Beruflich Dokumente
Kultur Dokumente
Nils Hallbäck
Department of Engineering and Physics, Karlstad University
2.1 Notation
The following notations are adopted throughout this series of lectures:
s : scalar
v : vector
u : matrix
If one of the vectors is a unit vector, then the scalar product is the projection of the other vector in the direction of the unit vector. Let e.g. v
be a unit vector so that v=1, then u · v=u·cosφuv .
A vector could be represented as a linear combination of the unit base vectors, e.g.
3
u = e1 u1 + e2 u2 + e3 u3 = ei ui = ei ui
i=1
where in the last expression Einstein’s summation convection is applied. If a term contains the same index twice, then Eintein’s summation
convection implies summation from 1 to 3 over that index. If an index appears in a term more than twice then the expression is corrupt. In
the above expression the ui are the componentes of the vector u in the ei basis. In fact they are the projections of u on the respective ei .
The length of any vector, say u, could be derived from the vector components as
The scalar product between two cartesian base vectors could be expressed as
ei · e j = δij (1.3)
This reflects the fact that the scalar product of two orthogonal vectors is zero, and that the projection of a unit vector onto itself is one.
Similar to the vector u, the vector v could be expressed as
v = e 1 v 1 + e2 v 2 + e3 v 3 = e i v i (1.4)
which gives the same result as the above calculation. Note here that
3
v j δij = v j δij = vi
j=1
since δij =1 only when j=i, where i could be any of the numbers 1, 2 or 3.
Hence the cross product between two vectors u and v is a vector perpendicular to the plane defined by u and v with a length equal to the
area spanned by u and v, see Fig. 2. Note also that the right hand rule according to Fig. 3 applies for the direction of the product. Hence
v ⨯ u = -u ⨯ v
The cross product between two different cartesian base vectors is the positive or negative of the third base vector, e.g.
e1 × e2 = e 3
e2 × e1 = -e3 ,
while the cross product of a base vector with itself obviously is zero.
The cross product between base vectors could generally be expressed as
ei × e j = εijk ek (note that summation over k from 1 to 3 is understood !)
The cross product between any two vectors u and v could be expressed in their components as
e1 e2 e3
u ⨯ v = u1 u2 u3 = e1 (u2 v3 - u3 v2 ) + e2 (u3 v1 - u1 v3 ) + e3 (u1 v2 - u2 v1 ) = εijk u j vk ei
v1 v2 v3
Note that the final expressions in the last two equations above are the same. Note also that ε ijk =εkij =εjki .
Useful identities between the permutation symbol and the kronecker delta
εijk εljk = 2 δil (1.7)
Fig. 2. Two vectors with spanned area and normal unit vector
Consider the volume V spanned by the vectors u, v and w. Using the symbols shown in Fig. 4 this volume could be expressed as
V = Auv · h (1.9)
but h is the projection of w in the direction of e⊥uv . Since e⊥uv is a unit vector then
h = w · e⊥uv (1.10)
Inserting Eq. 10 into Eq. 9 and utilizing the first expression of Eq. 6 gives
4 Ch2_Vectors and Tensors.nb
w1 w2 w3
V = w · (u ⨯ v) = el wl · εijk u j vk ei = εijk wl u j vk el · ei = εijk wl u j vk δil = εljk wl u j vk = u1 u2 u3
v1 v2 v3
Hence the volume spanned by the vectors u, v and w is obtained by the multiplication w · (u ⨯ v). This product is therefore sometimes
referred to as the box product. It can also be called the “triple product”.
So that the box product between three vectors could be more swiftly expressed as
w · (u ⨯ v) = wi u j vk ei · (e j ⨯ ek ) = wi u j vk εijk
The scalar components ui are the projections of the vector in the direction of the respective base vectors e i . As such, they represent the
embodiment of the vector u in the direction of ei .
Consider next a vector-like device τ where the components themselves are vectors, i.e. in each direction e j the device is represented by a
vectorial component. Analogeous to Eq. 11 the vectorial components could be expressed as (just replace u by τ j in Eq. 11)
τ j = e1 τj1 + e2 τj2 + e3 τj3 = ei τji (1.12)
The vectorial components τi of τ are the projections of τ onto the respective base vectors ei . Unlike Eq. 11, it is in the above expression not
self-evident how to interpret the multiplication operation. Each term is obviously a multiplication of two vectors, but it could not be the dot
product, since then the result would be a scalar. Nor could it be the cross product because then the result would be a vector. The multiplica-
tion operation here is another vector product referred to as the outer product. The outer product takes every component of the first vector
and multiplies it with every component of the second vector. Hence, since a vector is defined by 3 components, referred to a certain set of
base vectors, then the result of the outer product is described by 9 components. In order to distinguish the outer product from other
multiplication operations we denote it by ⊗. Devices created in this way are referred to as 2nd order tensors. In this case
τ = e1 ⊗ τ1 + e2 ⊗ τ2 + e3 ⊗ τ3 = e1 ⊗ (e1 τ11 + e2 τ12 + e3 τ13 ) + e2 ⊗ (e1 τ21 + e2 τ22 + e3 τ23 ) + e3 ⊗ (e1 τ31 + e2 τ32 + e3 τ33 ) =
e1 ⊗ e1 τ11 + e1 ⊗ e2 τ12 + e1 ⊗ e3 τ13 + e2 ⊗ e1 τ21 + e2 ⊗ e2 τ22 + e2 ⊗ e3 τ23 + e3 ⊗ e1 τ31 + e3 ⊗ e2 τ32 + e3 ⊗ e3 τ33 = e j ⊗ ei τji
The outer product of the base vectors ei ⊗ e j are referred to as unit dyads. The τji are the scalar components of the tensor τ. To derive a
certain scalar component τji from τ we first project τ on e j , to get the vector τ j , and then project this vector on ei . The tensor τ is a linear
Ch2_Vectors and Tensors.nb 5
combination of the dyads through their scalar components τji , as indicated above. The above reasoning can of course be extended to higher
order tensors. A device where the property in a certain direction is a 2nd order tensor is referred to as a 3rd order tensor. A 2nd order tensor
where each component is a 2nd order tensor is a 4th order tensor, and so on.
Hence the outer product of any two vectors u and v is a second order tensor
uv = u ⊗ v = ei ui ⊗ e j v j = ei ⊗ e j ui v j
The 9 components of the tensor uv could be represented as a matrix by expressing the vectors in the form of 3 by 1 matrices (i.e. as column
vectors). Referred to the cartesian basis this gives
1 0 0
e 1 = 0 , e 2 = 1 , e 3 = 0
0 0 1
Note here the difference between the dot (inner) product and the outer product. The inner product of two vectors is instead
v1
u · v = ( u1 u2 u3 ) v2 = u T v
v3
In particular
τ1 = e1 · τ = e1 · ei ⊗ e j τij = (e1 · ei ) e j τij = δ1 i e j τij = e j τ1 j
Note that it generally matters whether we premultiply or postmultiply a tensor with a vector. Postmultiplication yields
τ · n = ei ⊗ e j τij · nk ek = ei (e j · ek ) τij nk = ei δjk τij · nk = ei τij · n j
This is a vector where each component (τij · n j =τi1 · n1 + τi2 · n2 + τi3 · n3 ) is the projection on n of the vectorial component acting in the
direction ei . This vector is generally not the same as the vectorial component acting in direction n.
As seen above multiplications involving tensors and vectors boils down to multiplication operations involving the constituents base vectors
and dyads. Apart from the operations outlined above there are also some other common multiplication operations that sometimes needs to
be conducted. The dot product between two tensors σ and τ is computed as
σ · τ = ei ⊗ e j σij · ek ⊗ el τkl = σij τkl ei ⊗ e j · ek ⊗ el = σij τkl ei ⊗ (e j · ek ) el = σij τkl δjk ei ⊗ el = σik τkl ei ⊗ el
The result is a tensor. Therefore the dot product between tensors is sometimes referred to as the tensor product between two tensors. The
cross product between a tensor τ and a vector u is
τ × u = ei ⊗ e j τij × uk ek = τij uk ei ⊗ (e j × ek ) = τij uk ei ⊗ εjkl el = εjkl τij uk ei ⊗ el
The result is a tensor. Looking around a bit among the expressions presented so far in this section reveals that the dot and cross product
between dyads and base vectors follow a simple pattern: Simply take the dot, or cross, product of the nearest base vector on either side of
the dot, or cross, symbol.
The double dot product between two tensors σ and τ is somewhat different tough. It is defined as
6 Ch2_Vectors and Tensors.nb
σ : τ = ei ⊗ e j σij : ek ⊗ el τkl = σij τkl ei ⊗ e j : ek ⊗ el = σij τkl (ei · ek ) (e j · el ) = σij τkl δik δjl = σkj τkj
Note here how the double dot product between dyads are performed. It consists of the dot product between the first two base vectors in
each dyad, and by the dot product of the last two base vectors of the dyads. Since the result is a scalar, the double dot product is sometimes
referred to as the tensor scalar product.
Hence transposing a tensor implies a reverse summation order between the indices of the tensor components and the unit base vectors in
the dyad. Practically this can be accomplished by changing the order of the indices in the tensor component or in the dyad.
A tensor is symmetric if
τT = τ ⇔ τij = τji
A tensor is asymmetric if
τT = -τ ⇔ τij = -τji
where
1
τsym = τ + τT
2
and
1
τasym = τ - τT
2
The components of the unit tensor can be represented in matrix form by the so called unit matrix
1 0 0
I= 0 1 0
0 0 1
The component of this vector in e.g. the x′1 direction is found by projection as
u′1 = e′1 · ei ui
u′j = e′j · ei ui
This means that the vector components in the primed system could be derived from the components in the unprimed system by letting
coeffients involving projections of the base vectors operate on the original components
Define
αji = e′j · ei
Each coefficient αji is the cosine of the angle between e′j and ei . The transformation rule for vector components is thus
u′j = αji ui
Transformation of components from the second system to the first system is governed by
u j = e j · e′i u′i = αij u′i = αTji u′i
Consider now a tensor τ. If we would like to compute a certain (scalar) component τ′ij of τ in the primed coordinate system we should first
project the tensor to obtain the vectorial component τi′ of τ acting in the xi′ -direction. I.e.
τ′i = e′i · τ = e′i · ek ⊗ el τkl = (e′i · ek ) el τkl
This vectorial component of τ should then be projected in the x′j -direction to get τ′ij
τ′ij = e′j · τ′i = e′j · (e′i · ek ) el τkl = (e′i · ek ) (e′j · el ) τkl = e′i ⊗ e′j : ek ⊗ el τkl = e′i ⊗ e′j : τ
Hence the component of a tensor could be derived by projecting it on the corresponding dyad. This is accomplished by taking the double
dot product between the dyad and the tensor. In this way it acts very much like an ordinary vector. To get the component of a vector or a
tensor we project it on the corresponding base vector or dyad.
Expressed in α-coefficients the transformation could be expressed as
τ′ij = αik αjl τkl = αik τkl αjl = αik τkl αTlj
If the α-coeffiecents and the tensor components are represented as matrices then the transformation relation becomes
τ′ = α · τ · αT
An example of a tensor component transformation is given in the exercise section. Note that the rows of α are the components of the
corresponding primed base vector expressed in the unprimed system. Hence
e′1 · e1 e′1 · e2 e′1 · e3
α = e′ · e e′ · e e′ · e
2 1 2 2 2 3
e′3 · e1 e′3 · e2 e′3 · e3
8 Ch2_Vectors and Tensors.nb
where each product is the cosine of the angle between the two base vectors.
where τp is the magnitude of the vectorial component. In such a case the direction m is said to be a left eigenvector to τ (because it
multiplies τ from the left side) and τp is the corresponding eigenvalue.
By representing the components of m as a 1×3 matrix the above expression could be written in matrix form as
· τ = m
m ·τ ⇒
p
τ - I τ = 0 ⇒
m p
m1 ( τ11 - τp τ12 τ13 ) + m2 ( τ21 τ22 - τp τ23 ) + m3 ( τ31 τ32 τ33 - τp ) = ( 0 0 0 ) (1.14)
This is a third order equation with three solutions for τp . Each τp gives the magnitude of the vectorial component in a certain direction
given by the corresponding left eigenvector m. The left eigenvector components corresponding to a particular eigenvalue τp could be
computed by inserting the eigenvalue into Eq. 14 and solving for m1 , m2 and m3 . Note here that since det(A)=0 the equation system given
by Eq. 14 is singular. This means that only the relation between the components could be computed, i.e. the direction. Regarding the length
of the eigenvector it is common practice to set it to unity, i.e. m21 + m22 + m23 = 1.
Note however that the term “eigenvector” normally (e.g. in math program packages like Matlab and Mathematica among others) refers to
the so called right eigenvector defined by the equation
τ · n = n · τp
With the right eigenvector n represented by a 3×1 matrix, the above equation could be expressed in matrix notation as
τ · n = n · τp
Hence the left eigenvector is the transpose of the right eigenvector to τT . Since the determinant of a matrix is equal to the determinant of
the transpose of the matrix it follows that the eigenvalues to the left and right eigenvalues are the same. These relationships are valid as
long as all components in τ are real valued. For a symmetric matrix τ the left and right eigenvectors coincide. This is an important
Ch2_Vectors and Tensors.nb 9
observation since many matrices that appear in engineering applications are symmetric.
⇒ -τ3p + (τ11 + τ22 + τ33 ) τ2p + (τ11 τ33 - τ13 τ31 + τ22 τ33 - τ23 τ32 + τ11 τ22 - τ12 τ21 ) τp + det(τ) = 0
For the equation to be unique the coefficients in the equation has to be unique, i.e. invariant with respect coordinate transformation. Hence
the invariants of a tensor are
I1 = τ11 + τ22 + τ33 = trace(τ)
I2 = τ11 τ33 - τ13 τ31 + τ22 τ33 - τ23 τ32 + τ11 τ22 - τ12 τ21
I3 = det(τ)
where the notation “trace” has been used for the first invariant. The trace of a matrix is simply the sum of the diagonal elements of the
matrix.
where
∂ ∂ ∂ ∂
∇ = e1 +e2 +e3 = ei
∂ x1 ∂ x2 ∂ x3 ∂ xi
The result is a scalar. The divergence is measure of the inbalance of the vector field that must be compensated by sources or sinks. If e.g.
the vector field describes the velocity field of a fluid, a negative divergence implies the vanishing of the fluid into a sink.
The curl of the vector field is obtained as
∂ ∂vj ∂vj ∂vj ∂ v3 ∂ v2 ∂ v1 ∂ v3 ∂ v2 ∂ v1
curl(v) = ∇ × v = ei ×vj ej = ei × e j = εijk ek = εkij ek = e 1 - + e2 - + e3 -
∂ xi ∂ xi ∂ xi ∂ xi ∂ x2 ∂ x3 ∂ x3 ∂ x1 ∂ x1 ∂ x2
The result is a vector. Again, if the vector field is the velocity field of a fluid, the curl vector represents the rotational speed of that fluid in
each of the three coordinate directions (see Fig. 6).
10 Ch2_Vectors and Tensors.nb
The gradient of a vector field is the outer product of the vector differential operator and the vector field, i.e.
∂ ∂vj
∇ v = ∇ ⊗v = e i ⊗v j e j = ei ⊗ e j
∂ xi ∂ xi
The result is a tensor. This is not surprising, since the vector field is defined by 3 components - one in each of the three coordinate
directions. A full description of the variation of the vector field must involve the variation of each of these three components in each of the
three coordinate directions. This leaves 3×3=9 unique variations.
Finally we consider the tensor field τ = τ(x1 ,x2 ,x3 ). The divergence of a tensor field is obtained as
∂ ∂ τkl ∂ τkl ∂ τil
∇ · τ = ei · ek ⊗ el τkl = ei · e k ⊗ e l = δik el = el
∂ xi ∂ xi ∂ xi ∂ xi
which is a measure of the inbalance of the tensor field, sometimes literally speaking, as will be seen later in this book.
Another operator frequently seen in engineering mechanics applications is the Laplace operator (Δ). It is the divergence of the gradient, or
in other words, the inbalance of the variation. The laplacian of a scalar field is
∂ ∂ ∂ ∂2 s ∂2 s ∂2 s ∂2 s ∂2 s
∇ · ∇ s = Δs = ei ·ej s = ei · e j s = δij = = + +
∂ xi ∂xj ∂ xi ∂ xi ∂ x j ∂ xi ∂ xi ∂ x21 ∂ x22 ∂ x23
Suppose that s is the temperature field of a body. We know from basic thermodynamics that the flow of heat is proportional to the variation
of the temperature (heat flows from warmer to colder areas). If there is an inbalance in the variation then there would be a net increase or
loss of heat which would lead to an either increasing or decreasing temperature. Hence for a steady state temperatue field the laplacian of s
is equal to zero.
The laplacian of a vector field:
∂ ∂ ∂2 v k ∂2 v k ∂2 v k
∇ · ∇ v = Δ v = ei ·ej ⊗vk ek = ei · e j ⊗ e k = δij ek = ek
∂ xi ∂xj ∂ xi ∂ x j ∂ xi ∂ x j ∂ xi ∂ xi
∇ ·v ⅆV = n·v ⅆS (1.16)
V S
∇s ⅆV = n s ⅆS (1.17)
V S
∇ ·τ ⅆV = n·τ ⅆS (1.18)
V S
Note here that these relations in fact are quite easy to remember. Just replace ∇ in the volume integral with n in the surface integral. The
above relations can also be expressed in component form. The first (Eq. 16) becomes
∂ vi
ⅆ V = ni · v i ⅆ S
V ∂ xi S
Exercises
1. Let u, v, w and q all be vectors. Show that
(u × v) · (w × q) = (u · w) (v · q) - (u · q) (v · w)
3. Consider a tensor with components τ11 = 20, τ12 = -10, τ13 = 30, τ21 = 5, τ22 = 10, τ23 = 40, τ31 = -15, τ32 = 20, τ33 = -25 referred to
a certain cartesian coordinate system. What will the components of this tensor be in a cartesian coordinate system that is rotated 30° around
the x3 - axis of the original cartesian system.
4. Compute the eigenvalues and the left eigenvectors of the tensor defined in the previous exercise.
5. Let τ be a 2nd order tensor and v a vector. Derive a formula for the divergence of the dot product between τ and v, i.e.
∇ · τ · v
6. Consider the fluid flow described by the velocity field v. Consider an arbitrary control volume surrounding a point Q in the fluid. Let the
surface of the control volume be denoted S and the volume by V. Integrate the velocity field over the surface S to compute the net flow
12 Ch2_Vectors and Tensors.nb
(volume/s) of the fluid out of the control volume. Express the volume per unit time that flows out of each small part (dS) of the surface and
“sum” the contribution (see Fig. 8) from each dS over the whole surface S. Apply the Gauss integration theorem and imagine that the
control volume is shrunk towards the point Q. What do you conclude?
Fig. 8
Rotation around the 30° around the x3 - axis gives the transformation coefficient matrix as
= {{Cos[30. Degree], Cos[60. Degree], 0},
α
{Cos[120. Degree], Cos[30. Degree], 0}, {0, 0, 1}}
0.866025 0.5 0
-0.5 0.866025 0
0 0 1
Then
′ = α
τ .τ
.Transpose[α
]
The eigenvector components corresponding to τp = -31.2132 is derived by inserting this value, along with stress components into Eq. 14.
Since the system is singular we choose to use only the first two equations along with the constraint that the eigenvector shold be a unit
vector, i.e.
Solve[
n1 {20. - τp , - 10} + n2 {5, 10 - τp } + n3 {- 15, 20} == {0, 0} && n1 2 + n2 2 + n3 2 == 1 /. τps [[1]],
{n1 , n2 , n3 }]
Note two solutions are obtained reflecting the fact that the unit vector opposite to an eigenvector also is an eigenvector to the same
Ch2_Vectors and Tensors.nb 13
An so on. The simplest way to do the computation is however to employ the built-in routine for eigenvalue analysis in Mathematica. Note
here that this routine computes the right eigenvector, whence we need to call this routine with the transpose of the matrix as an argument,
i.e.
]]
Eigensystem[Transpose[τ