. To prove that its false for at least one F ,

we need to contrive a counterexample, but the right counterexample is not very evident.
This leads us to an alternative way to prove that there does not exist any tensor such that

D =

t2

t2

If = ( F ) , then every path satisfying F dt = 0 will also satisfy dt = 0

t1

(4.72)

t1

Stated in the converse,

t2

t2

If there exists any path such that F dt = 0 but

t1

and therefore

is not really a true rate.

dt 0 , then

t1

is not expressible as a function of F ,

(4.73)

Below, we prove that neither D nor D are true rates, which is why careful writers dont

often call D the stretching instead of the rate

typeset them as rates. Careful speakers
be confused with stretch.
of deformation, but we worry that such a term might
To demonstrate that D is not the time rate of any proper function of F , we need a path

that is closed in F for which D dt 0 . For this purpose, consider the following four-stage

deformation (illustrated in the figure):

[ F] = 1 0 (1 t) + 1 0 t
01
0 2

for 0 < t < 1

(4.74)

[ F] = 1 0 (2 t) + 1 1 ( t 1 )
02
0 2

for 1 < t < 2

(4.75)

[ F] = 1 1 (3 t) + 1 1 ( t 2 )
02
0 1

for 2 < t < 3

(4.76)

[ F] = 1 1 (4 t) + 1 0 ( t 3 )
01
0 1

for 3 < t < 4

(4.77)

41
Copyright is reserved. Individual copies may be made for personal use. No part of this document may be reproduced for profit.

T
F
A
DR
ann
ca Br
Rebec

December 7, 2008 10:51 pm

Strain

on

Note that [ F ] at t=0 equals [ F ] at t=4 . Therefore, the path is

4

closed in F . That is, F dt

= 0.

The corresponding symmetric part of the velocity gradient

can be computed using (4.57) to give
0

[D] =

[D] =

0
1
0 ----------1+t

for 0 < t < 1

(4.78)

0 14
14 0

for 1 < t < 2

(4.79)

for 2 < t < 3

(4.80)

for 3 < t < 4

(4.81)

0
1
0 ---------t4

[D] =

0 12
0

[D] =

12

The time integral of D is

D dt
0

t = 0

t = 1

t = 2

D dt + D dt + D dt + D dt
0

0 0 + 0 1 4 + 0 0 + 0 1 2
14 0
0 ln 2
0 ln 2
1 2 0

0
1 4
1 4
0

(4.82)

This result is non-zero, which [applying Eq. (4.73)], completes

the proof that D is not a true rate of any quantity that can be

expressed as a function
of the deformation gradient.
The preceding derivation used a four-stage deformation
sequence: 1. elongate vertically, 2. shear to the right, 3. compress vertically, and 4. shear to the left. This path was selected to
simplify the equations as much as possible for the counterexample.
Now that we know that D is not a true rate, a natural follow subclass of deformations, if any,
up activity is to determine what
will permit D to be a true rate. In other words, we are now
exploring thetime integral of D associated with closed paths in
F for which all values of F are restricted to belong to some par
ticular
subclass. It can be shown that D is a true rate for any
deformation that involves no rotation of the principal stretch
directions.

t = 3

t = 4

42
Copyright is reserved. Individual copies may be made for personal use. No part of this document may be reproduced for profit.

DRAF

December 7, 2008 10:51 pm

Strain

Rebec

In the preceding counterexample, the nonzero contributions

to the time integral of D might have appeared (at first glance) to

arise solely from the simple

shearing stages, so its natural to
question whether the time integral of D might turn out to be
zero whenever we limit ourselves to deformations that are
always pure stretches (i.e. deformations where the polar rotation
tensor R always equals the identity I . For this class of deforma
D and D . Hence, this
tions, there
is no distinction between
question is the same as asking whether or not D , defined in
Eq. (4.59), is a true rate. The answer again turns out no and
the following three-stage loading is a counterexample:
[ F] = 1 0 (1 t) + 1 0 t
01
0 2

for 0 < t < 1

(4.83)

[ F] = 1 0 (2 t) + 3 2 1 2 (t 1 )
12 32
02

for 1 < t < 2

(4.84)

for 2 < t < 3

(4.85)

[ F] =

32
12

1 2 (3 t) + 1 0 (t 2)
32
01

This problem can be analyzed in a manner similar to what was

done in the four-stage loading counterexample, but the math is
messy (so we recommend using a symbolic math program such
as mathematica). The time integral of D again turns out the be

nonzero even though the deformation consists

entirely of pure
stretches.

ca Br
ann

on

t = 0

t = 1

t = 2

t = 3

The reason why neither D nor D are true rates is linked to

of stretch. The above example illustrates that the
the changing of the principaldirections
principal directions of stretch can change without having any rotation (i.e. the polar rotation can remain R = I ).

43
Copyright is reserved. Individual copies may be made for personal use. No part of this document may be reproduced for profit.

T
F
A
DR
ann
ca Br
Rebec

December 7, 2008 10:51 pm

EXAMPLES: non-rotational deformations

on

5. EXAMPLES: non-rotational deformations

Uniaxial Strain
For uniaxial strain, all particle motion occurs in one direction. Because there is no lateral material motion, the shape of the cross-sectional area is irrelevant it might be a circular, square, or any other shape. For this reason, we may simplify the analysis without
any loss in generality by presuming that the cross-section is rectangular and is therefore
most easily described using a rectangular Cartesian system (RCS) for the coordinates.
Lets suppose that we set up the RCS basis so that the 1-direction is aligned with the
direction of motion. The initial location of a material particle can be expressed in terms of
our RCS as
X = X 1 e 1 + X 2 e2 + X3 e3

After deformation, this particle deforms to a new location

(5.1)

x = x1 e1 + x2 e2 + x3 e3

where, for uniaxial strain,

(5.2)

x1 = f ( X1 )

(5.3)

x2 = X2

(5.4)

x3 = X3

(5.5)

The last two of these equations state that no motion occurs transverse to the 1-direction.
All motion is characterized by the 1D deformation mapping function f . Here, the function
f depends on the nature of the loading. The deformation gradient is given by
x x x
--------1- --------1- --------1X 1 X 2 X 3

00
x x x
[ F ] = --------2- --------2- --------2- = 0 1 0

X 1 X 2 X 3
0 01
x 3 x 3 x 3
--------- --------- --------X 1 X 2 X 3

where

f ( x1 )

(5.6)

Note that the local stretch varies only along the axial direction. When the stretch is
constant, the deformation is said to be homogeneous uniaxial strain. Suppose, for example, that a piston contains a gas. If the piston is slowly compressed from a length L o to L ,
then the deformation will satisfy x 1 X 1 = L L o , or
x 1 = ( L Lo )X 1 , and therefore = L L o

(5.7)

44
Copyright is reserved. Individual copies may be made for personal use. No part of this document may be reproduced for profit.

DRAF

December 7, 2008 10:51 pm

EXAMPLES: non-rotational deformations

Rebec

ca Br
ann

on

If, on the other hand, the piston is suddenly compressed from a length L o to L , then the
material displacements at a given instant in time will be highly inhomogenous because of
wave motion back and forth. Even if a deformation is equilibrated, it might not be homogenous. Consider, for example, a long cylinder of a compressible gel sitting on a table so
that the axis of the cylinder points up. Then gravity will make material more compressed
at the bottom of the cylinder than at the top.

Inhomogenous uniaxial strain

Homogenous uniaxial strain

The left hand side of the above figure shows inhomogenous uniaxial strain. Note that the
material elements (i.e., the little squares of material) dont deform very much on the right
side of the bar, but they are stretched by more than 400% on the left hand side. You might
expect a deformation somewhat like this if you hang a long cylinder of cork* from the
ceiling. For homogenous uniaxial strain, each little material element stretches by the same
amount everywhere.

* Cork is a material whose Poissons ratio is approximately zero. Consequently, it would have no offaxis (transverse) contractions. This constraint is needed to ensure uniaxial strain conditions.

45
Copyright is reserved. Individual copies may be made for personal use. No part of this document may be reproduced for profit.

T
F
A
DR
ann
ca Br
Rebec

December 7, 2008 10:51 pm

EXAMPLES: non-rotational deformations

on

For uniaxial strain, the left-hand side is not required to remain motionless you
might additionally have rigid translation as well, resulting a displacement vector that is
longer by the amount of translation. The key is that the displacement vector always connects the initial and deformed locations of the same particle.
Inhomogenous uniaxial strain

Figure 5.1.

General uniaxial strain with translation.

Mapping function for uniaxial strain. For uniaxial strain, all motion is one-dimensional. Specifically, the mapping function is of the form
x1 = f ( X1 )

(5.8a)

x1 = X2

(5.8b)

x1 = X1

(5.8b)

which corresponds to the following deformation gradient matrix

J 0 0
df ( X 1 )
[ F ] = 0 1 0 , where J = ---------------dX 1
0 0 1

(5.9)

Since J is the Jacobian, it equals the ratio of deformed volume to initial volume. For this
1D uniaxial strain deformation (in which the cross-sectional area is constant), the volume
ratio is identically equal to the length ratio.

46
Copyright is reserved. Individual copies may be made for personal use. No part of this document may be reproduced for profit.

DRAF

December 7, 2008 10:51 pm

EXAMPLES: non-rotational deformations

Rebec

ca Br
ann

on

The function f quantifies the entire deformation

x1
except for rigid translation. It is the one-dimensional analog of the deformation gradient tensor. Since f is a single function of a single
variable, uniaxial strain is one of the few deformations that allows you to actually plot a graph
of the mapping function. For example, the
graph of the mapping corresponding to Fig. 5.1
can be approximated by picking up the picture of the deformed state to make it become the
vertical axis in a plot of x 1 vs. X1 , as illustrated
in Fig. 5.2. Note that the graph of the mapping
function has a steeper slope at the left side of
the bar than at the right; this indicates that material at the left side is stretched more than that at
the right. A similar plot for homogeneous
uniaxial strain would be a straight line, indicating that all elements are stretched by the same
amount. The intensity of stretching is quantified
X1
by the actual value of the slope. If the slope
equals unity at a point, then the material element there has not stretched at all. Slopes larger
than unity correspond to stretching. Slopes
Figure 5.2.
Graph of a 1D uniaxial
smaller than unity correspond to compression. strain deformation mapping function.
The vertical intercept of the mapping graph is
the amount of rigid translation that is superimposed on the deformation.
The slope of the mapping graph equals the material stretch (deformed length divided
by initial length). Therefore, since length is always positive, the slope must be everywhere
positive. In other words, a physically realizable uniaxial strain deformation must have a
monotonically increasing mapping graph it will never dip down.

Uniform spherical contraction/expansion

Consider a deformation for which
x = X
(5.10)

In other words, each initial material vector changes length by the same proportion. Then
F = I

(5.11)

47
Copyright is reserved. Individual copies may be made for personal use. No part of this document may be reproduced for profit.

T
F
A
DR
ann
ca Br
Rebec

December 7, 2008 10:51 pm

EXAMPLES: non-rotational deformations

on

Non-uniform spherical deformation

Consider a deformation in which spherical coordinates { R, , } are used to locate
the initial position of a point and spherical coordinates r, , identify the deformed location of that point. A spherical deformation is one for which
r = f(R) ,

= ,

and

(5.12)

Triaxial strain
Triaxial strain is a homogenous axially symmetric deformation. It differs from uniaxial
strain only in that lateral deformations are permitted. Specifically, the deformation gradient is of the form
F = A PA + T P T
(5.13)

Here, the subscripts A and T stand for axial and transverse respectively, and the
coefficient tensors are projection operations defined
PA n n

(5.14)

and
PT I n n
(5.15)

The vector n is a unit vector in the direction of the symmetry axis, and n n denotes the

ij compodyadic multiplication
of n with itself (the result is a second-order tensor with
P acts on an arbitrary vector v , the result is the part of that
nents given by n i n j . When
A
vector in the direction of the symmetry axis. When P T acts on the vector, the result is the
part transverse (perpendicular) to the symmetry axis. Stated mathematically, for any vector, v

P A v = ( v n )n
(5.16)

P A v = v ( v n )n
(5.17)

We have stated that triaxial strain is a being here described as a homogenous deformation,
which means that the deformation gradient tensor F is spatially constant and the definition of the deformation gradient, dx = F dX may therefore be integrated to give

x = F X + C,
(5.18)

where C is an integration constant representing rigid translation. In light of Eqs. (5.13),

(5.17), the deformation mapping function can be written
(5.16) and
x = A ( X n )n + T [ X ( X n )n ] + C
(5.19)

In terms of RCS coordinates with the 1-direction aligned with n , the deformation gradient

and the mapping function may be written

48
Copyright is reserved. Individual copies may be made for personal use. No part of this document may be reproduced for profit.

DRAF

December 7, 2008 10:51 pm

Axisymmetric deformations

Rebec

ca Br
ann

on

A 0 0
[F] =

0 T 0

(5.20)

0 0 T

and
x1 = A X1 + C 1

(5.21)

x2 = T X2 + C2

(5.22)

x3 = T X3 + C3

(5.23)

Here, X T denotes the distance of X from the symmetry axis.

Note that uniform spherical deformation (discussed on page 47) is a special case of triaxial deformation in which A = L = and C = 0 . Also, homogenous uniaxial strain
is a special case of triaxial deformation in which A = and T = 1 .

6. Axisymmetric deformations
An axisymmetric deformation is one for which all motion is axial and/or radial with
respect to a given unit vector N (the axis of symmetry). No angular motion is permitted.

Cylindrical coordinates are a natural

choice for this type of motion. Given an initial position vector
X = RE R + ZE Z

The displacement vector for axisymmetric deformations will be of the form

(6.1)

u = u R ( R, Z )E R + u Z ( R, Z )E Z
(6.2)

Note that the displacement has no angular component. The radial and axial components of
displacement depend only on the radial and axial coordinates. When the displacement
field is of this form, the displacement gradient becomes
uR, R 0 u R, Z
[H] =

0
u Z, R

u
-----R 0 with respect to cylindrical basis
R
0 u Z, Z

(6.3)

For axially symmetric deformations, the deformed position vector will have the same
radial base vector as the initial position vector, so it will be of the form
x = x R ( R, Z )E R + x Z ( R, Z )E Z ,

where x R = R + u R ( R, Z ) and x Z = Z + u Z ( R, Z )

(6.4)

Consequently, the deformation gradient will be of the form

49
Copyright is reserved. Individual copies may be made for personal use. No part of this document may be reproduced for profit.

T
F
A
DR
ann
ca Br
Rebec

December 7, 2008 10:51 pm

Axisymmetric deformations

on

x R, R 0 x R, Z
[F] =

0
x Z, R

xR
----- 0 with respect to cylindrical basis
R
0 x Z, Z

where
x R, R = 1 + u R, R
x Z, R = u Z, R

x R, Z = u R, Z
x Z, Z = 1 + u Z, Z

(6.5)

A cylindrical basis can be bypassed through the use of projection operations, as illustrated
below for the constricted rod problem.

Constricted rod
Consider a deformation in which a rod expands and/or contracts about an axis of symmetry, while planes that were initially perpendicular to the axis of symmetry remain planar.
Then, taking the base of the rod to be fixed, we may say that
x Z = Z

(6.6)

x R = R

(6.7)

and

Here, and are scalar-valued functions of Z. They do not vary with R. To avoid material inversion (i.e., to ensure local admissibility), both and must be positive. Applying Eq. (6.5),

[F] =

0 sh
0 0
0 0 h

<cylindrical basis>

(6.8)

where
h ( Z ) = + Z ( Z )
R ( Z )
and s ( Z ) = --------------------h(Z)

(6.9)

Note that the deformation gradient may be written

1 0 s 0 0
[F] = 0 1 0 0 0

0 0 1 0 0 h

<cylindrical basis>

(6.10)

50
Copyright is reserved. Individual copies may be made for personal use. No part of this document may be reproduced for profit.

DRAF

December 7, 2008 10:51 pm

Axisymmetric deformations

Rebec

ca Br
ann

on

In this form, it is clear that this deformation can be viewed as a triaxial stretch followed by
a simple shear. The function s ( Z ) is the deformed slope of the deformed shape of a line
initially defined by R = 1
Initial configuration
Special case: linear axisymmetry.
Recall that the general deformation analyzed
above corresponded to the mapping
x Z = Z

(6.11)

x R = R

(6.12)

Consider the special case of uniform axial

stretch and a linearly varying radial profile
= constant

(6.13)

xz
x R = c + k ---- R
L

(6.14)

Deformed configuration

and

Axisymmetric deformation.
where c and k are constants and L is the unde- Figure 6.1.
In this figure,
formed length of the bar. Note that cR is the
4
3 X 1-
deformed radial coordinate for a point whose ini- g ( X 1 ) = --5- X 1 and h ( X 1 ) = 1 --5- ----25
tial cylindrical coordinates are ( R, 0 ) , and
kR L is the slope of the line passing through this point. Substituting Eq. (6.11) into
Eq. (6.14) gives

Reb: what are the g and h functions referenced in Figure? Did I change notation and forget to update that caption? My notes.nb now has this deformation as an
example.
kZ
x R = c + ---------- R
L

(6.15)

from which it follows that

kZ
= c + ---------L

(6.16)

Then
h =

and

Rk
s = -----L

(6.17)

Note that s is the slope of the line originally defined by R = 1 . Fig. 6.1 shows this type
of deformation for = 4 5 , c = 1 , and k = 1 8 . With these choices, h = 4 5 ,
Z
s = R 8L , and = 1 --------- . Therefore the deformation gradient is
10L

51
Copyright is reserved. Individual copies may be made for personal use. No part of this document may be reproduced for profit.

T
F
A
DR
ann
ca Br
Rebec

December 7, 2008 10:51 pm

EXAMPLES: large rotation problems

on

1 0 s 0 0
[F] = 0 1 0 0 0 =

0 0 1 0 0 h

R
Z
0 ------ 1 --------8L
10L

Z
1 --------10L

4--5

<cylindrical basis>
For any point on the center line,

(6.18)

Z
1 --------10L

Z
1 --------10L

4--5

[F] =

<cylindrical basis>

(6.19)

which shows that points on the centerline undergo a triaxial stretching that becomes
increasingly compressive as Z increases.

7. EXAMPLES: large rotation problems

Bar bending
A bar, initially of height H , bends into
curved wedge segment as shown until the top
surface of the bar is oriented at angle max .
The geometry of this deformation can be alternatively described by specifying the radius R
of the centerline and the stretch c of the center line (i.e., its deformed length divided by its
initial length). One can easily switch back and
forth between these two specification methods
by using the relation
R max = c H ,

(7.1)

which follows from the definition of stretch.

( x 1, x 2 )

( X 1, X 2 )

52
Copyright is reserved. Individual copies may be made for personal use. No part of this document may be reproduced for profit.

DRAF

December 7, 2008 10:51 pm

EXAMPLES: large rotation problems

Rebec

ca Br
ann

on

For this deformation, we seek:

1. The mapping function.
2. The deformation gradient.
3. The polar rotation tensor R and reference (right) stretch U .

4. The Lagrange strain

Let X = X 1 E 1 + X 2 E 2 denote the initial position of a point in the body, and let

the deformed location of that same point.

x = x 1 E 1 + x2 E 2 denote

First lets find the mapping function. In other words, we seek formulas for the
deformed coordinates ( x 1, x 2 ) of a point expressed in terms of the initial coordinates
( X 1, X 2 ) of that same point. Any initially vertical line can be identified by its initial
abscissa value, X 1 . Each initially vertical line bends into the shape of an arc of radius X 1 ,
with an arc length given by X 1 , where is the arc angle. The arc angle increases in proportion to X 2 . Therefore, a constant exists such that
= X 2

(7.2)

The arc length of a point originally located at ( X1, X 2 ) must be given by

s = X 1 X 2

(7.3)

Applying this equation to the center line gives

R max = RH

(7.4)

Therefore
c
max
- =
---- = ---------H
R

(7.5)

and Eq. (7.2) and Eq. (7.3) becomes

X2
= ------ max
H

(7.6)

For example, if you are located 2/3 of the way up the bar in the undeformed configuration,
then you will be 2/3 of the way along the arc in the deformed configuration.
Consider a point ( X 1, X 2 ) , not necessarily on the center line. After deformation, this
point deforms to a new point ( x 1, x 2 ) located a distance r = X 1 from the origin. Consequently,
X
x 1 = r cos = X1 cos -----2- max
H

(7.7)

X2
x 2 = r sin = X 1 sin ------ max
H

(7.8)

Reb: my notes.nb now has this as an example.

The deformation gradient is given by

53
Copyright is reserved. Individual copies may be made for personal use. No part of this document may be reproduced for profit.

T
F
A
DR
ann
ca Br
Rebec

December 7, 2008 10:51 pm

EXAMPLES: large rotation problems

on

X
x
F 11 = --------1- = cos -----2- max
H

X1

x
X
X
F 12 = --------1- = c -----1- sin -----2- max
H

X2
R

X2
x 2
F 21 = --------- = sin ------ max
X1
H

x 2
X1
X2
F 22 = --------- = c ------ cos ------ max
X2
R
H

(7.9)

Assembled into matrix form,

[ F ] = c s ,
s c

(7.10)

To save ink in upcoming calculations, we have defined

X 1 max
X1
----------------- c ------ ,
H
R

(7.11)

X2
c
c cos ------ max cos X 2 -----
H

(7.12)

X
c
s sin -----2- max = sin X 2 -----
H

(7.13)

The Jacobian equals to the ratio of the deformed to undeformed volumes and can be computed by
J = detF =

The polar decomposition gives

[ R ] = c s
s c

and

(7.14)

[U] = 1 0
0

(7.15)

Note that this deformation consists of a stretch in the 2-direction, followed by a rotation
into the final deformed orientation.
The Lagrange strain is
0
0
1--1--T
2
[E] = {[F] [F] [I]} = {[U] [I]} =
1
2
2
0 --- ( 2 1 )
2

(7.16)

The reference logarithmic strain is

[ ] = ln [ U ] = 0 0
0 ln

(7.17)

The logarithmic volume strain is given by

v = tr = ln

The displacement gradient is given by

(7.18)

54
Copyright is reserved. Individual copies may be made for personal use. No part of this document may be reproduced for profit.

DRAF

December 7, 2008 10:51 pm

EXAMPLES: large rotation problems

Rebec

[ H ] = [ F ] [ I ] = ( c 1 ) s
s
( c 1 )

ca Br
ann

on

(7.19)

For this problem, the wrong-headed rotation measure becomes

[] =

(c 1)

c ---
2

(7.20)

Unlike the polar rotation, there is not really any physical interpretation for this tensor. For
pure stretch deformations, it equals the difference between the engineering strain (SethHill parameter k=1 ) and the Lagrange strain (Seth-Hill parameter k=2 ). Under sufficiently large stretches, this difference will be neither zero nor the identity even in the
absence of rotation.

Simple Shear
Simple shear involves two orthonormal unit vectors, m and n , often taken to be in the

1 and 2 directions, respectively.

Displacement u is in the direction of m . The magnitude of displacement varies as a

function of n X .Since u = x X , this means

that simple shear has a mapping of the

form
x = X + g ( n X )m

where the function g is specified.

(7.21)

linear g-function

Sinusoidal g-function
2
2

1
1

Figure 7.1.
Simple shear deformation of a unit square. When the g-function is nonlinear,
the mapping is nonlinear if the g-function is nonlinear. The local slope of formerly vertical lines
is quantified by the variable, which is a constant for simple linear shear.

In the simplest case, the g function is linear so that

55
Copyright is reserved. Individual copies may be made for personal use. No part of this document may be reproduced for profit.

T
F
A
DR
ann
ca Br
Rebec

December 7, 2008 10:51 pm

EXAMPLES: large rotation problems

on

g ( n X ) = 2 ( n X )
(7.22)


where is a constant (and the factor of 2 is introduced for convenience). Then the deformation gradient is
dx
dX
F = ------- = I + 2m n
F 1 = ------- = I 2m n
(7.23)
also
dX
dx

In the general case, where the g-function is nonlinear, the deformation gradient is of this
form except that is
1
= ( X ) = --- g ( n X )
2

(7.24)

Note the structure of the deformation gradient. A deformation is a shear whenever F is

the identity I plus a dyad a b for which a b = 0 , in which case a defines the shearing

direction, b defines the shearing plane, and = ( ab ) 2 is the shearing strain. Introducing a basis for which the 1 direction is parallel to m and the 2-direction is parallel to n ,

1 2 0
[F] = 0 1 0
0 0 1

and

[ F ] 1

1 2 0
= 0 1 0
0 0 1

(7.25)

Then the polar rotation is given by

c s 0
[R] = s c 0 ,
0 0 1

(7.26)

where
tan =

and

s = sin = -----------------1 + 2

and

1
c = cos = ------------------ (7.27)
1 + 2

In terms of ,
s
1 2 -- 0
c
[F] =
0 1 0
0 0 1

(7.28)

Note that a positive corresponds to a negative (clockwise) polar rotation angle, which
should be evident from Fig. 7.1. The unit rotation axis is a = m n . Letting A denote

form
as
the associated skew tensor, the rotation may be written in symbolic
R = ( cos ) ( I a a ) + a a + ( sin )A

The right stretch is

(7.29)

56
Copyright is reserved. Individual copies may be made for personal use. No part of this document may be reproduced for profit.

DRAF

December 7, 2008 10:51 pm

EXAMPLES: large rotation problems

Rebec

ca Br
ann

on

s2

+ [ U ] = [ R ] T [ F ] = s 1------------0
c
0
0 1

(7.30)

The left stretch is

1------------+ s 2c
[ V ] = [ F ][ R ]T =
s
0

1
0

0
1

(7.31)

In symbolic form,
2 c( c + 1)
U = I s ( m n + n m ) + ---------------------------- n n
c

(7.32)

2 c(c + 1)
V = I s ( m n + n m ) + ---------------------------- m m
c

For small distortions (i.e., small so that c 1 and s ), these become

(7.33)

U I (mn + nm)

(7.34)

V I (mn + nm )


For large distortions (i.e., 2 )

(7.35)

4
U = I ( m n + n m ) + --------------- n n ( )n n
2

(7.36)

4
(7.37)
V = I ( m n + n m ) + --------------- m m ( )m m
2

This result makes sense. As shear increases, fibers that were originally aligned with n in

the reference configuration become stretched without bound and approach alignment with
m . The n n component in the reference stretch U increases without bound because n is
original
orientation of the severely stretched fibers.

the
The m m component in the spatial
orientation approached by
stretch V is increases without bound because m is the final

these severely
stretched fibers.
Time effects in shear. Suppose that m and n are constant in time, but that varies with

time. Then

0 2 0

[F] = 0 0 0
0 0 0

F = 2 m n

(7.38)

57
Copyright is reserved. Individual copies may be made for personal use. No part of this document may be reproduced for profit.

T
F
A
DR
ann
ca Br
Rebec

December 7, 2008 10:51 pm

EXAMPLES: large rotation problems

on

L = F F 1 = 2 m n

0 2 0
[L] = 0 0 0
0 0 0

(7.39)

D = symL = ( m n + n m )

0 10

[D] = 1 0 0
0 00

(7.40)

W = skwL = ( m n n m )

0 1 0

[ W ] = 1 0 0
0 0 0

(7.41)

0 1 0

[ ] = c2 1 0 0
0 0 0

(7.42)

= R RT = c2(mn nm )

Note that, for constant , the vorticity, w = , remains constant for all time. However,

the polar angular velocity, = c 2 , approaches zero because c 0 . Thus, the polar
rotation gives a better description of material rotation since material fibers do not tumble
and tumble indefinitely. The horizontal fibers dont rotate at all, and the initially vertical
fibers rotate no more than 90 .

58
Copyright is reserved. Individual copies may be made for personal use. No part of this document may be reproduced for profit.

DRAF

December 7, 2008 10:51 pm

EXAMPLES: large rotation problems

Rebec

ca Br
ann

on

Torsion
An informative way to get started with torsion is to first recall some elementary concepts about rotation. The mapping of an initial vector X to become a rotated vector x is

given by
x = QX
(7.43)

in which the rotation tensor Q is proper orthogonal. For time-varying problems, the rota and a skew-symmetric tensor is introduced such that
tion tensor Q varies with time,

velocity vector . If
Q = Q . The axial vector associated with is the angular
is

angular velocity is constant in a time-varying rotation problem, the angle of rotation

= t . For time-varying rigid rotation, Eq. (7.43) is spatially linear because Q depends

only on time, not on position.

Torsion, on the other hand, is not rigid it involves material distortion. Nevertheless,
as suggested in Fig. 7.2, torsion still involves rotation of position vectors about the cylinder axis N = n . Therefore, Eq. (7.43) still applies, but it is no longer spatially linear
rotation

because the
angle and therefore Q , vary with distance along the cylinder axis.

N
n

max =H

Figure 7.2.
Torsion of a right-circular cylinder of radius R=1 and height H=4 by a twist
parameter = 0.2 radians per unit length.

A uniform twist corresponds to a rotation angle given by

= z

(7.44)

where the torsion parameter is the twist angle per unit length along the axis of the cylinder, and z is the axial coordinate, given in symbolic form by
z = NX = nx

(7.45)

59
Copyright is reserved. Individual copies may be made for personal use. No part of this document may be reproduced for profit.

T
F
A
DR
ann
ca Br
Rebec

December 7, 2008 10:51 pm

EXAMPLES: large rotation problems

on

In equation Eq. 7.44, the variable z is playing a role analogous to time t for time-varying
rotations with a constant angular velocity. Analogous to the introduction of an angular

velocity tensor such that Q = Q in time varying problems, we introduce an angu such that

lar twist tensor

dQ
------- = Q
(7.46)
dz

For the uniform twist problem, the axial vector associated with is

= n
(7.47)

For a constant twist parameter , the twist tensor is also constant even though Q var

ies with position.

It is useful to define a vector v by

v = x = x

Equivalently,

ez

(7.48)

v = r = r

(7.49)

er

where
r = ( I nn ) x
(7.50)

The vector r is the part of the position vector perpendicu of rotation. The vector v plays a role analolar to the axis

gous to velocity for time-varying rotations,

and it points in
the angular direction. Therefore
vN = 0

In a cylindrical system,

(7.51)

r = re r

(7.52)

Figure 7.3.

Helper vectors.

and
v = re
(7.53)

To find the deformation gradient, we must differentiate Eq. (7.43) with respect to X ,

using the product rule because Q itself varies with position by virtue of its dependence on

the axial coordinate z:

d(Q X)
dQ T
dQ T
dx
dX
dz

- = Q + X -------- - ------ = Q ------- + X --------F = ------- = --------------------dX

dX
dX
dX
dz
dX

From Eq. (7.46) and then Eq. (7.43) and ultimately Eq. (7.48), note that

(7.54)

60
Copyright is reserved. Individual copies may be made for personal use. No part of this document may be reproduced for profit.

DRAF

December 7, 2008 10:51 pm

EXAMPLES: large rotation problems

Rebec

ca Br
ann

dQ
dQ T
- = ------- X = Q X = x = v
X --------dz
dz

Also, from Eq. (7.45),
dz-----= N
dX

Therefore, Eq. (7.54) becomes

on

(7.55)

(7.56)

F = Q + vN

Incidentally, note that

(7.57)

F = ( I + vn ) Q

(7.58)

F = Q (I + uN)
where u Q T v
(7.59)

Recall from the previous section on shear that a deformation is a shear whenever its deformation gradient is the identity I plus a dyad a b for which a b = 0 , in which case a

= ( ab ) 2 is the shear
defines the shearing direction, b defines the shearing
plane, and
therefore note that the deformation gradient for torsion
ing strain. From Eq. (7.58), we
may be regarded as a rigid rotation Q followed by a tangent shear of shear strain magni
tude 1--2- r . Alternatively, from Eq. (7.59),
torsion may be seen as a tangent shear of the
same magnitude followed by rotation Q .

In terms of an orthonormal cylindrical basis { e r, e , e z } having the 3-direction

aligned with N ,

[F] =

cos ( z ) sin ( z ) 0
sin ( z ) cos ( z ) r
0
0
1

w.r.t. cylindrical basis { e r, e , e z }

(7.60)

The left Cauchy-Green tensor is given by

B = F FT = ( Q + v N ) ( QT + Nv )

Multiplying out gives

(7.61)

B = Q QT + vN QT + Q Nv + vN Nv

Recalling that Q represents rotation about the unit vector N , note that

T
QQ = I

Q N = N QT = n

NN = 1

Therefore Eq. (7.62) becomes

(7.62)

(7.63)
(7.64)
(7.65)

61
Copyright is reserved. Individual copies may be made for personal use. No part of this document may be reproduced for profit.

T
F
A
DR
ann
ca Br
Rebec

December 7, 2008 10:51 pm

EXAMPLES: large rotation problems

on

B = I + vn + nv + vv

The component matrix is
[B] =

0
0

2 r2

1+
r

where (recall)

v = x = x

(7.66)

0
w.r.t. cylindrical basis { e r, e , e z }

r
1

(7.67)

The right Cauchy-Green tensor is

C = FT F = I + Nu + uN + u2NN

The component matrix is

[C] =

1
0

0
1

0
r

1 + 2 r2

where u Q T v

(7.68)

w.r.t. cylindrical basis { E r, E , E z } (7.69)

Here, the E-basis is the un-rotation of the e-basis by Q T .

Application of a large deformation constitutive equation to torsion. Suppose that

you wish to know how much torque is required to achieve a given twist . Then you must
introduce a constitutive equation that relates deformation to stress. To illustrate the process, consider the following very simplistic constitutive model for an incompressible rubber-like material
= 2GE + q ( x )I
(7.70)

Here, q ( x ) is an undetermined function (discussed below), is the Cauchy stress tensor,

and E is the Signorini strain defined in Eq. (4.32),

1
1
E = --- [ V 2 I ] = --- ( B I ) ,
(7.71)
2
2

or, using Eq. (7.66)

1
where (recall) v = x = x .
E = --- ( v n + n v + v v )
(7.72)

The undetermined function q ( x ) is present because, for an incompressible material, the

pressure cannot be determined exclusively from knowledge of the deformation. Instead,
q ( x ) must be determined to satisfy the equations of equilibrium and the boundary condi Assuming no body forces, equilibrium requires
tions.
ij, j = 0

(7.73)

Substituting Eq. (7.72) into (7.70), and then into (7.73) gives
G ( v i, j n j + n i v j, j + v i, j v j + v i v j, j ) + q ,i = 0

(7.74)

62
Copyright is reserved. Individual copies may be made for personal use. No part of this document may be reproduced for profit.

DRAF

December 7, 2008 10:51 pm

EXAMPLES: large rotation problems

Rebec

ca Br
ann

Noting that v m, p = mp and recalling that is skew symmetric, it follows that

G ( ij n j + ij v j ) + q ,i = 0

on

(7.75)

Note that ij n j is the indicial form of

n = n = n n = 0

Therefore, Eq. (7.74) becomes, in symbolic form,
dq
G v + ------ = 0

dx

or, recalling that v = x ,

dq
------ = G x
dx

Integrating gives

(7.76)

(7.77)

(7.78)

1
q ( x ) = c --- Gx x
2

(7.79)

or
1
q ( x ) = c --- G r 2
(7.80)
2

where c is an integration constant and r is the radial vector defined in Eq. (7.50). The
above result may be alternatively written
1
q ( x ) = c --- G r 2
2

or, since and r are perpendicular to each other,

1
q ( x ) = c --- G 2 r 2 ,
2

(7.81)

(7.82)

Therefore, putting Eqs. (7.82) and (7.72) back into (7.70) gives
1
= G ( v n + n v + v v ) + c --- G 2 r 2 I where (recall) v = r .
(7.83)

This solution is complete except for the undetermined integration constant c, which must
be found by the boundary condition that the lateral surface of the cylinder must be traction
free. The outward normal to the lateral surface is radial and therefore perpendicular to
both v and n . Hence, the only way to satisfy the traction-free condition over the lateral
is if
boundary
1
c = --- G 2 R 2
2

(7.84)

so that the final solution for the stress field is

63
Copyright is reserved. Individual copies may be made for personal use. No part of this document may be reproduced for profit.

T
F
A
DR
ann
ca Br
Rebec

December 7, 2008 10:51 pm

EXAMPLES: large rotation problems

on

1
= G ( v n + n v + v v ) + --- G 2 ( R 2 r 2 ) I where (recall) v = r .
(7.85)

We are now in a position to determine the net force and torque required to achieve a specified twist . The traction at the top surface of the cylinder is
t = n

or, using Eq. (7.85) and recalling that v is perpendicular to n ,

1
t = Gv + --- G 2 ( R 2 r 2 ) n
2

(7.86)

(7.87)

or, introducing a cylindrical basis,

1
t = [ Gr ]e + --- G 2 ( R 2 r 2 ) e z
2

(7.88)

The first term is the tangential part of traction and the second term is the normal part of
traction. Note that the normal traction is nonzero. In fact, the net axial force is given by
R

F axial =

tz dA

1--1
G 2 ( R 2 r 2 ) 2r dr = --- G 2 R 4
2
4

(7.89)

This force is positive and therefore tension is required to achieve the specified deformation.
The torque is given by
R

T =

rt dA
A

r [ Gr ]2r dr

1
= --- GR 4
2

(7.90)

Keep in mind that these results are inextricably tied to the choice of constitutive model. If
a different model is used, the results will be different. However, it is to be expected that
the trends will be similar. For example, increasing the radius significantly increases the
force an torque (in this case by a power of 4).

Vortex
<need to add this section>

Finite element stretching

In finite element analysis, a quadrilateral element is described by a mapping from a
unit square. A reference square in logical coordinate space is defined by two parameters,
1 and 2 , each varying from a low value L to a high value H (usually either from 0
to 1 or from 1 to 1). The four vertices of the quadrilateral in physical space are known.

64
Copyright is reserved. Individual copies may be made for personal use. No part of this document may be reproduced for profit.

DRAF

December 7, 2008 10:51 pm

EXAMPLES: large rotation problems

Rebec

ca Br
ann

on

We will denote these known physical position vectors by { x LL, x HH, x LH, x HL } . Here, the
the location
of the point
subscripts H and L denote high and low. The vector x LL is
whose logical coordinates are ( L, L ) . The vector x HL is the location of the point whose

logical coordinates are ( H, L ) , and so on.

x HH

( L, H )

( H, H )

x HL

x LH

2
1
( L, L )

( H, L )

x LL

If ( 1, 2 ) are the logical coordinates of a point in the reference square, then the physical coordinates are given by
H 1 H 2
H 1 L 2
x ( 1, 2 ) = ------------------- ------------------- x LL + ------------------- ------------------- x LH
H L H L
H L L H

L 1 H 2
L 1 L 2
+ ------------------- ------------------- x HL + ------------------- ------------------- x HH
L H H L
L H L H

(7.91)

For the case that ranges from L = 0 to H = 1 , this becomes

x ( 1, 2 ) = ( 1 1 ) ( 1 2 )x LL +

+ ( 1 ) ( 1 2 )x HL
+

In this case, define

( 1 1 ) ( 2 )x LH

( 1 ) ( 2 )xHH

(7.92)

x
g 1 = -------- = ( x HL x LL ) ( 1 2 ) + ( x HH x LH ) 2
1

(7.93)

x
g 2 = -------- = ( x LH x LL ) ( 1 1 ) + ( x HH x HL ) 1
2

(7.94)

and the logical deformation gradient associated with this mapping is

dx
x d 1 x d 2
------ = -------- --------- + -------- --------- = g 1 e 1 + g2 e 2
d
1 d 2 d

(7.95)

65
Copyright is reserved. Individual copies may be made for personal use. No part of this document may be reproduced for profit.

T
F
A
DR
ann
ca Br
Rebec

December 7, 2008 10:51 pm

EXAMPLES: large rotation problems

on

where e 1 and e 2 are the orthonormal reference base vectors.

This mapping can be used to describe either the spatial or reference configuration for
an actual deformation. In that case, the actual reference configuration has vectors, G 1 and
G 2 , defined similarly to the g-vectors above. Then the actual deformation gradientcan be

evaluated
by
F = g1 G 1 + g2 G 2
(7.96)

where G 1 and G 2 , are the duals of G 1 and G2 . As a practical matter, it is usually easier
the deformation

to evaluate
gradient by
[ F ] = [ g ] [ G ] 1 ,
(7.97)

where the matrix [ g ] contains g 1 and g 2 in its columns and [ G ] contains G 1 and G 2 in

its columns.

66
Copyright is reserved. Individual copies may be made for personal use. No part of this document may be reproduced for profit.