G22.

3033-002: Topics in Computer Graphics: Lecture #7

Geometric Modeling
New York University

Elasticity Theory Basics

Lecture #7: 20 October 2003
Lecturer: Denis Zorin
Scribe: Adrian Secord, Yotam Gingold

Introduction
This lecture is the summary of the basics of the elasticity theory. We will discuss the funda-
mental concepts of stress and strain, and the equations describing the deformation of an elastic
object in equilibrium. We will focus on materials following the linear material law.
The linear material law can be thought of as a generalization Hooke’s law f = −kx to 3D
deformable objects. The most straightforward generalization would be to state that stretching
or compressing an object in any direction results in an elastic force in the same direction which
is proportional to the extension. However, this would match well only a small class of mate-
rials. In three dimensions in most cases a stretch in one direction changes the dimensions in
the perpendicular directions. The two aspects of deformation are characterized using two con-
stants: the Young’s modulus, which corresponds to the constant in Hooke’s law, and is denoted
E or Y , and the Poisson’s ratio, which is denoted ν. The Poisson ratio is the ratio of stretch to
perpendicular change. To formulate the equations of balance for elastic objects and define the
linear material law more precisely, we start with a discussion of the counterparts of forces and
displacements in the equations of elasticity: strain and stress tensors.

Stress and Strain

There are two kinds of forces acting on a small volume of a body:
1. Volume forces
• f is the volume force density;
• dFvol = f dV is the differential volume force, where dV is the differential volume
element.
2. Surface forces
• t(n) the vector surface force density, where n is the normal to the differential sur-
face element dA;
• t(n) is also called the stress vector;
• dFsurf = t(n)dA is the differential surface force, acting on the differential surface
element dA.
2 G22.3033-002: Lecture #7

Stress tensor
As stated, the stress vector t(n) can have any form at all, but it turns out that it can be expressed
in the special form t(n) = T n, where T is called the Cauchy stress tensor.

An aside on tensors. 2D tensors are matrices which follow certain transformation rules when
we change a coordinate system. Typically these rules follow from the definition of the quantity
described by the matrix. In the case of the stress tensor, this can be seen as follows. Let A be
the matrix representing a coordinate transformation x0 = Ax. Under an arbitrary linear change
of coordinates, normals are not generally transformed as points or displacement vectors. The
correct transformation rule for normals can be derived if we keep in mind that if n is the normal
to a plane P the transformed normal n0 should remain perpendicular to the transformed plane
P 0 . Then it is easy to derive that n0 = A−T n. Note that in the case of rigid transformations we
have A−1 = AT , so A−T = A. On the other hand force vectors are transformed as all other
vectors. Using these considerations, we have

t0 (n) = T 0 n0
At(n) = T 0 A−T n
t(n) = A−1 T 0 A−T n

which implies that T = A−1 T 0 A−T or T 0 = AT AT . So the stress tensor is transformed

using T 0 = AT AT .

Back to the stress tensor. In general, the stress tensor is

 
σ11 σ12 σ13
T =  σ21 σ22 σ23  , (1)
σ31 σ32 σ33

which means that, for example,

    
σ11 σ12 σ13 1 σ11
T î =  σ21 σ22 σ23   0  =  σ21  (2)
σ31 σ32 σ33 0 σ31

is the force if the normal to the surface area points along the x-axis.
We assume that the volume torques will balance, i.e. the volume forces create no net torque.
This is generally assumed in elasticity. One can derive from this constraint that T is symmetric
and has only six distinct elements.
G22.3033-002: Lecture #7 3

An aside on symmetric matrices. As an aside, some properties of a general symmetric

matrix S are

• The eigenvalues are always real (and as a consequence the eigenvectors have real com-
ponents);

• RT SR is diagonal for some (rotation) matrix R, which means that S = RT DR, where
D is a diagonal matrix with the eigenvalues of S along its diagonal.

This means that we can always find some change of basis that transforms T into a nice diagonal
form  
σ1
T = σ2 . (3)
σ3
The σi ’s are called the principal stresses. Choosing these coordinates brings us closer to our
goal of defining the generalization of Hooke’s law for 3D materials. The second ingredient that
we need is the measure of the deformation.

Strain tensor
The strain tensor measures the change of distances between close points in the deformed state
with respect to the distances in the undeformed state. Let ϕ : Ω → R3 be the mapping from
the reference domain Ω to the deformed state. Compare the distance between two points that
are close together in the reference frame, k(x + dx) − xk2 , with the distance between them in
the deformed frame kϕ(x + dx) − ϕ(x)k2 . We have

k(x + dx) − xk2 = kdxk2 = dxT dx

in the reference frame and

kϕ(x + dx) − ϕ(x)k2 = (ϕ(x + dx) − ϕ(x))T (ϕ(x + dx) − ϕ(x) (4)

in the deformed frame. For small deformations, we can approximate ϕ(x + dx) with ϕ(x) +
∇ϕdx, so (4) becomes

(ϕ(x + dx) − ϕ(x))T (ϕ(x + dx) − ϕ(x) ≈ (∇ϕdx)T (∇ϕdx)

= dxT ∇ϕT ∇ϕdx

Aside: the gradient of a vector is a matrix. Specifically, if ϕ(z) = (ϕ1 (z), ϕ2 (z), ϕ3 (z))T , then
 
∇ϕ1
∇ϕ =  ∇ϕ2  .
∇ϕ3
4 G22.3033-002: Lecture #7

The quantity ∇ϕT ∇ϕ is called the Cauchy-Green deformation tensor. Finally, if we take the
difference between the reference and deformed squared distances, we get

dxT ∇ϕT ∇ϕdx − dxT dx = dxT (∇ϕT ∇ϕ − I)dx

The quantity ∇ϕT ∇ϕ − I is twice the strain tensor. Note that strain is non-dimensional.
Note that it is a matrix, and it is easy to derive its transformation rules, which end up the
same as for the stress tensor. It follows directly from the definition that this tensor is symmetric:
 
11 12 13
 =  12 22 23  ,
13 23 33
We can also choose a coordinate system in which the strain is diagonal:
 
1
= 2 .
3
The diagonal elements are called principal strains. This mathematical fact has important phys-
ical meaning: any deformation locally can be regarded as a simple stretch/compression along
three perpendicular directions.
To get a better idea about the meaning of the principal strains, let’s consider the change in
length along a principal direction; according to the definition of strain, the new squared length
is given by (dx1 0, 0)(2 + I)(dx1 0, 0)T , i.e.

  
2ε1 + 1 dx1
  0  = (2ε1 + 1)dx2 ≡ dx02


dx1 0 0  2ε2 + 1 1
2ε3 + 1 0
√ |dx01 |−|dx1 |
Assuming small strains, 21 + 1 ≈ 1 + 1 . So |dx1 |
= ε1 , where the first quantity is
the relative change in length in the x1 direction.

Equations of balance
We will focus on elastostatic equations. Elasto-static equations are simpler to understand than
elastodynamic; at the same time, it is relatively easy to generalize elastostatic equations.
These equations are derived from the equations of the balance of force and the balance of
torques:
P
P fi = 0 sum of forces on body
ri × fi = 0 sum of moments of forces on body
and can be written either in the integral or differential form. We start with the the integral form.
We examine a small region of a deformed object dV . ∂dV denotes its surface.
G22.3033-002: Lecture #7 5

If fV is the volume force density,

Z Z Z Z
fV dV + t(n)dS = fV dV + T n dS (5)
∂dV ∂dV
| {z }
integrate over surface
which must be zero to stay balanced.
Given two volumes in contact along surface element dA,

the two surface forces with which the volumes act on each other are the same. In this way all
internal surface forces cancel out.
From equation (5), summing over all small volumes into which we have partitioned our
object, we get the balance of forces equation:
Z Z
fV dV + T n dS = 0
V ∂V

Similarly, the torques need to stay balanced:

Z Z
r × fV dV + r × T n dS = 0
V ∂V

To get the differential form of these equations, we use the Gauss Integral Theorem:
Z Z
T n dS = div T dV,
∂V V

where (div T ) is the divergence of T .

6 G22.3033-002: Lecture #7

An aside on tensor divergence. FOr a vector function

 
f1 (x)
f =  f2 (x)  ,
f3 (x)

divergence is scalar:
∂f1 ∂f2 ∂f3
div f = + + .
∂x1 ∂x2 ∂x3
For a tensor, divergence is vector (sum over each column).

Returning to our two equations,

Z Z
fV dV + T n dS = 0 (6)
Z V Z ∂V

r × fV dV + r × T n dS = 0 (7)
V ∂V

From equation (6),

Z
(fV + div T ) dV = 0
V
⇒ fV + div T = 0

The last equation is the differential form of the force balance. Recall that the symmetry of the
stress tensor was a consequence of the torque balance, so the torque equation is captured by
symmetry of the stress tensor and the boundary conditions.
The stress of the boundary is equal to the external stress:

T = Text on ∂V (stress continuity).

An important fact to mention is that these equations are written with respect to the coordi-
nates in the deformed object. In practice, it is much more convenient to use the coordinates in
the undeformed object to write these equations, as typically the deformation is unknown and
needs to be computed.

Material laws, a.k.a. constitutive equations.

As we have mentioned in the beginning, we are primarily interested in materials following a
generalization of Hooke’s law. In general the elastic material behavior is captured by writing
the strain as a function of stress.
We fix the coordinate system in which the stress tensor Σ is diagonal. In this coordinate
system the forces acting on the elementary volume act along the coordinate axes; for each axis,
the forces acting on opposite sides are equal in magnitude and opposite in direction.
G22.3033-002: Lecture #7 7

s2 s3

s1 s1

s3 s2
In this coordinate system, the linear material law has the form
1 ν
i = σi − (σj + σk ) (8)
E E
where E is the Young modulus and ν is the Poisson ratio. The first term in the formula reflects
the assumption that the strain in a given direction depend linearly on the stress in the same
direction. The second term takes into account the deformation due to the stresses in other
directions, taking into account the degree to which the object can compress. For example, for
incompressible objects ν = 1/2, and the volume is preserved.
While this is a simple an intuitive way to write these equations, this requires using a coor-
dinate system which might change from one point to another. In general, we need the form of
these equations in an arbitrarily chosen coordinate system.
A simple way to arrive at such a form is to express everything in terms of tensor invariants.

An aside on tensor invariants. What are invariants of a symmetric 3D tensor? (In other
words, functions of matrix elements that do not change when the coordinates change.) Eigen-
values can be regarded as invariants as we look at the whole set (the order may change as we
change the coordinate system). It is more convenient to use symmetric functions of eigenval-
ues:

X
λ1 + λ2 + λ3 = Tr  = ii
λ1 λ2 λ3 = det 
 
λ21
λ1 λ2 + λ2 λ3 + λ3 λ1 = Tr 2 = Tr   = 1 (Tr )2 − Tr 2


λ22
2
λ23
Intuitively, one can think of the action of the deformation on a small sphere. The result is an
ellipsoid with major axes lengths proportional to eigenvalues. Then

λ1 + λ2 + λ3 = 3 · average cross section

λ1 λ2 λ3 = volume
λ1 λ2 + λ2 λ3 + λ3 λ1 = 3 · length of diagonal
8 G22.3033-002: Lecture #7

Equation (8) can be rewritten as

1+ν ν
i = σi − (σi + σj + σk ) .
E E
Since σi + σj + σk = Tr Σ, we can write

1+ν ν
= σ − (Tr Σ) I (9)
E E
This is the coordinate-independent form of the material linear law. Materials that follow this
law are called St. Venant-Kirchhoff materials. Note that this law is more universal than it might
seem: most isotropic elastic materials are quite well described by it. The reason for this is that
as long as the strain is a function of stress, in most cases for small strains it is approximated
well by the linear term in its Taylor series expansion. But strains tend to be small in most cases,
excluding highly deformable objects such as rubber.

Lamé constants
We can invert equation (9) to write Σ as a function of . For this, we need to express the trace
of stress in terms of the trace of strain, which can be done by taking the trace of both sides of
the linear material law:
1+ν 3ν
Tr  = Tr Σ − Tr Σ
E E
Substituting the resulting expressions back, we obtain an expression of the form

Σ = λ(Tr )I + 2µ.

Where the constant λ and µ are called Lamé constants:

Summary
fV + div T = 0
and
Σ = λ(Tr )I + 2µ
The stress tensors in these equations are denoted with different letters for a reason. The first
equation is written with respect to the deformed body, and the constitutive law is written with
variables specified in the original body coordinates. So these stress tensors are different: the
second equation uses what is called the second Piola-Kirchhoff tensor and these two tensors
are related by

Σ = (det φ)−1 ∇φT ∇φT

G22.3033-002: Lecture #7 9

Fortunately, the form of the equation of the balance of forces does not change its form when
we switch to the undeformed body quantities.
The unknowns in these equations are the deformations (recall that the stress is defined using
the gradient of the deformation function) and the stress tensor. The deformations are defined
by three coordinate functions, and the stresses by six, so we have a total of nine unknown
functions. The force balance results in three equations with the constitutive law providing
additional six, so now the system of equations is complete. Of course, this simple counting
argument does not prove that the solution exists or is unique.