Stress and Strain As Tensor and Matrix Quantities Class Notes
Stress and Strain As Tensor and Matrix Quantities Class Notes
Lecture 3
Scalars
We are familiar with the idea of a scalar quantity; that is, a quantity that has only magnitude and
can be expressed by a single number. Examples include temperature and density. A scalar is
written as a single variable with no subscript, e.g. T for temperature.
Vectors
We are also familiar with the idea of a vector as a quantity that has both magnitude and
direction. A 3-D real-space vector requires 3 numbers to define it. For example, in a specific x, y,
z coordinate system we can write the vector 𝑉"⃗ as
"⃗ = 𝑉! 𝑥& + 𝑉" 𝑦& + 𝑉# 𝑧̂ ,
𝑉 (3.1)
where Vx, Vy, and Vz are the vector components and 𝑥&, 𝑦&, and 𝑧̂ are unit vectors in the x, y, and z
directions, respectively. We can also express the coordinate system as x1, x2, and x3, in which
case we write
"⃗ = 𝑉$ 𝑥&$ + 𝑉% 𝑥&% + 𝑉& 𝑥&& .
𝑉 (3.2)
This notation is convenient because we can use the summation convention to write
"⃗ = 𝑉' 𝑥&' .
𝑉 (3.3)
By convention, we drop the unit vector and write
"⃗ = 𝑉' .
𝑉 (3.4)
That is, we write a vector as a variable with a single subscript. Examples include force Fi,
position xi, electric field Ei, and electric current density ji.
We know that the same vector expressed in a different coordinate system will have different
components. That is, three different numbers will be used to express the same vector in a
different coordinate system. We can transform a vector from any coordinate system to any other
as follows: Suppose we have a vector Vi in the x1, x2, x3 coordinate system and we want to
express that same vector in a different coordinate system, say as 𝑉'( in the 𝑥$( , 𝑥%( , 𝑥&( system, as
shown in Figure 3.1. Let qij be the angle between the xi axis in the system we are transforming to
(the ‘new’ system) and the xj axis in the system we are transforming from (the ‘old’ system). For
example, q11 is the angle between 𝑥$( and 𝑥$ , q12 is the angle between 𝑥$( and 𝑥% , and so on.
θ11
θ21
θ12
Figure 3.1: A vector and its representation in the x1, x2, x3 and 𝑥$( , 𝑥%( , 𝑥&( coordinate
systems. The angles qij determine the relative orientations of the two coordinate systems
(only q11, q12, and q21 are shown—the other six are formed in the same way).
Now let aij = cos(qij). The aij values are the direction cosines and form a 3 × 3 matrix which we
can write using indicial or matrix notation,
𝛼$$ 𝛼$% 𝛼$&
𝛼') = ,𝛼%$ 𝛼%% 𝛼%& - . (3.5)
𝛼&$ 𝛼&% 𝛼&&
We can find 𝑉$( by projecting the components of Vi onto the 𝑥$( axis. That is
𝑉$( = 𝛼$$ 𝑉$ + 𝛼$% 𝑉% + 𝛼$& 𝑉& . (3.6)
Similarly
𝑉%( = 𝛼%$ 𝑉$ + 𝛼%% 𝑉% + 𝛼%& 𝑉& , (3.7)
and
𝑉&( = 𝛼&$ 𝑉$ + 𝛼&% 𝑉% + 𝛼&& 𝑉& . (3.8)
Using the summation convention, we can write this very efficiently as
𝑉'( = 𝛼') 𝑉) . (3.9)
Equation 3.9 is our coordinate transformation rule for vectors. If we were instead going from the
primed system to the unprimed system, it’s easy to see that the direction cosines, let’s call them
bij, would just be given by bij = aji. In linear algebra, we call the direction cosines the
transformation matrix and using typical notation, we might write V′ = [A]V and V = [A]TV′,
where vectors are indicated by bold typeface, [A] is the transformation matrix, and [A]T is the
transpose of [A].
It is important to note that not all quantities that can be expressed as magnitude and
direction are vectorsa! To be a vector, a quantity must have both magnitude and direction and
transform according to Eq. 3.9.
Tensors
As it turns out, it is very common for materials properties to be anisotropic. Consider electrical
conductivity, which expresses the current density 𝚥⃗ that is induced in a conductor by an electric
field 𝐸"⃗ . If the material were isotropic, the conductivity s would be represented by a scalar that
simply scales 𝚥⃗ with 𝐸"⃗ (see Figure 3.2a):
𝚥⃗ = 𝜎𝐸"⃗ (3.10)
You may have learned this form in an elementary physics class.
However, if the material is anisotropic, 𝚥⃗ will not always be in the direction of 𝐸"⃗ (see Figure
3.2b) and the relative magnitudes of 𝚥⃗ and 𝐸"⃗ will depend on the direction of 𝐸"⃗. This makes sense
because it is easier for electrons to move in some directions in a material than others. This means
that, if we apply an electric field along x1, we should expect to see components of ji along x1, x2,
and x3 (see Figure 3.1c). The components of ji scale linearly with E1: j1 = s11E1, j2 = s21E1, and j3
= s31E1.
x3 x3 x3
ji ji j3
Ei Ei
x2 x2 E1 j2 x2
j1
x1 x1 x1
(a) (b) (c)
Figure 3.2: An electric field Ei producing an electric current density ji in (a) isotropic and
(b) anisotropic materials. (c) In an anisotropic material, E1 may produce nonzero values
of j1, j2, and j3.
To specify the conductivity, we define a coordinate system and determine the relationships
between the components of 𝚥⃗ and the components of 𝐸"⃗. For example, E1, E2, and E3 all contribute
to j1 as follows:
𝑗$ = 𝜎$$ 𝐸$ + 𝜎$% 𝐸% + 𝜎$& 𝐸& . (3.11)
Similarly,
𝑗% = 𝜎%$ 𝐸$ + 𝜎%% 𝐸% + 𝜎%& 𝐸& (3.12)
a
Rotation is an example.
and
b
A multilinear relationship is a function of several variables that is linear separately in each variable.
c
The direction cosines are one example.
Stress as a Tensor
We can now proceed to see how stress is defined as a 2RT. Suppose we look at an infinitesimal
cube of material at a point P in a loaded body as shown in Figure 3.4a. We zoom in to this
element, and describe the net force acting on each face of the cube in terms its components, Dfij,
in the x1, x2, x3 coordinate system as shown in Figure 3.4b. In this notation the first subscript
indicates the direction in which the force component is acting and the second refers to the normal
to the plane it is acting on.
(a) (b)
Figure 3.4: (a) An infinitesimal element in a loaded body and (b) the forces acting on it.
Note that there must be equal and opposite forces (not shown) acting on the back faces of
the cube to prevent it from moving. We are interested in deformation, not translation!
(3.21)
We can write the stress tensor in indicial or matrix notationd,
𝜎$$ 𝜎$% 𝜎$&
𝜎') = , %$ 𝜎%% 𝜎%& - .
𝜎 (3.22)
𝜎&$ 𝜎&% 𝜎&&
And we can modify our sketch in Figure 3.4b to show that we are looking at stress components
as shown in Figure 3.5.
If we look directly along the x1 axis in Figure 3.5, we can immediately see that the stress
tensor must be symmetric. As shown in Figure 3.6, the s23 component applies a torque that acts
to rotate the cube clockwise while the s32 component applies a torque that acts to rotate the cube
counterclockwise. If the magnitudes of these components were different, the cube would rotate.
Since we are interested in the forces that cause deformation, not the forces that cause rotation, we
require that s23 = s32.
d
This of course looks identical to our description for electrical conductivity, but this is just because the variable s
was used in both cases.
Figure 3.6: A 2-D slice through the cube in Figure 3.5 showing stress components acting
in the x2-x3 plane.
Or, in general, for stresses we require that sij = sji. We can then write
(3.23)
or, by convention,
(3.24)
Let the force due to all the applied loads acting at point P be Dfi. We choose a small element of
the cut plane having area DA at P and define the traction vector, Ti, as a vector with magnitude
(3.25)
Ti = sijnj . (3.26)
Thus, the stress gives a relation between the traction acting at a point in a material and the
normal to the plane on which that traction is defined.
Stress States
Since stress is a symmetric 2RT, we know that there is always a coordinate system oriented such
that the shear components are zero. The axes in this system are the principal stress axes, 𝑥$0 , 𝑥%0 ,
and 𝑥&0 , and the values of the normal stresses along these axes are the principal stresses, 𝜎$$
0
,
0 0
𝜎%% , and 𝜎&& . We can write the stress tensor in this coordinate system as
e
The tensor formulation of stress was first worked out by Augustin-Louis Cauchy (1789-1857). Indeed, stress was
the first quantity to be described and used as a tensor. The word “tensor” is derived from the Latin “tensus” meaning
“stress” or “tension.” Stresses defined in this way are referred to as Cauchy stresses.
(3.27)
It is important to remember that the principal stresses are just one of infinitely many equivalent
representations of the stress state at a point. However, they are very convenient to use! The
principal stresses can always be found by solving Eq. 3.16 to find direction cosines aij such that
( ( (
𝜎$% = 𝜎%& = 𝜎%& = 0.
To simplify our consideration of mechanical properties, we will normally work from the
principal coordinate system. This allows us to quickly identify the maximum normal and shear
stress components and the planes on which they act. This will in turn allow us to relate specific
deformation mechanisms to the stress state; for example, fracture often occurs along planes of
maximum normal stress while plastic deformation often occurs along planes of maximum shear.
We use principal coordinates to define several different characteristic stress states:
Using Eq. 3.16, it is possible to show that the maximum normal stress for this stress state is just
the maximum principal stress. We can also use Eq. 3.16 to find the maximum shear stress, but
we will just use the simple geometry shown in Lecture 2 as follows:
Consider plane P in Fig 3.8b with normal 45˚ to 𝑥&0 and 𝑥%0 . As we saw in Lecture 2, 𝜎&& 0
0
would produce a shear stress of magnitude 𝜎&& /2 acting on plane P in the direction 45˚ from
0 0 0 0
𝑥& and -𝑥% , while 𝜎%% would produce a shear stress of magnitude 𝜎%% /2 on the same plane acting
0 0
in the direction 45˚ from 𝑥% and −𝑥& .
We sum these shears and write
0 0
𝜎&& − 𝜎%%
𝜏= (3.28)
2
Note that we cannot use subscripts for this shear component since it is in a different coordinate
system. Instead, we are specifying its plane and direction explicitly. We can do this for each pair
of principal stresses. The maximum shear will be
0 0
𝜎123 − 𝜎145
𝜏= (3.29)
2
0 0
where 𝜎123 and 𝜎145 are the algebraic maximum and minimumf principal stresses. In the
0 0 0 0
example shown in Fig. 3.8, 𝜎123 = 𝜎&& and 𝜎145 = 𝜎$$ and the maximum shear stress occurs on
0 0
plane Q (Fig. 3.8c) having normal 45˚ from 𝑥$ and 𝑥& because the shear stresses from these
components have the same sign.
A special triaxial stress state arises when 𝜎$$0 0
= 𝜎%% 0
= 𝜎&& = sH. This stress state is referred to
as hydrostatic and is characterized by the single stress component sH, the hydrostatic stress,
(3.30)
.
By comparison with Equation 3.24, we see that the shear stress in a hydrostatic stress state is
zero. This is true regardless of the coordinate system! Such stress states occur in materials
immersed in a uniformly pressurized fluid—for example at the bottom of the ocean.
f
The algebraic maximum and minimum are the most positive and negative values respectively.
(3.31)
The maximum normal and shear stresses are found in exactly the same way as for triaxial stress
states except that one of the principal stresses will be zero—which can also be the algebraic
maximum or minimum!
(3.32)
(3.33)
Where the hydrostatic stress is just the average of the normal stresses, that is sH = (s11 + s22 +
s33)/3. The deviatoric part of any stress state, 𝜎')6 , is given by the difference such that
(3.34)
Note the difference between a hydrostatic stress state where the hydrostatic stresses are the
principal stress components (3.30) and the hydrostatic part of any stress state given by Equation
3.33 where the hydrostatic stresses are the average of the normal stresses. As it turns out, it does
not matter what coordinate system we choose to determine the hydrostatic part of the stress state
since the sum of the normal components is a constant. This quantity is referred to as the first
stress invariant I1 = s11 + s22 + s33.
The deviatoric part of the stress state focuses on shear stresses and is used where shear stress
is important in deformation.
Q2 x2
∆u1
∆x2 ∆u2 Q′2
Q2
Q1
P x1 ∆x2
∆x1
Q1′ ∆u2
Q1
P, P′ x1
∆x1 ∆u1
We immediately remove the translation by subtracting the displacement from P to P′ from all
points and ignore it henceforth. We characterize the motions of Q1 to Q1′ and Q2 to Q2′ by their
displacements Du1 and Du2 parallel to the x1 and x2 axes respectively and separate the
displacements due to deformation from the displacements due to rotation as follows:
Let’s first look along the x1 axis at the motion of Q1 to Q1′. We consider only small
displacements such that Du1 << Dx1 and Du2 << Dx1. We can then define elements of the
displacement gradient tensor,
(3.35)
and
. (3.36)
We note that e11 is identical to our definition of the true normal strain in Lecture 2. We see that
in general,
𝜕𝑢' ∆𝑢'
𝑒') = = lim (3.37)
𝜕𝑥) ∆!! →9 ∆𝑥)
and by including the x3 dimension and point Q3, we can easily track the displacement Du3 to find
e31 along the x1 axis. We can then find the displacement gradients along the x2 and x3 axes in the
same way. This gives us the full displacement gradient tensor
𝑒$$ 𝑒$% 𝑒$&
𝑒') = ,𝑒%$ 𝑒%% 𝑒%& - . (3.38)
𝑒&$ 𝑒&% 𝑒&&
Consider the situation shown in Figure 3.10a, which shows the case of a pure rotation.
x2
∆u1
Q′2
Q2
∆x2
Q1′ ∆u2
P Q1 x1
∆x1
We see that ∂u1/∂x2 = – ∂u2/∂x1. Thus, we find that rotation wij is the antisymmetric part of eij.
. (3.39)
We thus define strain as the symmetric part of the displacement gradient tensor,
. (3.40)
An element that is subjected to shear strain without rotation is shown in Figure 3.12. Notice that
in this case, ∂u1/∂x2 = ∂u2/∂x1.
x2
∆u1
Q2
Q′2
∆x2
Q1′ ∆u2
P Q1 x1
∆x1
(3.41)
.
We have said that strain is a second-rank tensor quantity and that, as such, it specifies a relation
between two vector quantities. What two vectors are related by the strain tensor? Displacements
at a point arise from translations, rotations, and deformations. We can write this as
ui = (u0)i + ωijxj + ϵijxj . (3.42)
We see by inspection that the strain represents a multilinear relationship between that portion of
the displacement at a point that arises from deformation and the position of that point. That is
ui = 𝜀ijxj (3.43)
Strain also transforms according to Eq 3.16, so we conclude that it is also a tensor quantity.
Strain States
As a 2RT, any strain state has a principal coordinate system with principal strains which can be
written
(3.44)
and the description of the strain tensor can be found in any other coordinate system following the
transformation rule
. (3.45)
Like stress states, strain states can be said to be triaxial, biaxial, or uniaxial depending on
whether 3, 2, or 1 of the principal strains are non-zero, respectively.
We can also express the strain tensor as the sum of “hydrostatic” and deviatoric components, but
the ‘hydrostatic” part is usually referred to using the mean strain, em, defined as
∆
𝜀: = (3.47)
3
Thus, we can write
(3.48)
For sufficiently small strains, we can think of a strain state as the sum of a volume change given
6
by the dilatation Δ and a shape change given by the deviatoric strain 𝜖') .
Figure 3.14: A stress must be applied perpendicular to the plane of the applied strains to
maintain a biaxial strain state.
The same is true for a uniaxial strain state. A triaxial stress state must be applied to hold the
stains at zero transverse to the uniaxial strain axis.
to the line that goes through P and Q1 as shown in Figure 3.15. So Q1 does not move relative to P
and all of the displacement is counted as Du1 at Q2′.
x2
∆u1
Q2
Q′2
∆x2
P Q1 , Q1′ x 1
∆x1
(3.49)
While the tensor shear strains are defined using Equation 3.40, the engineering shear strains are
defined using
(3.50)
We see that the engineering shear strain values are just twice the tensor shear strain values. That
is, gij = 2eij. The engineering strains can be written in terms of the tensor strains in matrix form as
(3.51)
Comparing Figure 3.15 with Figures 3.11 and 3.12, we see that the engineering shear strain
includes a rotation and is not a true strain.
As we will see, this complicates analysis considerably. However, it is very important to note
that, by convention, strains are normally measured, recorded, and discussed using engineering
notation. Elastic constants are normally calculated using engineering strains. The engineering
strains are not tensor quantities and do not transform according to the tensor transformation rule.