Está en la página 1de 27

Chapter 1

Basic Concepts, Index Notation and Tensor Calculus


1.1 Introduction
A continuum is defined as a volume of material that is absolutely continuous in the
mathematical sense. Absolutely continuous means that it is not possible to subdivide a
continuum into volumes that are so small that they lose their identity. Hence, if you consider air
to be a continuum, it would not be possible to divide a container of air into sub-regions that were
so small that the composition of the gas defined as ``air" inside that volume varied appreciably
(i.e. there cant be too many or too few nitrogen or oxygen molecules, etc. or, at the most
extreme case, you cant divide a region into volumes that are so small that there are no molecules
in them at all). At Standard atmosphere conditions (a pressure of 101,325 N/m2 ) or 101,325
Pascals (Pa), and a temperature of 0 C (or 273.15 K ), there are 2.546 10 25 molecules of air
( N 2 , O2 , etc.) in a cubic meter of atmosphere. Consequently, one ``air molecule" occupies a
volume of
1
vMolecule = m3 ,
2.546 10 25

corresponding to a cube with dimensions:

9
v1/3
Molecule = 3.400 10 m 3.400 nanometers = 34 Angstroms ().

For comparison, visible light has wavelengths on the order of 0.5 microns (the wavelength of
green light is 0.52 m ), so that there would be approximately 147 air molecules along a tube
of diameter v1/3 , and of length equal to one wavelength of green light.
Lead (Pb) is one of the most dense naturally-occuring solids, with a density of 11,350
3
kg/m . Since the atomic weight of lead is 207.19 , a cubic meter of lead contains 54,781 gram-
moles or, multiplying by Avogadro's number (6.022141 x 1023 molecules per gram-mole), there
would be 3.299 10 28 Pb atoms in a cubic meter. Hence, the characteristic spacing between lead
atoms in its normal solid phase would be 3.11 10 10 m or 3.11 . There would be more than
a thousand lead atoms along a tube whose diameter was 3 , and whose length was one
wavelength of green light. Unless you are working with nanomaterials, in a particle physics
laboratory or in one of the silicon chip industries, the characteristic distances between atoms and
molecules in most physical systems are so small that we cannot detect atomic or molecular
effects using typical instrumentation. Hence, we will assume that normal materials are continua,
enabling mathematical limits to be employed.
If we are willing to exclude physical measurements in volumes that are so small that only
a few molecules or atoms of the material can be found, making statistical effects important, we
can treat fluid and solid systems as though they are perfectly continuous. Here, perfectly
continuous means, when we are modeling a solid or a fluid as a continuum, we can employ
rigorous mathematical concepts associated with the continuity of functions to invoke precise
mathematical relations to describe the behavior of a particular type of material; thus
differentiating between different types of gases. liquids and solids. An example would be the
density of a material at a specified point in space and time, where the density would be defined
as the total mass in a very small instantaneous volume surrounding that point, divided by that
very small occupied volume, taking the limit as the occupied volume approached zero. (This is
where real physical systems can start to exhibit erratic behavior if you really kept making the
volume smaller and smaller.)
Continuum mechanics is the study of material systems from a rigorous mathematical
point of view. Ignoring the fact that all real materials begin to behave as discrete particles before
the true ``zero'' mathematical limit is achieved, we choose instead to retain the mathematical
rigor associated with evaluating continuous systems in their mathematical limits. In essence, we
acknowledge that in the small particle limit, there are other disciplines such as solid-state physics,
rarefied gas dynamics, particle physics, nanomaterials and quantum mechanics1 that must be
considered in order to properly model system behavior at that level of resolution. We only state
that when particle effects become important, a continuum model fails to fully characterize those
systems. Otherwise, we are on solid ground when we assume that mathematical limits can be
taken to arbitrarily small values.
In order to utilize the mathematical power of the continuum mechanics approach, it is
necessary to study three-dimensional, time varying systems. As undergraduates, you have been
exposed to continuum mechanics in courses such as introductory fluid mechanics and solid
mechanics which probably used one- and two-dimensional, steady-state mathematical models
to analyze a variety of important everyday engineering problems; you were not required to
employ notation that enabled you to work in three-dimensional, time-varying space. In order to
utilize the power of vector and tensor analysis in continuum mechanics, a short-hand notation
has been developed to avoid writing down too many three-dimensional equations term-by-term,
because of their excessive length. Furthermore, it is possible to develop our theories and evolve
a solid fundamental understanding of the subject by relying almost exclusively on
representations in rectangular Cartesian coordinate systems (i.e. the traditional, right-handed
x, y, z coordinate system). We can employ mathematical rigor in the study of continuum
mechanics without getting bogged down in accounting for changes that result from using
curvilinear coordinates (such as cylindrical and spherical coordinate systems). Although
curvilinear coordinate systems are often the preferred or essential coordinates for a variety of
important problems, we will delay introducing those coordinate systems until after we have
begun to understand and appreciate the power of using continuum mechanics methodology to
study physical systems. We will utilize Cartesian Tensor Notation or index notation throughout
these lecture notes, because it greatly reduces the amount of equation-writing required to
represent fundamental physical concepts and also because it allows us to utilize additional
mathematical tools to exploit powerful continuum mechanics concepts.

1.2 Cartesian Tensor Notation


1
We wont be concerned with Higgs bosons or how sub atomic particles possess mass.
It is our desire to approach problems in the most general possible context (three-
dimensional, unsteady). All types of coordinate systems and coordinate motions are possible,
often very beneficial, in studying various types of actual physical problems. However, if
generalized coordinates and generalized coordinate accelerations (linear and angular) are
introduced when learning the fundamentals, the neophyte2 is likely to get lost in the equations.
Cartesian tensor notation is very useful, even though we will be restricted to a right-hand (only
to avoid confusion), rectangular coordinate system. However, rather than using the usual x-, y-,
and z-coordinate designations, we will be using subscripts (1, 2 and 3) to keep track of the vector
components or coordinate directions. In these lecture notes, we will use bold, lower case letters
to designate a vector3. Here, a vector is a three-dimensional quantity that has a specified
magnitude and direction. For example, the location of a point in a Cartesian coordinate system
could be specified using a position vector, r, where

r = xi yj zk ,

and x is the magnitude or distance in the direction along the unit vector i . However, we will not
use ( x, y, z ) or unit vectors i, j, k in these notes. Instead, we will use coordinates given by:

x1 x,
x2 y,
x3 z, (1-1.)
e1 i,
e2 j,
e 3 k.

This change in notation enables us to use a very convenient form of shorthand, called
index notation (which is part of Cartesian tensor notation). Instead of writing x, y, z
coordinates all of the time, we can write: Consider coordinate xi , where i is called an index, and
it is understood that i can be any of three values, 1 , 2 , or 3 . Similarly, we can talk about an
arbitrary unit vector, e i , where, depending on the integer assigned to the index i , (again taken
only as 1, 2 or 3 ), it can represent any one of our old unit vectors (i, j, or k ) . The concept is
most general when we try to let i represent any one of the three coordinate directionsall at
the same time. We can do this because we know that each othogonal component of a vector
equation must balance in the same way as the actual combined-component vector equation.
Recalling your undergraduate experience, you probably spent a great deal of time trying to
balance forces in the x-direction and then in the y-direction. You knew which quantities were
which because you used different symbols for the x-components and the y-components. When
you were forced to look at three-dimensional systems, you were stuck with three sets of

2Beginner or novice.
3
Since it is not possible to differentiate bold from unbold letters when taking notes or when writing on a white

board, you are encouraged to use arrows over vector symbols, i.e. a a , etc.
equations. Here, we are going to let the subscript numbers (indices) 1 , 2 , and 3 tell us whether
we are looking at the x (actually x1 ), y (actually x2 ), or z -component (actually x3 ) . Without
additional information, we can talk about some quantity ui , as any one of the three components
u1 , u2 or u3 of vector u. Notice that we have been using bold face type to indicate vectors and
will preserve this notation throughout these notes. We are also going to use lower case letters
(i,j,k, etc.) to represent arbitrary subscript indices (1,2, or 3).
Index notation does more than reduce the number of symbols used for simple quantities
(like the components of a force or velocity), it is also very compatible with matrix and vector
operations on a computer, and for our purposes, it enables us to utilize the Einstein Convention
(also called the Einstein Summation Convention). The Einstein Convention is a form of shorthand
that probably derived originally from the vector dot product. Recall that the dot product is an
operation performed on two vectors, when both are represented using the same unit base vector
components. The dot product operation produces a scaler (a single numerical value) utilizing a
specific combination of the components of the two vectors. Thus, if we represent vector a as
a = a1i a2 j a3k a1e1 a2 e 2 a3e 3 and vector b as b = b1i b2 j b3k b1e1 b2 e 2 b3e 3 ,
then the dot product a b is calculated as
a b = a1b1 a2b2 a3b3 (1-2.)
which we clearly see can be written in index notation as
3 3 3
a b = ai bi = a j b j ak bk , etc. (1-3.)
i 1 j 1 k 1

It is very important to note that using the Einstein Convention means that any time an index is
repeated in a product expression, it is to be summed from the index level of 1, through index level
3. Thus, if we look at the position vector r,
r = xi yj zk , (1-4.)

we see that we can use index notation to represent that vector in the following ways:
3
r = x1e1 x 2 e 2 x3e 3 xi e i xi e i . (1-5.)
i =1

It is very important to note that while we will restrict the use of Einstein Convention to
three-dimensional space in these notes, there are many other types of general coordinate
systems (such as phase spaces) that involve more than three dimensions, and it is possible to
define the Einstein Convention differently from our definition for those models. We will restrict
these notes to three-dimensional space so that
r = xi e i = x j e j etc.

1.3 Elementary Definitions and Operations Employing Index Notation


1.3.1 Kronecker Delta
We define the Kronecker delta, ij , as follows:

1, when i = j
ij = (1-6.)
0, when i j.
Note that 11 = 22 = 33 = 1 , but ii = jj = 3 , due to the Einstein summation convention. The
Kronecker delta is very useful in abbreviating actual vector dot product operations. It can also
be used to change the letter representing a live index. Consider the expression:

vi ij v j .

Because of the Einstein convention, we know that the right hand side of that equation must be
summed so that we have:
ij v j = i1v1 i 2v2 i 3v3 .

Since we haven't specified an index value for ``i" , we dont know which one of the three terms
on the right side of this equation will contain matching indices; therefore we dont know which
Kronecker delta term will be unity (for that term). The interesting idea is that whatever the
specification for ``i" , it can only be 1 or 2 or 3. Hence, if i is 1, then i1 = 11 = 1 , and i 2 = i 3 = 0,
leaving the right hand side of the above equation as : v1 0 0 . On the other hand, if i is either
a 2 or a 3 , the same process can be used to show that ij v j always reduces to vi and two zeros.
This seems like a great deal of work for something that is obvious, but this `` trick '' will be very
useful later.
Recall that the dot product between two base (unit) vectors is equal to unity when the
two base vectors are collinear and the dot product is zero when they are orthogonal (not the
same integer), hence
e i e j = ij . (1-7.)
Similarly, if vectors u and v are represented as:

u ui e i ,
and v v je j ,

since e i e j ij , (1-8.)

u v = ui ij v j = ui vi = u j v j = u k vk = etc. (1-9.)

Obviously, the repeated indices can be called dummy indices since their designation
letters can be chosen arbitrarily (but each dummy letter must only appear twice in any distinct
polymomial expression). You can use different dummy indices on the same side of an equal sign,
as long as the positions for the repeated indices in a specific term are preserved. When an index
is not repeated in a term, you can call that index a live index, because that index is controlling
the actual component of a vector (or, you will learn later, it can be a tensor component) and it is
only possible for an equation to be a valid vector equation if every term has the same letter
subscript as the live index representing the equation. That restriction will also be useful later in
this course. Now, suppose we have a vector relation:

u v.

Since the Cartesian base vectors (e i ) are orthogonal, we know that each component of the
vector pair must satisfy the relation:

ui = vi , where i = 1,2 , or 3 is inferred.

Since we cannot change the letter designating the subscript for the component of u without
changing the letter designating the subscript for component v , as stated before, that index is
called a live index. Vector equations are characterized by a single, recurring live index. Unlike
the dummy indices, the same symbol (letter) used to designate a live index in any term of an
equation must be preserved (used) in every term on both sides of the equal sign of that
equation. A scaler equation can contain multiple dummy indices, but no scalar term can contain
a live index. For example, if v is a velocity vector, we could define kinetic energy, q , noting that
energy is a scalar quantity, by:

1 vv
q= vv = k k .
2 2

Here, we see that the scalar is formed from the dot product of a vector with itself. You should
also realize that it is very convenient to consider only one base-vector direction in balancing
forces and other conservation operationsjust like you balanced forces in the x- and y-directions
separately in your undergraduate courses. However, in this course, you wont write down
statements like balancing forces in the x-, or y- or z-directions, you will simply write a force
balance on the ith component as follows:
N
0 FI i , i = 1, 2, or 3.
I 1

Notice that a capital letter subscript was employed to allow many different forces (more than
three) to be acting on a system such as a free-body diagram. Forces like friction or gravity or
magnetism could also be acting and thus I could be (at least two, but) any number. In any case,
a summation sign is required for the capital letter index because N does not represent the vector
components of a force, and is used instead to identify any number of different forces, greater
than or equal to 2.
1.3.2 Permutation Symbol (Alternator)

The vector cross product operation is more difficult to represent, partly because that
operation is restricted to right-hand coordinate systems. We first define the alternator or
permutation symbol, eijk as:

0, for i = j, j = k , or k = i

eijk 1, for (i, j, k ) (1,2,3), (2,3,1), (3,1,2) (1-10.)
1, for (i, j, k ) (3,2,1), (2,1,3), (1,3,2)

If you are interested in set theory, another way to represent this is that eijk has the value of 1
if (i, j , k ) is an even permutation4 of (1,2,3) and eijk has a 1 value if it is an odd permutation,
while anytime two or more indices are the same integers, eijk = 0 . A single permutation means
that two adjacent integers are swapped. For example, if the original alternator element is
e123 ( 1) , the following single permutations are possible by interchanging 1 with 2, or 2 with 3:

e 213 or e132 = -1.

To better remember when the permutation symbol is 1 or 1 , you can write out the sequence
123123123 = 123123123 = 123123123, and realize that any three consecutive integers from
that sequence is an even permutation of (1,2,3) and thus produces a permutation symbol value
of positive 1. Similarly, if you write out the backward sequence 321321321 = 321321321 =
321321321, any three consecutive integers from that sequence is an odd permutation of
(1,2,3) and thus has a value of negative 1. With this in mind, we can identify even and odd
permutations (being unconcerned with which permutation of i,j,k is actually positive or negative)
and interchange adjacent integer symbols (that operation is a single permutations) to write:

eijk = e jik = e jki = ekji . (1-11.)

You need to be sure to realize that the specific set of originally-specified integers can correspond
either to a positive or negative permutation value. Spending some time learning to understand
the permutation symbol will help you in understanding other topics later in these notes since the
permutation symbol will be used in many derivations. Equation (1-10.) contains three live
indicies that can be selected or specified separately. Of course if i = j , j = k and/or i = k ,
Equation (1-10.) is merely 0 = 0 0 . Now, you will recall that the vector cross product, u v ,
can be represented by the determinant:

e1 e 2 e3
u v = u1 u 2 u 3 u 2 v3 u 3 v 2 e1 u1v3 u 3 v1 e 2 u1v 2 u 2 v1 e 3 . (1-12.)
v1 v 2 v3

Employing the alternator, we can write:

4A permutation is an ordered arrangement of a finite number of elements from a set, in our case, (1,2,3) , where
each indice (1, 2, or 3) is listed exactly once.
u v eijk e i u j v k emnpe m u n v p = eijk u i v j e k ... , (1-13.)
e123u1v2 e 3 e132u1v3e 2 e213u 2 v1e 3 e231u 2 v3e1 e312u3 v1e 2 e321u3 v2 e1
(1)u1v2 e 3 (1)u1v3e 2 (1)u 2 v1e 3 (1)u 2 v3e1 (1)u3 v1e 2 (1)u3 v2 e1
u 2 v3 u3 v2 e1 u3 v1 u1v3 e 2 u1v2 u 2 v1 e 3 ,

where the alternator terms containing two or more identical indices that occur as a result of the
summations have been omitted since those terms are equal to zero. You can see from the final
version of this summation that it is exactly equal to Equation (1-12.), because the Einstein
summation convention requires that we sum over all three sets of repeated indices i , j , and k .
It will be to your benefit if, on your own, you expand all of the above summations using the
dummy indices in Eq. (1-13.), and noting all of the terms that go away because of matching
integers, in order to prove the equivalence with Eq. (1-12.) for yourself. It is instructive to
consider a vector definition involving the cross product. If we define w = u v , we can write the
vector-component relation:
wi = eijku j vk = e jik vk u j . (1-14.)
Notice that Eq. (1-14.) applies to the ith component, so that i is now a live index, rather than
contracting i with the base (unit) vectors. Just like you balanced forces acting in orthogonal
coordinate directions (i.e. x-, y-, and/or z-directions) in your undergraduate courses, you need to
realize that it is usually more convenient here to consider equations governing a single coordinate
direction component ( wi ), rather than the overall vector equation ( w wi e i ).

Before we leave the cross product operation, we can explore its connection with
determinants. In general, we will use subscripted capital letters like the symbol Aij to designate
the elements in a 3 3 matrix array, where the first index (i ) designates the row location of an
element and the second index ( j ) designates the column location5. Hence, if Aij represents the
elements in our cross-product-determinant (1-12.), we note that the elements in the first row of
Eq. (1-12.) can be represented as A1, j = e j , and they are the only vector elements in the cross
product matrix, A . Lets examine the expression eijk A1i A2 j A3k :

Since all three indices are repeated, this expression can be expanded, recognizing that if i = 1, j
must be either 2 or 3, and whichever index is assigned to j, leaves only one index for k for which
the alternator, eijk, is not zero, and so on. Hence,

eijk A1i A2 j A3k

e123 A11 A22 A33 e132 A11 A23 A32 e213 A12 A21 A33 e231 A12 A23 A31 e312 A13 A21 A32 e321 A13 A22 A31 .

Evaluating each alternator, this equation reduces to:

5An easy way to always remember Rows then Columns is the phrase ``Roman Catholic''.
eijk A1i A2 j A3k A11 A22 A33 A11 A23 A32 A12 A21 A33 A12 A23 A31 A13 A21 A32 A13 A22 A31

A11 A12 A13


A11 A22 A33 A23 A32 A12 A21 A33 A23 A31 A13 A21 A32 A22 A31 A21 A22 A23 .
A31 A32 A33
We can therefore write:


det Aij = eijk A1i A2 j A3k . (1-15.)

If we used this relation to calculate the determinant of Eq. (1-12.), we would show that it gives
us the cross product, which of course was the original definition. However, the use of the
alternator in Eq. (1-15.) can be used to develop additional ideas and concepts.
The homework material that follows is intended to cause you to prove these relationships
yourself. This homework is an essential part of the learning process. In order to insure that you
actually understand the subtleties of this material you must complete all of the homework. You
may want to create your own textbook for this course by incorporating your own words in the
basic text and by writing up explanations of the homework. You should write up the homework
in files that are suitable to be added to this text.

Exercise and Problem 1-1 (Sample Solution) Prove that

det Aij =
eijk emnp Aim Ajn Akp
6
. (1-16.)
This can be proven by expanding all of the elements, which is very tedious and of little value.
Alternatively, you should use words to describe either the eijk Aim A jn Akp operation or the
emnp Aim A jn Akp operation. Incidentally, you will also prove that:


det Aij = eijk A1i A2 j A3k = emnp Am1 An 2 Ap 3 . (1-17.)

Proof: We know from (Eq. 1-15.) that the determinant of a 3 3 matrix is given by eijk A1i A2 j A3k .
We also observe that i , j , k , m , n , and p are all repeated indices in (Eq. 1-16.) and therefore
must be summed from 1 to 3 (in any order). We choose to expand the summations of i , then j,
then k , recognizing that the alternator has only 6 non-zero values, where

eijk emnp Aim A jn Akp = emnp [e123 A1m A2 n A3 p e231 A2 m A3n A1 p e312 A3m A1n A2 p
e321 A3m A2 n A1 p e213 A2 m A1n A3 p e132 A1m A3n A2 p ] ,
which simplifies to
eijk emnp Aim A jn Akp =
emnp [ A1m A2 n A3 p A2 m A3n A1 p A3m A1n A2 p A3m A2 n A1 p A2 m A1n A3 p A1m A3n A2 p ] .

Now, m , n , and p are dummy indices in this last relation and since we have replaced all of the
original repeated indices with numbers, as the summations were developed in the previous
equation, we can now change the ``names'' of m, n, p to i, j , k , respectively. Then, we have:

eijk emnp Aim A jn Akp =


eijk [ A1i A2 j A3k A2i A3 j A1k A3i A1 j A2 k A3i A2 j A1k A2i A1 j A3k A1i A3 j A2 k ].

If we can manipulate this expression so that it looks like Eq. (1-15.) over and over, we can
complete the proof. We see that the first term is already identical with Eq. (1-15.) and returning
to the permutation concept of Eq. (1-11.), we can re-arrange our indices, carefully keeping track
of the signs of the alternators, based on whether the integer interchanges represent even or odd
numbers of adjacent integer exchanges, and recognizing that we are free to rearrange the
postions of the A elements to write:

eijk emnp Aim A jn Akp = eijk A1i A2 j A3k ekij A2i A3 j A1k eikj A3i A1 j A2 k
(ekji ) A3i A2 j A1k (e jik ) A2i A1 j A3k (eikj A1i A3 j A2 k ) .
Now
(ekji ) A3i A2 j A1k erjs A3s A2 j A1r erjs A1r A2 j A3s eijk A1i A2 j A3k ,
where, after converting the double negative to a positive, we changed the name of dummy index,
k, to r, and dummy index i, to s; then switched the locations of matrix elements A3 s and A1r (since
the positions of the matrix elements dont affect the result when using index notation and
perserving the locations of the dummy index pairs); and finally, we have changed the name of
dummy index, r, to i, and dummy index s, to k. You can use the same type of manipulations on
the last two double-negative terms to show finally that:

eijk emnp Aim A jn Akp = 2eijk A1i A2 j A3k 2ekij A1k A2i A3 j 2e jki A1 j A2 k A3i

Thus, from Eq. (1-15.), we see that

eijk emnp Aim A jn Akp = 2detA 2detA 2detA 6 det A .

To receive credit for this homework problem you must repeat this proof in your own words, and
then use a similar approach to prove that:

det[ Aij ] = eijk A1i A2 j A3k = emnp Am1 An 2 A p 3 .

Upon examination of Eq. (1-16.), an interesting operation has been defined by eijk emnp .
Because of the definition of the permutation symbol (alternator), you should realize that when
eijk and emnp are both non-zero, i.e. each alternator must have a value of either +1 or -1, then the
set of indices ( i, j , k ) must be a unique combination of 1,2, and 3 , as must the integers m, n and
p. Consequently, index i must be equal to one and only one of the indices m, n, or p . If i = m ,
we can employ the Kronecker delta to make that statement, i.e., writing im is equivalent to
saying: i equals m, because that Kronecker delta term is zero unless i equals m. Then, in that
case it follows that j must either equal n or p , since all other relationships between the
numerical index value of j and the index values of n or p would cause one of the alternators
to equal zero. Using this logic process show:

Problem 1-2 Prove that

eijk emnp = im jn kp in jp km ip jm kn im jp kn in jm kp ip jn km (1-18.)

Hint: As was just noted, if i = m , then we can represent that caseone out of six non-zero
combinations of eijk emnp by writing im , to enforce the statement, i = m . By writing down that
Kronecker delta statement, you can then take a look at the other four indices (j, k, n and p), noting
that if k = n ( kn ) , then j must equal p ( jp ) , and one possible non-zero alternator combination
involves the index equalities represented by im kn jp . However, we know that it is possible for
eijk and emnp to be either 1 or 1 , independently and since the Kronecker deltas are not always
equal to plus one, we have to do a little bit of additional work in order to keep track of the sign of
the alternator products. Suppose that we take the case when eijk is given by e123 (in other words,
i = 1, j = 2 , k = 3 ), then the presence of the Kronecker deltas im , kn , jp , tell us that the
corresponding representation of emnp must be e132 . Now, for this example case, eijk (e123 ) is an
even permutation and that alternator must therefore have the value of 1, while emnp is an odd
permutation ( e132 ) and must therefore have the value of 1 . Hence, for the case when eijk emnp is
given by e123e132 , we have one situation (out of three possible values for index i) where i = m ,
j = p , and k = n . This would take forever! What if we said that if indices i,j,k,m,n,p, satisfy the
condition, i = m, j = p, and k = n, represented by im jp kn , we have a case where emnp = eikj = eijk
? Now, this situation is true whether we pick i = 1 or 2 or 3 and we see that if eijk is 1 , emnp
is 1 and visa versa, so that this triple product of Kronecker deltas always represents a case when
the first alternator and the second alternator have opposite signs ( 1 and 1 ) and therefore
that combination is always negative. We can thus write:
eijk emnp = im kn jp

i.e. all other possible combinations, when the two alternators are not equal to zero. You need to
fill in the remaining details to complete this proof. Could you have used this relation to prove the
determinant identity? If this approach is easier for you, work the two problems in reverse.

At this time, it may be beneficial to see an alternate proof of Problem 1-2. We begin by
writing out the alternator in terms of the Kronecker delta, namely if eijk e123 , then that case
can be represented as i1 j 2 k 3 . Why would e 213 be represented as j1 i 2 k 3 ? You should
show that

eijk = i1 j 2 k 3 i 2 j 3 k1 i 3 j1 k 2 i 3 j 2 k1 i 2 j1 k 3 i1 j 3 k 2 . (1-19.)

If you cannot see why this relationship is correct, simply choose values for the indices of Eq. (1-
19.) and see that you will always get a true statement. We can quickly see that Eq. (1-19.) can be
written as:
i1 j1 k1
eijk = i 2 j 2 k 2 = det ij .
(1-20.)
i3 j3 k 3
Now, if we follow the same process for emnp , we get
m1 n1 p1
emnp = m 2 n 2 p 2 = det jm . (1-21.)
m3 n3 p 3

Using Eqs. (1-20.) and (1-21.), we can write out the product of eijk and emnp as:

eijk emnp = det[ ij ]det[ jm ]. (1-22.)

Recall that when dealing with determinants of matrices, we can apply the following properties:

det[A] = det[A T ] , (1-23.)

where AT is the transpose of matrix A, meaning that the rows and columns are interchanged, i.e.

A11 A12 A13 A11 A21 A31


If A A21 A22 A23 , then A A12
T
A22 A32 . (1-24.)
A31 A32 A33 A13 A23 A33

Also, determinants have another useful property:

det[ A]det[B] = det[ A B]. (1-25.)

Therefore we can write Eq. (1-22) as:

eijk emnp = det[ ij mj ] = det[ ij mj


T
] = det[ ij jm ]. (1-26.)
We already know that ij jm = im so our final step is to expand the determinant of im which
will give us
im jm km
eijk emnp = in jn kn
ip jp kp

= im jn kp in jp km ip jm kn im jp kn in jm kp ip jn km . (1-27.)

Problem 1-3 Show that


eijk eimn = eijk emin eijk emni e jki emni = etc. = jm kn jn km (1-28.)

Problems 1 through 3 will enable you to develop some very useful relationships when we
start digging more deeply into continuum mechanics. Even though most graduate students in
engineering either think these exercises are too elementary or wonder what this has to do with
the subject, you need to use them to familiarize yourself with the notation and to be able to
exploit the three sets of relationships when they become important in later derivations.
Index notation is different from matrix algebra. For example, consider the matrix
equation:

u1 A11 A12 A13 v1


u = A A22 A23 v2 , (1-29.)
2 21
u3 A31 A32 A33 v3
which can be written:
u = A v . (1-30.)

In this course, by convention, we will employ bold-face capital letter symbols to designate a three
by three matrix array. In index notation, Equation (1-30.) is represented by the relation:

ui e i = Aij v j e i , (1-31.)
Or, representing the i th component
ui = Aij v j = v j Aij ,

in index notation. This last expression demonstrates the index notation property where
intechanging the positions of the vector components for v, and the three by three matrix
components for A, does not affect the meaning of the equation, whereas that property does not
exist using matrix notation. That is,
u v A.
When a particular set of symbols and associated operational definitions can be shown to preserve
the meaning of an equation even when the positions of the symbols are interchanged, that
operational set is considered to be commutative. Obviously, matrix algebra multiplication
operations are not commutative, whereas index notation multiplication operations are
cummutative because the index subscript locations, rather than the function symbol locations,
control the multiplications.
In order to interchange the position of v and A using matrix notation, we must utilize
the transpose of matrix, A T , where we have already noted that the transpose interchanges the
rows and columns. Using index notation to represent the matrix notation components, we can
represent Eq. (1-24.) as:
AijT Aji , (1-32.)
which can be read as: Element (row) i, (column) j, of the transpose matrix, AT, of the original
matrix, A, is defined as the element corresponding to (row) j, (column) i of the original matrix, A.
The appropriate modified matrix algebra equation would be:

uT = vT AT . (1-33.)

You should realize that we had a row equation representing Eq. (1-31.), but our new matrix
equation represents the vectors as columns. Both matrix notation and index notation are
commutative with respect to additionthe order of addition doesnt matter.

1.3.3 Index Representations of Differential Operators

Differential operations are represented quite easily in index notation. For example, the
gradient and Laplacian operators are represented:

2
= ei and 2 = . (1-34.)
xi xi xi

The divergence of a vector is represented:


v
v = k . (1-35.)
xk

The curl of a vector is:


v v
v = eijk k ei = eijk i e k . (1-36.)
x j x j

In order to make sure that you understand the subtleties of these operations, it is necessary for
you to demonstrate your knowledge. The problem (identity, actually) that follows is extremely
important in many areas of mechanicsparticularly in fluid mechanics.

Problem 1-4 Show that


1
( v ) v = ( v v) v ( v) , (1-37.)
2
using index notation.

If you have taken vector calculus, you should realize that Eq. (1-37.) is a special case of the
gradient of a vector product rule, i.e.:

(a b) = a ( b) b ( a) (a )b (b )a, (1-38.)

for the special case when a = b = v . Thus, if you can prove Eq. (1-38.) using index notation, then
you have proved Eq. (1-37.).

SAMPLE USE OF INDEX MANIPULATIONS


Many of you will need to calculate inertial velocities and accelerations in rotating, non-
inertial coordinates. You already know that the inertial velocity, at some position, r, in a rotating
coordinate system is given by:
dr
r
v dt r r r r ,
r r


where is the vector representing the rate of rotation of the non-inertial coordinate system
r
with respect to an inertial reference coordinate system, r r r , and defines a unit vector
r
r
in the direction of r. If our original position vector has magnitude, r, and direction
, then the
r
above equation suggests that we can define the inertial time derivative (in our non-inertial
coordinates), using index notation as:

dri
dt ri eijk j rk .
inertial

If you dont already know this, you need to realize that the vector components in a Cartesian
coordinate system can be either inertial components or non-inertial components, depending on
whether the origin of the coordinate system is rigid and fixed or moving (translating and rotating).
This time derivative gets more interesting if we compute the acceleration, recognizing that non-
inertial derivatives with respect to time can be represented as the change in the scalar magnitude
of the vector with respect to time, plus the cross product of the angular rate of rotation of the
non-inertial coordinate system with the vector on which we are differentiating with respect to
time. Hence if we want to know the time rate of change of the inertial velocity vector, it should
be written:
dr
dr r
dt dt r r r r .
inertial r r
This equation is a bit cumbersome. As we will learn later in this course, we dont need to restrict
our base unit vectors to Cartesian coordinate components. Even though we could define a non-
inertial x-, y-, z-coordinate system to proceed with this example, we will define a unit base vector
in the radial-direction to be:
r
er so that r re r ,
r
dr
and dt re r r .
inertial
Then, we see that

d 2 ri
2
ri eijk j rk j rk eimn m rn enjk j rk . (S-1.)
dt inertial

The vector form of this acceleration equation is written in most dynamics texts as:


d 2r r d
2 r r r 2 r . (S-2.)
dt inertial r dt

We will use this as an opportunity to exercise some of our index notation properties. Since we
can change the symbols for dummy indices in Eq. (S-1.) any time we want (as long as the dummy
indices only appear twice in any single term), we will write:

d 2 ri
2
ri eijk j rk j rk eijk j rk ekjm j rm .
dt inertial
Thus,
d 2 ri
2 ri 2eijk j rk eijk j rk eimnenjk m j rk . (S-3.)
dt inertial

Now,
eimn enjk m j rk eimn m enjk j rk r ,

and, by inspection, we see that Eqs. (S-3.) and (S-2.) are equivalent. However, if we are using
these acceleration expressions in some sort of computer code, it is not practical to program the

symbolic operation r . It is useful to simplify the unmodified index notation form of the
last term in Eq. (S-3.). That is, we start by rearranging the product of the two alternators, then
use the result from Homework Problem 1-3, to write:
eimn enjk einmenjk enimenjk ij mk ik mj .
Hence,

eimn enjk m j rk ij mk ik mj m j rk i k rk m m ri .
Leaving:
d 2 ri
2 ri 2eijk j rk eijk j rk i k rk m m ri . (S-4.)
dt inertial

This expression (for each component of acceleration) is easier to program than the vector
equation (S-2.). .

1.4 Rotational Coordinate Transformations and the


Transformation Matrix
In order to develop the ideas of tensor analysis, we need to investigate simple rotational
coordinate transformations. Recall that a counterclockwise coordinate system rotationsay
about the x3 (or z ) axis has the effect of increasing the magnitude of the new x1 -component
representation of a vector located in the first quadrant and decreasing the new x2 - component.
Referring to Figure 1-1, if we designate the new components of some vector u, in a new, rotated
coordinate system using primes, we see that the new components of vector u are given in the
new ( x1 ' , x2 ' , x3 ' ) coordinate system by:
u1 ' = u1 cos u2 sin
u2 ' = u1 sin u2 cos
It will be convenient for us to define an element of an coordinate rotation matrix, ij , as the
angle between some new coordinate, xi ' , and the old coordinate x j . Then, we have the angle
matrix representation for this particular case of a counterclockwise rotation about the x3-axis,
given by:

( /2 ) /2

ij

= ( /2 ) /2 (1-39.)
/2 /2 0

for the angles between the new and old coordinates. As you can see, nothing has happened to
the original x3-axis, and that means that the angles between the new x1- and, x2-coordinates and
the old x3-axis, remains 90o (/2 radians) and the angle between the old and new x3-coordinate
is zero degrees. Furthermore, if the original components of u are ui , the new components are
given as:
u1 ' = u1 cos 11 u 2 cos 12 u3 cos 13 , etc.
Note here that vector u is vector u , regardless of the coordinate system (a vector has the same
length and direction in any coordinate system, regardless or our choice of coordinate
orientation). We are only keeping track of the components of u in a different coordinate
orientation.

Figure 1-1: Representation of vector u in x1 , x2 and x1 ' , x2 ' coordinates

Now, we can define a transformation matrix, Qij , ( Q Qij ), by:

Qij cos ij , (1-40.)

where the row-location (subscript i) is the new coordinate direction (1, 2, or 3) and the column-
location (subscript j) represents the old coordinate direction (1, 2, or 3), and we have

u i ' = Qij u j . (1-41.)

Similarly, we can transform the new components of ui ' back into the old coordinate
representation, where:

u1 = u1 ' cos 11 u 2 ' cos 21 u3 ' cos 31 , etc.

or, we can utilize the definition for the transpose of a matrix to write:

ui = Q jiu j ' = QijT u j '. (1-42.)

You may wish to verify the fact that the transformation matrix for the example in Figure 1-1 is:
cos sin 0

Qij = sin cos 0. (1-43.)
0 0 1

Aside: We have just developed one of three successive rotations commonly used in Flight
Mechanics, where the angles of the rotations are known as Euler Angles. If we were to continue
on and develop another right-handed rotation about the x2 axis and then develop the final right-
handed-rotation about the x1 we would have the complete transformation matrix from an Earth
reference coordinate system to the aircraft body-fixed coordinate system. The order of the
rotations is typically ``yaw'', followed by ``pitch', followed by' ``roll''. Thus, the transformation
from the reference to fixed body coordinate system is

1 0 0 cos 0 sin cos sin 0



ubody
= 0 cos sin 0 1
0 sin cos 0 ureference . (1-44.)
0 sin cos sin
0 cos 0 0 1
As you should already know, the product of a three by three matrix with another three by three
matrix yields a single three by three matrix. Therefore, you can see that the product of three,
three by three matrices reduces to a product of two three by three matrices and is thus a single
three by three matrix. If we let Q correspond to a counterclockwise rotation about the x3 (z)
axis, Q correspond to a counterclockwise rotation about the x2 (y) axis, and Q correspond to a
counterclockwise rotation about the x1 (x) axis, we see that the overall coordinate transformation
can be written using index notation as:
Qij Qik Qkm Qmj Qmj Qik Qkm QT ki Qkm Qmj ... ,
in any order, because the positions of the index subscripts control the multiplication (summing
over the dummy indices). However, the matrix-notation representation requires that the order of
the transformations be fixed.

By transforming u from its old coordinate components to the new coordinates, and then
back to the old coordinates, we find that if:

u j ' = Q jmu m

and

ui = Q jiu j ' = Q jiQ jmu m , (1-45.)

then, we can utilize the ``change of index trick'' to write:

ui = um im = Q jiQ jmum ,
or, moving the last two equality expressions to the same side of the equation, and realizing that
Q ji can be written as QijT , we can write:

im
QijT Q jm um = 0. (1-46.)

Now, in general, none of the components, u m , are required to equal zero, and since this equation
must always be true, it can only be true in all possible cases if:

im = QijT Q jm = Q ji Q jm . (1-47.)

This transformation matrix relationship, Eq. (1-47.) is considered to be an identity.

Problem 1-5 Prove that


ij = QimQmj
T
= QimQ jm . (1-48.)

Since we have shown that the transformation matrix obeys transformation rules given by
Eqs. (1-47.) and (1-48.), we can exploit the definition of the inverse of a matrix, A-1, where:

1 0 0
AA 1
= A A = 0 1 0 1 .
1
(1-49.)
0 0 1
Or, in index notation:

Aij Ajk1 = Aij1 A jk = ik , (1-50.)

thus, allowing us to conclude from Eq. (1-47.) that

Qij1 = QijT . (1-51.)


In addition, because det QijQTjk = 1 , the transformation matrix is a called a proper orthogonal
matrix. If left-hand coordinates were allowed, you would find that the right- to left-hand
transformation would yield a negative determinant, violating the proper orthogonal
transformation requirement. We will only consider ``improper'' orthogonal transformations
when we investigate material symmetries later in this course.

1.4.1 Second Rank Tensors

Suppose we define some vector v as the most general linear function of the components
of some input vector u . Then, each component, vi , could be represented as a linear function of
all of the components of u , i.e.:

vi = Aij u j bi . (1-52.)

For the continuum mechanics problems of greatest interest to us, we consider vector, b, to be
some sort of offset vector. A constant vector offset can be problematic since we arent that
interested in systems where an output occurs even when the actual input vector, u, is a null
vector (zero). Hence, we will require that b equal zero, and we are interested in the structure of
a coefficient matrix, A, that employs the input vector components to produce an appropriate
output vector. We require that, when operating on some input vector, u, with the three by three
coeffient matrix A, we will get precisely the same output vector, v, regardless of the orientation
of an arbitrarily-specified inertial, Cartesian coordinate system. We already know how vi ' and
u j ' , transform with respect to arbitrary coordinate rotations, so that (with bi = 0 ) the vector
relationship between u and v in some new (transformed) coordinate system can be represented:

v'i Qim v m = Aij ' u ' j Aij ' Q jn u n , (1-53.)

and, from Eq. (1-52.), we will change the name of our live index and write vm = Amr ur , allowing
us to substitute that relationship for the components of v into Eq. (1-53.) and write:

Qim Amr u r = Aij ' Q jnu n . (1-54.)

We can multiply both sides of this equation by Qik , so that:

Qik Qim Amr u r = Qik Q jn Aij ' u n . (1-55.)

From Eq. (1-47), we know that the product of the two transformation matrices on the left side of
the equation is a Kronecker delta, and we can write6:

km Amr u r = Qik Q jr Aij ' u r .

Then, we can contract the resulting dummy index to write:

Akr u r = Qik Q jr Aij ' u r .

Finally, since we have the same vector component (ur) on both sides of this equation, we can
move everything to the left side, factoring out ur, and write:

6
You will note in the equation that follows that dummy index n has been replaced by dummy index r. This was
done for convenience and to demonstrate that you can change the symbols used for your dummy indices any time
you feel like itas long as the subscript positions of the dummy index pairs are not altered.
A kr
Qik Q jr Aij ' u r = 0.

Again, recognizing that the ur -component need not be zero, it is necessary to require that this
equation hold for all possible cases. That is only possible if:

Akr = Qik Q jr Aij ' , (1-56.)

in order for our three by three matrix, A, to preserve the specific vector relationship between
input vector, u , and output vector, v , during coordinate transformations from some ``new''
coordinate system back to ``old'' coordinates.

Problem 1-6 Prove that


Aij ' = QimQ jn Amn . (1-57.)

After this homework problem is completed, you will have shown that when three by three matrix,
A, satisfies the transformation relation

Aij ' = QimQ jn Amn


and Aij = Qmi Qnj Amn ' ,
this is the definition of a SECOND RANK/SECOND ORDER TENSOR. (1-58.)

This transformation pair represents specific and rigid restrictions on any type of three-by-three
matrix relationship between input and output vectors, preserving the relationship between
the two vectors, regardless of the coordinate orientation. We define the matrices that obey
these transformation relations as second rank tensors or second order tensors. By analogy, it
should be obvious that vectors are first order tensors, obeying the transformation relations:

ui ' = Qij u j and ui = Q jiu j ' , FIRST ORDER TENSOR

and second rank tensors obey relations given by Eq. (1-58.); we can generalize to write that
tensors of rank N must obey transformation rules with N transformation matrices being
employed, i.e. an Nth rank/order tensor must obey the transformation rules:

Ai... j ' = Qim ...Q jn Am... n ,


and Ai... j = Qmi ...Qnj Am... n ' ,

where there are N transformation matrices in front of the 3 by N tensor matrices.


Problem 1-7 Employing Q to designate a transformation matrix, write out the definition of a
second rank tensor using matrix algebra notation.

A symmetric tensor is a tensor whose elements obey the equation:

Aij = A ji (1-59.)

An antisymmetric or skew symmetric tensor is defined by

Aij = A ji (1-60.)

Obviously, the diagonal elements of an anti symmetric tensor must be zero. In general, any
second rank tensorsay Tcan be decomposed into symmetric and anti-symmetric parts where

Tij = S ij Aij , (1-61.)


with

1
S ij = (Tij T ji ), (1-62.)
2
and

1
Aij = (Tij T ji ). (1-63.)
2

We also define the tensor product, AB by:

A B = [ Aij B jk ]. (1-64.)
You should keep in mind that if the elements of A B are calculated using the index notation
operation, i.e. Aij B jk B jk Aij Bkj Aij Aij Bkj A ji Bkj ... , the positions of the matrix
T T T T

elements in the index products do not affect the result if the elements represented by the live
indices (i and k) in the two matrix arrays are preserved. Obviously, that isnt true using regular
matrix operations.

Problem 1-8 If A and B are both second rank tensors, prove that A B is also a second rank
tensor.
Problem 1-9 If Aij are the elements of a second rank tensor, prove that Aij ij is also a
second rank tensor, for any arbitrary constant, .

1.4.2 Invariance
Another important idea is the idea that some quantity is invariant with respect to
coordinate transformations. We have defined a vector as a quantity whose direction and
magnitude in an inertial coordinate frame are not affected by coordinate transformations. We
can say that the length of a vector is invariant with respect to coordinate rotations, even though
the components of the vector are not invariant. Another example is the vector dot product u v ,
which we can be prove is invariant with respect to coordinate rotations as follows:
u v = uk vk ,
which we can represent in some new (primed) coordinate system (equating it to the original
represenation using the transformation matrices) as:

uk ' vk ' = QkmumQkn vn .

But, we know from Eq. (1-47.) that:


QkmQkn = mn .
Therefore,

uk ' vk ' = um mnvn um vm , (1-65.)

demonstrating that the scalar quantity produced by a vector dot product is an invariant with
respect to coordinate orientation.
The trace of a tensor is defined as the sum of its diagonal elements. For example if A is
a second rank tensor, then
tr A Akk . (1-66.)


Problem 1-10 Show that trA, tr A 2 and tr A n are invariant with respect to coordinate
transformations. (Use index notation)

Problem 1-11 Use Eqs. (1-16.), (1-18.), (1-56.) and (1-57.) to show that the determinant of a

second rank tensor is invariant. Explain why det Aijn is invariant.

After you have completed these problems, you have shown that
u v


tr A n ,
and


det A n ,
for n = 1,2, , are invariants with respect to coordinate transformations.

1.4.3 Anisotropic Heat Conduction Example

Fourier's Law of Heat Conduction relates the heat flux vector7say qto the
temperature gradient, according to:

q = q1e1 q2e 2 q3e3 = qi ei ,


where
T
qi = k (1-67.)
xi

However, wood, composites and crystals all exhibit directionally-dependent heat transfer,
leading to the Generalized Fourier's Law:

T
qi = Kij , (1-68.)
x j

where you should realize that K is a second rank tensor since it must preserve the vector
relationship between the output, heat flux vector, q , and the input, temperature gradient vector,

T .

Now, it is often possible to model laminated materials like plywood as having two thermal
conductivitiesa parallel-ply value, K P , and a perpendicular (to the plies) value K . If the
plywood is used in a usual manner and coordinate direction, x3 , is perpendicular to the plies, the
thermal conductivity in the x- and y-directions would be given by the parallel-ply value and the
thermal conductivity in the z-direction would be the perpendicular value, so that the tensor for
that plywood could be represented as:

KP 0 0
K = 0 KP 0 . (1-69.)
0 0 K

Problem 1-12 (a) Using matrix methods, show that this thermal conductivity tensor for plywood,
i.e. Eq. (1-69.) is unaltered by a coordinate rotation of 3 about the x3 -axis. (b) Find the thermal
conductivity tensor for a rotation of the coordinate system by 1 , counterclockwise about the x1

7
Heat flux is a measure of the energy flow rate per unit area (e.g. W/m2). It is sometimes called heat flux density or
heat flow rate intensity. Regardless of its name, it is a vector pointing in the direction along which energy must flow
in order to be consistent with the Second Law of Thermodynamics (energy can only flow from a hot reservoir to a
cold reservoir).
-axis.

Problem 1-13 Using the untransformed thermal conductivity tensor for ``plywood,'' suppose that
KP = 2K . If the temperature gradient is e1 e 2 e 3 (degrees/m), (a) prove that the
temperature gradient vector and the heat flux vector are not collinear. (b) List the most general
temperature gradient cases where the heat flux and temperature gradient will be along the same
line (collinear).

1.5 Additional Problems


Problem 1-14 Represent Gauss'divergence theorem for a vector v in index notation.
Problem 1-15 Using index notation, prove that
(a) a b = b a ,
(b) (a u) (b v) = (a b)(u v) (a v)(b u)


(c) a b b a b a a b a b

(d) ( v) = ( v) 2 v
Problem 1-16 Represent the following vector operations in index notation and indicate their

tensor rank (a scalar is a tensor of rank 0). (a) 2 (b) 2 v (c) (d) v (e) v .
Problem 1-17 Explain why successive coordinate transformations are not commutative. You
may want to use a toy block or some other rectangular object with different markings on each
face to demonstrate this fact to yourself, i.e. rotate the block 90 about one axis, then another,
then return to the original orientation and reverse the order. Why would successive infinitesimal
transformations be commutative?
Problem 1-18 Some authors employ index notation that utilizes a comma to designate a partial
derivative. That is,

, i . (1-70.)
xi

Repeat problem 1-16, using this notation.


Problem 1-19 If A is a second rank tensor, prove that A2 and An are second rank tensors where

a tensor product is defined by: A B Aik Bkj A 2 Aik Akj , etc.
A11 A12 0
Problem 1-20 If A A12 A22 0 , and the transformation matrix for counter clockwise
0 0 A33
cos sin 0 C S 0
rotation about the x3-axis is Q = sin cos 0 S C 0 , show that the new

0 0 1 0 0 1
components of second rank tensor A in the rotated coordinate system are:
A11C 2 A22 S 2 2 A12 SC A12 C 2 S 2 A22 A11 CS 0
A' A12 C 2 S 2 A22 A11 CS A11 S 2 A22C 2 2 A12 SC

0 .
0 0 A33

También podría gustarte