Tensor Calculus
and
Continuum Mechanics
by J.H. Heinbockel
Department of Mathematics and Statistics
Old Dominion University
PREFACE
This is an introductory text which presents fundamental concepts from the subject
areas of tensor calculus, differential geometry and continuum mechanics. The material
presented is suitable for a two semester course in applied mathematics and is flexible
enough to be presented to either upper level undergraduate or beginning graduate students
majoring in applied mathematics, engineering or physics. The presentation assumes the
students have some knowledge from the areas of matrix theory, linear algebra and advanced
calculus. Each section includes many illustrative worked examples. At the end of each
section there is a large collection of exercises which range in difficulty. Many new ideas
are presented in the exercises and so the students should be encouraged to read all the
exercises.
The purpose of preparing these notes is to condense into an introductory text the basic
definitions and techniques arising in tensor calculus, differential geometry and continuum
mechanics. In particular, the material is presented to (i) develop a physical understanding
of the mathematical concepts associated with tensor calculus and (ii) develop the basic
equations of tensor calculus, differential geometry and continuum mechanics which arise
in engineering applications. From these basic equations one can go on to develop more
sophisticated models of applied mathematics. The material is presented in an informal
manner and uses mathematics which minimizes excessive formalism.
The material has been divided into two parts. The first part deals with an introduc
tion to tensor calculus and differential geometry which covers such things as the indicial
notation, tensor algebra, covariant differentiation, dual tensors, bilinear and multilinear
forms, special tensors, the Riemann Christoffel tensor, space curves, surface curves, cur
vature and fundamental quadratic forms. The second part emphasizes the application of
tensor algebra and calculus to a wide variety of applied areas from engineering and physics.
The selected applications are from the areas of dynamics, elasticity, fluids and electromag
netic theory. The continuum mechanics portion focuses on an introduction of the basic
concepts from linear elasticity and fluids. The Appendix A contains units of measurements
from the Système International d’Unitès along with some selected physical constants. The
Appendix B contains a listing of Christoffel symbols of the second kind associated with
various coordinate systems. The Appendix C is a summary of useful vector identities.
A scalar field describes a onetoone correspondence between a single scalar number and a point. An n
dimensional vector field is described by a onetoone correspondence between nnumbers and a point. Let us
generalize these concepts by assigning nsquared numbers to a single point or ncubed numbers to a single
point. When these numbers obey certain transformation laws they become examples of tensor fields. In
general, scalar fields are referred to as tensor fields of rank or order zero whereas vector fields are called
tensor fields of rank or order one.
Closely associated with tensor calculus is the indicial or index notation. In section 1 the indicial
notation is defined and illustrated. We also define and investigate scalar, vector and tensor fields when they
are subjected to various coordinate transformations. It turns out that tensors have certain properties which
are independent of the coordinate system used to describe the tensor. Because of these useful properties,
we can use tensors to represent various fundamental laws occurring in physics, engineering, science and
mathematics. These representations are extremely useful as they are independent of the coordinate systems
considered.
~ = A1 b
A e1 + A2 b
e2 + A3 b
e3 and ~ = B1 b
B e1 + B2 b
e2 + B3 b
e3 ,
where be1 , b
e2 and b ~ and
e3 are orthogonal unit basis vectors. Often when no confusion arises, the vectors A
~ are expressed for brevity sake as number triples. For example, we can write
B
~ = (A1 , A2 , A3 )
A and ~ = (B1 , B2 , B3 )
B
~ and B
where it is understood that only the components of the vectors A ~ are given. The unit vectors would
be represented
b
e1 = (1, 0, 0), b
e2 = (0, 1, 0), b
e3 = (0, 0, 1).
~ and B
A still shorter notation, depicting the vectors A ~ is the index or indicial notation. In the index notation,
the quantities
Ai , i = 1, 2, 3 and Bp , p = 1, 2, 3
~ and B.
represent the components of the vectors A ~ This notation focuses attention only on the components of
the vectors and employs a dummy subscript whose range over the integers is specified. The symbol Ai refers
~ simultaneously. The dummy subscript i can have any of the integer
to all of the components of the vector A
~ Setting i = 2 focuses
values 1, 2 or 3. For i = 1 we focus attention on the A1 component of the vector A.
attention on the second component A2 of the vector A ~ and similarly when i = 3 we can focus attention on
the third component of A.~ The subscript i is a dummy subscript and may be replaced by another letter, say
p, so long as one specifies the integer values that this dummy subscript can have.
2
It is also convenient at this time to mention that higher dimensional vectors may be defined as ordered
n−tuples. For example, the vector
~ = (X1 , X2 , . . . , XN )
X
where
b
e1 , b
e2 , . . . , b
eN
are linearly independent unit base vectors. Note that many of the operations that occur in the use of the
index notation apply not only for three dimensional vectors, but also for N −dimensional vectors.
In future sections it is necessary to define quantities which can be represented by a letter with subscripts
or superscripts attached. Such quantities are referred to as systems. When these quantities obey certain
transformation laws they are referred to as tensor systems. For example, quantities like
The subscripts or superscripts are referred to as indices or suffixes. When such quantities arise, the indices
must conform to the following rules:
1. They are lower case Latin or Greek letters.
2. The letters at the end of the alphabet (u, v, w, x, y, z) are never employed as indices.
The number of subscripts and superscripts determines the order of the system. A system with one index
is a first order system. A system with two indices is called a second order system. In general, a system with
N indices is called a N th order system. A system with no indices is called a scalar or zeroth order system.
The type of system depends upon the number of subscripts or superscripts occurring in an expression.
For example, Aijk and Bst
m
, (all indices range 1 to N), are of the same type because they have the same
number of subscripts and superscripts. In contrast, the systems Aijk and Cpmn are not of the same type
because one system has two superscripts and the other system has only one superscript. For certain systems
the number of subscripts and superscripts is important. In other systems it is not of importance. The
meaning and importance attached to sub and superscripts will be addressed later in this section.
In the use of superscripts one must not confuse “powers ”of a quantity with the superscripts. For
example, if we replace the independent variables (x, y, z) by the symbols (x1 , x2 , x3 ), then we are letting
y = x2 where x2 is a variable and not x raised to a power. Similarly, the substitution z = x3 is the
replacement of z by the variable x3 and this should not be confused with x raised to a power. In order to
write a superscript quantity to a power, use parentheses. For example, (x2 )3 is the variable x2 cubed. One
of the reasons for introducing the superscript variables is that many equations of mathematics and physics
can be made to take on a concise and compact form.
There is a range convention associated with the indices. This convention states that whenever there
is an expression where the indices occur unrepeated it is to be understood that each of the subscripts or
superscripts can take on any of the integer values 1, 2, . . . , N where N is a specified integer. For example,
3
the Kronecker delta symbol δij , defined by δij = 1 if i = j and δij = 0 for i 6= j, with i, j ranging over the
values 1,2,3, represents the 9 quantities
The symbol δij refers to all of the components of the system simultaneously. As another example, consider
the equation
b
em · b
en = δmn m, n = 1, 2, 3 (1.1.1)
the subscripts m, n occur unrepeated on the left side of the equation and hence must also occur on the right
hand side of the equation. These indices are called “free ”indices and can take on any of the values 1, 2 or 3
as specified by the range. Since there are three choices for the value for m and three choices for a value of
n we find that equation (1.1.1) represents nine equations simultaneously. These nine equations are
e1 · b
b e1 = 1 b
e1 · b
e2 = 0 b
e1 · b
e3 = 0
e2 · b
b e1 = 0 e2 · b
b e2 = 1 e2 · b
b e3 = 0
e3 · b
b e1 = 0 e3 · b
b e2 = 0 e3 · b
b e3 = 1.
A system defined by subscripts and superscripts ranging over a set of values is said to be symmetric
in two of its indices if the components are unchanged when the indices are interchanged. For example, the
third order system Tijk is symmetric in the indices i and k if
A system defined by subscripts and superscripts is said to be skewsymmetric in two of its indices if the
components change sign when the indices are interchanged. For example, the fourth order system Tijkl is
skewsymmetric in the indices i and l if
As another example, consider the third order system aprs , p, r, s = 1, 2, 3 which is completely skew
symmetric in all of its indices. We would then have
It is left as an exercise to show this completely skew symmetric systems has 27 elements, 21 of which are
zero. The 6 nonzero elements are all related to one another thru the above equations when (p, r, s) = (1, 2, 3).
This is expressed as saying that the above system has only one independent component.
4
Summation Convention
The summation convention states that whenever there arises an expression where there is an index which
occurs twice on the same side of any equation, or term within an equation, it is understood to represent a
summation on these repeated indices. The summation being over the integer values specified by the range. A
repeated index is called a summation index, while an unrepeated index is called a free index. The summation
convention requires that one must never allow a summation index to appear more than twice in any given
expression. Because of this rule it is sometimes necessary to replace one dummy summation symbol by
some other dummy symbol in order to avoid having three or more indices occurring on the same side of
the equation. The index notation is a very powerful notation and can be used to concisely represent many
complex equations. For the remainder of this section there is presented additional definitions and examples
to illustrated the power of the indicial notation. This notation is then employed to define tensor components
and associated operations with tensors.
y1 = a11 x1 + a12 x2
y2 = a21 x1 + a22 x2
can be represented as one equation by introducing a dummy index, say k, and expressing the above equations
as
yk = ak1 x1 + ak2 x2 , k = 1, 2.
The range convention states that k is free to have any one of the values 1 or 2, (k is a free index). This
equation can now be written in the form
2
X
yk = aki xi = ak1 x1 + ak2 x2
i=1
where i is the dummy summation index. When the summation sign is removed and the summation convention
is adopted we have
yk = aki xi i, k = 1, 2.
Since the subscript i repeats itself, the summation convention requires that a summation be performed by
letting the summation subscript take on the values specified by the range and then summing the results.
The index k which appears only once on the left and only once on the right hand side of the equation is
called a free index. It should be noted that both k and i are dummy subscripts and can be replaced by other
letters. For example, we can write
yn = anm xm n, m = 1, 2
where m is the summation index and n is the free index. Summing on m produces
yn = an1 x1 + an2 x2
and letting the free index n take on the values of 1 and 2 we produce the original two equations.
5
EXAMPLE 1.12. For yi = aij xj , i, j = 1, 2, 3 and xi = bij zj , i, j = 1, 2, 3 solve for the y variables in
terms of the z variables.
Solution: In matrix form the given equations can be expressed:
y1 a11 a12 a13 x1 x1 b11 b12 b13 z1
y2 = a21 a22 a23 x2 and x2 = b21 b22 b23 z2 .
y3 a31 a32 a33 x3 x3 b31 b32 b33 z3
Now solve for the y variables in terms of the z variables and obtain
y1 a11 a12 a13 b11 b12 b13 z1
y2 = a21 a22 a23 b21 b22 b23 z2 .
y3 a31 a32 a33 b31 b32 b33 z3
The index notation employs indices that are dummy indices and so we can write
Here we have purposely changed the indices so that when we substitute for xm , from one equation into the
other, a summation index does not repeat itself more than twice. Substituting we find the indicial form of
the above matrix equation as
yn = anm bmj zj , m, n, j = 1, 2, 3
where n is the free index and m, j are the dummy summation indices. It is left as an exercise to expand
both the matrix equation and the indicial equation and verify that they are different ways of representing
the same thing.
EXAMPLE 1.13. The dot product of two vectors Aq , q = 1, 2, 3 and Bj , j = 1, 2, 3 can be represented
with the index notation by the product Ai Bi = AB cos θ i = 1, 2, 3, A = A, ~ ~ Since the
B = B.
subscript i is repeated it is understood to represent a summation index. Summing on i over the range
specified, there results
A1 B1 + A2 B2 + A3 B3 = AB cos θ.
Observe that the index notation employs dummy indices. At times these indices are altered in order to
conform to the above summation rules, without attention being brought to the change. As in this example,
the indices q and j are dummy indices and can be changed to other letters if one desires. Also, in the future,
if the range of the indices is not stated it is assumed that the range is over the integer values 1, 2 and 3.
To systems containing subscripts and superscripts one can apply certain algebraic operations. We
present in an informal way the operations of addition, multiplication and contraction.
6
Here the symbol C mnl is used to represent the third order system that results when the contraction is
performed. Whenever a contraction is performed, the resulting system is always of order 2 less than the
original system. Under certain special conditions it is permissible to perform a contraction on two lower case
indices. These special conditions will be considered later in the section.
The above operations will be more formally defined after we have explained what tensors are.
EXAMPLE 1.14. The total number of possible ways of arranging the digits 1 2 3 is six. We have
three choices for the first digit. Having chosen the first digit, there are only two choices left for the second
digit. Hence the remaining number is for the last digit. The product (3)(2)(1) = 3! = 6 is the number of
permutations of the digits 1, 2 and 3. These six permutations are
1 2 3 even permutation
1 3 2 odd permutation
3 1 2 even permutation
3 2 1 odd permutation
2 3 1 even permutation
2 1 3 odd permutation.
Here a permutation of 1 2 3 is called even or odd depending upon whether there is an even or odd number
of transpositions of the digits. A mnemonic device to remember the even and odd permutations of 123
is illustrated in the figure 1.11. Note that even permutations of 123 are obtained by selecting any three
consecutive numbers from the sequence 123123 and the odd permutations result by selecting any three
consecutive numbers from the sequence 321321.
In general, the number of permutations of n things taken m at a time is given by the relation
By selecting a subset of m objects from a collection of n objects, m ≤ n, without regard to the ordering is
called a combination of n objects taken m at a time. For example, combinations of 3 numbers taken from
the set {1, 2, 3, 4} are (123), (124), (134), (234). Note that ordering of a combination is not considered. That
is, the permutations (123), (132), (231), (213), (312), (321) are considered equal. In general, the number of
n n!
n
combinations of n objects taken m at a time is given by C(n, m) = = where m are the
m m!(n − m)!
binomial coefficients which occur in the expansion
n
X n n−m m
(a + b)n = a b .
m=0
m
8
In figure 1.12, there are seven intersections of the lines connecting like numbers. The number of
intersections is an odd number and shows that an odd number of transpositions must be performed. These
results imply e612453 = −1.
Another definition used quite frequently in the representation of mathematical and engineering quantities
is the Kronecker delta which we now define in terms of both subscripts and superscripts.
EXAMPLE 1.16. Some examples of the e−permutation symbol and Kronecker delta are:
EXAMPLE 1.17. When an index of the Kronecker delta δij is involved in the summation convention,
the effect is that of replacing one index with a different index. For example, let aij denote the elements of an
N × N matrix. Here i and j are allowed to range over the integer values 1, 2, . . . , N. Consider the product
aij δik
where the range of i, j, k is 1, 2, . . . , N. The index i is repeated and therefore it is understood to represent
a summation over the range. The index i is called a summation index. The other indices j and k are free
indices. They are free to be assigned any values from the range of the indices. They are not involved in any
summations and their values, whatever you choose to assign them, are fixed. Let us assign a value of j and
k to the values of j and k. The underscore is to remind you that these values for j and k are fixed and not
to be summed. When we perform the summation over the summation index i we assign values to i from the
range and then sum over these values. Performing the indicated summation we obtain
In this summation the Kronecker delta is zero everywhere the subscripts are different and equals one where
the subscripts are the same. There is only one term in this summation which is nonzero. It is that term
where the summation index i was equal to the fixed value k This gives the result
where the underscore is to remind you that the quantities have fixed values and are not to be summed.
Dropping the underscores we write
aij δik = akj
Here we have substituted the index i by k and so when the Kronecker delta is used in a summation process
it is known as a substitution operator. This substitution property of the Kronecker delta can be used to
simplify a variety of expressions involving the index notation. Some examples are:
Some texts adopt the notation that if indices are capital letters, then no summation is to be performed.
For example,
aKJ δKK = aKJ
10
as δKK represents a single term because of the capital letters. Another notation which is used to denote no
summation of the indices is to put parenthesis about the indices which are not to be summed. For example,
since δ(k)(k) represents a single term and the parentheses indicate that no summation is to be performed.
At any time we may employ either the underscore notation, the capital letter notation or the parenthesis
notation to denote that no summation of the indices is to be performed. To avoid confusion altogether, one
can write out parenthetical expressions such as “(no summation on k)”.
EXAMPLE 1.18. In the Kronecker delta symbol δji we set j equal to i and perform a summation. This
operation is called a contraction. There results δii , which is to be summed over the range of the index i.
Utilizing the range 1, 2, . . . , N we have
δii = 1 + 1 + · · · + 1
δii = N.
In certain circumstances the Kronecker delta can be written with only subscripts. For example,
δij , i, j = 1, 2, 3. We shall find that these circumstances allow us to perform a contraction on the lower
indices so that δii = 3.
EXAMPLE 1.19. The determinant of a matrix A = (aij ) can be represented in the indicial notation.
Employing the epermutation symbol the determinant of an N × N matrix is expressed
where the summation is over the range 1,2 and the epermutation symbol is of order 2. In the special case
of a 3 × 3 matrix we have
a11 a12 a13
A = a21 a22 a23 = eijk ai1 aj2 ak3 = eijk a1i a2j a3k
a31 a32 a33
where i, j, k are the summation indices and the summation is over the range 1,2,3. Here eijk denotes the
epermutation symbol of order 3. Note that by interchanging the rows of the 3 × 3 matrix we can obtain
11
more general results. Consider (p, q, r) as some permutation of the integers (1, 2, 3), and observe that the
determinant can be expressed
ap1 ap2 ap3
∆ = aq1 aq2 aq3 = eijk api aqj ark .
ar1 ar2 ar3
If (p, q, r) is an even permutation of (1, 2, 3) then ∆ = A
If (p, q, r) is an odd permutation of (1, 2, 3) then ∆ = −A
If (p, q, r) is not a permutation of (1, 2, 3) then ∆ = 0.
We can then write
eijk api aqj ark = epqr A.
Each of the above results can be verified by performing the indicated summations. A more formal proof of
the above result is given in EXAMPLE 1.125, later in this section.
EXAMPLE 1.110. The expression eijk Bij Ci is meaningless since the index i repeats itself more than
twice and the summation convention does not allow this. If you really did want to sum over an index which
occurs more than twice, then one must use a summation sign. For example the above expression would be
X
n
written eijk Bij Ci .
i=1
EXAMPLE 1.111.
The cross product of the unit vectors e1 , b
b e2 , b
e3 can be represented in the index notation by
b
ek if (i, j, k) is an even permutation of (1, 2, 3)
ei × b
b ej = − b ek if (i, j, k) is an odd permutation of (1, 2, 3)
0 in all other cases
ej = ekij b
ei × b
This result can be written in the form b ek . This later result can be verified by summing on the
index k and writing out all 9 possible combinations for i and j.
EXAMPLE 1.112. Given the vectors Ap , p = 1, 2, 3 and Bp , p = 1, 2, 3 the cross product of these two
vectors is a vector Cp , p = 1, 2, 3 with components
Ci = eijk Aj Bk , i, j, k = 1, 2, 3. (1.1.2)
~ =A
C ~ = C1 b
~×B e1 + C2 b
e2 + C3 b
e3 .
We next sum on the index j which repeats itself in each term of equation (1.1.3). This gives
Now we are left with i being a free index which can have any of the values of 1, 2 or 3. Letting i = 1, then
letting i = 2, and finally letting i = 3 produces the cross product components
C1 = A2 B3 − A3 B2
C2 = A3 B1 − A1 B3
C3 = A1 B2 − A2 B1 .
Solution: The array i k j represents an odd number of transpositions of the indices i j k and to each
transposition there is a sign change of the epermutation symbol. Similarly, j k i is an even transposition
of i j k and so there is no sign change of the epermutation symbol. The above holds regardless of the
numerical values assigned to the indices i, j, k.
An identity relating the epermutation symbol and the Kronecker delta, which is useful in the simpli
fication of tensor expressions, is the eδ identity. This identity can be expressed in different forms. The
subscript form for this identity is
where i is the summation index and j, k, m, n are free indices. A device used to remember the positions of
the subscripts is given in the figure 1.13.
The subscripts on the four Kronecker delta’s on the righthand side of the eδ identity then are read
(first)(second)(outer)(inner).
This refers to the positions following the summation index. Thus, j, m are the first indices after the sum
mation index and k, n are the second indices after the summation index. The indices j, n are outer indices
when compared to the inner indices k, m as the indices are viewed as written on the lefthand side of the
identity.
13
Another form of this identity employs both subscripts and superscripts and has the form
One way of proving this identity is to observe the equation (1.1.5) has the free indices j, k, m, n. Each
of these indices can have any of the values of 1, 2 or 3. There are 3 choices we can assign to each of j, k, m
or n and this gives a total of 34 = 81 possible equations represented by the identity from equation (1.1.5).
By writing out all 81 of these equations we can verify that the identity is true for all possible combinations
that can be assigned to the free indices.
An alternate proof of the e − δ identity is to consider the determinant
1
δ1 δ21 δ31 1 0 0
2
δ δ2 δ32 = 0 1 0 = 1.
13 2
δ1 δ23 δ33 0 0 1
By performing a permutation of the rows of this matrix we can use the permutation symbol and write
i
δ1 δ2i δ3i
j
δ j j ijk
k1 δk2 δk3 = e .
δ1 δ2 δ3
Summing on i we have δii = δ11 + δ22 + δ33 = 3 and expand the determinant to obtain the desired result
Performing a contraction on the indices k and p we obtain the fourth order system
rs
δmn rsp
= δmnp = ersp emnp = eprs epmn = δm δn − δnr δm
r s s
.
As an exercise one can verify that the definition of the epermutation symbol can also be defined in terms
of the generalized Kronecker delta as
ej1 j2 j3 ···jN = δj11 j22 j33 ···j
··· N
N
.
Additional definitions and results employing the generalized Kronecker delta are found in the exercises.
In section 1.3 we shall show that the Kronecker delta and epsilon permutation symbol are numerical tensors
which have fixed components in every coordinate system.
Additional Applications of the Indicial Notation
The indicial notation, together with the e − δ identity, can be used to prove various vector identities.
We desire to show that Di = −Ci for all values of i. Consider the following manipulations: Let Bj = Bs δsj
and Ak = Am δmk and write
Di = eijk Bj Ak = eijk Bs δsj Am δmk (1.1.6)
where all indices have the range 1, 2, 3. In the expression (1.1.6) note that no summation index appears
more than twice because if an index appeared more than twice the summation convention would become
meaningless. By rearranging terms in equation (1.1.6) we have
In this expression the indices s and m are dummy summation indices and can be replaced by any other
letters. We replace s by k and m by j to obtain
~ = −C
Consequently, we find that D ~ or B
~ ×A
~ = −A
~ × B. ~ = Di b
~ That is, D ei = −Ci b ~
ei = −C.
Note 1. The expressions
Ci = eijk Aj Bk and Cm = emnp An Bp
with all indices having the range 1, 2, 3, appear to be different because different letters are used as sub
scripts. It must be remembered that certain indices are summed according to the summation convention
and the other indices are free indices and can take on any values from the assigned range. Thus, after
summation, when numerical values are substituted for the indices involved, none of the dummy letters
used to represent the components appear in the answer.
Note 2. A second important point is that when one is working with expressions involving the index notation,
the indices can be changed directly. For example, in the above expression for Di we could have replaced
j by k and k by j simultaneously (so that no index repeats itself more than twice) to obtain
Note 3. Be careful in switching back and forth between the vector notation and index notation. Observe that a
~ can be represented
vector A
~ = Ai b
A ei
~· b
A ei = Ai , i = 1, 2, 3.
~ = Ai as this is a
Do not set a vector equal to a scalar. That is, do not make the mistake of writing A
misuse of the equal sign. It is not possible for a vector to equal a scalar because they are two entirely
different quantities. A vector has both magnitude and direction while a scalar has only magnitude.
~ · (B
A ~ × C)
~ =B
~ · (C
~ × A)
~
Solution: Let
~ ×C
B ~ = Di b
~ =D ei where Di = eijk Bj Ck and let
~ = F~ = Fi b
~ ×A
C ei where Fi = eijk Cj Ak
where all indices have the range 1, 2, 3. To prove the above identity, we have
~ · (B
A ~ × C)
~ =A
~ ·D
~ = Ai Di = Ai eijk Bj Ck
= Bj (eijk Ai Ck )
= Bj (ejki Ck Ai )
16
Fi = eijk Cj Ak
Fj = ejki Ck Ai .
A physical interpretation that can be assigned to this triple scalar product is that its absolute value represents
the volume of the parallelepiped formed by the three noncoplaner vectors A, ~ B,
~ C.~ The absolute value is
needed because sometimes the triple scalar product is negative. This physical interpretation can be obtained
from an analysis of the figure 1.14.
~ × C
In figure 1.14 observe that: (i) B ~ is the area of the parallelogram P QRS. (ii) the unit vector
~ ×C
B ~
b
en =
~ ~
B × C
~ and C.
is normal to the plane containing the vectors B ~ (iii) The dot product
~
~· B×C =h
~
A en = A
~· b
~
~ × C
B
~ on b
equals the projection of A en which represents the height of the parallelepiped. These results demonstrate
that
~ ~ ~ = B
A · (B × C) ~ × C
~ h = (area of base)(height) = volume.
~ × B)
(A ~ × (C
~ × D)
~ = C(
~ D~ ·A
~ × B)
~ − D(
~ C~ ·A
~ × B)
~
Solution: Let F~ = A ~ = Fi b
~×B ~ =C
ei and E ~ = Ei b
~ ×D ei . These vectors have the components
which is now in a form where we can use the e − δ identity applied to the term in parentheses to produce
Gq = eijk [(Dp δip )(Cn δqn )Aj Bk − (Dp δqp )(Cn δin )Aj Bk ]
= eijk [Di Cq Aj Bk − Dq Ci Aj Bk ]
= Cq [Di eijk Aj Bk ] − Dq [Ci eijk Aj Bk ]
~ D
C( ~ ·A
~ × B)
~ − D(
~ C~ ·A
~ × B).
~
18
Transformation Equations
Consider two sets of N independent variables which are denoted by the barred and unbarred symbols
xi and xi with i = 1, . . . , N. The independent variables xi , i = 1, . . . , N can be thought of as defining
the coordinates of a point in a N −dimensional space. Similarly, the independent barred variables define a
point in some other N −dimensional space. These coordinates are assumed to be real quantities and are not
complex quantities. Further, we assume that these variables are related by a set of transformation equations.
xi = xi (x1 , x2 , . . . , xN ) i = 1, . . . , N. (1.1.7)
It is assumed that these transformation equations are independent. A necessary and sufficient condition that
these transformation equations be independent is that the Jacobian determinant be different from zero, that
is 1
∂x1 ∂x1
··· ∂x1
∂x2 ∂xN
i ∂x 2
∂x2 ∂x2
x ∂x ∂x ···
∂x1 ∂x2 ∂xN
J( ) = j = .
x ∂ x̄ .. .. .. 6= 0.
.. . . .
N
∂x 1 ∂xN
··· ∂xN
∂x ∂x2 ∂x N
xi = xi (x1 , x2 , . . . , xN ) i = 1, . . . , N, (1.1.8)
where the x0 s are determined in terms of the x0 s. Throughout our discussions it is to be understood that the
given transformation equations are real and continuous. Further all derivatives that appear in our discussions
are assumed to exist and be continuous in the domain of the variables considered.
EXAMPLE 1.117. The following is an example of a set of transformation equations of the form
defined by equations (1.1.7) and (1.1.8) in the case N = 3. Consider the transformation from cylindrical
coordinates (r, α, z) to spherical coordinates (ρ, β, α). From the geometry of the figure 1.15 we can find the
transformation equations
r = ρ sin β
α=α 0 < α < 2π
z = ρ cos β 0<β<π
with inverse transformation p
ρ= r2 + z 2
α=α
r
β = arctan( )
z
Now make the substitutions
The resulting transformations then have the forms of the equations (1.1.7) and (1.1.8).
Calculation of Derivatives
We now consider the chain rule applied to the differentiation of a function of the bar variables. We
represent this differentiation in the indicial notation. Let Φ = Φ(x1 , x2 , . . . , xn ) be a scalar function of the
variables xi , i = 1, . . . , N and let these variables be related to the set of variables xi , with i = 1, . . . , N by
the transformation equations (1.1.7) and (1.1.8). The partial derivatives of Φ with respect to the variables
xi can be expressed in the indicial notation as
∂Φ ∂Φ ∂xj ∂Φ ∂x1 ∂Φ ∂x2 ∂Φ ∂xN
= = + + · · · + (1.1.9)
∂xi ∂xj ∂xi ∂x1 ∂xi ∂x2 ∂xi ∂xN ∂xi
for any fixed value of i satisfying 1 ≤ i ≤ N.
The second partial derivatives of Φ can also be expressed in the index notation. Differentiation of
equation (1.1.9) partially with respect to xm produces
∂2Φ ∂Φ ∂ 2 xj ∂ ∂Φ ∂xj
= + . (1.1.10)
∂xi ∂xm ∂xj ∂xi ∂xm ∂xm ∂xj ∂xi
This result is nothing more than an application of the general rule for differentiating a product of two
quantities. To evaluate the derivative of the bracketed term in equation (1.1.10) it must be remembered that
the quantity inside the brackets is a function of the bar variables. Let
∂Φ
G= = G(x1 , x2 , . . . , xN )
∂xj
to emphasize this dependence upon the bar variables, then the derivative of G is
∂G ∂G ∂xk ∂ 2 Φ ∂xk
= = . (1.1.11)
∂xm ∂xk ∂xm ∂xj ∂xk ∂xm
This is just an application of the basic rule from equation (1.1.9) with Φ replaced by G. Hence the derivative
from equation (1.1.10) can be expressed
∂2Φ ∂Φ ∂ 2 xj ∂ 2 Φ ∂xj ∂xk
= j + (1.1.12)
i
∂x ∂x m i
∂x ∂x ∂x m
∂xj ∂xk ∂xi ∂xm
where i, m are free indices and j, k are dummy summation indices.
20
EXAMPLE 1.118. Let Φ = Φ(r, θ) where r, θ are polar coordinates related to the Cartesian coordinates
∂Φ ∂2Φ
(x, y) by the transformation equations x = r cos θ y = r sin θ. Find the partial derivatives and
∂x ∂x2
Solution: The partial derivative of Φ with respect to x is found from the relation (1.1.9) and can be written
∂Φ ∂Φ ∂r ∂Φ ∂θ
= + . (1.1.13)
∂x ∂r ∂x ∂θ ∂x
The second partial derivative is obtained by differentiating the first partial derivative. From the product
rule for differentiation we can write
∂2Φ ∂Φ ∂ 2 r ∂r ∂ ∂Φ ∂Φ ∂ 2 θ ∂θ ∂ ∂Φ
= + + + . (1.1.14)
∂x2 ∂r ∂x2 ∂x ∂x ∂r ∂θ ∂x2 ∂x ∂x ∂θ
To further simplify (1.1.14) it must be remembered that the terms inside the brackets are to be treated as
functions of the variables r and θ and that the derivative of these terms can be evaluated by reapplying the
∂Φ ∂Φ
basic rule from equation (1.1.13) with Φ replaced by ∂r and then Φ replaced by ∂θ . This gives
∂ 2Φ ∂Φ ∂ 2 r ∂r ∂ 2 Φ ∂r ∂ 2 Φ ∂θ
= + +
∂x2 ∂r ∂x2 ∂x ∂r2 ∂x ∂r∂θ ∂x
(1.1.15)
∂Φ ∂ 2 θ ∂θ ∂ 2 Φ ∂r ∂ 2 Φ ∂θ
+ + + .
∂θ ∂x2 ∂x ∂θ∂r ∂x ∂θ2 ∂x
y
From the transformation equations we obtain the relations r2 = x2 + y 2 and and from
tan θ =
x
these relations we can calculate all the necessary derivatives needed for the simplification of the equations
(1.1.13) and (1.1.15). These derivatives are:
∂r ∂r x
2r = 2x or = = cos θ
∂x ∂x r
∂θ y ∂θ y sin θ
sec2 θ = − 2 or =− 2 =−
∂x x ∂x r r
∂2r ∂θ sin2 θ ∂2θ −r cos θ ∂x
∂θ ∂r
+ sin θ ∂x 2 sin θ cos θ
= − sin θ = = = .
∂x2 ∂x r ∂x2 r2 r2
Therefore, the derivatives from equations (1.1.13) and (1.1.15) can be expressed in the form
∂Φ ∂Φ ∂Φ sin θ
= cos θ −
∂x ∂r ∂θ r
2
2
∂ Φ ∂Φ sin θ ∂Φ sin θ cos θ ∂ 2 Φ 2 ∂ 2 Φ cos θ sin θ ∂ 2 Φ sin2 θ
= + 2 + cos θ − 2 + .
∂x2 ∂r r ∂θ r2 ∂r2 ∂r∂θ r ∂θ2 r2
∂φ ∂φ ∂φ
grad φ = b
e1 + b
e2 + b
e3 .
∂x ∂y ∂z
The index notation focuses attention only on the components of the gradient. In Cartesian coordinates these
components are represented using a comma subscript to denote the derivative
∂φ
b
ej · grad φ = φ,j = , j = 1, 2, 3.
∂xj
The comma notation will be discussed in section 4. For now we use it to denote derivatives. For example
∂φ ∂2φ
φ ,j = j
, φ ,jk = , etc.
∂x ∂xj ∂xk
Bi = ei12 A2,1 + ei13 A3,1 + ei21 A1,2 + ei23 A3,2 + ei31 A1,3 + ei32 A2,3
Here i is a free index which can take on any of the values 1, 2 or 3. Consequently, we have
∂A3 ∂A2
For i = 1, B1 = A3,2 − A2,3 = −
∂x2 ∂x3
∂A1 ∂A3
For i = 2, B2 = A1,3 − A3,1 = −
∂x3 ∂x1
∂A2 ∂A1
For i = 3, B3 = A2,1 − A1,2 = −
∂x1 ∂x2
~ in Cartesian coordinates.
which verifies the index notation representation of curl A
22
Other Operations. The following examples illustrate how the index notation can be used to represent
additional vector operators in Cartesian coordinates.
~ · ∇)A
1. In index notation the components of the vector (B ~ are
~ · ∇)A}
{(B ~ ·b
ep = Ap,q Bq p, q = 1, 2, 3
This can be verified by performing the indicated summations. We have by summing on the repeated
index q
Ap,q Bq = Ap,1 B1 + Ap,2 B2 + Ap,3 B3 .
The index p is now a free index which can have any of the values 1, 2 or 3. We have:
~ · ∇)φ has the following form when expressed in the index notation:
2. The scalar (B
This can be verified by performing the indicated summations and is left as an exercise.
~ × ∇) · A
4. The scalar (B ~ may be expressed in the index notation. It has the form
~ × ∇) · A
(B ~ = eijk Bj Ai,k .
This can also be verified by performing the indicated summations and is left as an exercise.
5. The vector components of ∇2 A ~ in the index notation are represented
ep · ∇2 A
b ~ = Ap,qq .
~ = ∇ × (f A)
curl (f A) ~ = (∇f ) × A
~ + f (∇ × A).
~
~ = curl (f A)
Solution: Let B ~ and write the components as
Bi = eijk (f Ak ),j
= eijk [f Ak,j + f,j Ak ]
= f eijk Ak,j + eijk f,j Ak .
~ = curl (f A)
B ~ = f (∇ × A)
~ + (∇f ) × A
~
~ + B)
EXAMPLE 1.120. Prove the vector identity ∇ · (A ~ = ∇·A ~ +∇·B ~
~+B
Solution: Let A ~ =C
~ and write this vector equation in the index notation as Ai + Bi = Ci . We then
have
~ + ∇ · B.
~ = Ci,i = (Ai + Bi ),i = Ai,i + Bi,i = ∇ · A
∇·C ~
~ · ∇)f = A
EXAMPLE 1.121. In Cartesian coordinates prove the vector identity (A ~ · ∇f
Solution: In the index notation we write
~ · ∇)f = Ai f,i = A1 f,1 + A2 f,2 + A3 f,3
(A
∂f ∂f ∂f ~ · ∇f.
= A1 1 + A2 2 + A3 3 = A
∂x ∂x ∂x
~ × B)
∇ × (A ~ = A(∇
~ ~ − B(∇
· B) ~ ~ + (B
· A) ~ · ∇)A
~ − (A
~ · ∇)B
~
~ × B)
Solution: The pth component of the vector ∇ × (A ~ is
b ~ × B)]
ep · [∇ × (A ~ = epqk [ekji Aj Bi ],q
By applying the e − δ identity, the above expression simplifies to the desired result. That is,
b ~ × B)]
ep · [∇ × (A ~ = (δpj δqi − δpi δqj )Aj Bi,q + (δpj δqi − δpi δqj )Aj,q Bi
~ × B)
∇ × (A ~ = A(∇
~ ~ − (A
· B) ~ · ∇)B
~ + (B
~ · ∇)A
~ − B(∇
~ ~
· A)
24
~ − ∇2 A.
~ = ∇(∇ · A)
Expressing this result in vector form we have ∇ × (∇ × A) ~
The divergence theorem, in both vector and indicial notation, can be written
ZZZ ZZ Z Z
div · F~ dτ = b dσ
F~ · n Fi,i dτ = Fi ni dσ i = 1, 2, 3 (1.1.16)
V S V S
where ni are the direction cosines of the unit exterior normal to the surface, dτ is a volume element and dσ
is an element of surface area. Note that in using the indicial notation the volume and surface integrals are
to be extended over the range specified by the indices. This suggests that the divergence theorem can be
applied to vectors in n−dimensional spaces.
The vector form and indicial notation for the Stokes theorem are
ZZ Z Z Z
b dσ =
(∇ × F~ ) · n F~ · d~r eijk Fk,j ni dσ = Fi dxi i, j, k = 1, 2, 3 (1.1.17)
S C S C
and the Green’s theorem in the plane, which is a special case of the Stoke’s theorem, can be expressed
ZZ Z Z Z
∂F2 ∂F1
− dxdy = F1 dx + F2 dy e3jk Fk,j dS = Fi dxi i, j, k = 1, 2 (1.1.18)
∂x ∂y C S C
is obtained by first letting F~ = φ∇ψ in the divergence theorem and then letting F~ = ψ∇φ in the divergence
theorem and then subtracting the results.
Determinants, Cofactors
This gives a summation of the n! permutations of products formed from the elements of the matrix A. The
result is a single number called the determinant of A.
Brst = eijk air ajs akt = ekji akr ajs ait = ekji ait ajs akr = −eijk ait ajs akr = −Btsr ,
and by considering other permutations of the indices, one can establish that Brst is completely skew
symmetric. In the exercises it is shown that any third order completely skewsymmetric system satisfies
Brst = B123 erst . But B123 = det A and so we arrive at the identity
eijk ari asj atk = Aerst and eijk air ajs akt = Aerst . (1.1.19)
by use of the indicial notation. By column expansions, this determinant can be represented
A = ar1 A1r = a11 A11 + a21 A12 + a31 A13 . (1.1.23)
That is, the determinant A is obtained by multiplying each element in the first column by its corresponding
cofactor and summing the result. Observe also that from the equation (1.1.20) we find the additional
cofactors
A2s = erst ar1 at3 and A3t = erst ar1 as2 . (1.1.24)
Hence, the equation (1.1.20) can also be expressed in one of the forms
The results from equations (1.1.22) and (1.1.24) can be written in a slightly different form with the indicial
notation. From the notation for a generalized Kronecker delta defined by
ijk
eijk elmn = δlmn ,
1 ijk s t
Air = δ a a (1.1.25)
2! rst j k
which represents the cofactor of ari . When the elements from any row (or column) are multiplied by their
corresponding cofactors, and the results summed, we obtain the value of the determinant. Whenever the
elements from any row (or column) are multiplied by the cofactor elements from a different row (or column),
and the results summed, we get zero. This can be illustrated by considering the summation
1 ijk s t m 1
am i
r Am = δmst aj ak ar = eijk emst am s t
r aj ak
2! 2!
1 1 ijk
= eijk erjk A = δrjk A = δri A
2! 2!
Here we have used the e − δ identity to obtain
ijk
δrjk = eijk erjk = ejik ejrk = δri δkk − δki δrk = 3δri − δri = 2δri
Observe that
δkj11jk22...jN
...kN e
k1 k2 ...kN
= (N !) ej1 j2 ...jN
This follows because ek1 k2 ...kN is skewsymmetric in all pairs of its superscripts. The lefthand side denotes
a summation of N ! terms. The first term in the summation has superscripts j1 j2 . . . jN and all other terms
have superscripts which are some permutation of this ordering with minus signs associated with those terms
having an odd permutation. Because ej1 j2 ...jN is completely skewsymmetric we find that all terms in the
summation have the value +ej1 j2 ...jN . We thus obtain N ! of these terms.
28
EXERCISE 1.1
I 1. Simplify each of the following by employing the summation property of the Kronecker delta. Perform
sums on the summation indices only if your are unsure of the result.
(a) eijk δkn (c) eijk δis δjm δkn (e) δij δjn
(b) eijk δis δjm (d) aij δin (f ) δij δjn δni
I 3. ~ = Ai
Express each of the following in index notation. Be careful of the notation you use. Note that A
is an incorrect notation because a vector can not equal a scalar. The notation A~· b
ei = Ai should be used to
express the ith component of a vector.
~ · (B
(a) A ~ × C)
~ ~ A
(c) B( ~ · C)
~
~ × (B
(b) A ~ × C)
~ (d) ~ A
B( ~ · C)
~ − C(
~ A~ · B)
~
I 4. Show the e permutation symbol satisfies: (a) eijk = ejki = ekij (b) eijk = −ejik = −eikj = −ekji
I 5. ~ × (B
Use index notation to verify the vector identity A ~ × C)
~ = B(
~ A~ · C)
~ − C(
~ A~ · B)
~
(a) Solve for yi in terms of zi using the indicial notation and check your result
to be sure that no index repeats itself more than twice.
(b) Perform the indicated summations and write out expressions
for y1 , y2 in terms of z1 , z2
(c) Express the above equations in matrix form. Expand the matrix
equations and check the solution obtained in part (b).
I 7. Use the e − δ identity to simplify (a) eijk ejik (b) eijk ejki
(a) ~ · (B
A ~ × C)
~ =B
~ · (C
~ × A)
~ =C
~ · (A
~ × B)
~ triple scalar product
~ × B)
(b) (A ~ ×C
~ = B(
~ A~ · C)
~ − A(
~ B~ · C)
~
(c) ~ × B)
(A ~ × (C
~ × D)
~ = B(
~ A~·C
~ × D)
~ − A(
~ B~ ·C
~ × D)
~
29
(a) Ci = eijk Aj Bk , i = 1, 2, 3
(b) Ai Bi
(c) What do the results in (a) and (b) represent?
dy1 dy2
I 11. Represent the differential equations = a11 y1 + a12 y2 and = a21 y1 + a22 y2
dt dt
using the index notation.
I 12.
Let Φ = Φ(r, θ) where r, θ are polar coordinates related to Cartesian coordinates (x, y) by the transfor
mation equations x = r cos θand y = r sin θ.
∂Φ ∂2Φ
(a) Find the partial derivatives , and
∂y ∂y 2
(b) Combine the result in part (a) with the result from EXAMPLE 1.118 to calculate the Laplacian
∂2Φ ∂2Φ
∇2 Φ = +
∂x2 ∂y 2
in polar coordinates.
I 15. Determine if the following relation is true or false. Justify your answer.
ei · ( b
b ej × b
ek ) = ( b
ei × b
ej ) · b
ek = eijk , i, j, k = 1, 2, 3.
Hint: Let b
em = (δ1m , δ2m , δ3m ).
I 16. Without substituting values for i, l = 1, 2, 3 calculate all nine terms of the given quantities
I 17. Let Amn xm y n = 0 for arbitrary xi and y i , i = 1, 2, 3, and show that Aij = 0 for all values of i, j.
30
I 18.
(a) For amn , m, n = 1, 2, 3 skewsymmetric, show that amn xm xn = 0.
(b) Let amn xm xn = 0, m, n = 1, 2, 3 for all values of xi , i = 1, 2, 3 and show that amn must be skew
symmetric.
I 19. Let A and B denote 3 × 3 matrices with elements aij and bij respectively. Show that if C = AB is a
matrix product, then det(C) = det(A) · det(B).
Hint: Use the result from example 1.19.
I 20.
(a) Let u1 , u2 , u3 be functions of the variables s1 , s2 , s3 . Further, assume that s1 , s2 , s3 are in turn each
∂um ∂(u1 , u2 , u3 )
functions of the variables x1 , x2 , x3 . Let n = denote the Jacobian of the u0 s with
∂x ∂(x1 , x2 , x3 )
respect to the x0 s. Show that
i i
∂u ∂u ∂sj ∂ui ∂sj
= = ·
∂xm ∂sj ∂xm ∂sj ∂xm .
I 21. A third order system a`mn with `, m, n = 1, 2, 3 is said to be symmetric in two of its subscripts if the
components are unaltered when these subscripts are interchanged. When a`mn is completely symmetric then
a`mn = am`n = a`nm = amn` = anm` = an`m . Whenever this third order system is completely symmetric,
then: (i) How many components are there? (ii) How many of these components are distinct?
Hint: Consider the three cases (i) ` = m = n (ii) ` = m 6= n (iii) ` 6= m 6= n.
I 22. A third order system b`mn with `, m, n = 1, 2, 3 is said to be skewsymmetric in two of its subscripts
if the components change sign when the subscripts are interchanged. A completely skewsymmetric third
order system satisfies b`mn = −bm`n = bmn` = −bnm` = bn`m = −b`nm . (i) How many components does
a completely skewsymmetric system have? (ii) How many of these components are zero? (iii) How many
components can be different from zero? (iv) Show that there is one distinct component b123 and that
b`mn = e`mn b123 .
Hint: Consider the three cases (i) ` = m = n (ii) ` = m 6= n (iii) ` 6= m 6= n.
I 23. Let i, j, k = 1, 2, 3 and assume that eijk σjk = 0 for all values of i. What does this equation tell you
about the values σij , i, j = 1, 2, 3?
I 24. Assume that Amn and Bmn are symmetric for m, n = 1, 2, 3. Let Amn xm xn = Bmn xm xn for arbitrary
values of xi , i = 1, 2, 3, and show that Aij = Bij for all values of i and j.
I 25. Assume Bmn is symmetric and Bmn xm xn = 0 for arbitrary values of xi , i = 1, 2, 3, show that Bij = 0.
31
I 26. (Generalized Kronecker delta) Define the generalized Kronecker delta as the n × n determinant
i
δm δni · · · δpi
j
δm δnj · · · δpj
ij...k
δmn...p = . .. .. . where δsr is the Kronecker delta.
.. . ..
.
δk δnk · · · δpk
m
123
(a) Show eijk = δijk
ijk
(b) Show eijk = δ123
ij
(c) Show δmn = eij emn
rs rsp
(d) Define δmn = δmnp (summation on p)
and show rs
δmn = δm δn − δnr δm
r s s
Note that by combining the above result with the result from part (c)
we obtain the two dimensional form of the e − δ identity ers emn = δm δn − δnr δm
r s s
.
r
(e) Define δm = 12 δmn
rn
(summation on n) and show rst
δpst = 2δpr
rst
(f ) Show δrst = 3!
1
a1 a12 a13
I 27. Let Ar denote the cofactor of ai in the determinant a21
i r
a22 a23 as given by equation (1.1.25).
a31 a32 a33
(a) Show erst Air = eijk asj atk (b) Show erst Ari = eijk ajs akt
I 28. (a) Show that if Aijk = Ajik , i, j, k = 1, 2, 3 there is a total of 27 elements, but only 18 are distinct.
(b) Show that for i, j, k = 1, 2, . . . , N there are N 3 elements, but only N 2 (N + 1)/2 are distinct.
I 29. Let aij = Bi Bj for i, j = 1, 2, 3 where B1 , B2 , B3 are arbitrary constants. Calculate det(aij ) = A.
I 30.
(a) For A = (aij ), i, j = 1, 2, 3, show A = eijk ai1 aj2 ak3 .
(b) For A = (aij ), i, j = 1, 2, 3, show A = eijk ai1 aj2 ak3 .
(c) For A = (aij ), i, j = 1, 2, 3, show A = eijk a1i a2j a3k .
(d) For I = (δji ), i, j = 1, 2, 3, show I = 1.
I 31. Let A = eijk ai1 aj2 ak3 and define Aim as the cofactor of aim . Show the determinant can be
expressed in any of the forms:
I 33. Use the results in problems 31 and 32 to prove that apm Aim = Aδip .
1 2 1
I 34. Let (aij ) = 1 0 3 and calculate C = aij aij , i, j = 1, 2, 3.
2 3 2
I 35. Let
a111 = −1, a112 = 3, a121 = 4, a122 = 2
a211 = 1, a212 = 5, a221 = 2, a222 = −2
and calculate the quantity C = aijk aijk , i, j, k = 1, 2.
I 36. Let
a1111 = 2, a1112 = 1, a1121 = 3, a1122 = 1
a1211 = 5, a1212 = −2, a1221 = 4, a1222 = −2
a2111 = 1, a2112 = 0, a2121 = −2, a2122 = −1
a2211 = −2, a2212 = 1, a2221 = 2, a2222 = 2
and calculate the quantity C = aijkl aijkl , i, j, k, l = 1, 2.
∂xi
(a) (Aijkl + Ajkli + Aklij + Alijk )xi xj xk xl (c)
∂xj
(b) (Pijk + Pjki + Pkij )xi xj xk ∂ 2 xi ∂xj ∂ 2 xm ∂xi
(d) aij r − ami
t s
∂x ∂x ∂x ∂xs ∂xt ∂xr
I 38. Let g denote the determinant of the matrix having the components gij , i, j = 1, 2, 3. Show that
g1r g1s g1t gir gis git
(a) g erst = g2r g2s g2t (b) g erst eijk = gjr gjs gjt
g3r g3s g3t gkr gks gkt
i
δm δni δpi
j
I 39. Show that eijk emnp = δmnp
ijk
= δm δnj δpj
δk δnk δpk
m
I 40. Show that eijk emnp Amnp = Aijk − Aikj + Akij − Ajik + Ajki − Akji
Hint: Use the results from problem 39.
I 42. Determine if the following statement is true or false. Justify your answer. eijk Ai Bj Ck = eijk Aj Bk Ci .
(c) Let aij , i, j = 1, 2, 3 denote the components of a 3 × 3 matrix A, which are functions of time t. Develop
appropriate relations, expand them and verify, similar to parts (a) and (b) above, the representation of
a determinant and its derivative.
I 44. For f = f (x1 , x2 , x3 ) and φ = φ(f ) differentiable scalar functions, use the indicial notation to find a
formula to calculate grad φ .
I 46. If Aij is symmetric and Bij is skewsymmetric, i, j = 1, 2, 3, then calculate C = Aij Bij .
I 47. Assume Aij = Aij (x1 , x2 , x3 ) and Aij = Aij (x1 , x2 , x3 ) for i, j = 1, 2, 3 are related by the expression
∂xi ∂xj ∂Amn
Amn = Aij m n . Calculate the derivative .
∂x ∂x ∂xk
I 48. Prove that if any two rows (or two columns) of a matrix are interchanged, then the value of the
determinant of the matrix is multiplied by minus one. Construct your proof using 3 × 3 matrices.
I 49. Prove that if two rows (or columns) of a matrix are proportional, then the value of the determinant
of the matrix is zero. Construct your proof using 3 × 3 matrices.
I 50. Prove that if a row (or column) of a matrix is altered by adding some constant multiple of some other
row (or column), then the value of the determinant of the matrix remains unchanged. Construct your proof
using 3 × 3 matrices.
I 52. Let Aijk denote a third order system where i, j, k = 1, 2. (a) How many components does this system
have? (b) Let Aijk be skewsymmetric in the last pair of indices, how many independent components does
the system have?
I 53. Let Aijk denote a third order system where i, j, k = 1, 2, 3. (a) How many components does this
system have? (b) In addition let Aijk = Ajik and Aikj = −Aijk and determine the number of distinct
nonzero components for Aijk .
34
I 54. Show that every second order system Tij can be expressed as the sum of a symmetric system Aij and
skewsymmetric system Bij . Find Aij and Bij in terms of the components of Tij .
I 57. The angle 0 ≤ θ ≤ π between two skew lines in space is defined as the angle between their direction
vectors when these vectors are placed at the origin. Show that for two lines with direction numbers ai and
bi i = 1, 2, 3, the cosine of the angle between these lines satisfies
ai b i
cos θ = √ √
ai ai b i b i
I 58. Let aij = −aji for i, j = 1, 2, . . . , N and prove that for N odd det(aij ) = 0.
∂λ ∂2λ
I 59. Let λ = Aij xi xj where Aij = Aji and calculate (a) (b)
∂xm ∂xm ∂xk
I 60. Given an arbitrary nonzero vector Uk , k = 1, 2, 3, define the matrix elements aij = eijk Uk , where eijk
is the epermutation symbol. Determine if aij is symmetric or skewsymmetric. Suppose Uk is defined by
the above equation for arbitrary nonzero aij , then solve for Uk in terms of the aij .
I 61. If Aij = Ai Bj 6= 0 for all i, j values and Aij = Aji for i, j = 1, 2, . . . , N , show that Aij = λBi Bj
where λ is a constant. State what λ is.
I 62. Assume that Aijkm , with i, j, k, m = 1, 2, 3, is completely skewsymmetric. How many independent
components does this quantity have?
I 63. Consider Rijkm , i, j, k, m = 1, 2, 3, 4. (a) How many components does this quantity have? (b) If
Rijkm = −Rijmk = −Rjikm then how many independent components does Rijkm have? (c) If in addition
Rijkm = Rkmij determine the number of independent components.
I 64. Let xi = aij x̄j , i, j = 1, 2, 3 denote a change of variables from a barred system of coordinates to an
unbarred system of coordinates and assume that Āi = aij Aj where aij are constants, Āi is a function of the
∂ Āi
x̄j variables and Aj is a function of the xj variables. Calculate .
∂ x̄m
35
e1 , b
For b e2 , b ~ as
e3 independent orthogonal unit vectors (base vectors), we may write any vector A
~ = A1 b
A e1 + A2 b
e2 + A3 b
e3
~ = (A
A e1 ) b
~· b ~· b
e1 + (A e2 ) b ~·b
e2 + (A e3 ) b
e3 .
Reciprocal Basis
This can be done by taking appropriate projections and obtaining three equations and three unknowns from
which the components are determined. A much easier way to find the components (A1 , A2 , A3 ) is to construct
~ 1, E
a reciprocal basis (E ~ 2, E
~ 3 ). Recall that two bases (E
~ 1, E
~ 2, E ~ 1, E
~ 3 ) and (E ~ 2, E
~ 3 ) are said to be reciprocal
if they satisfy the condition
~ j = δj =
~i · E 1 if i = j
E i .
0 6 j
if i =
Note that E ~ 1 = δ21 = 0 and E
~2 · E ~ 1 = δ31 = 0 so that the vector E
~3 · E ~ 1 is perpendicular to both the
~ 2 and E
vectors E ~ 3 . (i.e. A vector from one basis is orthogonal to two of the vectors from the other basis.)
We can therefore write E ~ 1 = V −1 E
~2 × E ~ 3 where V is a constant to be determined. By taking the dot
product of both sides of this equation with the vector E ~ 1 we find that V = E ~ 1 · (E
~2 × E
~ 3 ) is the volume
of the parallelepiped formed by the three vectors E ~ 2, E
~ 1, E ~ 3 when their origins are made to coincide. In a
36
~ 1, E
similar manner it can be demonstrated that for (E ~ 2, E
~ 3 ) a given set of basis vectors, then the reciprocal
basis vectors are determined from the relations
E~1 = 1 E ~2 × E~ 3, ~2 = 1 E
E ~3 × E
~ 1, ~3 = 1 E
E ~1 × E~ 2,
V V V
where V = E ~2 × E
~ 1 · (E ~ 3 ) 6= 0 is a triple scalar product and represents the volume of the parallelepiped
having the basis vectors for its sides.
Let (E ~ 2, E
~ 1, E ~ 1, E
~ 3 ) and (E ~ 2, E
~ 3 ) denote a system of reciprocal bases. We can represent any vector A
~
with respect to either of these bases. If we select the basis (E ~ 2, E
~ 1, E ~ 3 ) and represent A
~ in the form
~ = A1 E
A ~ 1 + A2 E
~ 2 + A3 E
~ 3, (1.2.1)
A ~ 1 = A1 ,
~ ·E ~ 2 = A2 ,
~·E
A ~ 3 = A3 .
~·E
A
~ 1, E
Similarly, if we choose the reciprocal basis (E ~ 2, E
~ 3 ) and represent A
~ in the form
~ 1 + A2 E
~ = A1 E
A ~ 2 + A3 E
~ 3, (1.2.2)
~ 1, E
then the components (A1 , A2 , A3 ) relative to the basis (E ~ 2, E
~ 3 ) are called the covariant components of
~ These components can be determined from the relations
A.
~ ·E
A ~ 1 = A1 , ~ ·E
A ~ 2 = A2 , ~ ·E
A ~ 3 = A3 .
The contravariant and covariant components are different ways of representing the same vector with respect
to a set of reciprocal basis vectors. There is a simple relationship between these components which we now
develop. We introduce the notation
E ~ j = gij = gji ,
~i · E and ~i · E
E ~ j = g ij = g ji (1.2.3)
where gij are called the metric components of the space and g ij are called the conjugate metric components
of the space. We can then write
~ ·E
A ~1 · E
~ 1 = A1 (E ~2 · E
~ 1 ) + A2 (E ~3 · E
~ 1 ) + A3 (E ~ 1 ) = A1
A ~ 1 = A1 (E
~ ·E ~ 1 ) + A2 (E
~1 · E ~ 1 ) + A3 (E
~2 · E ~3 · E
~ 1 ) = A1
or
A1 = A1 g11 + A2 g12 + A3 g13 . (1.2.4)
~·E
In a similar manner, by considering the dot products A ~ ·E
~ 2 and A ~ 3 one can establish the results
Ai = gik Ak . (1.2.6)
The equations (1.2.6) and (1.2.7) are relations which exist between the contravariant and covariant compo
nents of the vector A. ~ Similarly, if for some value j we have E ~1 + β E
~j = αE ~2 + γ E~ 3 , then one can show
~ =g E
that E j ij ~ i . This is left as an exercise.
37
Coordinate Transformations
Consider a coordinate transformation from a set of coordinates (x, y, z) to (u, v, w) defined by a set of
transformation equations
x = x(u, v, w)
y = y(u, v, w) (1.2.8)
z = z(u, v, w)
It is assumed that these transformations are single valued, continuous and possess the inverse transformation
u = u(x, y, z)
v = v(x, y, z) (1.2.9)
w = w(x, y, z).
These transformation equations define a set of coordinate surfaces and coordinate curves. The coordinate
surfaces are defined by the equations
u(x, y, z) = c1
v(x, y, z) = c2 (1.2.10)
w(x, y, z) = c3
where c1 , c2 , c3 are constants. These surfaces intersect in the coordinate curves
where
e1 + y(u, v, w) b
~r(u, v, w) = x(u, v, w) b e2 + z(u, v, w) b
e3 .
~ 1 = grad u = ∇u,
E ~ 2 = grad v = ∇v,
E ~ 3 = grad w = ∇w
E (1.2.12)
evaluated at the common point of intersection (c1 , c2 , c3 ) of the coordinate surfaces. The system of vectors
~ 1, E
(E ~ 2, E
~ 3 ) can be selected as a system of basis vectors which are normal to the coordinate surfaces.
Similarly, the vectors
~ 1 = ∂~r ,
E ~ 2 = ∂~r ,
E ~ 3 = ∂~r
E (1.2.13)
∂u ∂v ∂w
~ 1, E
when evaluated at the common point of intersection (c1 , c2 , c3 ) forms a system of vectors (E ~ 2, E
~ 3 ) which
we can select as a basis. This basis is a set of tangent vectors to the coordinate curves. It is now demonstrated
that the normal basis (E ~ 1, E
~ 2, E
~ 3 ) and the tangential basis (E
~ 1, E
~ 2, E
~ 3 ) are a set of reciprocal bases.
e1 + y b
Recall that ~r = x b e2 + z b
e3 denotes the position vector of a variable point. By substitution for
x, y, z from (1.2.8) there results
e1 + y(u, v, w) b
~r = ~r(u, v, w) = x(u, v, w) b e2 + z(u, v, w) b
e3 . (1.2.14)
38
du = grad u · d~r
∂~r ∂~r ∂~r
du = grad u · du + dv + dw
∂u ∂v ∂w (1.2.18)
∂~r ∂~r ∂~r
du = grad u · du + grad u · dv + grad u · dw.
∂u ∂v ∂w
By comparing like terms in this last equation we find that
~1 · E
E ~ 1 = 1, ~1 · E
E ~ 2 = 0, ~1 · E
E ~ 3 = 0. (1.2.19)
Similarly, from the other equations in equation (1.2.9) which define v = v(x, y, z), and w = w(x, y, z) it
can be demonstrated that
∂~r ∂~r ∂~r
dv = grad v · du + grad v · dv + grad v · dw (1.2.20)
∂u ∂v ∂w
39
and
∂~r ∂~r ∂~r
dw = grad w · du + grad w · dv + grad w · dw. (1.2.21)
∂u ∂v ∂w
By comparing like terms in equations (1.2.20) and (1.2.21) we find
~2 · E
E ~ 1 = 0, ~2 · E
E ~ 2 = 1, ~2 · E
E ~3 = 0
(1.2.22)
~3 · E
E ~ 1 = 0, ~3 · E
E ~ 2 = 0, ~3 · E
E ~ 3 = 1.
The equations (1.2.22) and (1.2.19) show us that the basis vectors defined by equations (1.2.12) and (1.2.13)
are reciprocal.
Introducing the notation
where the x0 s denote the generalized coordinates and the y 0 s denote the rectangular Cartesian coordinates,
the above equations can be expressed in a more concise form with the index notation. For example, if
~ i = grad xi ,
E i = 1, 2, 3 (1.2.25)
and
~ i = ∂~r ,
E i = 1, 2, 3. (1.2.26)
∂xi
We now show that these basis vectors are reciprocal. Observe that ~r = ~r(x1 , x2 , x3 ) with
∂~r
d~r = dxm (1.2.27)
∂xm
and consequently
∂~r
dxi = grad xi · d~r = grad xi · dxm
= E~i · E
~ m dxm = δm
i
dxm , i = 1, 2, 3 (1.2.28)
∂xm
Comparing like terms in this last equation establishes the result that
E ~ m = δi ,
~i · E i, m = 1, 2, 3 (1.2.29)
m
Tensors are quantities which obey certain transformation laws. That is, scalars, vectors, matrices
and higher order arrays can be thought of as components of a tensor quantity. We shall be interested in
finding how these components are represented in various coordinate systems. We desire knowledge of these
transformation laws in order that we can represent various physical laws in a form which is independent of
the coordinate system chosen. Before defining different types of tensors let us examine what we mean by a
coordinate transformation.
Coordinate transformations of the type found in equations (1.2.8) and (1.2.9) can be generalized to
higher dimensions. Let xi , i = 1, 2, . . . , N denote N variables. These quantities can be thought of as
representing a variable point (x1 , x2 , . . . , xN ) in an N dimensional space VN . Another set of N quantities,
call them barred quantities, xi , i = 1, 2, . . . , N, can be used to represent a variable point (x1 , x2 , . . . , xN ) in
an N dimensional space V N . When the x0 s are related to the x0 s by equations of the form
xi = xi (x1 , x2 , . . . , xN ), i = 1, 2, . . . , N (1.2.30)
then a transformation is said to exist between the coordinates xi and xi , i = 1, 2, . . . , N. Whenever the
relations (1.2.30) are functionally independent, single valued and possess partial derivatives such that the
Jacobian of the transformation
∂x1
1 ∂x1
... ∂x1
x 1 2 N ∂x ∂x2 ∂xN
x ,x ,...,x ..
J =J = ... .. (1.2.31)
x x1 , x2 , . . . , xN N . ... .
∂x ∂xN
... ∂xN
∂x1 ∂x2 ∂xN
xi = xi (x1 , x2 , . . . , xN ), i = 1, 2, . . . , N. (1.2.32)
For brevity the transformation equations (1.2.30) and (1.2.32) are sometimes expressed by the notation
T1 : y i = y i (x1 , . . . , xN ) i = 1, . . . , N or T1 x = y
i i 1 N
T2 : z = z (y , . . . , y ) i = 1, . . . , N or T2 y = z
Then the transformation T3 obtained by substituting T1 into T2 is called the product of two successive
transformations and is written
T3 : z i = z i (y 1 (x1 , . . . , xN ), . . . , y N (x1 , . . . , xN )) i = 1, . . . , N or T3 x = T2 T1 x = z.
A group G is a nonempty set of elements together with a law, for combining the elements. The combined
elements are denoted by a product. Thus, if a and b are elements in G then no matter how you define the
law for combining elements, the product combination is denoted ab. The set G and combining law forms a
group if the following properties are satisfied:
(i) For all a, b ∈ G, then ab ∈ G. This is called the closure property.
(ii) There exists an identity element I such that for all a ∈ G we have Ia = aI = a.
(iii) There exists an inverse element. That is, for all a ∈ G there exists an inverse element a−1 such that
a a−1 = a−1 a = I.
(iv) The associative law holds under the combining law and a(bc) = (ab)c for all a, b, c ∈ G.
For example, the set of elements G = {1, −1, i, −i}, where i2 = −1 together with the combining law of
ordinary multiplication, forms a group. This can be seen from the multiplication table.
× 1 1 i i
1 1 1 i i
1 1 1 i i
i i i 1 1
i i i 1 1
The set of all coordinate transformations of the form found in equation (1.2.30), with Jacobian different
from zero, forms a group because:
(i) The product transformation, which consists of two successive transformations, belongs to the set of
transformations. (closure)
(ii) The identity transformation exists in the special case that x and x are the same coordinates.
(iii) The inverse transformation exists because the Jacobian of each individual transformation is different
from zero.
(iv) The associative law is satisfied in that the transformations satisfy the property T3 (T2 T1 ) = (T3 T2 )T1 .
When the given transformation equations contain a parameter the combining law is often times repre
sented as a product of symbolic operators. For example, we denote by Tα a transformation of coordinates
having a parameter α. The inverse transformation can be denoted by Tα−1 and one can write Tα x = x or
x = Tα−1 x. We let Tβ denote the same transformation, but with a parameter β, then the transitive property
is expressed symbolically by Tα Tβ = Tγ where the product Tα Tβ represents the result of performing two
successive transformations. The first coordinate transformation uses the given transformation equations and
uses the parameter α in these equations. This transformation is then followed by another coordinate trans
formation using the same set of transformation equations, but this time the parameter value is β. The above
symbolic product is used to demonstrate that the result of applying two successive transformations produces
a result which is equivalent to performing a single transformation of coordinates having the parameter value
γ. Usually some relationship can then be established between the parameter values α, β and γ.
42
In this symbolic notation, we let Tθ denote the identity transformation. That is, using the parameter
value of θ in the given set of transformation equations produces the identity transformation. The inverse
transformation can then be expressed in the form of finding the parameter value β such that Tα Tβ = Tθ .
Cartesian Coordinates
y i = y i (x1 , x2 , . . . , xN ), i = 1, 2, . . . , N, (1.2.34)
with Jacobian different from zero. Similarly, if there is some other set of generalized coordinates, say a barred
system xi , i = 1, . . . , N where the x0 s are independent functions of the y 0 s, then there will exist another set
of independent transformation equations
y i = y i (x1 , x2 , . . . , xN ), i = 1, 2, . . . , N, (1.2.35)
with Jacobian different from zero. The transformations found in the equations (1.2.34) and (1.2.35) imply
that there exists relations between the x0 s and x0 s of the form (1.2.30) with inverse transformations of the
form (1.2.32). It should be remembered that the concepts and ideas developed in this section can be applied
to a space VN of any finite dimension. Two dimensional surfaces (N = 2) and three dimensional spaces
(N = 3) will occupy most of our applications. In relativity, one must consider spaces where N = 4.
from rectangular coordinates (x, y, z) to cylindrical coordinates (r, θ, z), illustrated in the figure 1.22. By
letting
y 1 = x, y 2 = y, y3 = z x1 = r, x2 = θ, x3 = z
the above set of equations are examples of the transformation equations (1.2.8) with u = r, v = θ, w = z as
the generalized coordinates.
43
y 1 = x, y 2 = y, y 3 = z x1 = ρ, x2 = θ , x3 = φ
the above set of equations has the form found in equation (1.2.8) with u = ρ, v = θ, w = φ the generalized
coordinates. One could place bars over the x0 s in this example in order to distinguish these coordinates from
the x0 s of the previous example. The spherical coordinates (ρ, θ, φ) are illustrated in the figure 1.23.
We are now at a point where we can begin to define what tensor quantities are. The first definition is
for a scalar invariant or tensor of order zero.
44
That is, an absolute scalar field is an invariant object in the space VN with respect to the group of
coordinate transformations. It has a single component in each coordinate system. For any scalar function
of the type defined by equation (1.2.36), we can substitute the transformation equations (1.2.30) and obtain
C: xi = xi (t), i = 1, . . . , N
In index notation, which focuses attention on the components, this tangent vector is denoted
dxi
Ti = , i = 1, . . . , N.
dt
For a coordinate transformation of the type defined by equation (1.2.30) with its inverse transformation
defined by equation (1.2.32), the curve C is represented in the barred space by
with t unchanged. The tangent to the curve in the barred system of coordinates is represented by
i
Letting T , i = 1, . . . , N denote the components of this tangent vector in the barred system of coordinates,
the equation (1.2.39) can then be expressed in the form
i ∂xi j
T = T , i, j = 1, . . . , N. (1.2.40)
∂xj
This equation is said to define the transformation law associated with an absolute contravariant tensor of
rank or order one. In the case N = 3 the matrix form of this transformation is represented
1
1 ∂x ∂x1 ∂x1
T ∂x1 ∂x2 ∂x3 T1
2
T = ∂x21 ∂x2 ∂x2 2
∂x ∂x2 ∂x3 T (1.2.41)
3
T ∂x3
1
∂x3 ∂x3 T3
∂x ∂x2 ∂x3
i ∂xi j
A = JW A
∂xj
We see that the above transformation law satisfies the group properties.
i ∂xi j
A (x) = A (x). (1.2.42)
∂xj
Here we have used the notation Aj (x) to emphasize the dependence of the components Aj upon the x
coordinates. Changing indices and substituting equation (1.2.42) into (1.2.43) we find
i
i ∂x ∂xj m
A (x) = A (x). (1.2.44)
∂xj ∂xm
46
as these expressions are incorrect. Note that there are no summations in these terms, whereas there is a
summation index in the representation of the chain rule.
Consider a scalar invariant A(x) = A(x) which is a shorthand notation for the equation
A(x1 , x2 , . . . , xn ) = A(x1 , x2 , . . . , xn )
involving the coordinate transformation of equation (1.2.30). By the chain rule we differentiate this invariant
and find that the components of the gradient must satisfy
∂A ∂A ∂xj
i = ∂xj . (1.2.46)
∂x ∂xi
Let
∂A ∂A
Aj = and Ai = ,
∂xj ∂xi
then equation (1.2.46) can be expressed as the transformation law
∂xj
Ai = Aj . (1.2.47)
∂xi
This is the transformation law for an absolute covariant tensor of rank or order one. A more general definition
is
47
∂xj
Ai = J W Aj (1.2.48)
∂xi
Again we note that the above transformation satisfies the group properties. Absolute tensors of rank or
order one are referred to as vectors while absolute tensors of rank or order zero are referred to as scalars.
EXAMPLE 1.24. (Transitive Property of Covariant Transformation)
Consider a sequence of transformation laws of the type defined by the equation (1.2.47)
∂xj
x→x Ai (x) = Aj (x)
∂xi
x→x ∂xm
Ak (x) = Am (x) k
∂x
We can therefore express the transformation of the components associated with the coordinate transformation
x → x and
∂xj ∂xm ∂xj
Ak (x) = Aj (x) m k
= Aj (x) k
,
∂x ∂x ∂x
which demonstrates the transitive property of a covariant transformation.
We have shown that first order tensors are quantities which obey certain transformation laws. Higher
order tensors are defined in a similar manner and also satisfy the group properties. We assume that we are
given transformations of the type illustrated in equations (1.2.30) and (1.2.32) which are single valued and
continuous with Jacobian J different from zero. Further, the quantities xi and xi , i = 1, . . . , n represent the
coordinates in any two coordinate systems. The following transformation laws define second order and third
order tensors.
48
mn ∂xm ∂xn
A (x) = Aij (x)J W (1.2.49)
∂xi ∂xj
is satisfied, then these quantities are called components of a relative contravariant tensor of
rank or order two with weight W . Whenever W = 0 these quantities are called the components
of an absolute contravariant tensor of rank or order two.
∂xi ∂xj
Amn (x) = Aij (x)J W (1.2.50)
∂xm ∂xn
is satisfied, then these quantities are called components of a relative covariant tensor
of rank or order two with weight W . Whenever W = 0 these quantities are called
the components of an absolute covariant tensor of rank or order two.
m ∂xm ∂xj
An (x) = Aij (x)J W (1.2.51)
∂xi ∂xn
is satisfied, then these quantities are called components of a relative mixed tensor of
rank or order two with weight W . Whenever W = 0 these quantities are called the
components of an absolute mixed tensor of rank or order two. It is contravariant
of order one and covariant of order one.
Higher order tensors are defined in a similar manner. For example, if we can find Ncubed quantities
Am
np such that
i ∂xi ∂xα ∂xβ
Ajk (x) = Aγαβ (x)J W (1.2.52)
∂xγ ∂xj ∂xk
then this is a relative mixed tensor of order three with weight W . It is contravariant of order one and
covariant of order two.
49
General Definition
Tji11ji22...j
...im
n
(1.2.53)
where x
∂x ∂(x1 , x2 , . . . , xN )
J = =
x ∂x ∂(x1 , x2 , . . . , xN )
is the Jacobian of the transformation. When W = 0 the tensor is called an absolute tensor, otherwise it is
called a relative tensor of weight W.
Here superscripts are used to denote contravariant components and subscripts are used to denote covari
ant components. Thus, if we are given the tensor components in one coordinate system, then the components
in any other coordinate system are determined by the transformation law of equation (1.2.54). Throughout
the remainder of this text one should treat all tensors as absolute tensors unless specified otherwise.
A = Ai Ei
This notation can also be generalized to tensor quantities. Higher order tensors can also be denoted by bold
face type. For example the tensor components Tij and Bijk can be represented in terms of the basis vectors
Ei , i = 1, . . . , N by using a notation which is similar to that for the representation of vectors. For example,
T = Tij Ei Ej
B = Bijk Ei Ej Ek .
Here T denotes a tensor with components Tij and B denotes a tensor with components Bijk . The quantities
Ei Ej are called unit dyads and Ei Ej Ek are called unit triads. There is no multiplication sign between the
basis vectors. This notation is called a polyad notation. A further generalization of this notation is the
representation of an arbitrary tensor using the basis and reciprocal basis vectors in bold type. For example,
a mixed tensor would have the polyadic representation
ij...k
T = Tlm...n Ei Ej . . . Ek El Em . . . En .
A dyadic is formed by the outer or direct product of two vectors. For example, the outer product of the
vectors
a = a 1 E 1 + a2 E 2 + a3 E 3 and b = b1 E1 + b2 E2 + b3 E3
50
A = Aij Ei Ej i, j = 1, . . . , N
where the summation convention is in effect for the repeated indices. The coefficients Aij are called the
coefficients of the dyad. When the coefficients are written as an N × N array it is called a matrix. Every
second order tensor can be written as a linear combination of dyads. The dyads form a basis for the second
order tensors. As the example above illustrates, the nine dyads {E1 E1 , E1 E2 , . . . , E3 E3 }, associated with
the outer products of three dimensional base vectors, constitute a basis for the second order tensor A = ab
having the components Aij = ai bj with i, j = 1, 2, 3. Similarly, a triad has the form
where i, j, k have the range 1, 2, . . . , N. The set of outer or direct products { Ei Ej Ek }, with i, j, k = 1, . . . , N
i
constitutes a basis for all third order tensors. Tensor components with mixed suffixes like Cjk are associated
with triad basis of the form
i
C = Cjk Ei Ej Ek
where i, j, k have the range 1, 2, . . . N. Dyads are associated with the outer product of two vectors, while triads,
tetrads,... are associated with higherorder outer products. These higherorder outer or direct products are
referred to as polyads.
The polyad notation is a generalization of the vector notation. The subject of how polyad components
transform between coordinate systems is the subject of tensor calculus.
If a dyadic equals its conjugate A = Ac , then Aij = Aji and the dyadic is called symmetric. If a dyadic
equals the negative of its conjugate A = −Ac , then Aij = −Aji and the dyadic is called skewsymmetric. A
special dyadic called the identical dyadic or idemfactor is defined by
e1 b
J= b e2 b
e1 + b e3 b
e2 + b e3 .
~ produces the
This dyadic has the property that pre or post dot product multiplication of J with a vector V
same vector V~ . For example,
~ · J = (V1 b
V e1 + V2 b
e2 + V3 b
e3 ) · J
= V1 b e1 b
e1 · b e1 + V2 b e2 b
e2 · b e2 + V3 b e3 b
e3 · b ~
e3 = V
~ = J · (V1 b
and J · V e1 + V2 b
e2 + V3 b
e3 )
= V1 b e1 · b
e1 b e1 + V2 b e2 · b
e2 b e2 + V3 b e3 · b
e3 b ~
e3 = V
A dyadic operation often used in physics and chemistry is the double dot product A : B where A and
B are both dyadics. Here both dyadics are expanded using the distributive law of multiplication, and then
ej : b
ei b
each unit dyad pair b em b
en are combined according to the rule
ei b
b em b
ej : b en = ( b
ei · b
em )( b
ej · b
en ).
For example, if A = Aij b
ei b
ej and B = Bij b
ei b
ej , then the double dot product A : B is calculated as follows.
A : B = (Aij b
ei b
ej ) : (Bmn b
em b ei b
en ) = Aij Bmn ( b em b
ej : b en ) = Aij Bmn ( b
ei · b
em )( b
ej · b
en )
= Aij Bmn δim δjn = Amj Bmj
= A11 B11 + A12 B12 + A13 B13
+ A21 B21 + A22 B22 + A23 B23
+ A31 B31 + A32 B32 + A33 B33
When operating with dyads, triads and polyads, there is a definite order to the way vectors and polyad
~ = Ai b
components are represented. For example, for A ~ = Bi b
ei and B ei vectors with outer product
~B
A ~ = Am Bn b
em b
en = φ
there is produced the dyadic φ with components Am Bn . In comparison, the outer product
~A
B ~ = Bm An b
em b
en = ψ
produces the dyadic ψ with components Bm An . That is
φ=A ~B~ =A1 B1 be1 b
e1 + A1 B2 b
e1 b
e2 + A1 B3 b
e1 b
e3
A2 B1 b
e2 b
e1 + A2 B2 b
e2 b
e2 + A2 B3 b
e2 b
e3
A3 B1 b
e3 b
e1 + A3 B2 b
e3 b
e2 + A3 B3 b
e3 b
e3
~A
and ψ = B ~ =B1 A1 b
e1 b
e1 + B1 A2 b
e1 b
e2 + B1 A3 b
e1 b
e3
B2 A1 b
e2 b
e1 + B2 A2 b
e2 b
e2 + B2 A3 b
e2 b
e3
B3 A1 b
e3 b
e1 + B3 A2 b
e3 b
e2 + B3 A3 b
e3 b
e3
are different dyadics.
~ is defined for both pre and post multiplication as
The scalar dot product of a dyad with a vector C
φ·C~ =A ~B~ ·C
~ =A(~ B~ · C)
~
C~ ·φ=C
~ ·A
~B~ =(C
~ · A)
~ B~
These products are, in general, not equal.
52
The following are some important tensor operations which are used to derive special equations and to
prove various identities.
The product of two tensors is also a tensor. The rank or order of the resulting tensor is the sum of
the ranks of the tensors occurring in the multiplication. As an example, let Aijk denote a mixed third order
l
tensor and let Bm denote a mixed second order tensor. The outer product of these two tensors is the fifth
order tensor
il
Cjkm = Aijk Bm
l
, i, j, k, l, m = 1, 2, . . . , N.
i l
Here all indices are free indices as i, j, k, l, m take on any of the integer values 1, 2, . . . , N. Let Ajk and B m
il
denote the components of the given tensors in the barred system of coordinates. We define C jkm as the
il
outer product of these components. Observe that Cjkm is a tensor for by hypothesis Aijk and Bm
l
are tensors
and hence obey the transformation laws
∂xα ∂xj ∂xk
α
Aβγ = Aijk
∂xi ∂xβ ∂xγ (1.2.55)
δ m
δ l ∂x ∂x
B = Bm .
∂xl ∂x
The outer product of these components produces
αδ α δ ∂xα ∂xj ∂xk ∂xδ ∂xm
C βγ = Aβγ B = Aijk Bm
l
∂xi ∂xβ ∂xγ ∂xl ∂x
(1.2.56)
il ∂xα ∂xj ∂xk ∂xδ ∂xm
= Cjkm i
∂x ∂xβ ∂xγ ∂xl ∂x
il
which demonstrates that Cjkm transforms as a mixed fifth order absolute tensor. Other outer products are
analyzed in a similar way.
53
Contraction
The operation of contraction on any mixed tensor of rank m is performed when an upper index is
set equal to a lower index and the summation convention is invoked. When the summation is performed
over the repeated indices the resulting quantity is also a tensor of rank or order (m − 2). For example, let
Aijk , i, j, k = 1, 2, . . . , N denote a mixed tensor and perform a contraction by setting j equal to i. We obtain
Aiik = A11k + A22k + · · · + AN
N k = Ak (1.2.57)
i
where k is a free index. To show that Ak is a tensor, we let Aik = Ak denote the contraction on the
transformed components of Aijk . By hypothesis Aijk is a mixed tensor and hence the components must
satisfy the transformation law
i∂xi ∂xn ∂xp
Ajk = Am
np .
∂xm ∂xj ∂xk
Now execute a contraction by setting j equal to i and perform a summation over the repeated index. We
find
i ∂xi ∂xn ∂xp ∂xn ∂xp
Aik = Ak = Am
np i k
= Am np
m
∂x ∂x ∂x ∂xm ∂xk (1.2.58)
p p
m n ∂x n ∂x ∂xp
= Anp δm k = Anp k = Ap k .
∂x ∂x ∂x
Hence, the contraction produces a tensor of rank two less than the original tensor. Contractions on other
mixed tensors can be analyzed in a similar manner.
New tensors can be constructed from old tensors by performing a contraction on an upper and lower
index. This process can be repeated as long as there is an upper and lower index upon which to perform the
contraction. Each time a contraction is performed the rank of the resulting tensor is two less than the rank
of the original tensor.
Note that in some situations the inner product is performed by employing only subscript indices. For
example, the above inner product is sometimes expressed as
C = Ai Bi = A1 B1 + A2 B2 + · · · AN BN .
This notation is discussed later when Cartesian tensors are considered.
54
Quotient Law
Assume Brqs and Cps are arbitrary absolute tensors. Further assume we have a quantity A(ijk) which
we think might be a third order mixed tensor Aijk . By showing that the equation
is satisfied, then it follows that Arqp must be a tensor. This is an example of the quotient law. Obviously,
this result can be generalized to apply to tensors of any order or rank. To prove the above assertion we shall
show from the above equation that Aijk is a tensor. Let xi and xi denote a barred and unbarred system of
coordinates which are related by transformations of the form defined by equation (1.2.30). In the barred
system, we assume that
r qs s
Aqp B r = C p (1.2.59)
Because Bkin is an arbitrary tensor, the quantity inside the brackets is zero and therefore
r ∂xq ∂xk k ∂x
m
Aqp r − Aim = 0.
i
∂x ∂x ∂xp
∂xi ∂xl
This equation is simplified by inner multiplication by ∂xj ∂xk
to obtain
EXERCISE 1.2
Treat α as a parameter and show this set of transformations constitutes a group by finding the value of α
which:
(i) gives the identity transformation.
(ii) gives the inverse transformation.
(iii) show the transformation is transitive in that a transformation with α = θ1 followed by a transformation
with α = θ2 is equivalent to the transformation using α = θ1 + θ2 .
I 2. Show the transformation
x1 = αx1
Tα :
x2 = α1 x2
forms a group with α as a parameter. Find the value of α such that:
(i) the identity transformation exists.
(ii) the inverse transformation exists.
(iii) the transitive property is satisfied.
I 3. Show the given transformation forms a group with parameter α.
( x1
x1 = 1−αx1
Tα :
x2
x2 = 1−αx1
I 4. Consider the Lorentz transformation from relativity theory having the velocity parameter V, c is the
speed of light and x4 = t is time. 1 1 4
x −V x
x = p
V2
1−
c2
x2 = x2
TV :
x3 = x3
x4 x4 − Vcx2
1
= p 2
1− V2
c
I 5. ~ 1, E
For (E ~ 2, E
~ 3 ) an arbitrary independent basis, (a) Verify that
~1 = 1 E
E ~2 × E
~ 3, ~2 = 1 E
E ~3 × E
~ 1, ~3 = 1 E
E ~1 × E
~2
V V V
~ 1 · (E
is a reciprocal basis, where V = E ~2 × E
~ 3) (b) Show that E ~ i.
~ j = g ij E
56
~ = A1 E
A ~ 1 + A2 E
~ 2 + A3 E
~3
~ 1 + A2 E
~ = A1 E
A ~ 2 + A3 E
~3
~ = Ar êr + Aβ êβ + Az êz
A
~ = A1 E
A ~ 1 + A2 E
~ 2 + A3 E
~3
~ 1 + A2 E
~ = A1 E
A ~ 2 + A3 E
~3
~ = Aρ êρ + Aα êα + Aβ êβ
A
depending upon the basis vectors selected . Calculate, in terms of the coordinates (ρ, α, β) and the
components Ax , Ay , Az
(i) The contravariant components A1 , A2 , A3 .
(ii) The covariant components A1 , A2 , A3 .
(iii) The components Aρ , Aα , Aβ which are called physical components.
I 8. Work the problems 6,7 and then let (x1 , x2 , x3 ) = (r, β, z) denote the coordinates in the cylindrical
system and let (x1 , x2 , x3 ) = (ρ, α, β) denote the coordinates in the spherical system.
(a) Write the transformation equations x → x from cylindrical to spherical coordinates. Also find the
inverse transformations. ( Hint: See the figures 1.24 and 1.25.)
(b) Use the results from part (a) and the results from problems 6,7 to verify that
∂xj
Ai = Aj for i = 1, 2, 3.
∂xi
(c) Use the results from part (a) and the results from problems 6,7 to verify that
i ∂xi
A = Aj for i = 1, 2, 3.
∂xj
~1 = 5 b
V e1 + b
e2 ~2 = b
and V e1 + 5 b
e2
~3 = b
and let V e3 be a unit vector perpendicular to both V~1 and V
~2 . The vectors V
~1 and V~2 can be thought of
as defining an oblique coordinate system, as illustrated in the figure 1.26.
~ 1 , V~ 2 , V~ 3 ).
(a) Find the reciprocal basis (V
(b) Let
e1 + y b
~r = x b e2 + z b
e3 = αV~1 + β V~2 + γ V
~3
x = x(u, v, w)
y = y(u, v, w)
z = z(u, v, w)
~1 = 1 E
E ~2 × E
~ 3, ~2 = 1 E
E ~3 × E
~ 1, ~3 = 1 E
E ~1 × E
~ 2.
V V V
where
V =E ~2 × E
~ 1 · (E ~ 3 ).
~ 1 · (E
Let v = E ~2 × E
~ 3 ) and show that v · V = 1.
I 11. Given the coordinate transformation
x = −u − 2v y = −u − v z=z
~ 1 = ∂~r ,
E ~ 2 = ∂~r ,
E ~ 3 = ∂~r .
E
∂u ∂v ∂z
~ 1, E
Calculate these vectors and then calculate the reciprocal basis E ~ 2, E
~ 3.
(c) With respect to the basis vectors in (b) find the contravariant components Ai associated with the vector
~ = α1 b
A e1 + α2 b
e2 + α3 b
e3
i m
~ j = ∂~r = ∂y b
E ei . From this relation show that ~ m = ∂x b
E ej
∂xj ∂xj ∂y j
and consequently
m m i j
gij = E ~ j = ∂y ∂y ,
~i · E ~ j = ∂x ∂x ,
~i · E
and g ij = E i, j, m = 1, . . . , 3
∂xi ∂xj ∂y m ∂y m
y i = aij xj + bi
where aij and bi are constants and the determinant of aij is different from zero. Show this set of transforma
tions forms a group.
I 15. A surface can be represented using two parameters u, v by introducing the parametric equations
The parameters u, v are called the curvilinear coordinates of a point on the surface. A point on the surface
can be represented by the position vector ~r = ~r(u, v) = x1 (u, v) b
e1 + x2 (u, v) b
e2 + x3 (u, v) b
e3 . The vectors ∂~
r
∂u
∂~
r
and ∂v are tangent vectors to the coordinate surface curves ~r(u, c2 ) and ~r(c1 , v) respectively. An element of
surface area dS on the surface is defined as the area of the elemental parallelogram having the vector sides
∂~
r ∂~
r
∂u du and ∂v dv. Show that
∂~r ∂~r p
dS =  ×  dudv = g11 g22 − (g12 )2 dudv
∂u ∂v
where
∂~r ∂~r ∂~r ∂~r ∂~r ∂~r
g11 = · g12 = · g22 = · .
∂u ∂u ∂u ∂v ∂v ∂v
~ × B)
Hint: (A ~ · (A
~ × B)
~ = A ~ 2 See Exercise 1.1, problem 9(c).
~ × B
I 16.
(a) Use the results from problem 15 and find the element of surface area of the circular cone
I 17. The equation of a plane is defined in terms of two parameters u and v and has the form
xi = αi u + βi v + γi i = 1, 2, 3,
where αi βi and γi are constants. Find the equation of the plane which passes through the points (1, 2, 3),
(14, 7, −3) and (5, 5, 5). What does this problem have to do with the position vector ~r(u, v), the vectors
∂~
r ∂~ r
∂u , ∂v and ~r(0, 0)? Hint: See problem 15.
I 18. Determine the points of intersection of the curve x1 = t, x2 = (t)2 , x3 = (t)3 with the plane
8 x1 − 5 x2 + x3 − 4 = 0.
I 19. ~k = E
Verify the relations V eijk E ~i × E
~j and v −1 eijk E
~k = E
~i × E ~ 1 · (E
~ j where v = E ~2 × E
~ 3 ) and
V =E ~2 × E
~ 1 · (E ~ 3 )..
I 20. Let x̄i and xi , i = 1, 2, 3 be related by the linear transformation x̄i = cij xj , where cij are constants
such that the determinant c = det(cij ) is different from zero. Let γm
n
denote the cofactor of cm
n divided by
the determinant c.
(a) Show that cij γkj = γji cjk = δki .
(b) Show the inverse transformation can be expressed xi = γji x̄j .
(c) Show that if Ai is a contravariant vector, then its transformed components are Āp = cpq Aq .
(d) Show that if Ai is a covariant vector, then its transformed components are Āi = γip Ap .
I 21. Show that the outer product of two contravariant vectors Ai and B i , i = 1, 2, 3 results in a second
order contravariant tensor.
I 24. Let Aij denote an absolute covariant tensor of order 2. Show that the determinant A = det(Aij ) is
p
an invariant of weight 2 and (A) is an invariant of weight 1.
I 25. Let B ij denote an absolute contravariant tensor of order 2. Show that the determinant B = det(B ij )
√
is an invariant of weight −2 and B is an invariant of weight −1.
I 26.
(a) Write out the contravariant components of the following vectors
~1
(i) E ~2
(ii) E ~3
(iii) E where ~ i = ∂~r
E for i = 1, 2, 3.
∂xi
~1
(i) E ~2
(ii) E ~3
(ii) E ~ i = grad xi ,
where E for i = 1, 2, 3.
62
I 27. Let Aij and Aij denote absolute second order tensors. Show that λ = Aij Aij is a scalar invariant.
I 28. Assume that aij , i, j = 1, 2, 3, 4 is a skewsymmetric second order absolute tensor. (a) Show that
is a third order tensor. (b) Show bijk is skewsymmetric in all pairs of indices and (c) determine the number
of independent components this tensor has.
I 29. Show the linear forms A1 x + B1 y + C1 and A2 x + B2 y + C2 , with respect to the group of rotations
and translations x = x cos θ − y sin θ + h and y = x sin θ + y cos θ + k, have the forms A1 x + B 1 y + C 1 and
A2 x + B 2 y + C 2 . Also show that the quantities A1 B2 − A2 B1 and A1 A2 + B1 B2 are invariants.
I 30. Show that the curvature of a curve y = f (x) is κ = ± y 00 (1 + y 02 )−3/2 and that this curvature remains
dy dy dx
invariant under the group of rotations given in the problem 1. Hint: Calculate dx = dx dx .
I 31. Show that when the equation of a curve is given in the parametric form x = x(t), y = y(t), then
ẋÿ − ẏẍ
the curvature is κ = ± 2 and remains invariant under the change of parameter t = t(t), where
(ẋ + ẏ 2 )3/2
ẋ = dx
dt , etc.
I 33. Let φ = φ(x1 , x2 , . . . , xN ) denote an absolute scalar invariant. (a) Is the quantity ∂φ
∂xi a tensor? (b)
2
∂ φ
Is the quantity ∂xi ∂xj a tensor?
I 34. Consider the second order absolute tensor aij , i, j = 1, 2 where a11 = 1, a12 = 2, a21 = 3 and a22 = 4.
Find the components of aij under the transformation of coordinates x1 = x1 + x2 and x2 = x1 − x2 .
I 35. Let Ai , Bi denote the components of two covariant absolute tensors of order one. Show that
Cij = Ai Bj is an absolute second order covariant tensor.
I 36. Let Ai denote the components of an absolute contravariant tensor of order one and let Bi denote the
components of an absolute covariant tensor of order one, show that Cji = Ai Bj transforms as an absolute
mixed tensor of order two.
I 37. (a) Show the sum and difference of two tensors of the same kind is also a tensor of this kind. (b) Show
that the outer product of two tensors is a tensor. Do parts (a) (b) in the special case where one tensor Ai
is a relative tensor of weight 4 and the other tensor Bkj is a relative tensor of weight 3. What is the weight
of the outer product tensor Tkij = Ai Bkj in this special case?
I 39. Let Aij denote the components of an absolute mixed tensor of order two. Show that the scalar
contraction S = Aii is an invariant.
63
I 40. Let Ai = Ai (x1 , x2 , . . . , xN ) denote the components of an absolute contravariant tensor. Form the
∂Ai
quantity Bji = ∂xj and determine if Bji transforms like a tensor.
∂Ai ∂Aj
I 41. Let Ai denote the components of a covariant vector. (a) Show that aij = j
− are the
∂x ∂xi
∂aij ∂ajk ∂aki
components of a second order tensor. (b) Show that + + = 0.
∂xk ∂xi ∂xj
I 42. Show that xi = K eijk Aj Bk , with K 6= 0 and arbitrary, is a general solution of the system of equations
Ai xi = 0, Bi xi = 0, i = 1, 2, 3. Give a geometric interpretation of this result in terms of vectors.
A : Ac = A11 A11 + A12 A21 + A13 A31 + A21 A12 + A22 A22 + A23 A32 + A31 A13 + A32 A23 + A23 A33
I 45. ~ = Ai b
Let A ~ = Bi b
ei , B ~ = Ci b
ei , C ~ = Di b
ei , D ~ B,
ei denote vectors and let φ = A ~ ψ =C
~D~ denote
dyadics which are the outer products involving the above vectors. Show that the double dot product satisfies
~B
φ:ψ=A ~ :C
~D~ = (A
~ · C)(
~ B ~ · D)
~
I 46. Show that if aij is a symmetric tensor in one coordinate system, then it is symmetric in all coordinate
systems.
I 47. Write the transformation laws for the given tensors. (a) Akij (b) Aij
k (c) Aijk
m
∂xj ∂xj
I 48. Show that if Ai = Aj i , then Ai = Aj ∂xi . Note that this is equivalent to interchanging the bar
∂x
and unbarred systems.
I 49.
(a) Show that under the linear homogeneous transformation
x1 =a11 x1 + a21 x2
x2 =a12 x1 + a22 x2
Q(x1 , x2 ) = g11 (x1 )2 + 2g12 x1 x2 + g22 (x2 )2 becomes Q(x1 , x2 ) = g11 (x1 )2 + 2g12 x1 x2 + g 22 (x2 )2
where g ij = g11 aj1 ai1 + g12 (ai1 aj2 + aj1 ai2 ) + g22 ai2 aj2 .
(b) Show F = g11 g22 − (g12 )2 is a relative invariant of weight 2 of the quadratic form Q(x1 , x2 ) with respect
to the group of linear homogeneous transformations. i.e. Show that F = ∆2 F where F = g 11 g22 −(g12 )2
and ∆ = (a11 a22 − a21 a12 ).
64
I 50. Let ai and bi for i = 1, . . . , n denote arbitrary vectors and form the dyadic
Φ = a1 b1 + a2 b2 + · · · + an bn .
φ1 = a1 · b1 + a2 · b2 + · · · + an · bn
where a dot product operator has been placed between the vectors. The first vector invariant of Φ is defined
~ = a1 × b1 + a2 × b2 + · · · + an × bn
φ
where a vector cross product operator has been placed between the vectors.
(a) Show that the first scalar and vector invariant of
e1 b
Φ= b e2 b
e2 + b e3 b
e3 + b e3
I 51. Let Φ denote the dyadic given in problem 50. The dyadic Φ2 defined by
1X
Φ2 = ai × aj bi × bj
2 i,j
is called the Gibbs second dyadic of Φ, where the summation is taken over all permutations of i and j. When
i = j the dyad vanishes. Note that the permutations i, j and j, i give the same dyad and so occurs twice
in the final sum. The factor 1/2 removes this doubling. Associated with the Gibbs dyad Φ2 are the scalar
invariants
1X
φ2 = (ai × aj ) · (bi × bj )
2 i,j
1X
φ3 = (ai × aj · ak )(bi × bj · bk )
6
i,j,k
Show that the dyad
Φ = as + tq + cu
has
the first scalar invariant φ1 = a · s + b · t + c · u
~ = a×s+b×t+c×u
the first vector invariant φ
Gibbs second dyad Φ2 = b × ct × u + c × au × s + a × bs × t
second scalar of Φ φ2 = (b × c) · (t · u) + (c × a) · (u × s) + (a × b) · (s × t)
third scalar of Φ φ3 = (a × b · c)(s × t · u)
65
I 52. (Spherical Trigonometry) Construct a spherical triangle ABC on the surface of a unit sphere with
sides and angles less than 180 degrees. Denote by a,b c the unit vectors from the origin of the sphere to the
vertices A,B and C. Make the construction such that a · (b × c) is positive with a, b, c forming a righthanded
system. Let α, β, γ denote the angles between these unit vectors such that
The great circles through the vertices A,B,C then make up the sides of the spherical triangle where side α
is opposite vertex A, side β is opposite vertex B and side γ is opposite the vertex C. The angles A,B and C
between the various planes formed by the vectors a, b and c are called the interior dihedral angles of the
spherical triangle. Note that the cross products
define unit vectors a, b and c perpendicular to the planes determined by the unit vectors a, b and c. The
dot products
a · b = cos γ b · c = cos α c · a = cos β (3)
define the angles α,β and γ which are called the exterior dihedral angles at the vertices A,B and C and are
such that
α=π−A β =π−B γ = π − C. (4)
(a) Using appropriate scaling, show that the vectors a, b, c and a, b, c form a reciprocal set.
(b) Show that a · (b × c) = sin α a · a = sin β b · b = sin γ c · c
(c) Show that a · (b × c) = sin α a · a = sin β b · b = sin γ c · c
(d) Using parts (b) and (c) show that
sin α sin β sin γ
= =
sin α sin β sin γ
(e) Use the results from equation (4) to derive the law of sines for spherical triangles
sin α sin β sin γ
= =
sin A sin B sin C
(f) Using the equations (2) show that
(g) Using part (f) derive the law of cosines for spherical triangles
cos α = cos β cos γ + sin β sin γ cos A
cos A = − cos B cos C + sin B sin C cos α
A cyclic permutation of the symbols produces similar results involving the other angles and sides of the
spherical triangle.
65
Knowing how tensors are defined and recognizing a tensor when it pops up in front of you are two
different things. Some quantities, which are tensors, frequently arise in applied problems and you should
learn to recognize these special tensors when they occur. In this section some important tensor quantities
are defined. We also consider how these special tensors can in turn be used to define other tensors.
Metric Tensor
Assume that the coordinates y i are related to a set of independent generalized coordinates xi , i = 1, . . . , N
by a set of transformation equations
y i = y i (x1 , x2 , . . . , xN ), i = 1, . . . , N. (1.3.2)
To emphasize that each y i depends upon the x coordinates we sometimes use the notation y i = y i (x), for
i = 1, . . . , N. The differential of each coordinate can be written as
∂y m j
dy m = dx , m = 1, . . . , N, (1.3.3)
∂xj
and consequently in the xgeneralized coordinates the distance squared, found from the equation (1.3.1),
becomes a quadratic form. Substituting equation (1.3.3) into equation (1.3.1) we find
∂y m ∂y m i j
ds2 = dx dx = gij dxi dxj (1.3.4)
∂xi ∂xj
where
∂y m ∂y m
gij = , i, j = 1, . . . , N (1.3.5)
∂xi ∂xj
are called the metrices of the space defined by the coordinates xi , i = 1, . . . , N. Here the gij are functions of
the x coordinates and is sometimes written as gij = gij (x). Further, the metrices gij are symmetric in the
indices i and j so that gij = gji for all values of i and j over the range of the indices. If we transform to
another coordinate system, say xi , i = 1, . . . , N , then the element of arc length squared is expressed in terms
of the barred coordinates and ds2 = g ij dxi dxj , where gij = g ij (x) is a function of the barred coordinates.
The following example demonstrates that these metrices are second order covariant tensors.
66
EXAMPLE 1.31. Show the metric components gij are covariant tensors of the second order.
Solution: In a coordinate system xi , i = 1, . . . , N the element of arc length squared is
while in a coordinate system xi , i = 1, . . . , N the element of arc length squared is represented in the form
Here it is assumed that there exists a coordinate transformation of the form defined by equation (1.2.30)
together with an inverse transformation, as in equation (1.2.32), which relates the barred and unbarred
coordinates. In general, if xi = xi (x), then for i = 1, . . . , N we have
∂xi ∂xj
dxi = dxm and dxj = dxn (1.3.9)
∂xm ∂xn
∂xi ∂xj
For arbitrary changes in dxm this equation implies that g mn = gij and consequently gij transforms
∂xm ∂xn
as a second order absolute covariant tensor.
EXAMPLE 1.32. (Curvilinear coordinates) Consider a set of general transformation equations from
rectangular coordinates (x, y, z) to curvilinear coordinates (u, v, w). These transformation equations and the
corresponding inverse transformations are represented
x = x(u, v, w) u = u(x, y, z)
y = y(u, v, w) v = v(x, y, z) (1.3.10)
z = z(u, v, w). w = w(x, y, z)
~ 1 = ∂~r =
E
∂x
b
e1 +
∂y
b
e2 +
∂z
b
e3
∂u ∂u ∂u ∂u
E~ 2 = ∂~r = ∂x
b
e1 +
∂y
b
e2 +
∂z
be3 (1.3.11)
∂v ∂v ∂v ∂v
~3 = r =
E
∂~ ∂x
b
e1 +
∂y
b
e2 +
∂z
b
e3 .
∂w ∂w ∂w ∂w
are tangent vectors to the coordinate curves. The element of arc length in the curvilinear coordinates is
ds2 = E
~i · E
~ j dxi dxj = gij dxi dxj , i, j = 1, 2, 3
where
m m
gij = E ~ j = ∂~r · ∂~r = ∂y ∂y ,
~i · E i, j free indices (1.3.13)
∂xi ∂xj ∂xi ∂xj
are called the metric components of the curvilinear coordinate system. The metric components may be
thought of as the elements of a symmetric matrix, since gij = gji . In the rectangular coordinate system
x, y, z, the element of arc length squared is ds2 = dx2 + dy 2 + dz 2 . In this space the metric components are
1 0 0
gij = 0 1 0.
0 0 1
68
~ 1 = ∂~r = cos θ b
E e1 + sin θ b
e2 , ~ 2 = ∂~r = −r sin θ b
E e1 + r cos θ b
e2 , ~ 3 = ∂~r = b
E e3 .
∂r ∂θ ∂z
From the results in equation (1.3.13), the metric components of this space are
1 0 0
gij = 0 r2 0.
0 0 1
Given a set of transformations of the form found in equation (1.3.10), one can readily determine the
metric components associated with the generalized coordinates. For future reference we list several differ
ent coordinate systems together with their metric components. Each of the listed coordinate systems are
orthogonal and so gij = 0 for i 6= j. The metric components of these orthogonal systems have the form
h21 0 0
gij = 0 h22 0
0 0 h23
x = r cos θ r≥0 h1 = 1
y = r sin θ 0 ≤ θ ≤ 2π h2 = r
z=z −∞<z <∞ h3 = 1
The coordinate curves, illustrated in the figure 1.31, are formed by the intersection of the coordinate
surfaces
x2 + y 2 = r2 , Cylinders
y/x = tan θ Planes
z = Constant Planes.
The coordinate curves, illustrated in the figure 1.32, are formed by the intersection of the coordinate
surfaces
x2 + y 2 + z 2 = ρ2 Spheres
x2 + y 2 = tan2 θ z 2 Cones
y = x tan φ Planes.
4. Parabolic cylindrical coordinates (ξ, η, z)
p
x = ξη −∞<ξ <∞ h1 = ξ 2 + η2
1 p
y = (ξ 2 − η 2 ) −∞<z <∞ h2 = ξ 2 + η 2
2
z=z η≥0 h3 = 1
70
The coordinate curves, illustrated in the figure 1.33, are formed by the intersection of the coordinate
surfaces
ξ2
x2 = −2ξ 2 (y − ) Parabolic cylinders
2
η2
x2 = 2η 2 (y + ) Parabolic cylinders
2
z = Constant Planes.
The coordinate curves, illustrated in the figure 1.34, are formed by the intersection of the coordinate
surfaces
ξ2
x2 + y 2 = −2ξ 2 (z − ) Paraboloids
2
η2
x2 + y 2 = 2η 2 (z + ) Paraboloids
2
y = x tan φ Planes.
The coordinate curves, illustrated in the figure 1.35, are formed by the intersection of the coordinate
surfaces
x2 y2
+ =1 Elliptic cylinders
cosh2 ξ sinh2 ξ
x2 y2
− =1 Hyperbolic cylinders
cos2 η sin2 η
z = Constant Planes.
72
The coordinate curves, illustrated in the figure 1.36, are formed by the intersection of the coordinate
surfaces
x2 y2 z2
+ + =1 Prolate ellipsoid
ξ2 − 1 ξ2 − 1 ξ2
z2 x2 y2
− − =1 Twosheeted hyperboloid
η2 1 − η2 1 − η2
y = x tan φ Planes
8. Bipolar coordinates (u, v, z)
The coordinate curves, illustrated in the figure 1.37, are formed by the intersection of the coordinate
surfaces
a2
(x − a coth v)2 + y 2 = Cylinders
sinh2 v
a2
x2 + (y − a cot u)2 = Cylinders
sin2 u
z = Constant Planes.
74
The coordinate curves, illustrated in the figure 1.38, are formed by the intersection of the coordinate
surfaces
x2 + y 2 + z 2 = u2 Spheres
2 2 2
x y z
+ 2 + 2 =0, Cones
v2 v − a2 v − b2
x2 y2 z2
+ + = 0, Cones.
w2 w 2 − a2 w 2 − b2
The coordinate curves, illustrated in the figure 1.39, are formed by the intersection of the coordinate
surfaces
x2 y2 z2
2
+ 2
+ = 1, Prolate ellipsoids
(a sinh u) (a sinh u) (a cosh u)2
z2 x2 y2
2
− 2
− = 1, Twosheeted hyperboloid
(a cos v) (a sin v) (a sin v)2
y = x tan φ, Planes.
75
The coordinate curves, illustrated in the figure 1.310, are formed by the intersection of the coordinate
surfaces
x2 y2 z2
2
+ 2
+ = 1, Oblate ellipsoids
(a cosh ξ) (a cosh ξ) (a sinh ξ)2
x2 y2 z2
2
+ 2
− = 1, Onesheet hyperboloids
(a cos η) (a cos η) (a sin η)2
y = x tan φ, Planes.
12. Toroidal coordinates (u, v, φ)
x=
a sinh v cos φ
, 0 ≤ u < 2π h21 = h22
cosh v − cos u a2
a sinh v sin φ h22 =
y= , −∞ < v < ∞ (cosh v − cos u)2
cosh v − cos u
a sin u a2 sinh2 v
z= , 0 ≤ φ < 2π h23 =
cosh v − cos u (cosh v − cos u)2
The coordinate curves, illustrated in the figure 1.311, are formed by the intersection of the coordinate
surfaces a cos u 2 a2
x2 + y 2 + z − = , Spheres
sin u sin2 u
p 2
cosh v a2
x2 + y 2 − a + z2 = , Torus
sinh v sinh2 v
y = x tan φ, planes
76
EXAMPLE 1.34. Show the Kronecker delta δji is a mixed second order tensor.
Solution: Assume we have a coordinate transformation xi = xi (x), i = 1, . . . , N of the form (1.2.30) and
i
possessing an inverse transformation of the form (1.2.32). Let δ j and δji denote the Kronecker delta in the
barred and unbarred system of coordinates. By definition the Kronecker delta is defined
i 0, if i 6= j
δj = δji = .
1, if i=j
77
Let g denote the determinant of the matrix having the metric tensor gij , i, j = 1, . . . , N as its elements.
In our study of cofactor elements of a matrix we have shown that
cof (g1j )g1k + cof (g2j )g2k + . . . + cof (gN j )gN k = gδkj . (1.3.15)
We can use this fact to find the elements in the inverse matrix associated with the matrix having the
components gij . The elements of this inverse matrix are
1
g ij = cof (gij ) (1.3.16)
g
and are called the conjugate metric components. We examine the summation g ij gik and find:
The equation
g ij gik = δkj (1.3.17)
is an example where we can use the quotient law to show g ij is a second order contravariant tensor. Because
of the symmetry of g ij and gij the equation (1.3.17) can be represented in other forms.
EXAMPLE 1.35. Let Ai and Ai denote respectively the covariant and contravariant components of a
~ Show these components are related by the equations
vector A.
Ai = gij Aj (1.3.18)
k jk
A = g Aj (1.3.19)
where gij and g ij are the metric and conjugate metric components of the space.
78
Solution: We multiply the equation (1.3.18) by g im (inner product) and use equation (1.3.17) to simplify
the results. This produces the equation g im Ai = g im gij Aj = δjm Aj = Am . Changing indices produces the
result given in equation (1.3.19). Conversely, if we start with equation (1.3.19) and multiply by gkm (inner
product) we obtain gkm Ak = gkm g jk Aj = δm
j
Aj = Am which is another form of the equation (1.3.18) with
the indices changed.
Notice the consequences of what the equations (1.3.18) and (1.3.19) imply when we are in an orthogonal
Cartesian coordinate system where
1 0 0 1 0 0
gij = 0 1 0 and g ij = 0 1 0.
0 0 1 0 0 1
EXAMPLE 1.36. We have previously shown that if Ai is a covariant tensor of rank 1 its components in
a barred system of coordinates are
∂xj
Ai = Aj . (1.3.20)
∂xi
Solve for the Aj in terms of the Aj . (i.e. find the inverse transformation).
∂xi
Solution: Multiply equation (1.3.20) by ∂xm (inner product) and obtain
∂xi j
Ai = Aj δm = Am (1.3.22)
∂xm
Associated Tensors
Associated tensors can be constructed by taking the inner product of known tensors with either the
metric or conjugate metric tensor.
Associated tensors are different ways of representing a tensor. The multiplication of a tensor by the
metric or conjugate metric tensor has the effect of lowering or raising indices. For example the covariant
and contravariant components of a vector are different representations of the same vector in different forms.
These forms are associated with one another by way of the metric and conjugate metric tensor and
g ij Ai = Aj gij Aj = Ai .
Aj = g ij Ai Aj = gij Ai
Am
.jk = g
mi
Aijk Ai.k
m = gmj A
ijk
A.nm
i.. = g
mk nj
g Aijk Amjk = gim Ai.jk
Sometimes ‘dots’are used as indices in order to represent the location of the index that was raised or lowered.
If a tensor is symmetric, the position of the index is immaterial and so a dot is not needed. For example, if
Amn is a symmetric tensor, then it is easy to show that An.m and A.n
m are equal and therefore can be written
as Anm without confusion.
Higher order tensors are similarly related. For example, if we find a fourth order covariant tensor Tijkm
we can then construct the fourth order contravariant tensor T pqrs from the relation
T pqrs = g pi g qj g rk g sm Tijkm .
This fourth order tensor can also be expressed as a mixed tensor. Some mixed tensors associated with
the given fourth order covariant tensor are:
p pq p
T.jkm = g pi Tijkm , T..km = g qj T.jkm .
80
Riemann Space VN
A Riemannian space VN is said to exist if the element of arc length squared has the form
where the metrices gij = gij (x1 , x2 , . . . , xN ) are continuous functions of the coordinates and are different
from constants. In the special case gij = δij the Riemannian space VN reduces to a Euclidean space EN .
The element of arc length squared defined by equation (1.3.23) is called the Riemannian metric and any
geometry which results by using this metric is called a Riemannian geometry. A space VN is called flat if
it is possible to find a coordinate transformation where the element of arclength squared is ds2 = i (dxi )2
where each i is either +1 or −1. A space which is not flat is called curved.
Geometry in VN
~·B
A ~i · E
~ = Ai B j E ~ j = gij Ai B j = Aj B j = Ai Bi = g ij Aj Bi = A
~ B~ cos θ. (1.3.24)
~ and B
Consequently, in an N dimensional Riemannian space VN the dot or inner product of two vectors A ~
is defined:
gij Ai B j = Aj B j = Ai Bi = g ij Aj Bi = AB cos θ. (1.3.25)
In this definition A is the magnitude of the vector Ai , the quantity B is the magnitude of the vector Bi and
θ is the angle between the vectors when their origins are made to coincide. In the special case that θ = 90◦
we have gij Ai B j = 0 as the condition that must be satisfied in order that the given vectors Ai and B i are
orthogonal to one another. Consider also the special case of equation (1.3.25) when Ai = B i and θ = 0. In
this case the equations (1.3.25) inform us that
From this equation one can determine the magnitude of the vector Ai . The magnitudes A and B can be
1 1
written A = (gin Ai An ) 2 and B = (gpq B p B q ) 2 and so we can express equation (1.3.24) in the form
gij Ai B j
cos θ = 1 . (1.3.27)
(gmn Am An ) 2 (g p B q ) 12
pq B
An import application of the above concepts arises in the dynamics of rigid body motion. Note that if a
dAi
vector Ai has constant magnitude and the magnitude of dt is different from zero, then the vectors Ai and
i j
dA
dt must be orthogonal to one another due to the fact that gij Ai dA
dt = 0. As an example, consider the unit
e1 , b
vectors b e2 and b
e3 on a rotating system of Cartesian axes. We have for constants ci , i = 1, 6 that
The above definition of a dot product in VN can be used to define unit vectors in VN .
Associated with the e−permutation symbols there are the epsilon permutation symbols defined by the
relations
√ 1
ijk = geijk and ijk = √ eijk (1.3.30)
g
where g is the determinant of the metrices gij .
It can be demonstrated that the eijk permutation symbol is a relative tensor of weight −1 whereas the
ijk permutation symbol is an absolute tensor. Similarly, the eijk permutation symbol is a relative tensor of
weight +1 and the corresponding ijk permutation symbol is an absolute tensor.
EXAMPLE 1.310. ( permutation symbol)
Show that eijk is a relative tensor of weight −1 and the corresponding ijk permutation symbol is an
absolute tensor.
Solution: Examine the Jacobian 1
∂x1 ∂x1 ∂x1
x ∂x ∂x2 ∂x3
2 ∂x2 ∂x2
J = ∂x
x ∂x13 ∂x2 ∂x3
∂x1 ∂x3 ∂x3
∂x ∂x2 ∂x3
and make the substitution
∂xi
aij = , i, j = 1, 2, 3.
∂xj
From the definition of a determinant we may write
x
eijk aim ajn akp = J( )emnp . (1.3.31)
x
By definition, emnp = emnp in all coordinate systems and hence equation (1.3.31) can be expressed in the
form h x i−1 ∂xi ∂xj ∂xk
J( ) eijk m n p = emnp (1.3.32)
x ∂x ∂x ∂x
which demonstrates that eijk transforms as a relative tensor of weight −1.
We have previously shown the metric tensor gij is a second order covariant tensor and transforms
∂xm ∂xn
according to the rule gij = gmn . Taking the determinant of this result we find
∂xi ∂xj
m 2 h x i2
∂x
g = gij  = gmn  i = g J( ) (1.3.33)
∂x x
where g is the determinant of (gij ) and g is the determinant of (g ij ). This result demonstrates that g is a
scalar invariant of weight +2. Taking the square root of this result we find that
p √ x
g= gJ( ). (1.3.34)
x
√
Consequently, we call g a scalar invariant of weight +1. Now multiply both sides of equation (1.3.32) by
√
g and use (1.3.34) to verify the relation
In a similar manner one can show eijk is a relative tensor of weight +1 and ijk = √1 eijk is an absolute
g
tensor. This is left as an exercise.
Another exercise found at the end of this section is to show that a generalization of the e − δ identity
is the epsilon identity
g ij ipt jrs = gpr gts − gps gtr . (1.3.36)
Cartesian Tensors
Consider the motion of a rigid rod in two dimensions. No matter how complicated the movement of
the rod is we can describe the motion as a translation followed by a rotation. Consider the rigid rod AB
illustrated in the figure 1.313.
In this figure there is a before and after picture of the rod’s position. By moving the point B to B 0 we
have a translation. This is then followed by a rotation holding B fixed.
85
A similar situation exists in three dimensions. Consider two sets of Cartesian axes, say a barred and
unbarred system as illustrated in the figure 1.314. Let us translate the origin 0 to 0 and then rotate the
(x, y, z) axes until they coincide with the (x, y, z) axes. We consider first the rotation of axes when the
origins 0 and 0 coincide as the translational distance can be represented by a vector bk , k = 1, 2, 3. When
the origin 0 is translated to 0 we have the situation illustrated in the figure 1.315, where the barred axes
can be thought of as a transformation due to rotation.
Let
e1 + y b
~r = x b e2 + z b
e3 (1.3.37)
denote the position vector of a variable point P with coordinates (x, y, z) with respect to the origin 0 and the
e1 , b
unit vectors b e2 , b
e3 . This same point, when referenced with respect to the origin 0 and the unit vectors
ê1 , ê2 , ê3 , has the representation
~r = x ê1 + y ê2 + z ê3 . (1.3.38)
By considering the projections of ~r upon the barred and unbarred axes we can construct the transformation
equations relating the barred and unbarred axes. We calculate the projections of ~r onto the x, y and z axes
and find:
e1 = x = x( ê1 · b
~r · b e1 ) + y( ê2 · b
e1 ) + z( ê3 · b
e1 )
e2 = y = x( ê1 · b
~r · b e2 ) + y( ê2 · b
e2 ) + z( ê3 · b
e2 ) (1.3.39)
e3 = z = x( ê1 · b
~r · b e3 ) + y( ê2 · b
e3 ) + z( ê3 · b
e3 ).
We also calculate the projection of ~r onto the x, y, z axes and find:
~r · ê1 = x = x( b
e1 · ê1 ) + y( b
e2 · ê1 ) + z( b
e3 · ê1 )
~r · ê2 = y = x( b
e1 · ê2 ) + y( b
e2 · ê2 ) + z( b
e3 · ê2 ) (1.3.40)
~r · ê3 = z = x( b
e1 · ê3 ) + y( b
e2 · ê3 ) + z( b
e3 · ê3 ).
By introducing the notation (y1 , y2 , y3 ) = (x, y, z) (y 1 , y2 , y3 ) = (x, y, z) and defining θij as the angle
between the unit vectors b
ei and êj , we can represent the above transformation equations in a more concise
86
`11 = b
e1 · ê1 = cos θ11 `12 = b
e1 · ê2 = cos θ12 `13 = b
e1 · ê3 = cos θ13
`21 = b
e2 · ê1 = cos θ21 `22 = b
e2 · ê2 = cos θ22 `23 = b
e2 · ê3 = cos θ23 (1.3.41)
`31 = b
e3 · ê1 = cos θ31 `32 = b
e3 · ê2 = cos θ32 `33 = b
e3 · ê3 = cos θ33
which enables us to write the equations (1.3.39) and (1.3.40) in the form
êr = `pr b
ep or b
ep = `pr êr (1.3.43)
where `pr are the direction cosines. In both the barred and unbarred system the unit vectors are orthogonal
and consequently we must have the dot products
where δij is the Kronecker delta. Substituting equation (1.3.43) into equation (1.3.44) we find the direction
cosines `ij must satisfy the relations:
The relations
`mr `ms = δrs and `rm `sm = δrs , (1.3.45)
with summation index m, are important relations which are satisfied by the direction cosines associated with
a rotation of axes.
Combining the rotation and translation equations we find
yi = `ij y j + bi . (1.3.46)
 {z } {z}
rotation translation
We multiply this equation by `ik and make use of the relations (1.3.45) to find the inverse transformation
where ~ω = ωi bei is called an angular velocity vector of the rotating system. The term ~ω × A ~ represents the
dA~ i
velocity of the rotating system relative to the fixed system and = dA b ei represents the derivative with
dt r dt
respect to the rotating system.
Employing the special transformation equations (1.3.46) let us examine how tensor quantities transform
when subjected to a translation and rotation of axes. These are our special transformation laws for Cartesian
tensors. We examine only the transformation laws for first and second order Cartesian tensor as higher order
transformation laws are easily discerned. We have previously shown that in general the first and second order
tensor quantities satisfy the transformation laws:
∂yj
Ai = Aj (1.3.48)
∂y i
i ∂y
A = Aj i (1.3.49)
∂yj
mn ∂y ∂y
A = Aij m n (1.3.50)
∂yi ∂yj
∂yi ∂yj
Amn = Aij (1.3.51)
∂y m ∂yn
m ∂y ∂yj
An = Aij m (1.3.52)
∂yi ∂y n
For the special case of Cartesian tensors we assume that yi and y i , i = 1, 2, 3 are linearly independent. We
differentiate the equations (1.3.46) and (1.3.47) and find
∂yi ∂yj ∂y k ∂yi
= `ij = `ij δjk = `ik , and = `ik = `ik δim = `mk .
∂yk ∂yk ∂ym ∂ym
Substituting these derivatives into the transformation equations (1.3.48) through (1.3.52) we produce the
transformation equations
Ai = Aj `ji
i
A = Aj `ji
mn
A = Aij `im `jn
Amn = Aij `im `jn
m
An = Aij `im `jn .
88
These are the transformation laws when moving from one orthogonal system to another. In this case the
direction cosines `im are constants and satisfy the relations given in equation (1.3.45). The transformation
laws for higher ordered tensors are similar in nature to those given above.
In the unbarred system (y1 , y2 , y3 ) the metric tensor and conjugate metric tensor are:
where δij is the Kronecker delta. In the barred system of coordinates, which is also orthogonal, we have
∂ym ∂ym
g ij = .
∂y i ∂yj
Ai = g ij Aj Ai = gij Aj
Aij = g im g jn Amn Amn = gmi gnj Aij
Ain = g im Amn Ain = gnj Aij
and find that the contravariant and covariant components are identical to one another. This holds also in
the barred system of coordinates. Also note that these special circumstances allow the representation of
contractions using subscript quantities only. This type of a contraction is not allowed for general tensors. It
is left as an exercise to try a contraction on a general tensor using only subscripts to see what happens. Note
that such a contraction does not produce a tensor. These special situations are considered in the exercises.
Physical Components
~ as
In the Cartesian coordinate system we can represent a vector A
~ = Ax b
A e1 + Ay b
e2 + Az b
e3
where ( b
e1 , b
e2 , b
e3 ) are the basis vectors. Consider a coordinate transformation to a more general coordinate
system, say (x , x2 , x3 ). The vector A
1 ~ can be represented with contravariant components as
~ = A1 E
A ~ 1 + A2 E
~ 2 + A3 E
~3 (1.3.53)
~ 1, E
having covariant components with respect to the gradient basis vectors (E ~ 2, E
~ 3 ). These equations are
just different ways of representing the same vector. In the above representations the basis vectors need not
be orthogonal and they need not be unit vectors. In general, the physical dimensions of the components Ai
and Aj are not the same.
The physical components of the vector A ~ in a direction is defined as the projection of A ~ upon a unit
vector in the desired direction. For example, the physical component of A ~ in the direction E
~ 1 is
~
~ · E1 = A1 = projection of A
A ~ on E
~ 1. (1.3.58)
~ 1
E ~ 1
E
~ 1 is
~ in the direction E
Similarly, the physical component of A
~1 1
~ · E = A = projection of A
A ~ 1.
~ on E (1.3.59)
~ 1
E ~ 1
E
EXAMPLE 1.311. (Physical components) Let α, β, γ denote nonzero positive constants such that the
product relation αγ = 1 is satisfied. Consider the nonorthogonal basis vectors
~1 = α b
E e1 , ~2 = β b
E e1 + γ b
e2 , ~3 = b
E e3
~1 = γ b
E e1 − β b
e2 , ~2 = αb
E e2 , ~3 = b
E e3 .
~ = Ax b
Consider the problem of representing the vector A e1 + Ay b
e2 in the contravariant vector form
~ = A1 E
A ~ 1 + A2 E
~2 or tensor form Ai , i = 1, 2.
A1 = A ~ 1 = γAx − βAy
~ ·E and A2 = A ~ 2 = αAy .
~ ·E
A ~ 1 + A2 E
~ = A1 E ~2 which has the tensor form Ai , i = 1, 2.
~ ·E
A1 = A ~ 1 = αAx ~ ·E
A2 = A ~ 2 = βAx + γAy .
~ 1 and E
~ in the directions E
The physical components of A ~ 2 are found to be:
~1 1
x − βAy
~ · E = A = γA
A p = A(1)
~ 1
E  ~ 1
E  γ2 + β2
~2 2
~ · E = A = αAy = Ay = A(2).
A
~ 2
E ~ 2
E α
~
Note that these same results are obtained from the dot product relations using either form of the vector A.
For example, we can write
~1 ~1 ~1 ~2 ~1
~ · E = A1 (E · E ) + A2 (E · E ) = A(1)
A
~ 1
E E~ 1
~2 ~1 ~2 ~2 ~2
and ~ · E = A1 (E · E ) + A2 (E · E ) = A(2).
A
~ 2
E E~ 2
~ in direction of λi
gij Ai λj = Ai λi = Ai λi = projection of A
91
ds2 = gij dxi dxj = (h1 )2 (dx1 )2 + (h2 )2 (dx2 )2 + (h3 )2 (dx3 )2
where
(h1 )2 0 0
gij = 0 (h2 )2 0 . (1.3.60)
0 0 (h3 )2
In this case the curvilinear coordinates are orthogonal and
At an arbitrary point in this coordinate system we take λi , i = 1, 2, 3 as a unit vector in the direction
of the coordinate x1 . We then obtain
dx1
λ1 = , λ2 = 0, λ3 = 0.
ds
In summary, we can say that in an orthogonal coordinate system the physical components of a contravariant
tensor of order one can be determined from the equations
√
A(i) = h(i) A(i) = g(i)(i) A(i) , i = 1, 2 or 3 no summation on i,
which is a short hand notation for the physical components (h1 A1 , h2 A2 , h3 A3 ). In an orthogonal coordinate
system the nonzero conjugate metric components are
1
g (i)(i) = , i = 1, 2, or 3 no summation on i.
g(i)(i)
92
These components are needed to calculate the physical components associated with a covariant tensor of
order one. For example, in the x1 −direction, we have the covariant components
1
λ1 = g11 λ1 = h21 = h1 , λ2 = 0, λ3 = 0
h1
and consequently the projection in V3 can be represented
1 A1
gij Ai λj = gij Ai g jm λm = Aj g jm λm = A1 λ1 g 11 = A1 h1 = = A(1).
h21 h1
for the other physical components in the directions x2 and x3 . These physical components can be represented
in the short hand notation
A(i) A(i)
A(i) = =√ , i = 1, 2 or 3 no summation on i.
h(i) g(i)(i)
In an orthogonal coordinate system the physical components associated with both the contravariant and
covariant components are the same. To show this we note that when Ai gij = Aj is summed on i we obtain
√ A(i)
A(i) = A(i) g(i)(i) = √ i not summed,
g(i)(i)
which demonstrates that the physical components associated with the contravariant and covariant compo
nents are identical.
NOTATION The physical components are sometimes expressed by symbols with subscripts which represent
the coordinate curve along which the projection is taken. For example, let H i denote the contravariant
components of a first order tensor. The following are some examples of the representation of the physical
components of H i in various coordinate systems:
orthogonal coordinate tensor physical
coordinates system components components
The physical components associated with higher ordered tensors are defined by projections in VN just
like the case with first order tensors. For an nth ordered tensor Tij...k we can select n unit vectors λi , µi , . . . , ν i
and form the inner product (projection)
Tij...k λi µj . . . ν k .
When projecting the tensor components onto the coordinate curves, there are N choices for each of the unit
vectors. This produces N n physical components.
The above inner product represents the physical component of the tensor Tij...k along the directions of
the unit vectors λi , µi , . . . , ν i . The selected unit vectors may or may not be orthogonal. In the cases where
the selected unit vectors are all orthogonal to one another, the calculation of the physical components is
greatly simplified. By relabeling the unit vectors λi(m) , λi(n) , . . . , λi(p) where (m), (n), ..., (p) represent one of
the N directions, the physical components of a general nth order tensor is represented
A(ij) = Amn λm n
(i) µ(j) i, j = 1, 2, 3,
where the subscripts (i) and (j) represent one of the coordinate directions. Dropping the subscripts (i) and
(j), we make the observation that in an orthogonal curvilinear coordinate system there are three choices for
the direction of the unit vector λi and also three choices for the direction of the unit vector µi . These three
choices represent the directions along the x1 , x2 or x3 coordinate curves which emanate from a point of the
curvilinear coordinate system. This produces a total of nine possible physical components associated with
the tensor Amn .
For example, we can obtain the components of the unit vector λi , i = 1, 2, 3 in the x1 direction directly
from an examination of the element of arc length squared
dx1 1
= = λ1 , λ2 = 0, λ3 = 0.
ds h1
This is the vector λi(1) , i = 1, 2, 3. Similarly, if we choose to select the unit vector λi , i = 1, 2, 3 in the x2
direction, we set dx1 = dx3 = 0 in the element of arc length squared and find the components
dx2 1
λ1 = 0, λ2 = = , λ3 = 0.
ds h2
94
This is the vector λi(2) , i = 1, 2, 3. Finally, if we select λi , i = 1, 2, 3 in the x3 direction, we set dx1 = dx2 = 0
in the element of arc length squared and determine the unit vector
dx3 1
λ1 = 0, λ2 = 0, λ3 = = .
ds h3
This is the vector λi(3) , i = 1, 2, 3. Similarly, the unit vector µi can be selected as one of the above three
directions. Examining all nine possible combinations for selecting the unit vectors, we calculate the physical
components in an orthogonal coordinate system as:
A(i)(j)
A(ij) = no summation on i or j . (1.3.61)
h(i) h(j)
From the fact g ij = 0 for i 6= j, together with the physical components from equation (1.3.61), the equation
(1.3.62) reduces to
(i) 1
A(j) = g (i)(i) A(i)(j) = · h(i) h(j) A(ij) no summation on i and i, j = 1, 2 or 3.
h2(i)
(i) h(i)
A(ij) = A(j) no summation on i or j. (1.3.63)
h(j)
Hence, the physical components associated with the mixed tensor Aij in an orthogonal coordinate system
can be expressed as
A(11) = A11 h1 h1
A(12) = A12 A(13) = A13
h2 h2 h3
A(21) = A21 A(22) = A22 h2
h1 A(23) = A23
h3 h3 h3
A(31) = A31 A(32) = A32 A(33) = A33 .
h1 h2
For second order contravariant tensors we may write
We use the fact gij = 0 for i 6= j together with the physical components from equation (1.3.63) to reduce the
(i)
above equation to the form A(m) = A(i)(m) g(m)(m) no summation on m . In terms of physical components
we have
h(m)
A(im) = A(i)(m) h2(m) or A(im) = A(i)(m) h(i) h(m) . no summation i, m = 1, 2, 3 (1.3.64)
h(i)
Examining the results from equation (1.3.64) we find that the physical components associated with the
contravariant tensor Aij , in an orthogonal coordinate system, can be written as:
In an orthogonal curvilinear coordinate system, the physical components associated with the nth order
tensor Tij...kl along the curvilinear coordinate directions can be represented:
T(i)(j)...(k)(l)
T (ij . . . kl) = no summations.
h(i) h(j) . . . h(k) h(l)
These physical components can be related to the various tensors associated with Tij...kl . For example, in
ij...m
an orthogonal coordinate system, the physical components associated with the mixed tensor Tn...kl can be
expressed as:
(i)(j)...(m) h(i) h(j) . . . h(m)
T (ij . . . m n . . . kl) = T(n)...(k)(l) no summations. (1.3.65)
h(n) . . . h(k) h(l)
EXAMPLE 1.313. (Physical components) Let xi = xi (t), i = 1, 2, 3 denote the position vector of a
particle which moves as a function of time t. Assume there exists a coordinate transformation xi = xi (x), for
i = 1, 2, 3, of the form given by equations (1.2.33). The position of the particle when referenced with respect
to the barred system of coordinates can be found by substitution. The generalized velocity of the particle
in the unbarred system is a vector with components
dxi
vi = , i = 1, 2, 3.
dt
The generalized velocity components of the same particle in the barred system is obtained from the chain
rule. We find this velocity is represented by
are tensor quantities. These quantities are called the components of the generalized velocity. The coordinates
x1 , x2 , x3 are generalized coordinates. This means we can select any set of three independent variables for
the representation of the motion. The variables selected might not have the same dimensions. For example,
in cylindrical coordinates we let (x1 = r, x2 = θ, x3 = z). Here x1 and x3 have dimensions of distance but x2
has dimensions of angular displacement. The generalized velocities are
Here v 1 and v 3 have units of length divided by time while v 2 has the units of angular velocity or angular
change divided by time. Clearly, these dimensions are not all the same. Let us examine the physical
components of the generalized velocities. We find in cylindrical coordinates h1 = 1, h2 = r, h3 = 1 and the
physical components of the velocity have the forms:
dr dθ dz
vr = v(1) = v 1 h1 = , vθ = v(2) = v 2 h2 = r , vz = v(3) = v 3 h3 = .
dt dt dt
Now the physical components of the velocity all have the same units of length divided by time.
Additional examples of the use of physical components are considered later. For the time being, just
remember that when tensor equations are derived, the equations are valid in any generalized coordinate
system. In particular, we are interested in the representation of physical laws which are to be invariant and
independent of the coordinate system used to represent these laws. Once a tensor equation is derived, we
can chose any type of generalized coordinates and expand the tensor equations. Before using any expanded
tensor equations we must replace all the tensor components by their corresponding physical components in
order that the equations are dimensionally homogeneous. It is these expanded equations, expressed in terms
of the physical components, which are used to solve applied problems.
Tensors can be thought of as being created by multilinear forms defined on some vector space V. Let
us define on a vector space V a linear form, a bilinear form and a general multilinear form. We can then
illustrate how tensors are created from these forms.
for all arbitrary vectors ~x1 , ~x2 in V and all real numbers µ.
97
~ · ~x
ϕ(~x) = A (1.3.67)
ei = x1 b
~x = xi b e1 + x2 b
e2 + x3 b
e3 ,
~i
~x = xi E ~ i ) = xi ϕ(E
and ϕ(~x) = ϕ(xi E ~ i ).
The linear form ϕ defines a new tensor ai = ϕ(E ~ i ) so that ϕ(~x) = xi ai . Whenever there is a definite relation
e1 , b
between the basis vectors ( b e2 , b ~ 1, E
e3 ) and (E ~ 2, E
~ 3 ), say,
j
~ i = ∂x b
E ej ,
∂xi
then there exists a definite relation between the tensors ai and ai . This relation is
∂xj ∂xj ∂xj
~ i ) = ϕ(
ai = ϕ(E b
i ej ) =
b
i ϕ( ej ) = aj .
∂x ∂x ∂xi
This is the transformation law for an absolute covariant tensor of rank or order one.
The above idea is now extended to higher order tensors.
for arbitrary vectors ~x1 , ~x2 , ~y1 , ~y2 in the vector space V and for all
real numbers µ.
98
Note in the definition of a bilinear form that the scalar function ϕ is linear in both the arguments ~x and
~y . An example of a bilinear form is the dot product relation
is a scalar function
ϕ(~x1 , ~x2 , . . . , ~xM )
of M vector arguments which satisfies the property that it is a linear form in each of its
arguments. That is, ϕ must satisfy for each j = 1, 2, . . . , M the properties:
(i) ϕ(~x1 , . . . , ~xj1 + ~xj2 , . . . ~xM ) = ϕ(~x1 , . . . , ~xj1 , . . . , ~xM ) + ϕ(~x1 , . . . , ~xj2 , . . . , ~xM )
(ii) ϕ(~x1 , . . . , µ~xj , . . . , ~xM ) = µϕ(~x1 , . . . , ~xj , . . . , ~xM )
(1.3.70)
for all arbitrary vectors ~x1 , . . . , ~xM in the vector space V and all real numbers µ.
An example of a third degree multilinear form or trilinear form is the triple scalar product
Note that multilinear forms are independent of the coordinate system selected and depend only upon the
e1 , b
vector arguments. In a three dimensional vector space we select the basis vectors ( b e2 , b
e3 ) and represent
all vectors with respect to this basis set. For example, if ~x, ~y , ~z are three vectors we can represent these
vectors in the component forms
~x = xi b
ei , ~y = y j b
ej , ~z = z k b
ek (1.3.72)
where we have employed the summation convention on the repeated indices i, j and k. Substituting equations
(1.3.72) into equation (1.3.71) we obtain
ϕ(xi b
ei , y j b
ej , z k b
ek ) = xi y j z k ϕ( b
ei , b
ej , b
ek ), (1.3.73)
ei , b
ϕ( b ej , b
ek ) = eijk (1.3.74)
99
(See exercise 1.1, problem 15) the trilinear form, given by equation (1.3.71), with vectors from equations
(1.3.72), can be expressed as
ϕ(~x, ~y , ~z) = eijk xi y j z k , i, j, k = 1, 2, 3. (1.3.75)
The coefficients eijk of the trilinear form is called a third order tensor. It is the familiar permutation symbol
considered earlier.
In a multilinear form of degree M , ϕ(~x, ~y , . . . , ~z), the M arguments can be represented in a component
e1 , b
form with respect to a set of basis vectors ( b e2 , b
e3 ). Let these vectors have components xi , y i , z i , i = 1, 2, 3
with respect to the selected basis vectors. We then can write
~x = xi b
ei , ~y = y j b
ej , ~z = z k b
ek .
ϕ(xi b
ei , y j b
ej , . . . , z k b
ek ) = xi y j · · · z k ϕ( b
ei , b
ej , . . . , b
ek ). (1.3.76)
aij...k = ϕ( b
ei , b
ej , . . . , b
ek ) (1.3.77)
which are referred to as the components of a tensor of order M. The tensor is thus created by the multilinear
form and has M indices if ϕ is of degree M.
Note that if we change to a different set of basis vectors, say, (E ~ 2, E
~ 1, E ~ 3 ) the multilinear form defines
a new tensor
~i, E
aij...k = ϕ(E ~j, . . . , E
~ k ). (1.3.78)
This new tensor has a bar over it to distinguish it from the previous tensor. A definite relation exists between
the new and old basis vectors and consequently there exists a definite relation between the components of
the barred and unbarred tensors components. Recall that if we are given a set of transformation equations
y i = y i (x1 , x2 , x3 ), i = 1, 2, 3, (1.3.79)
from rectangular to generalized curvilinear coordinates, we can express the basis vectors in the new system
by the equations
j
E~ i = ∂y bej , i = 1, 2, 3. (1.3.80)
∂xi
For example, see equations (1.3.11) with y 1 = x, y 2 = y, y 3 = z, x1 = u, x2 = v, x3 = w. Substituting
equations (1.3.80) into equations (1.3.78) we obtain
∂y α ∂y β ∂y γ
b
eα , j b
aij...k = ϕ(
i
eβ , . . . , k b eγ ).
∂x ∂x ∂x
By the linearity property of ϕ, this equation is expressible in the form
∂y α ∂y β ∂y γ
aij...k =
i j
. . . k ϕ( b eα , b eβ , . . . , b
eγ )
∂x ∂x ∂x
∂y α ∂y β ∂y γ
aij...k = . . . aαβ...γ
∂xi ∂xj ∂xk
This is the familiar transformation law for a covariant tensor of degree M. By selecting reciprocal basis
vectors the corresponding transformation laws for contravariant vectors can be determined.
The above examples illustrate that tensors can be considered as quantities derivable from multilinear
forms defined on some vector space.
100
Dual Tensors
The epermutation symbol is often used to generate new tensors from given tensors. For Ti1 i2 ...im a
skewsymmetric tensor, we define the tensor
1 j1 j2 ...jn−m i1 i2 ...im
T̂ j1 j2 ...jn−m = e Ti1 i2 ...im m≤n (1.3.81)
m!
as the dual tensor associated with Ti1 i2 ...im . Note that the epermutation symbol or alternating tensor has
a weight of +1 and consequently the dual tensor will have a higher weight than the original tensor.
The epermutation symbol has the following properties
δkj11jk22...jm
...km Tj1 j2 ...jm = m!Tk1 k2 ...km .
Using the above properties we can solve for the skewsymmetric tensor in terms of the dual tensor. We find
1
Ti1 i2 ...im = ei i ...i j j ...j T̂ j1 j2 ...jn−m . (1.3.83)
(n − m)! 1 2 m 1 2 n−m
For example, if Aij i, j = 1, 2, 3 is a skewsymmetric tensor, we may associate with it the dual tensor
1 ijk
Vi = e Ajk ,
2!
which is a first order tensor or vector. Note that Aij has the components
0 A12 A13
−A12 0 A23 (1.3.84)
−A13 −A23 0
~ are
and consequently, the components of the vector V
Note that the vector components have a cyclic order to the indices which comes from the cyclic properties
of the epermutation symbol.
As another example, consider the fourth order skewsymmetric tensor Aijkl , i, j, k, l = 1, . . . , n. We can
associate with this tensor any of the dual tensor quantities
1 ijkl
V = e Aijkl
4!
1
V i = eijklm Ajklm
4!
1
V ij = eijklmn Aklmn (1.3.86)
4!
1
V ijk = eijklmnp Almnp
4!
1
V ijkl = eijklmnpr Amnpr
4!
Applications of dual tensors can be found in section 2.2.
101
EXERCISE 1.3
I 1.
∂xa ∂xb
(a) From the transformation law for the second order tensor g ij = gab
∂xi ∂xj
solve for the gab in terms of gij .
(b) Show that if gij is symmetric in one coordinate system it is symmetric in all coordinate systems.
p √ x
(c) Let g = det(gij ) and g = det(gij ) and show that g = gJ 2 ( xx ) and consequently g = gJ( ). This
√ x
shows that g is a scalar invariant of weight 2 and g is a scalar invariant of weight 1.
I 2. For
∂y m ∂y m ∂xi ∂xj
gij = show that g ij =
∂xi ∂xj ∂y m ∂y m
i 2
∂y
I 4. Show that g = det(gij ) = j = J 2 , where J is the Jacobian.
∂x
(c) Show that the volume of the elemental parallelepiped having d~r as diagonal can be represented
√ ∂(x, y, z)
dτ = g dudvdw = J dudvdw = dudvdw.
∂(u, v, w)
Hint:
A1 A2 A3
A · (B × C) = B1
~ ~ ~ B2 B3
C1 C2 C3
102
I 6. For the change d~r given inqproblem 5, show the elemental parallelepiped with diagonal d~r has:
(a) the element of area dS1 = g22 g33 − g232 dvdw in the u =constant surface.
q
(b) The element of area dS2 = 2 dudw in the v =constant surface.
g33 g11 − g13
q
(c) the element of area dS3 = 2 dudv in the w =constant surface.
g11 g22 − g12
I 7. In Cartesian coordinates you are given the affine transformation. xi = `ij xj where
1 1 1
x1 = (5x1 − 14x2 + 2x3 ), x2 = − (2x1 + x2 + 2x3 ), x3 = (10x1 + 2x2 − 11x3 )
15 3 15
I 8. Calculate the metric and conjugate metric tensors in cylindrical coordinates (r, θ, z).
I 9. Calculate the metric and conjugate metric tensors in spherical coordinates (ρ, θ, φ).
I 10. Calculate the metric and conjugate metric tensors in parabolic cylindrical coordinates (ξ, η, z).
I 11. Calculate the metric and conjugate metric components in elliptic cylindrical coordinates (ξ, η, z).
I 12. Calculate the metric and conjugate metric components for the oblique cylindrical coordinates (r, φ, η),
illustrated in figure 1.318, where x = r cos φ, y = r sin φ + η cos α, z = η sin α and α is a parameter
0<α≤ π
2. Note: When α = π
2 cylindrical coordinates result.
103
I 13. Calculate the metric and conjugate metric tensor associated with the toroidal surface coordinates
(ξ, η) illustrated in the figure 1.319, where
I 14. Calculate the metric and conjugate metric tensor associated with the spherical surface coordinates
(θ, φ), illustrated in the figure 1.320, where
I 16. Let Ax , Ay , Az denote the constant components of a vector in Cartesian coordinates. Using the
transformation laws (1.2.42) and (1.2.47) to find the contravariant and covariant components of this vector
upon changing to (a) cylindrical coordinates (r, θ, z). (b) spherical coordinates (ρ, θ, φ) and (c) Parabolic
cylindrical coordinates.
I 17. Find the relationship which exists between the given associated tensors.
(a) Apqk
r. and Apq
rs (c) Ai.j.
.l.m and A.s.p
r.t.
I 18. Given the fourth order tensor Cikmp = λδik δmp + µ(δim δkp + δip δkm ) + ν(δim δkp − δip δkm ) where λ, µ
and ν are scalars and δij is the Kronecker delta. Show that under an orthogonal transformation of rotation of
axes with xi = `ij xj where `rs `is = `mr `mi = δri the components of the above tensor are unaltered. Any
tensor whose components are unaltered under an orthogonal transformation is called an ‘isotropic’ tensor.
Another way of stating this problem is to say “Show Cikmp is an isotropic tensor.”
I 19. Assume Aijl is a third order covariant tensor and B pqmn is a fourth order contravariant tensor. Prove
that Aikl B klmn is a mixed tensor of order three, with one covariant and two contravariant indices.
I 20. Assume that Tmnrs is an absolute tensor. Show that if Tijkl + Tijlk = 0 in the coordinate system xr
then T ijkl + T ijlk = 0 in any other coordinate system xr .
I 22. Determine if the tensor equation mnp mij + mnj mpi = mni mpj is true or false. Justify your answer.
I 23. Prove the epsilon identity g ij ipt jrs = gpr gts − gps gtr . Hint: See problem 38, Exercise 1.1
1
I 24. Let Ars denote a skewsymmetric contravariant tensor and let cr = rmn Amn where
√ 2
rmn = germn . Show that cr are the components of a covariant tensor. Write out all the components.
1 rmn 1
I 25. Let Ars denote a skewsymmetric covariant tensor and let cr = Amn where rmn = √ ermn .
2 g
Show that cr are the components of a contravariant tensor. Write out all the components.
105
√ i
I 27. When Aij is an absolute tensor prove that gAj is a relative tensor of weight +1.
I 28. When Aij is an absolute tensor prove that √1 Ai is a relative tensor of weight −1.
g j
I 29.
(a) Show eijk is a relative tensor of weight +1.
(b) Show ijk = √1 eijk is an absolute tensor. Hint: See example 1.125.
g
I 30. The equation of a surface can be represented by an equation of the form Φ(x1 , x2 , x3 ) = constant.
Show that a unit normal vector to the surface can be represented by the vector
∂Φ
g ij ∂x j
ni = ∂Φ ∂Φ 2 1 .
(g mn ∂x m ∂xn )
I 31. Assume that gij = λgij with λ a nonzero constant. Find and calculate gij in terms of g ij .
I 32. Determine if the following tensor equation is true. Justify your answer.
I 33. Show that for Ci and C i associated tensors, and C i = ijk Aj Bk , then Ci = ijk Aj B k
I 34. Prove that ijk and ijk are associated tensors. Hint: Consider the determinant of gij .
I 36. Let Tji , i, j = 1, 2, 3 denote a second order mixed tensor. Show that the given quantities are scalar
invariants.
(i) I1 = Tii
1 i 2
(ii) I2 = (Ti ) − Tm
i
Tim
2
(iii) I3 = detTji 
I 37.
(a) Assume Aij and B ij , i, j = 1, 2, 3 are absolute contravariant tensors, and determine if the inner product
C ik = Aij B jk is an absolute tensor?
∂xj ∂xj
(b) Assume that the condition = δnm is satisfied, and determine whether the inner product in
∂xn ∂xm
part (a) is a tensor?
(c) Consider only transformations which are a rotation and translation of axes y i = `ij yj + bi , where `ij are
∂y j ∂yj
direction cosines for the rotation of axes. Show that = δnm
∂yn ∂ym
106
I 38. For Aijk a Cartesian tensor, determine if a contraction on the indices i and j is allowed. That
is, determine if the quantity Ak = Aiik , (summation on i) is a tensor. Hint: See part(c) of the previous
problem.
I 39. Prove the eδ identity eijk eimn = δm δn − δnj δm
j k k
.
I 40. Consider the vector Vk , k = 1, 2, 3 and define the matrix (aij ) having the elements aij = eijk Vk ,
where eijk is the e−permutation symbol.
(a) Solve for Vi in terms of amn by multiplying both sides of the given equation by eijl and note the e − δ
identity allows us to simplify the result.
(b) Sum the given expression on k and then assign values to the free indices (i,j=1,2,3) and compare your
results with part (a).
(c) Is aij symmetric, skewsymmetric, or neither?
I 41. It can be shown that the continuity equation of fluid dynamics can be expressed in the tensor form
1 ∂ √ ∂%
√ r
( g%V r ) + = 0,
g ∂x ∂t
where % is the density of the fluid, t is time, V r , with r = 1, 2, 3 are the velocity components and g = gij 
is the determinant of the metric tensor. Employing the summation convention and replacing the tensor
components of velocity by their physical components, express the continuity equation in
I 42. Let x1 , x2 , x3 denote a set of skewed coordinates with respect to the Cartesian coordinates y 1 , y 2 , y 3 .
Assume that E~ 1, E ~ 3 are unit vectors in the directions of the x1 , x2 and x3 axes respectively. If the unit
~ 2, E
vectors satisfy the relations
E ~1 = 1
~1 · E ~1 · E
E ~ 2 = cos θ12
E ~2 = 1
~2 · E ~1 · E
E ~ 3 = cos θ13
E ~3 = 1
~3 · E ~2 · E
E ~ 3 = cos θ23 ,
where a, b, c, d, e, f are complex constants. Calculate the components of the dual tensor
1 ijkl
V ij = e Akl .
2
107
I 44. In Cartesian coordinates the vorticity tensor at a point in a fluid medium is defined
1 ∂Vj ∂Vi
ωij = −
2 ∂xi ∂xj
where Vi are the velocity components of the fluid at the point. The vorticity vector at a point in a fluid
1
medium in Cartesian coordinates is defined by ω i = eijk ωjk . Show that these tensors are dual tensors.
2
I 45. Write out the relation between each of the components of the dual tensors
1
T̂ ij = eijkl Tkl i, j, k, l = 1, 2, 3, 4
2
and show that if ijkl is an even permutation of 1234, then T̂ ij = Tkl .
I 46. Consider the general affine transformation x̄i = aij xj where (x1 , x2 , x3 ) = (x, y, z) with inverse
transformation xi = bij x̄j . Determine (a) the image of the plane Ax + By + Cz + D = 0 under this
transformation and (b) the image of a second degree conic section
Ax2 + 2Bxy + Cy 2 + Dx + Ey + F = 0.
I 47. Using a multilinear form of degree M, derive the transformation law for a contravariant vector of
degree M.
∂g ∂gij
I 48. Let g denote the determinant of gij and show that = gg ij k .
∂xk ∂x
I 49. We have shown that for a rotation of xyz axes with respect to a set of fixed x̄ȳ z̄ axes, the derivative
~ with respect to an observer on the barred axes is given by
of a vector A
dA~ ~
= dA + ω~ × A.~
dt f dt r
Introduce the operators
~
dA
~=
Df A = derivative in fixed system
dt f
~
~ = dA = derivative in rotating system
Dr A
dt r
(a) Show that Df A ~ = (Dr + ~ ~
ω ×)A.
(b) Consider the ~ is the position vector ~r. Show that Df ~r = (Dr + ~ω×)~r
special case that the vector
A
produces V ~ = V~ + ~ ω × ~r where V~ represents the velocity of a particle relative to the fixed system
f r f
and V~ represents the velocity of a particle with respect to the rotating system of coordinates.
r
(c) Show that ~a = ~a + ~ ω × (~ω × ~r) where ~a represents the acceleration of a particle relative to the
f r f
fixed system and ~a represents the acceleration of a particle with respect to the rotating system.
r
(d) Show in the special case ~
ω is a constant that
~a = 2~ω × V
~ +ω
~ × (~ω × ~r)
f
~ is the velocity of the particle relative to the rotating system. The term 2~ω × V
where V ~ is referred to
as the Coriolis acceleration and the term ~ω × (~ω × ~r) is referred to as the centripetal acceleration.
108
In this section we develop some additional operations associated with tensors. Historically, one of the
basic problems of the tensor calculus was to try and find a tensor quantity which is a function of the metric
∂gij ∂ 2 gij
tensor gij and some of its derivatives , , . . . . A solution of this problem is the fourth order
∂xm ∂xm ∂xn
Riemann Christoffel tensor Rijkl to be developed shortly. In order to understand how this tensor was arrived
at, we must first develop some preliminary relationships involving Christoffel symbols.
Christoffel Symbols
Let us consider the metric tensor gij which we know satisfies the transformation law
∂xa ∂xb
g αβ = gab .
∂xα ∂xβ
1
and form the combination of terms [(α, β, γ) + (β, γ, α) − (γ, α, β)] to obtain the result
2
1 ∂gαβ ∂gβγ ∂g γα 1 ∂gab ∂gbc ∂gca ∂xa ∂xb ∂xc ∂xb ∂ 2 xa
+ − = + − + g ab . (1.4.1)
2 ∂xγ ∂xα ∂xβ 2 ∂xc ∂xa α
∂xb ∂x ∂xβ ∂x γ
∂xβ ∂xα ∂xγ
In this equation the combination of derivatives occurring inside the brackets is called a Christoffel symbol
of the first kind and is defined by the notation
1 ∂gab ∂gbc ∂gac
[ac, b] = [ca, b] = + − . (1.4.2)
2 ∂xc ∂xa ∂xb
The equation (1.4.1) defines the transformation for a Christoffel symbol of the first kind and can be expressed
as
∂xa ∂xb ∂xc ∂ 2 xa ∂xb
[α γ, β] = [ac, b] + g ab . (1.4.3)
∂xα ∂xβ ∂xγ ∂xα ∂xγ ∂xβ
Observe that the Christoffel symbol of the first kind [ac, b] does not transform like a tensor. However, it is
symmetric in the indices a and c.
At this time it is convenient to use the equation (1.4.3) to develop an expression for the second derivative
term which occurs in that equation as this second derivative term arises in some of our future considerations.
∂xβ de
To solve for this second derivative we can multiply equation (1.4.3) by g and simplify the result to the
∂xd
form
∂ 2 xe ∂xa ∂xc ∂xβ de
= −g de
[ac, d] + [α γ, β] g . (1.4.4)
∂xα ∂xγ ∂xα ∂xγ ∂xd
∂xd ∂xe
The transformation g de = g λµ allows us to express the equation (1.4.4) in the form
∂xλ ∂xµ
This Christoffel symbol of the second kind is symmetric in the indices j and k and from equation (1.4.5) we
see that it satisfies the transformation law
µ ∂xe e ∂xa ∂xc ∂ 2 xe
= γ + . (1.4.7)
αγ ∂xµ ac α
∂x ∂x ∂xα ∂xγ
Observe that the Christoffel symbol of the second kind does not transform like a tensor quantity. We can use
the relation defined by equation (1.4.7) to express the second derivative of the transformation equations in
terms of the Christoffel symbols of the second kind. At times it will be convenient to represent the Christoffel
symbols with a subscript to indicate the metric from which they are calculated. Thus, an alternative notation
i i
for j k is the notation j k .
g
EXAMPLE 1.41. (Christoffel symbols) Solve for the Christoffel symbol of the first kind in terms of
the Christoffel symbol of the second kind.
Solution: By the definition from equation (1.4.6) we have
i
= g iα [jk, α].
jk
and so
i 1 N
[jk, α] = gαi = gα1 + · · · + gαN .
jk jk jk
[11, 2], [11, 3], [22, 1], [22, 3], [33, 1], [33, 2].
CASE III Let a = c = i 6= b, and noting that gib = 0 for i 6= b, it can be verified that the equation (1.4.8)
simplifies to the form
1 ∂gii
[ab, c] = [ib, i] = [bi, i] = (no summation on i and i 6= b). (1.4.11)
2 ∂xb
From this equation we can calculate any of the twelve Christoffel symbols
[12, 1] = [21, 1] [31, 3] = [13, 3]
[32, 3] = [23, 3] [21, 2] = [12, 2]
[13, 1] = [31, 1] [23, 2] = [32, 2]
CASE IV Let a 6= b 6= c and show that the equation (1.4.8) reduces to
[ab, c] = 0, (a 6= b 6= c.)
From the Cases I,II,III,IV all twenty seven Christoffel symbols of the first kind can be determined. In
practice, only the nonzero Christoffel symbols are listed.
EXAMPLE 1.43. (Christoffel symbols of the first kind)Find the nonzero Christoffel symbols of the
first kind in cylindrical coordinates.
Solution: From the results of example 1.42 we find that for x1 = r, x2 = θ, x3 = z and
the nonzero Christoffel symbols of the first kind in cylindrical coordinates are:
1 ∂g22
[22, 1] = − = −x1 = −r
2 ∂x1
1 ∂g22
[21, 2] = [12, 2] = = x1 = r.
2 ∂x1
111
1
g ij = 0 for i 6= j and g ii = i not summed.
gii
The only nonzero term in the equation (1.4.12) occurs when m = i and consequently
i [jk, i]
= g ii [jk, i] = no summation on i. (1.4.13)
jk gii
We can now consider the four cases considered in the example 1.42.
CASE I Let j = k = i and show
i [ii, i] 1 ∂gii 1 ∂
= = = ln gii no summation on i. (1.4.14)
ii gii 2gii ∂xi 2 ∂xi
EXAMPLE 1.45. (Notation) In the case of cylindrical coordinates we can use the above relations and
find the nonzero Christoffel symbols of the second kind:
1 1 ∂g22
=− = −x1 = −r
22 2g11 ∂x1
2 2 1 ∂g22 1 1
= = = 1 =
12 21 2g22 ∂x1 x r
Note 1: The notation for the above Christoffel symbols are based upon the assumption that x1 = r, x2 = θ
and x3 = z. However, in tensor calculus the choice of the coordinates can be arbitrary. We could just as well
have defined x1 = z, x2 = r and x3 = θ. In this latter case, the numbering system of the Christoffel symbols
changes. To avoid confusion, an alternate method of writing the Christoffel symbols is to use coordinates in
place of the integers 1,2 and 3. For example, in cylindrical coordinates we can write
θ θ 1 r
= = and = −r.
rθ θr r θθ
Note 2: Some textbooks use the notation Γa,bc for Christoffel symbols of the first kind and Γdbc = g da Γa,bc for
Christoffel symbols of the second kind. This notation is not used in these notes since the notation suggests
that the Christoffel symbols are third order tensors, which is not true. The Christoffel symbols of the first
and second kind are not tensors. This fact is clearly illustrated by the transformation equations (1.4.3) and
(1.4.7).
Covariant Differentiation
Let Ai denote a covariant tensor of rank 1 which obeys the transformation law
∂xi
Aα = Ai . (1.4.17)
∂xα
Differentiate this relation with respect to xβ and show
Employing the equation (1.4.17), with α replaced by σ, the equation (1.4.19) is expressible in the form
∂Aα σ ∂Aj ∂xj ∂xk i ∂xj ∂xk
− A σ = α − Ai (1.4.20)
∂xβ αβ k
∂x ∂x ∂x β j k ∂xα ∂xβ
or alternatively " #
∂Aα σ ∂Aj i ∂xj ∂xk
− Aσ = − Ai . (1.4.21)
∂xβ αβ ∂xk jk ∂xα ∂xβ
∂xj ∂xk
Aα,β = Aj,k . (1.4.23)
∂xα ∂xβ
In the special case where gij are constants the Christoffel symbols of the second kind are zero, and conse
∂Aj
quently the covariant derivative reduces to Aj,k = . That is, under the special circumstances where the
∂xk
Christoffel symbols of the second kind are zero, the covariant derivative reduces to an ordinary derivative.
i ∂xi
A contravariant tensor Ai obeys the transformation law A = Aα which can be expressed in the
∂xα
form
α ∂xi
Ai = A (1.4.24)
∂xα
by interchanging the barred and unbarred quantities. We write the transformation law in the form of equation
(1.4.24) in order to make use of the second derivative relation from the previously derived equation (1.4.7).
Differentiate equation (1.4.24) with respect to xj to obtain the relation
2 i α
∂Ai α ∂ x ∂xβ ∂A ∂xβ ∂xi
= A + . (1.4.25)
∂xj ∂xα ∂xβ ∂xj ∂xβ ∂xj ∂xα
Changing the indices in equation (1.4.25) and substituting for the second derivative term, using the relation
from equation (1.4.7), produces the equation
" # α
∂Ai α σ ∂xi i ∂xm ∂xk ∂xβ ∂A ∂xβ ∂xi
=A − + . (1.4.26)
∂xj αβ ∂xσ mk ∂xα ∂xβ ∂xj ∂xβ ∂xj ∂xα
Applying the relation found in equation (1.4.24), with i replaced by m, together with the relation
∂xβ ∂xk
= δjk ,
∂xj ∂xβ
114
∂Ai
Again it should be observed that for the condition where gij are constants we have Ai ,j = and the
∂xj
covariant derivative of a contravariant tensor reduces to an ordinary derivative in this special case.
In a similar manner the covariant derivative of second rank tensors can be derived. We find these
derivatives have the forms:
∂Aij σ σ
Aij,k = k
− Aσj − Aiσ
∂x ik jk
∂Aij i σ
Aij ,k = + Aσj − Aiσ (1.4.30)
∂xk σk jk
∂Aij i j
Aij ,k = + A σj
+ A iσ
.
∂xk σk σk
In general, the covariant derivative of a mixed tensor
Aij...k
lm...p
and this derivative is a tensor of rank n + 1. Note the pattern of the + signs for the contravariant indices
and the − signs for the covariant indices.
Observe that the covariant derivative of an nth order tensor produces an n+ 1st order tensor, the indices
of these higher order tensors can also be raised and lowered by multiplication by the metric or conjugate
metric tensor. For example we can write
The rules for covariant differentiation are the same as for ordinary differentiation. That is:
(i) The covariant derivative of a sum is the sum of the covariant derivatives.
(ii) The covariant derivative of a product of tensors is the first times the covariant derivative of the second
plus the second times the covariant derivative of the first.
(iii) Higher derivatives are defined as derivatives of derivatives. Be careful in calculating higher order deriva
tives as in general
Ai,jk 6= Ai,kj .
EXAMPLE 1.46. (Covariant differentiation) Calculate the second covariant derivative Ai,jk .
Solution: The covariant derivative of Ai is
∂Ai σ
Ai,j = − Aσ .
∂xj ij
By definition, the second covariant derivative is the covariant derivative of a covariant derivative and hence
∂ ∂Ai σ m m
Ai,jk = (Ai,j ) ,k = − Aσ − Am,j − Ai,m .
∂xk ∂xj ij ik jk
Rearranging terms, the second covariant derivative can be expressed in the form
∂ 2 Ai ∂Aσ σ ∂Am m ∂Ai m
Ai,jk = − − −
∂xj ∂xk ∂xk i j ∂xj i k ∂xm j k
(1.4.32)
∂ σ σ m m σ
− Aσ − − .
∂xk i j im jk ik mj
116
where
∂ σ ∂ σ m σ m σ
σ
Rijk = − + − (1.4.33)
∂xj ik ∂xk ij ik mj ij mk
is called the Riemann Christoffel tensor. The covariant form of this tensor is
i
Rhjkl = gih Rjkl . (1.4.34)
It is an easy exercise to show that this covariant form can be expressed in either of the forms
∂ ∂ s s
Rinjk = [nk, i] − k [nj, i] + [ik, s] − [ij, s]
∂xj ∂x nj nk
2
1 ∂ gil ∂ 2 gjl ∂ 2 gik ∂ 2 gjk
or Rijkl = − i k − j l + i l + g αβ ([jk, β][il, α] − [jl, β][ik, α]) .
2 ∂xj ∂xk ∂x ∂x ∂x ∂x ∂x ∂x
From these forms we find that the Riemann Christoffel tensor is skew symmetric in the first two indices
and the last two indices as well as being symmetric in the interchange of the first pair and last pairs of
indices and consequently
In a two dimensional space there are only four components of the Riemann Christoffel tensor to consider.
These four components are either +R1212 or −R1212 since they are all related by
In a Cartesian coordinate system Rhijk = 0. The Riemann Christoffel tensor is important because it occurs
in differential geometry and relativity which are two areas of interest to be considered later. Additional
properties of this tensor are found in the exercises of section 1.5.
117
E ~ i (x1 , x2 , x3 ) = ∂~r
~i = E i = 1, 2, 3.
∂xi
Associated with these basis vectors we have the reciprocal basis vectors
E ~ i (x1 , x2 , x3 ),
~i = E i = 1, 2, 3
~ = A1 E
A ~ 1 + A2 E
~ 2 + A3 E
~ 3 = Aj E
~j (1.4.35)
~ 1 + A2 E
~ = A1 E
A ~ 2 + A3 E
~ 3 = Aj E
~ j. (1.4.36)
~ is represented as
A change in the vector A
∂A~
~=
dA dxk
∂xk
where from equation (1.4.35) we find
∂A~ ~ ∂Aj ~
j ∂ Ej
= A + Ej (1.4.37)
∂xk ∂xk ∂xk
or alternatively from equation (1.4.36) we may write
∂A~ ~j
∂E ∂Aj ~ j
k
= Aj k
+ E . (1.4.38)
∂x ∂x ∂xk
We define the covariant derivative of the covariant components as
∂A~ ~j
Ai,k = ~ i = ∂Ai + Aj ∂ E · E
·E ~ i. (1.4.39)
∂xk ∂xk ∂xk
The covariant derivative of the contravariant components are defined by the relation
∂A~ i ~
Ai ,k = ~ i = ∂A + Aj ∂ Ej · E
·E ~ i. (1.4.40)
∂xk ∂xk ∂xk
Introduce the notation
~j ~j
∂E m ~ ∂E j
= Em and =− E~ m. (1.4.41)
∂xk jk ∂xk mk
We then have
~
~ i · ∂ Ej =
E
m ~ ~ i = m δi =
Em · E
i
(1.4.42)
∂xk jk jk m jk
118
and
~j
~ i · ∂E = − j E
E ~ i = − j δim = − j .
~m · E (1.4.43)
∂xk mk mk ik
Then equations (1.4.39) and (1.4.40) become
∂Ai j
Ai,k = − Aj
∂xk ik
∂Ai i
Ai ,k = k
+ Aj ,
∂x jk
which is consistent with our earlier definitions from equations (1.4.22) and (1.4.28). Here the first term of
the covariant derivative represents the rate of change of the tensor field as we move along a coordinate curve.
The second term in the covariant derivative represents the change in the local basis vectors as we move
along the coordinate curves. This is the physical interpretation associated with the Christoffel symbols of
the second kind.
We make the observation that the derivatives of the basis vectors in equations (1.4.39) and (1.4.40) are
related since
~ j = δj
~i · E
E i
and consequently
∂ ~ ~j ~j ~
~ i · ∂ E + ∂ Ei · E
(Ei · E ) = E ~j = 0
∂x k ∂x k ∂xk
~j ~
or ~ i · ∂ E = −E
E ~ j · ∂ Ei
∂x k ∂xk
Hence we can express equation (1.4.39) in the form
∂Ai ~
Ai,k = ~ j · ∂ Ei .
− Aj E (1.4.44)
∂xk ∂xk
We write the first equation in (1.4.41) in the form
~j
∂E m ~ i = [jk, i]E
~i
= gim E (1.4.45)
∂xk jk
and consequently
~j
∂E i ~ ~m i m
k
~
·E =m
Ei · E = m
δi =
∂x jk jk jk
(1.4.46)
~
∂ Ej ~
and ~ ·E
· Em =[jk, i]E i ~ m = [jk, i]δ = [jk, m].
i
m
∂xk
These results also reduce the equations (1.4.40) and (1.4.44) to our previous forms for the covariant deriva
tives.
~i
∂E ~j
∂E
The equations (1.4.41) are representations of the vectors ∂xk
and ∂xk
in terms of the basis vectors and
reciprocal basis vectors of the space. The covariant derivative relations then take into account how these
vectors change with position and affect changes in the tensor field.
The Christoffel symbols in equations (1.4.46) are symmetric in the indices j and k since
~j ~k
∂E ∂ ∂~r ∂ ∂~r ∂E
= = = . (1.4.47)
∂xk ∂xk ∂xj ∂xj ∂xk ∂xj
119
Ricci’s Theorem
Ricci’s theorem states that the covariant derivative of the metric tensor vanishes and gik,l = 0.
Proof: We have
∂gik m m
gik,l = − gim − gmk
∂xl kl il
∂gik
gik,l = − [kl, i] − [il, k]
∂xl
∂gik 1 ∂gik ∂gil ∂gkl 1 ∂gik ∂gkl ∂gil
gik,l = − + k − − + − k = 0.
∂xl 2 ∂xl ∂x ∂xi 2 ∂xl ∂xi ∂x
Because of Ricci’s theorem the components of the metric tensor can be regarded as constants during covariant
differentiation.
i
EXAMPLE 1.47. (Covariant differentiation) Show that δj,k = 0.
Solution
∂δji i σ i i
i
δj,k = σ
+ δj − δσ
i
= − = 0.
∂xk σk jk jk jk
120
But gij,l = 0 by Ricci’s theorem and hence gij g jk,l = 0. We multiply this expression by g im and obtain
which demonstrates that the covariant derivative of the conjugate metric tensor is also zero.
The intrinsic or absolute derivative of a covariant vector Ai taken along a curve xi = xi (t), i = 1, . . . , N
is defined as the inner product of the covariant derivative with the tangent vector to the curve. The intrinsic
derivative is represented
δAi dxj
= Ai,j
δt dt
j
δAi ∂Ai α dx
= − Aα (1.4.50)
δt ∂xj ij dt
j
δAi dAi α dx
= − Aα .
δt dt i j dt
Similarly, the absolute or intrinsic derivative of a contravariant tensor Ai is represented
δAi dxj dAi i dxj
= Ai ,j = + Ak .
δt dt dt jk dt
The intrinsic or absolute derivative is used to differentiate sums and products in the same manner as used
in ordinary differentiation. Also if the coordinate system is Cartesian the intrinsic derivative becomes an
ordinary derivative.
The intrinsic derivative of higher order tensors is similarly defined as an inner product of the covariant
derivative with the tangent vector to the given curve. For example,
δAij dxp
klm
= Aij
klm,p
δt dt
EXAMPLE 1.410. (Generalized velocity and acceleration) Let t denote time and let xi = xi (t)
for i = 1, . . . , N , denote the position vector of a particle in the generalized coordinates (x1 , . . . , xN ). From
the transformation equations (1.2.30), the position vector of the same particle in the barred system of
coordinates, (x1 , x2 , . . . , xN ), is
dxi
The generalized velocity is v i = dt , i = 1, . . . , N. The quantity v i transforms as a tensor since by definition
Let us now find an expression for the generalized acceleration. Write equation (1.4.51) in the form
∂xj
vj = v i (1.4.52)
∂xi
dv j ∂ 2 xj dxk dv i ∂xj
= vi i k + (1.4.53)
dt ∂x ∂x dt dt ∂xi
dv i
The equation (1.4.53) demonstrates that dt does not transform like a tensor. From the equation (1.4.7)
previously derived, we change indices and write equation (1.4.53) in the form
" #
dv j dxk σ ∂xj j ∂xa ∂xc ∂xj dv i
= vi − + .
dt dt ik ∂xσ i
a c ∂x ∂x k
∂xi dt
y i = y i (x1 , x2 , x3 ), i = 1, 2, 3
The space curve C in the new coordinates is obtained directly from the transformation equations and can
be written
xi = xi (y 1 (t), y 2 (t), y 3 (t)) = xi (t), i = 1, 2, 3.
∂y i
Y i = Xj . (1.4.55)
∂xj
Since the components of Y i are constants, their derivatives will be zero and consequently we obtain by
differentiating the equation (1.4.55), with respect to the parameter t, that the field of parallel vectors X i
must satisfy the differential equation
dX j ∂y i ∂ 2 y i dxm dY i
j
+ Xj j m = = 0. (1.4.56)
dt ∂x ∂x ∂x dt dt
Changing symbols in the equation (1.4.7) and setting the Christoffel symbol to zero in the Cartesian system
of coordinates, we represent equation (1.4.7) in the form
i
∂ 2yi α ∂y
=
∂xj ∂xm j m ∂xα
The equation (1.4.57) is the differential equation which must be satisfied by a parallel field of vectors X i
along an arbitrary curve xi (t).
123
EXERCISE 1.4
I 1. Find the nonzero Christoffel symbols of the first and second kind in cylindrical coordinates
(x , x , x3 ) = (r, θ, z), where x = r cos θ,
1 2
y = r sin θ, z = z.
I 2. Find the nonzero Christoffel symbols of the first and second kind in spherical coordinates
(x , x , x3 ) = (ρ, θ, φ), where x = ρ sin θ cos φ,
1 2
y = ρ sin θ sin φ, z = ρ cos θ.
I 3. Find the nonzero Christoffel symbols of the first and second kind in parabolic cylindrical coordinates
1
(x , x , x3 ) = (ξ, η, z), where x = ξη, y = (ξ 2 − η 2 ), z = z.
1 2
2
I 4. Find the nonzero Christoffel symbols of the first and second kind in parabolic coordinates
1
(x , x , x3 ) = (ξ, η, φ), where x = ξη cos φ, y = ξη sin φ, z = (ξ 2 − η 2 ).
1 2
2
I 5. Find the nonzero Christoffel symbols of the first and second kind in elliptic cylindrical coordinates
(x , x , x3 ) = (ξ, η, z), where x = cosh ξ cos η,
1 2
y = sinh ξ sin η, z = z.
I 6. Find the nonzero Christoffel symbols of the first and second kind for the oblique cylindrical coordinates
(x , x2 , x3 ) = (r, φ, η), where x = r cos φ,
1
y = r sin φ+η cos α, z = η sin α with 0 < α < π
2 and α constant.
Hint: See figure 1.318 and exercise 1.3, problem 12.
∂gik
I 7. Show [ij, k] + [kj, i] = .
∂xj
I 8.
r
(a) Let = g ri [st, i] and solve for the Christoffel symbol of the first kind in terms of the Christoffel
st
symbol of the secondkind.
n
(b) Assume [st, i] = gni and solve for the Christoffel symbol of the second kind in terms of the
st
Christoffel symbol of the first kind.
I 9.
(a) Write down the transformation law satisfied by the fourth order tensor ijk,m .
(b) Show that ijk,m = 0 in all coordinate systems.
√
(c) Show that ( g),k = 0.
∂φ
I 12. The gradient of a scalar field φ(x1 , x2 , x3 ) is the vector grad φ = E
. ~i
∂xi
(a) Find the physical components associated with the covariant components φ ,i
dφ Ai φ,i
(b) Show the directional derivative of φ in a direction Ai is = .
dA (gmn Am An )1/2
124
I 13.
√
(a) Show g is a relative scalar of weight +1.
(b) Use the results from
problem
9(c) and problem 44, Exercise 1.4, to show that
√
√ ∂ g m √
( g),k = k
− g = 0.
∂x
km
m ∂ √ 1 ∂g
(c) Show that = k
ln( g) = .
km ∂x 2g ∂xk
m ∂ √ 1 ∂g
I 14. Use the result from problem 9(b) to show = ln( g) = .
km ∂xk 2g ∂xk√
Hint: Expand the covariant derivative rst,p and then substitute rst = gerst . Simplify by inner
rst
e√
multiplication with g and note the Exercise 1.1, problem 26.
I 15. Calculate the covariant derivative Ai,m and then contract on m and i to show that
1 ∂ √ i
Ai,i = √ gA .
g ∂xi
1 ∂ √ ij i
I 16. Show √ gg + g pq = 0. Hint: See problem 14.
g ∂xj pq
I 17. Prove that the covariant derivative of a sum equals the sum of the covariant derivatives.
Hint: Assume Ci = Ai + Bi and write out the covariant derivative for Ci,j .
I 18. Let Cji = Ai Bj and prove that the covariant derivative of a product equals the first term times the
covariant derivative of the second term plus the second term times the covariant derivative of the first term.
∂xα ∂xβ
I 19. Start with the transformation law Āij = Aαβ and take an ordinary derivative of both sides
∂ x̄i ∂ x̄j
k
with respect to x̄ and hence derive the relation for Aij,k given in (1.4.30).
∂xi ∂xj
I 20. Start with the transformation law Aij = Āα β and take an ordinary derivative of both sides
∂ x̄α ∂ x̄β
with respect to xk and hence derive the relation for Aij,k given in (1.4.30).
I 22. Find the intrinsic derivative along the curve xi = xi (t), i = 1, . . . , N for
I 23.
(a) Assume A ~ i and show that dA
~ = Ai E ~ = Ai dxk E ~ i.
,k
(b) Assume A ~ and show that dA
~ = Ai E i ~ = Ai,k dx E
k ~ i.
125
I 24. (parallel vector field) Imagine a vector field Ai = Ai (x1 , x2 , x3 ) which is a function of position.
Assume that at all points along a curve xi = xi (t), i = 1, 2, 3 the vector field points in the same direction,
we would then have a parallel vector field or homogeneous vector field. Assume A ~ is a constant, then
~= ∂A~
dA ∂xk dxk = 0. Show that for a parallel vector field the condition Ai,k = 0 must be satisfied.
∂[ik, n] ∂ σ σ
I 25. Show that = gnσ j + ([nj, σ] + [σj, n]) .
∂xj ∂x ik ik
∂Ar ∂As
I 26. Show Ar,s − As,r = − .
∂xs ∂xr
I 27. In cylindrical coordinates you are given the contravariant vector components
A1 = r A2 = cos θ A3 = z sin θ
I 28. Find the covariant form of the contravariant tensor C i = ijk Ak,j .
Express your answer in terms of Ak,j .
1
I 29. In Cartesian coordinates let x denote the magnitude of the position vector xi . Show that (a) x ,j = xj
x
1 1 2 1 −δij 3xi xj
(b) x ,ij = δij − 3 xi xj (c) x ,ii = . (d) LetU = , x 6= 0, and show that U ,ij = 3
+ and
x x x x x x5
U ,ii = 0.
I 30. Consider a two dimensional space with element of arc length squared
2 1 2 2 2 g11 0
ds = g11 (du ) + g22 (du ) and metric gij =
0 g22
I 31. Find the metric tensor and Christoffel symbols of the first and second kind associated with the
two dimensional space describing points on a cylinder of radius a. Let u1 = θ and u2 = z denote surface
coordinates where
x = a cos θ = a cos u1
y = a sin θ = a sin u1
z = z = u2
126
I 32. Find the metric tensor and Christoffel symbols of the first and second kind associated with the
two dimensional space describing points on a sphere of radius a. Let u1 = θ and u2 = φ denote surface
coordinates where
x = a sin θ cos φ = a sin u1 cos u2
y = a sin θ sin φ = a sin u1 sin u2
z = a cos θ = a cos u1
I 33. Find the metric tensor and Christoffel symbols of the first and second kind associated with the
two dimensional space describing points on a torus having the parameters a and b and surface coordinates
u1 = ξ, u2 = η. illustrated in the figure 1.319. The points on the surface of the torus are given in terms
of the surface coordinates by the equations
x = (a + b cos ξ) cos η
y = (a + b cos ξ) sin η
z = b sin ξ
I 34. Prove that eijk am bj ck ui,m + eijk ai bm ck uj,m + eijk ai bj cm uk,m = ur,r eijk ai bj ck . Hint: See Exercise 1.3,
problem 32 and Exercise 1.1, problem 21.
I 37. Find the contravariant, covariant and physical components of velocity and acceleration in (a) Cartesian
coordinates and (b) cylindrical coordinates.
I 38. Find the contravariant, covariant and physical components of velocity and acceleration in spherical
coordinates.
I 39. In spherical coordinates (ρ, θ, φ) show that the acceleration components can be represented in terms
of the velocity components as
I 40. The divergence of a vector Ai is Ai,i . That is, perform a contraction on the covariant derivative
Ai,j to obtain Ai,i . Calculate the divergence in (a) Cartesian coordinates (b) cylindrical coordinates and (c)
spherical coordinates.
I 41. If S is a scalar invariant of weight one and Aijk is a third order relative tensor of weight W , show
that S −W Aijk is an absolute tensor.
127
I 42. Let Ȳ i ,i = 1, 2, 3 denote the components of a field of parallel vectors along the curve C defined by
dȳ i
the equations y i = ȳ i (t), i = 1, 2, 3 in a space with metric tensor ḡij , i, j = 1, 2, 3. Assume that Ȳ i and dt
are unit vectors such that at each point of the curve C̄ we have
dȳ j
ḡij Ȳ i = cos θ = Constant.
dt
(i.e. The field of parallel vectors makes a constant angle θ with the tangent to each point of the curve C̄.)
Show that if Ȳ i and ȳ i (t) undergo a transformation xi = xi (ȳ 1 , ȳ 2 , ȳ 3 ), i = 1, 2, 3 then the transformed
m
vector X m = Ȳ i ∂x
∂ ȳ j makes a constant angle with the tangent vector to the transformed curve C given by
xi = xi (ȳ 1 (t), ȳ 2 (t), ȳ 3 (t)).

∂xi
I 43. Let J denote the Jacobian determinant  . Differentiate J with respect to xm and show that
∂xj
∂J α ∂xp r
= J − J .
∂xm α p ∂xm rm
I 44. Assume that φ is a relative scalar of weight W so that φ = J W φ. Differentiate this relation with
respect to xk . Use the result from problem 43 to obtain the transformation law:
" # m
∂φ α ∂φ r ∂x
−W φ =J W
−W φ .
∂xk αk ∂xm mr ∂xk
The quantity inside the brackets is called the covariant derivative of a relative scalar of weight W. The
covariant derivative of a relative scalar of weight W is defined as
∂φ r
φ ,k = −W φ
∂xk kr
When the weight term is zero these covariant derivatives reduce to the results given in our previous definitions.
dxi
I 45. Let dt = v i denote a generalized velocity and define the scalar function of kinetic energy T of a
particle with mass m as
1 1
T = m gij v i v j = m gij ẋi ẋj .
2 2
δT dT
Show that the intrinsic derivative of T is the same as an ordinary derivative of T. (i.e. Show that δT = dt .)
128
1 ∂ √ ijk
I 47. Assume that B ijk is an absolute tensor. Is the quantity T jk = √ gB a tensor? Justify
g ∂xi
your answer. If your answer is “no”, explain your answer and determine if there any conditions you can
impose upon B ijk such that the above quantity will be a tensor?
I 48. The epermutation symbol can be used to define various vector products. Let Ai , Bi , Ci , Di
i = 1, . . . , N denote vectors, then expand and verify the following products:
(a) In two dimensions
R =eij Ai Bj a scalar determinant.
Ri =eij Aj a vector (rotation).
(b) In three dimensions
S =eijk Ai Bj Ck a scalar determinant.
Si =eijk Bj Ck a vector cross product.
Sij =eijk Ck a skewsymmetric matrix
(c) In four dimensions
For xi = xi (s),i = 1, 2, 3, a 3dimensional space curve in a Riemannian space Vn with metric tensor gij ,
dxi
and arc length parameter s, the vector T i = ds represents a tangent vector to the curve at a point P on
i
the curve. The vector T is a unit vector because
dxi dxj
gij T i T j = gij = 1. (1.5.1)
ds ds
Differentiate intrinsically, with respect to arc length, the relation (1.5.1) and verify that
δT j δT i j
gij T i + gij T = 0, (1.5.2)
δs δs
which implies that
δT i
gij T j = 0. (1.5.3)
δs
δT i
Hence, the vector δs is perpendicular to the tangent vector T i . Define the unit normal vector N i to the
δT i
space curve to be in the same direction as the vector δs and write
1 δT i
Ni = (1.5.4)
κ δs
where κ is a scale factor, called the curvature, and is selected such that
δT i δT j
gij N i N j = 1 which implies gij = κ2 . (1.5.5)
δs δs
130
The reciprocal of curvature is called the radius of curvature. The curvature measures the rate of change of
the tangent vector to the curve as the arc length varies. By differentiating intrinsically, with respect to arc
length s, the relation gij T i N j = 0 we find that
δN j δT i j
gij T i + gij N = 0. (1.5.6)
δs δs
Consequently, the curvature κ can be determined from the relation
δN j δT i j
gij T i = −gij N = −gij κN i N j = −κ (1.5.7)
δs δs
which defines the sign of the curvature. In a similar fashion we differentiate the relation (1.5.5) and find that
δN j
gij N i = 0. (1.5.8)
δs
δN j
This later equation indicates that the vector δs is perpendicular to the unit normal N i . The equation
(1.5.3) indicates that T i is also perpendicular to N i and hence any linear combination of these vectors will
also be perpendicular to N i . The unit binormal vector is defined by selecting the linear combination
δN j
+ κT j (1.5.9)
δs
and then scaling it into a unit vector by defining
j 1 δN j
B = + κT j (1.5.10)
τ δs
where τ is a scalar called the torsion. The sign of τ is selected such that the vectors T i , N i and B i form a
right handed system with ijk T i N j B k = 1 and the magnitude of τ is selected such that B i is a unit vector
satisfying
gij B i B j = 1. (1.5.11)
The triad of vectors T i , N i , B i at a point on the curve form three planes. The plane containing T i and B i is
called the rectifying plane. The plane containing N i and B i is called the normal plane. The plane containing
T i and N i is called the osculating plane. The reciprocal of the torsion is called the radius of torsion. The
torsion measures the rate of change of the osculating plane. The vectors T i , N i and B i form a righthanded
orthogonal system at a point on the space curve and satisfy the relation
B i = ijk Tj Nk . (1.5.12)
By using the equation (1.5.10) it can be shown that B i is perpendicular to both the vectors T i and N i since
which contain two independent parameters u, v called surface coordinates. For example, the equations
are the parametric equations which define a spherical surface of radius a with parameters u = θ and v = φ.
See for example figure 1.320 in section 1.3. By eliminating the parameters u, v one can derive the implicit
form of the surface and by solving for z one obtains the explicit form of the surface. Using the parametric
form of a surface we can define the position vector to a point on the surface which is then represented in
terms of the parameters u, v as
e1 + y(u, v) b
~r = ~r(u, v) = x(u, v) b e2 + z(u, v) b
e3 . (1.5.14)
The coordinates (u, v) are called the curvilinear coordinates of a point on the surface. The functions
x(u, v), y(u, v), z(u, v) are assumed to be real and differentiable such that ∂~
r
∂u × ∂~
r
∂v 6= 0. The curves
with c1 , c2 constants, then define two surface curves called coordinate curves, which intersect at the surface
coordinates (c1 , c2 ). The family of curves defined by equations (1.5.15) with equally spaced constant values
∂~
r ∂~
r
ci , ci + ∆ci , ci + 2∆ci , . . . define a surface coordinate grid system. The vectors ∂u and ∂v evaluated at the
surface coordinates (c1 , c2 ) on the surface, are tangent vectors to the coordinate curves through the point
and are basis vectors for any vector lying in the surface. Letting (x, y, z) = (y 1 , y 2 , y 3 ) and (u, v) = (u1 , u2 )
and utilizing the summation convention, we can write the position vector in the form
~r = ~r(u1 , u2 ) = y i (u1 , u2 ) b
ei . (1.5.16)
The tangent vectors to the coordinate curves at a point P can then be represented as the basis vectors
i
~ α = ∂~r = ∂y b
E ei , α = 1, 2 (1.5.17)
∂u α ∂u α
where the partial derivatives are to be evaluated at the point P where the coordinate curves on the surface
intersect. From these basis vectors we construct a unit normal vector to the surface at the point P by
∂~
r ∂~
r
calculating the cross product of the tangent vector ~ru = ∂u and ~rv = ∂v . A unit normal is then
~1 × E
E ~2 ~ru × ~rv
b=n
n b(u, v) = = (1.5.18)
~ ~
E1 × E2  ~ru × ~rv 
132
~ 1, E
and is such that the vectors E ~ 2 and n
b form a righthanded system of coordinates.
If we transform from one set of curvilinear coordinates (u, v) to another set (ū, v̄), which are determined
by a set of transformation laws
u = u(ū, v̄), v = v(ū, v̄),
∂y i ∂y i ∂uβ
α
= .
∂ ū ∂uβ ∂ ūα
This is the transformation law connecting the two systems of basis vectors on the surface.
A curve on the surface is defined by a relation f (u, v) = 0 between the curvilinear coordinates. Another
way to represent a curve on the surface is to represent it in a parametric form where u = u(t) and v = v(t),
where t is a parameter. The vector
d~r ∂~r du ∂~r dv
= +
dt ∂u dt ∂v dt
is tangent to the curve on the surface.
An element of arc length with respect to the surface coordinates is represented by
∂~r ∂~r
ds2 = d~r · d~r = α
· duα duβ = aαβ duα duβ (1.5.19)
∂u ∂uβ
where aαβ = ∂~
r
∂uα · ∂~
r
∂uβ with α, β = 1, 2 defines a surface metric. This element of arc length on the surface is
often written as the quadratic form
1 EG − F 2 2
A = ds2 = E(du)2 + 2F du dv + G(dv)2 = (E du + F dv)2 + dv (1.5.20)
E E
and called the first fundamental form of the surface. Observe that for ds2 to be positive definite the quantities
E and EG − F 2 must be positive.
The surface metric associated with the two dimensional surface is defined by
~α · E
~β = ∂~r ∂~r ∂y i ∂y i
aαβ = E α
· β
= , α, β = 1, 2 (1.5.21)
∂u ∂u ∂uα ∂uβ
with conjugate metric tensor aαβ defined such that aαβ aβγ = δγα . Here the surface is embedded in a three
dimensional space with metric gij and aαβ is the two dimensional surface metric. In the equation (1.5.20)
the quantities E, F, G are functions of the surface coordinates u, v and are determined from the relations
∂~r ∂~r ∂y i ∂y i
E =a11 = · =
∂u ∂u ∂u1 ∂u1
∂~r ∂~r ∂y i ∂y i
F =a12 = · = (1.5.22)
∂u ∂v ∂u1 ∂u2
∂~r ∂~r ∂y i ∂y i
G =a22 = · =
∂v ∂v ∂u2 ∂u2
133
Here and throughout the remainder of this section, we adopt the convention that Greek letters have the
range 1,2, while Latin letters have the range 1,2,3.
b at this point. Also construct a
Construct at a general point P on the surface the unit normal vector n
b. Observe that there are an infinite number of planes
plane which contains this unit surface normal vector n
which contain this unit surface normal. For now, select one of these planes, then later on we will consider
all such planes. Let ~r = ~r(s) denote the position vector defining a curve C which is the intersection of the
selected plane with the surface, where s is the arc length along the curve, which is measured from some fixed
point on the curve. Let us find the curvature of this curve of intersection. The vector Tb = d~r , evaluated
ds
at the point P, is a unit tangent vector to the curve C and lies in the tangent plane to the surface at the
point P. Here we are using ordinary differentiation rather than intrinsic differentiation because we are in
a Cartesian system of coordinates. Differentiating the relation Tb · Tb = 1, with respect to arc length s we
find that Tb · dTb = 0 which implies that the vector dTb is perpendicular to the tangent vector Tb. Since the
ds ds
coordinate system is Cartesian we can treat the curve of intersection C as a space curve, then the vector
K~ = dTb , evaluated at point P, is defined as the curvature vector with curvature K ~ = κ and radius of
ds
curvature R = 1/κ. A unit normal N b to the space curve is taken in the same direction as dTb so that the
ds
~ b dTb
curvature will always be positive. We can then write K = κN = . Consider the geometry of figure 1.51
ds
b=n
and define on the surface a unit vector u b × Tb which is perpendicular to both the surface tangent vector
Tb and the surface normal vector nb, such that the vectors T i ,ui and ni forms a righthanded system.
Figure 1.51 Surface curve with tangent plane and a normal plane.
134
b
~ = dT = κ(n) n
K b + κ(g) u ~n +K
b=K ~g (1.5.23)
ds
where κ(n) is called the normal curvature and κ(g) is called the geodesic curvature. The subscripts are not
b · Tb = 0 we obtain
indices. These curvatures can be calculated as follows. From the orthogonality condition n
b
dT db
n
b·
by differentiation with respect to arc length s the result n + Tb · = 0. Consequently, the normal
ds ds
curvature is determined from the dot product relation
~ = κ(n) = −Tb · db
b·K
n
n d~r db
=− ·
n
. (1.5.24)
ds ds ds
dTb dTb
b·
κ(g) = u n × Tb) ·
= (b . (1.5.25)
ds ds
Normal Curvature
∂~r ∂~r ∂b
n ∂b
n
d~r = du + dv and db
n= du + dv. (1.5.27)
∂u ∂v ∂u ∂v
where
∂~r ∂b
n ∂~r ∂b
n ∂b n ∂~r ∂~r ∂b
n
e=− · , 2f = − · + · , g=− · (1.5.29)
∂u ∂u ∂u ∂v ∂u ∂v ∂v ∂v
and bαβ α, β = 1, 2 is called the curvature tensor and aαγ bαβ = bγβ is an associated curvature tensor.
The quadratic form of equation (1.5.28) is called the second fundamental form of the surface. Alternative
methods for calculating the coefficients of this quadratic form result from the following considerations. The
unit surface normal is perpendicular to the tangent vectors to the coordinate curves at the point P and
therefore we have the orthogonality relationships
∂~r ∂~r
·n
b=0 and ·n
b = 0. (1.5.30)
∂u ∂v
135
Observe that by differentiating the relations in equation (1.5.30), with respect to both u and v, one can
derive the results
∂ 2~r ∂~r ∂b n
e=
2
·n
b=− · = b11
∂u ∂u ∂u
∂ 2~r ∂~r ∂b n ∂b
n ∂~r
f= ·nb=− · =− · = b21 = b12 (1.5.31)
∂u∂v ∂u ∂v ∂u ∂v
∂ 2~r ∂~r ∂b n
g= 2
·n
b=− · = b22
∂v ∂v ∂v
and consequently the curvature tensor can be expressed as
∂~r ∂b
n
bαβ = − α
· . (1.5.32)
∂u ∂uβ
The quadratic forms from equations (1.5.20) and (1.5.28) enable us to represent the normal curvature
in the form of a ratio of quadratic forms. We find from equation (1.5.26) that the normal curvature in the
du
direction dv is
B e(du)2 + 2f du dv + g(dv)2
κ(n) = = . (1.5.33)
A E(du)2 + 2F du dv + G(dv)2
If we write the unit tangent vector to the curve in the form Tb = d~r r duα
∂~
ds = ∂uα ds and express the derivative
of the unit surface normal with respect to arc length as ddsb
n ∂b
n duβ
= ∂uβ ds , then the normal curvature can be
e + 2f λ + gλ2
κ(n) = (1.5.35)
E + 2F λ + Gλ2
dv
where λ = du . From the theory of proportions we can also write this equation in the form
(e + f λ) + λ(f + gλ) f + gλ e + fλ
κ(n) = = = . (1.5.36)
(E + F λ) + λ(F + Gλ) F + Gλ E + Fλ
dκ(n)
The maximum and minimum curvatures occur in those directions λ where dλ = 0. Calculating the deriva
tive of κ(n) with respect to λ and setting the derivative to zero we obtain a quadratic equation in λ
Eg − Ge Ef − F e
λ1 + λ2 = − and λ1 λ2 = , (1.5.38)
F g − Gf F g − Gf
where F g − Gf 6= 0. The curvatures κ(1) ,κ(2) corresponding to the roots λ1 and λ2 are called the principal
curvatures at the point P. Several quantities of interest that are related to κ(1) and κ(2) are: (1) the principal
1
radii of curvature Ri = 1/κi ,i = 1, 2; (2) H = 2 (κ(1) + κ(2) ) called the mean curvature and K = κ(1) κ(2)
called the total curvature or Gaussian curvature of the surface. Observe that the roots λ1 and λ2 determine
two directions on the surface
It is left as an exercise to verify that this is indeed the case and so the directions determined by the principal
curvatures must be orthogonal. In the case where F g − Gf = 0 we have that F = 0 and f = 0 because the
coordinate curves are orthogonal and G must be positive. In this special case there are still two directions
determined by the differential equations (1.5.37) with dv = 0, du arbitrary, and du = 0, dv arbitrary. From
the differential equations (1.5.37) we find these directions correspond to
e g
κ(1) = and κ(2) = .
E G
duα
We let λα = ds denote a unit vector on the surface satisfying aαβ λα λβ = 1. Then the equation (1.5.34)
can be written as κ(n) = bαβ λα λβ or we can write (bαβ − κ(n) aαβ )λα λβ = 0. The maximum and minimum
normal curvature occurs in those directions λα where
and so κ(n) must be a root of the determinant equation bαβ − κ(n) aαβ  = 0 or
1
b −κ b12
a αγ
bαβ − κ(n) δβγ  = 1 2 (n) = κ2 − bαβ aαβ κ(n) + b = 0. (1.5.40)
b1 b22 − κ(n) (n)
a
This is a quadratic equation in κ(n) of the form κ2(n) − (κ(1) + κ(2) )κ(n) + κ(1) κ(2) = 0. In other words the
principal curvatures κ(1) and κ(2) are the eigenvalues of the matrix with elements bγβ = aαγ bαβ . Observe that
from the determinant equation in κ(n) we can directly find the total curvature or Gaussian curvature which
is an invariant given by K = κ(1) κ(2) = bα
β  = a bγβ  = b/a. The mean curvature is also an invariant
αγ
1 1 αβ
obtained from H = 2 (κ(1) + κ(2) ) = 2a bαβ , where a = a11 a22 − a12 a21 and b = b11 b22 − b12 b21 are the
determinants formed from the surface metric tensor and curvature tensor components.
137
At each point on a space curve we can construct a unit tangent T~ , a unit normal N
~ and unit binormal
~ The derivatives of these vectors, with respect to arc length, can also be represented as linear combinations
B.
of the base vectors T~ , N
~ , B.
~ See for example the FrenetSerret formulas from equations (1.5.13). In a similar
b form a basis and the derivatives of these basis vectors with respect to
fashion the surface vectors ~ru , ~rv , n
b. For
the surface coordinates u, v can also be expressed as linear combinations of the basis vectors ~ru , ~rv , n
b. We can write
example, the derivatives ~ruu , ~ruv , ~rvv can be expressed as linear combinations of ~ru , ~rv , n
b
~ruu = c1~ru + c2~rv + c3 n
b
~ruv = c4~ru + c5~rv + c6 n (1.5.41)
b
~rvv = c7~ru + c8~rv + c9 n
where c1 , . . . , c9 are constants to be determined. It is an easy exercise (see exercise 1.5, problem 8) to show
that these equations can be written in the indicial notation as
∂ 2~r γ ∂~r
α β
= b.
+ bαβ n (1.5.42)
∂u ∂u α β ∂uγ
These equations are known as the Gauss equations.
In a similar fashion the derivatives of the normal vector can be represented as linear combinations of
the surface basis vectors. If we write
∂b
n ∂~r ∂b
n ∂b
n
= c1~ru + c2~rv = c∗1 + c∗2
∂u or ∂u ∂u ∂v (1.5.43)
∂b
n ∂~r ∂b
n ∂b
n
= c3~ru + c4~rv = c∗3 + c∗4
∂v ∂v ∂u ∂v
where c1 , . . . , c4 and c∗1 , . . . , c∗4 are constants. These equations are known as the Weingarten equations. It
is easily demonstrated (see exercise 1.5, problem 9) that the Weingarten equations can be written in the
indicial form
∂b
n ∂~r
= −bβα β (1.5.44)
∂uα ∂u
where bβα = aβγ bγα is the mixed second order form of the curvature tensor.
The equations of Gauss produce a system of partial differential equations defining the surface coordinates
i
x as a function of the curvilinear coordinates u and v. The equations are not independent as certain
compatibility conditions must be satisfied. In particular, it is required that the mixed partial derivatives
must satisfy
∂ 3~r ∂ 3~r
= .
∂uα ∂uβ ∂uδ ∂uα ∂uδ ∂uβ
We calculate
∂
γ
∂ 3~r γ ∂ 2~r αβ ∂~r ∂b
n ∂bαβ
= + + bαβ δ + b
n
∂uα ∂uβ ∂uδ αβ ∂uγ ∂uδ ∂uδ ∂uγ ∂u ∂uδ
and use the equations of Gauss and Weingarten to express this derivative in the form
∂
ω
∂ 3~r αβ γ ω r
ω ∂~ γ ∂bαβ
= + − b b + b + b.
n
∂uα ∂uβ ∂uδ ∂uδ αβ γδ
αβ δ
∂uω αβ
γδ
∂uδ
138
These equations are sometimes referred to as the MainardiCodazzi equations. Equating to zero the coefficient
of ∂~
r
∂uω we find that Rδαγβ = bαβ bδγ − bαγ bδβ or changing indices we have the covariant form
where
∂ δ ∂ δ ω δ ω δ
Rδαγβ = − + − (1.5.47)
∂uγ αβ ∂uβ αγ αβ ωγ αγ ωβ
is the mixed Riemann curvature tensor.
EXAMPLE 1.51
Show that the Gaussian or total curvature K = κ(1) κ(2) depends only upon the metric aαβ and is
R1212
K= where a = det(aαβ ).
a
Solution:
Utilizing the twodimensional alternating tensor eαβ and the property of determinants we can write
eγδ K = eαβ bγα bδβ where from page 137, K = bγβ  = aαγ bαβ . Now multiply by eγζ and then contract on
ζ and δ to obtain
eγδ eγδ K = eγδ eαβ bγα bδβ = 2K
2K = eγδ eαβ (aγµ bαu ) aδν bβν
√
But eγδ aγµ aδν = aeµν so that 2K = eαβ a eµν bαµ bβν . Using aeµν = µν we have 2K = µν αβ bαµ bβν .
Interchanging indices we can write
Adding these last two results we find that 4K = βγ ωγ (bωβ bαγ − bωγ bαβ ) = βγ ωγ Rωαβγ . Now multiply
βγ ωα
both sides by στ λν to obtain 4Kστ λν = δστ δλν Rωαβγ . From exercise 1.5, problem 16, the Riemann
curvature tensor Rijkl is skew symmetric in the (i, j), (k, l) as well as being symmetric in the (ij), (kl) pair
βγ ωα
of indices. Consequently, δστ δλν Rωαβγ = 4Rλνστ and hence Rλνστ = Kστ λν and we have the special case
√ √ R1212 b
where K ae12 ae12 = R1212 or K = . A much simpler way to obtain this result is to observe K =
a a
(bottom of page 137) and note from equation (1.5.46) that R1212 = b11 b22 − b12 b21 = b.
Note that on a surface ds2 = aαβ duα duβ where aαβ are the metrices for the surface. This metric is a
∂uα ∂uβ
tensor and satisfies āγδ = aαβ γ and by taking determinants we find
∂ ū ∂ ūδ
∂uα ∂uβ
ā = āγδ γ δ = aJ 2
∂ ū ∂ ū
139
where J is the Jacobian of the surface coordinate transformation. Here the curvature tensor for the surface
Rαβγδ has only one independent component since R1212 = R2121 = −R1221 = −R2112 (See exercises 20,21).
From the transformation law
∂uα ∂uβ ∂uγ ∂uδ
R̄ηλµ = Rαβγδ
∂ ū ∂ ūη ∂ ūλ ∂ ūµ
one can sum over the repeated indices and show that R̄1212 = R1212 J 2 and consequently
R̄1212 R1212
= =K
ā a
Geodesic Curvature
d~r du dv
T~ = = ~ru + ~rv
ds ds ds
~
dT
=K~ = ~ruu (u 0 )2 + 2~ruv u 0 v 0 + ~rvv (v 0 )2 + ~ru u 00 + ~rv v 00 ,
ds
0 d
where = ds , and by utilizing the results from problem 10 of the exercises following this section, we find
that the geodesic curvature can be represented as
2 0 3 2 1
κ(g) = (u ) + 2 − (u 0 )2 v 0 +
11 12 11
p (1.5.48)
2 1 1
−2 u 0 (v 0 )2 − (v 0 )3 + (u 0 v 00 − u 00 v 0 ) EG − F 2 .
22 12 22
This equation indicates that the geodesic curvature is only a function of the surface metrices E, F, G and
the derivatives u 0 , v 0 , u 00 , v 00 . When the geodesic curvature is zero the curve is called a geodesic curve. Such
curves are often times, but not always, the lines of shortest distance between two points on a surface. For
example, the great circle on a sphere which passes through two given points on the sphere is a geodesic curve.
If you erase that part of the circle which represents the shortest distance between two points on the circle
you are left with a geodesic curve connecting the two points, however, the path is not the shortest distance
between the two points.
For plane curves we let u = x and v = y so that the geodesic curvature reduces to
dφ
kg = u 0 v 00 − u 00 v 0 =
ds
140
where φ is the angle between the tangent T~ to the curve and the unit vector b
e1 .
Geodesics are curves on the surface where the geodesic curvature is zero. Since kg = 0 along a geodesic
surface curve, then at every point on this surface curve the normal N ~ to the curve will be in the same
b to the surface. In this case, we have ~ru · n
direction as the normal n b = 0 and ~rv · n
b = 0 which reduces to
dT~ dT~
· ~ru = 0 and · ~rv = 0, (1.5.49)
ds ds
~
b and
since the vectors n dT
ds have the same direction. In particular, we may write
dT~
· ~ru = (~ruu · ~ru ) (u 0 )2 + 2(~ruv · ~ru ) u 0 v 0 + (~rvv · ~ru ) (v 0 )2 + Eu 00 + F v 00 = 0
ds . (1.5.50)
dT~ 0 2 0 0 0 2 00 00
· ~rv = (~ruu · ~rv ) (u ) + 2(~ruv · ~rv ) u v + (~rvv · ~rv ) (v ) + F u + Gv = 0.
ds
Utilizing the results from exercise 1.5,(See problems 4,5 and 6), we can eliminate v 00 from the equations
(1.5.50) to obtain
2 2
d2 u 1 du 1 du dv 1 dv
+ +2 + =0
ds2 11 ds 12 ds ds 22 ds
and eliminating u 00 from the equations (1.5.50) produces the equation
2 2
d2 v 2 du 2 du dv 2 dv
+ +2 + = 0.
ds2 11 ds 12 ds ds 22 ds
where u = u1 and v = u2 . The equations (1.5.51) are the differential equations defining a geodesic curve on
a surface. We will find that these same type of equations arise in considering the shortest distance between
two points in a generalized coordinate system. See for example problem 18 in exercise 2.2.
141
Tensor Derivatives
Let uα = uα (t) denote the parametric equations of a curve on the surface defined by the parametric
equations xi = xi (u1 , u2 ). We can then represent the surface curve in the spatial geometry since the surface
curve can be represented in the spatial coordinates through the representation xi = xi (u1 (t), u2 (t)) = xi (t).
Recall that for xi = xi (t) a given curve C , the intrinsic derivative of a vector field Ai along C is defined as
the inner product of the covariant derivative of the vector field with the tangent vector to the curve. This
intrinsic derivative is written
" #
δAi i dx
j
∂Ai i k dx
j
= A ,j = j
+ A
δt dt ∂x jk g dt
or
δAi dAi i dxj
= + Ak
δt dt jk g dt
where the subscript g indicates that the Christoffel symbol is formed from the spatial metric gij . If Aα is a
surface vector defined along the curve C, the intrinsic derivative is represented
α β
δAα duβ ∂A α γ du
= Aα,β = + A
δt dt ∂uβ βγ a dt
or
δAα dAα α duβ
= + Aγ
δt dt βγ a dt
where the subscript a denotes that the Christoffel is formed from the surface metric aαβ .
Similarly, the formulas for the intrinsic derivative of a covariant spatial vector Ai or covariant surface
vector Aα are given by
δAi dAi k dxj
= − Ak
δt dt ij g dt
and
δAα dAα γ duβ
= − Aα .
δt dt αβ a dt
Consider a mixed tensor Tαi which is contravariant with respect to a transformation of space coordinates
x and covariant with respect to a transformation of surface coordinates uα . For Tαi defined over the surface
i
curve C, which can also be viewed as a space curve C, define the scalar invariant Ψ = Ψ(t) = Tαi Ai B α where
Ai is a parallel vector field along the curve C when it is viewed as a space curve and B α is also a parallel
vector field along the curve C when it is viewed as a surface curve. Recall that these parallel vector fields
must satisfy the differential equations
δAi dAi k dxj δB α dB α α duβ
= − Ak = 0 and = + Bγ = 0. (1.5.52)
δt dt ij g dt δt dt βγ a dt
The scalar invariant Ψ is a function of the parameter t of the space curve since both the tensor and the
parallel vector fields are to be evaluated along the curve C. By differentiating the function Ψ with respect
to the parameter t there results
dΨ dT i dAi α dB α
= α Ai B α + Tαi B + Tαi Ai . (1.5.53)
dt dt dt dt
142
But the vectors Ai and B α are parallel vector fields and must satisfy the relations given by equations (1.5.52).
This implies that equation (1.5.53) can be written in the form
" #
dΨ dTαi i k dx
j
γ i du
β
= + T − T Ai B α . (1.5.54)
dt dt k j g α dt β α a γ dt
The quantity inside the brackets of equation (1.5.54) is defined as the intrinsic tensor derivative with respect
to the parameter t along the curve C. This intrinsic tensor derivative is written
δTαi dTαi i k dx
j
γ duβ
= + Tα − Tγi . (1.5.55)
dt dt kj g dt βα a dt
The spatial representation of the curve C is related to the surface representation of the curve C through the
defining equations. Therefore, we can express the equation (1.5.55) in the form
" #
δTαi ∂Tαi i k ∂x
j
γ i du
β
= + T − T (1.5.56)
dt ∂uβ k j g α ∂uβ β α a γ dt
The quantity inside the brackets is a mixed tensor which is defined as the tensor derivative of Tαi with
respect to the surface coordinates uβ . The tensor derivative of the mixed tensor Tαi with respect to the
surface coordinates uβ is written
∂Tαi i ∂xj γ
i
Tα,β = + Tαk − Tγi .
∂uβ kj g ∂uβ βα a
i...j
In general, given a mixed tensor Tα...β which is contravariant with respect to transformations of the
space coordinates and covariant with respect to transformations of the surface coordinates, then we can
define the scalar field along the surface curve C as
i...j
Ψ(t) = Tα...β Ai · · · Aj B α · · · B β (1.5.57)
where Ai , . . . , Aj and B α , . . . , B β are parallel vector fields along the curve C. The intrinsic tensor derivative
is then derived by differentiating the equation (1.5.57) with respect to the parameter t.
Tensor derivatives of the metric tensors gij , aαβ and the alternating tensors ijk , αβ and their associated
tensors are all zero. Hence, they can be treated as constants during the tensor differentiation process.
Generalizations
In a Riemannian space Vn with metric gij and curvilinear coordinates xi , i = 1, 2, 3, the equations of a
surface can be written in the parametric form xi = xi (u1 , u2 ) where uα , α = 1, 2 are called the curvilinear
coordinates of the surface. Since
∂xi α
dxi = du (1.5.58)
∂uα
then a small change duα on the surface results in change dxi in the space coordinates. Hence an element of
arc length on the surface can be represented in terms of the curvilinear coordinates of the surface. This same
element of arc length can also be represented in terms of the curvilinear coordinates of the space. Thus, an
element of arc length squared in terms of the surface coordinates is represented
where aαβ is the metric of the surface. This same element when viewed as a spatial element is represented
∂xi ∂xj α β
gij dxi dxj = gij du du = aαβ duα duβ . (1.5.61)
∂uα ∂uβ
The equation (1.5.61) shows that the surface metric is related to the spatial metric and can be calculated
∂xi ∂xj
from the relation aαβ = gij α β . This equation reduces to the equation (1.5.21) in the special case of
∂u ∂u
Cartesian coordinates. In the surface coordinates we define the quadratic form A = aαβ duα duβ as the first
fundamental form of the surface. The tangent vector to the coordinate curves defining the surface are given
∂xi
by ∂uα and can be viewed as either a covariant surface vector or a contravariant spatial vector. We define
this vector as
∂xi
xiα = , i = 1, 2, 3, α = 1, 2. (1.5.62)
∂uα
Any vector which is a linear combination of the tangent vectors to the coordinate curves is called a surface
vector. A surface vector Aα can also be viewed as a spatial vector Ai . The relation between the spatial
representation and surface representation is Ai = Aα xiα . The surface representation Aα , α = 1, 2 and the
spatial representation Ai , i = 1, 2, 3 define the same direction and magnitude since
Consider any two surface vectors Aα and B α and their spatial representations Ai and B i where
These vectors are tangent to the surface and so a unit normal vector to the surface can be defined from the
cross product relation
ni AB sin θ = ijk Aj B k (1.5.64)
where A, B are the magnitudes of Ai , B i and θ is the angle between the vectors when their origins are made
to coincide. Substituting equations (1.5.63) into the equation (1.5.64) we find
In terms of the surface metric we have AB sin θ = αβ Aα B β so that equation (1.5.65) can be written in the
form
(ni αβ − ijk xjα xkβ )Aα B β = 0 (1.5.66)
is now demonstrated. By using the results from equation (1.5.50), the tensor derivative of equation (1.5.59),
with respect to the surface coordinates, produces
∂ 2 xi i σ
xiα,β = + xpα xqβ − xiσ (1.5.68)
∂uα ∂uβ pq g αβ a
where the subscripts on the Christoffel symbols refer to the metric from which they are calculated. Also the
tensor derivative of the equation (1.5.57) produces the result
Interchanging the indices α, β, γ cyclically in the equation (1.5.69) one can verify that
The equation (1.5.70) indicates that in terms of the space coordinates, the vector xiα,β is perpendicular to
the surface tangent vector xiγ and so must have the same direction as the unit surface normal ni . Therefore,
there must exist a second order tensor bαβ such that
By using the relation gij ni nj = 1 we can transform equation (1.5.71) to the form
1 γδ
bαβ = gij nj xiα,β = ijk xiα,β xjγ xkδ . (1.5.72)
2
The second order symmetric tensor bαβ is called the curvature tensor and the quadratic form
Taking the tensor derivative of gij ni nj = 1 with respect to the surface coordinates produces the result
gij ni nj,α = 0 which shows that the vector nj,α is perpendicular to ni and must lie in the tangent plane to the
surface. It can therefore be expressed as a linear combination of the surface tangent vectors xiα and written
in the form
ni,α = ηαβ xiβ (1.5.75)
where the coefficients ηαβ can be written in terms of the surface metric components aαβ and the curvature
components bαβ as follows. The unit vector ni is normal to the surface so that
The tensor derivative of this equation with respect to the surface coordinates gives
Substitute into equation (1.5.77) the relations from equations (1.5.57), (1.5.71) and (1.5.75) and show that
Now substituting equation (1.5.79) into the equation (1.5.75) produces the Weingarten formula
This is a relation for the derivative of the unit normal in terms of the surface metric, curvature tensor and
surface tangents.
A third fundamental form of the surface is given by the quadratic form
By using the Weingarten formula in the equation (1.5.81) one can verify that
Geodesic Coordinates
In a Cartesian coordinate system the metric tensor gij is a constant and consequently the Christoffel
symbols are zero at all points of the space. This is because the Christoffel symbols are dependent upon
the derivatives of the metric tensor which is constant. If the space VN is not Cartesian then the Christoffel
symbols do not vanish at all points of the space. However, it is possible to find a coordinate system where
the Christoffel symbols will all vanish at a given point P of the space. Such coordinates are called geodesic
coordinates of the point P.
Consider a two dimensional surface with surface coordinates uα and surface metric aαβ . If we transform
to some other two dimensional coordinate system, say ūα with metric āαβ , where the two coordinates are
related by transformation equations of the form
uα = uα (ū 1 , ū 2 ), α = 1, 2, (1.5.84)
146
then from the transformation equation (1.4.7) we can write, after changing symbols,
δ ∂uα α ∂uδ ∂u ∂ 2 uα
= + . (1.5.85)
β γ ā ∂ ū δ δ a ∂ ū β ∂ ū γ ∂ ū β ∂ ū γ
δ
This is a relationship between the Christoffel symbols in the two coordinate systems. If βγ
vanishes at
ā
a point P , then for that particular point the equation (1.5.85) reduces to
∂ 2 uα α ∂uδ ∂u
β γ
= − (1.5.86)
∂ ū ∂ ū δ a ∂ ū β ∂ ū γ
and
∂ 2 uα α
=− (1.5.89)
∂ ū τ ∂ ū σ τ σ a uα =0
where these derivative are evaluated at ū α = 0. We find the derivative equations (1.5.88) and (1.5.89) do
satisfy the equation (1.5.86) locally at the point P. Hence, the Christoffel symbols will all be zero at this
particular point. The new coordinates can then be called geodesic coordinates.
(see example 1.51) which is an invariant of the surface and called the Gaussian curvature or total curvature.
In the exercises following this section it is shown that the Riemann Christoffel tensor of the surface can be
expressed in terms of the total curvature and the alternating tensors as
Using the relation (1.5.96) we can now derive some interesting properties relating to the tensors aαβ , bαβ ,
cαβ , Rαβγδ , the mean curvature H and the total curvature K.
Consider the tensor derivative of the equation (1.5.71) which can be written
where
∂bαβ σ σ
bαβ,γ = − bσβ − bασ . (1.5.98)
∂uα αγ a βγ a
By using the Weingarten formula, given in equation (1.5.80), the equation (1.5.97) can be expressed in the
form
xiα,βγ = bαβ,γ ni − bαβ aτ σ bτ γ xiσ (1.5.99)
and by using the equations (1.5.98) and (1.5.99) it can be established that
Now by equating the results from the equations (1.5.96) and (1.5.100) we arrive at the relation
Multiplying the equation (1.5.101) by nr and using the results from the equation (1.5.76) there results the
Codazzi equations
bαβ,γ − bαγ,β = 0. (1.5.102)
By using the Gauss equations (1.5.103) the equation (1.5.94) can be written as
Another form of equation (1.5.104) is obtained by using the equation (1.5.83) together with the relation
aαβ = −aσγ σα βγ . It is left as an exercise to verify the resulting form
C − 2H B + K A = 0 (1.5.109)
EXAMPLE 1.52
In a two dimensional space the Riemann Christoffel tensor has only one nonzero independent component
R1212 . ( See Exercise 1.5, problem number 21.) Consequently, the equation (1.5.104) can be written in the
√ √
form K ae12 ae12 = b22 b11 − b21 b12 and solving for the Gaussian curvature K we find
Surface Curvature
For a surface curve uα = uα (s),α = 1, 2 lying upon a surface xi = xi (u1 , u2 ),i = 1, 2, 3, we have a two
duα
dimensional space embedded in a three dimensional space. Thus, if tα = is a unit tangent vector to
ds
α β
du du
the surface curve then aαβ = aαβ tα tβ = 1. This same vector can be represented as the unit tangent
ds ds
dxi dxi dxj
vector to the space curve xi = xi (u1 (s), u2 (s)) with T i = . That is we will have gij = gij T i T j = 1.
ds ds ds
The surface vector tα and the space vector T i are related by
∂xi duα
Ti = = xiα tα . (1.5.111)
∂uα ds
The surface vector tα is a unit vector so that aαβ tα tβ = 1. If we differentiate this equation intrinsically with
β
δtα
respect to the parameter s, we find that aαβ tα δtδs = 0. This shows that the surface vector δs is perpendicular
α α
to the surface vector t . Let u denote a unit normal vector in the surface plane which is orthogonal to the
tangent vector tα . The direction of uα is selected such that αβ tα uβ = 1. Therefore, there exists a scalar κ(g)
such that
δtα
= κ(g) uα (1.5.112)
δs
149
δuα
where κ(g) is called the geodesic curvature of the curve. In a similar manner it can be shown that δs
δuα
is a surface vector orthogonal to tα . Let δs = αtα where α is a scalar constant to be determined. By
differentiating the relation aαβ tα uβ = 0 intrinsically and simplifying we find that α = −κ(g) and therefore
δuα
= −κ(g) tα . (1.5.113)
δs
The equations (1.5.112) and (1.5.113) are sometimes referred to as the FrenetSerret formula for a curve
relative to a surface.
Taking the intrinsic derivative of equation (1.5.111), with respect to the parameter s, we find that
δT i δtα duβ α
= xiα + xiα,β t . (1.5.114)
δs δs ds
Treating the curve as a space curve we use the Frenet formulas (1.5.13). If we treat the curve as a surface
curve, then we use the Frenet formulas (1.5.112) and (1.5.113). In this way the equation (1.5.114) can be
written in the form
κN i = xiα κ(g) uα + xiα,β tβ tα . (1.5.115)
where ui is the space vector counterpart of the surface vector uα . Let θ denote the angle between the surface
normal ni and the principal normal N i , then we have that cos θ = ni N i . Hence, by multiplying the equation
(1.5.116) by ni we obtain
κ cos θ = bαβ tα tβ . (1.5.117)
Consequently, for all curves on the surface with the same tangent vector tα , the quantity κ cos θ will remain
constant. This result is known as Meusnier’s theorem. Note also that κ cos θ = κ(n) is the normal component
of the curvature and κ sin θ = κ(g) is the geodesic component of the curvature. Therefore, we write the
equation (1.5.117) as
κ(n) = bαβ tα tβ (1.5.118)
which represents the normal curvature of the surface in the direction tα . The equation (1.5.118) can also be
written in the form
duα duβ B
κ(n) = bαβ = (1.5.119)
ds ds A
which is a ratio of quadratic forms.
The surface directions for which κ(n) has a maximum or minimum value is determined from the equation
(1.5.119) which is written as
(bαβ − κ(n) aαβ )λα λβ = 0. (1.5.120)
The direction giving a maximum or minimum value to κ(n) must then satisfy
b
κ2(n) − aαβ bαβ κ(n) + =0 (1.5.123)
a
where a = a11 a22 − a12 a21 and b = b11 b22 − b12 b21 . Using the definition given in equation (1.5.107) and using
the result from equation (1.5.110), the equation (1.5.123) can be expressed in the form
The roots κ(1) and κ(2) of the equation (1.5.124) then satisfy the relations
1
H= (κ(1) + κ(2) ) (1.5.125)
2
and
K = κ(1) κ(2) . (1.5.126)
Here H is the mean value of the principal curvatures and K is the Gaussian or total curvature which is the
product of the principal curvatures. It is readily verified that
Eg − 2f F + eG eg − f 2
H= and K =
2(EG − F 2 ) EG − F 2
are invariants obtained from the surface metric and curvature tensor.
Relativity
Sir Isaac Newton and Albert Einstein viewed the world differently when it came to describing gravity and
the motion of the planets. In this brief introduction to relativity we will compare the Newtonian equations
with the relativistic equations in describing planetary motion. We begin with an examination of Newtonian
systems.
Newton’s viewpoint of planetary motion is a multiple bodied problem, but for simplicity we consider
only a two body problem, say the sun and some planet where the motion takes place in a plane. Newton’s
law of gravitation states that two masses m and M are attracted toward each other with a force of magnitude
GmM
ρ2 , where G is a constant, ρ is the distance between the masses, m is the mass of the planet and M is the
mass of the sun. One can construct an x, y plane containing the two masses with the origin located at the
eρ = cos φ b
center of mass of the sun. Let b e1 + sin φ b
e2 denote a unit vector at the origin of this coordinate
system and pointing in the direction of the mass m. The vector force of attraction of mass M on mass m is
given by the relation
−GmM
F~ = b
eρ . (1.5.127)
ρ2
151
The equation of motion of mass m with respect to mass M is obtained from Newton’s second law. Let
~ = ρb
ρ eρ denote the position vector of mass m with respect to the origin. Newton’s second law can then be
written in any of the forms
−GmM d2 ρ
~ ~
dV −GmM
F~ = b
e ρ = m = m = ρ~ (1.5.128)
ρ2 dt2 dt ρ3
and from this equation we can show that the motion of the mass m can be described as a conic section.
Recall that a conic section is defined as a locus of points p(x, y) such that the distance of p from a fixed
point (or points), called a focus (foci), is proportional to the distance of the point p from a fixed line, called
a directrix, that does not contain the fixed point. The constant of proportionality is called the eccentricity
and is denoted by the symbol . For = 1 a parabola results; for 0 ≤ ≤ 1 an ellipse results; for > 1 a
hyperbola results; and if = 0 the conic section is a circle.
FP
With reference to figure 1.52, a conic section is defined in terms of the ratio PD
= where F P = ρ and
P D = 2q − ρ cos φ. From the ratio we solve for ρ and obtain the polar representation for the conic section
p
ρ= (1.5.129)
1 + cos φ
152
where p = 2q and the angle φ is known as the true anomaly associated with the orbit. The quantity p is
π
called the semiparameter of the conic section. (Note that when φ = 2, then ρ = p.) A more general form
of the above equation is
p 1
ρ= or u = = A[1 + cos(φ − φ0 )], (1.5.130)
1 + cos(φ − φ0 ) ρ
where φ0 is an arbitrary starting anomaly. An additional symbol a, known as the semimajor axes of an
elliptical orbit can be introduced where q, p, , a are related by
p
= q = a(1 − ) or p = a(1 − 2 ). (1.5.131)
1+
To show that the equation (1.5.128) produces a conic section for the motion of mass m with respect to
mass M we will show that one form of the solution of equation (1.5.128) is given by the equation (1.5.129).
To verify this we use the following vector identities:
~× b
ρ eρ =0
d d~
ρ d2 ρ
~
~×
ρ ρ× 2
=~
dt dt dt
dbeρ (1.5.132)
b
eρ · =0
dt
dbeρ dbeρ
eρ × b
b eρ × =− .
dt dt
d~
ρ ~
~×
ρ = h = constant. (1.5.134)
dt
The quantity H ~ = ρ ~ = ρ
~ × mV ~ × m d~
ρ ~
dt is the angular momentum of the mass m so that the quantity h
represents the angular momentum per unit mass. The equation (1.5.134) tells us that ~h is a constant for our
two body system. Note that because ~h is constant we have
d ~ ~ dV
~ GM d~
ρ
V ×h = ×h=− 2 b
~ eρ × ρ ~×
dt dt ρ dt
GM dbeρ dρ
=− 2 b ρb
eρ × [~ eρ × (ρ + b
eρ )]
ρ dt dt
GM dbeρ 2 dbeρ
=− 2 b
eρ × ( b
eρ × )ρ = GM
ρ dt dt
~ × ~h = GM b
V ~
eρ + C
153
where p = h2 /GM and = C/GM. This result is known as Kepler’s first law and implies that when < 1
the mass m describes an elliptical orbit with the sun at one focus.
We present now an alternate derivation of equation (1.5.130) for later use. From the equation (1.5.128)
we have
ρ d2 ρ
d~ ~ d d~
ρ d~ρ GM d~
ρ GM d
2 · 2 = · = −2 3
~·
ρ =− 3 ρ·ρ
(~ ~) . (1.5.137)
dt dt dt dt dt ρ dt ρ dt
Consider the equation (1.5.137) in spherical coordinates ρ, θ, φ. The tensor velocity components are V 1 = dρ
dt ,
V2 = dθ
dt , V3 = dφ
dt and the physical components of velocity are given by Vρ = dρ
dt , Vθ = ρ dθ
dt , Vφ = ρ sin θ dφ
dt
so that the velocity can be written
d~
ρ dρ dθ dφ
V~ = = b
eρ + ρ b eθ + ρ sin θ b
eφ . (1.5.138)
dt dt dt dt
Substituting equation (1.5.138) into equation (1.5.137) gives the result
" 2 2 #
2
d dρ 2 dθ 2 2 dφ GM d 2 2GM dρ d 1
+ρ + ρ sin θ =− 3 (ρ ) = − 2 = 2GM
dt dt dt dt ρ dt ρ dt dt ρ
1
The substitution ρ = u can be used to represent the equation (1.5.142) in the form
2
du 2GM E
+ u2 − 2
u+ 2 =0 (1.5.143)
dφ h h
which is a form we will return to later in this section. Note that we can separate the variables in equations
(1.5.142) or (1.5.143). The results can then be integrate to produce the equation (1.5.130).
Newton also considered the relative motion of two inertial systems, say S and S. Consider two such
systems as depicted in the figure 1.53 where the S system is moving in the x−direction with speed v relative
to the system S.
For a Newtonian system, if at time t = 0 we have clocks in both systems which coincide, than at time t
a point P (x, y, z) in the S system can be described by the transformation equations
x =x + vt x =x − vt
y =y y =y
or (1.5.144)
z =z z =z
t =t t =t.
These are the transformation equation of Newton’s relativity sometimes referred to as a Galilean transfor
mation.
Before Einstein the principle of relativity required that velocities be additive and obey Galileo’s velocity
addition rule
VP/R = VP/Q + VQ/R . (1.5.145)
155
That is, the velocity of P with respect to R equals the velocity of P with respect to Q plus the velocity of Q
with respect to R. For example, a person (P ) running north at 3 km/hr on a train (Q) moving north at 60
km/hr with respect to the ground (R) has a velocity of 63 km/hr with respect to the ground. What happens
when (P ) is a light wave moving on a train (Q) which is moving with velocity V relative to the ground? Are
the velocities still additive? This type of question led to the famous MichelsonMorley experiment which
has been labeled as the starting point for relativity. Einstein’s answer to the above question was ”NO” and
required that VP/R = VP/Q = c =speed of light be a universal constant.
In contrast to the Newtonian equations, Einstein considered the motion of light from the origins 0 and
0 of the systems S and S. If the S system moves with velocity v relative to the S system and at time t = 0
a light signal is sent from the S system to the S system, then this light signal will move out in a spherical
wave front and lie on the sphere
x2 + y 2 + z 2 = c2 t2 (1.5.146)
where c is the speed of light. Conversely, if a light signal is sent out from the S system at time t = 0, it will
lie on the spherical wave front
2
x2 + y 2 + z 2 = c2 t . (1.5.147)
Observe that the Newtonian equations (1.5.144) do not satisfy the equations (1.5.146) and (1.5.147) identi
cally. If y = y and z = z then the space variables (x, x) and time variables (t, t) must somehow be related.
Einstein suggested the following transformation equations between these variables
dx dx
When = = c, the speed of light, the equation (1.5.150) requires that
dt dt
v 2 −1 v 2 −1/2
γ 2 = (1 − ) or γ = (1 − ) . (1.5.151)
c2 c2
From the equations (1.5.148) we eliminate x and find
v
t = γ(t − x). (1.5.152)
c2
We can now replace the Newtonian equations (1.5.144) by the relativistic transformation equations
where γ is given by equation (1.5.151). These equations are also known as the Lorentz transformation.
v
Note that for v << c, then 2 ≈ 0, γ ≈ 1 , then the equations (1.5.153) closely approximate the equations
c
(1.5.144). The equations (1.5.153) also satisfy the equations (1.5.146) and (1.5.147) identically as can be
readily verified by substitution. Further, by using chain rule differentiation we obtain from the relations
(1.5.148) that
dx
dx dt
+v
= dx
. (1.5.154)
dt v
1+ dt
c c
The equation (1.5.154) is the Einstein relative velocity addition rule which replaces the previous Newtonian
rule given by equation (1.5.145). We can rewrite equation (1.5.154) in the notation of equation (1.5.145) as
VP/Q + VQ/R
VP/R = VP/Q VQ/R
. (1.5.155)
1+ c c
Observe that when VP/Q << c and VQ/R << c then equation (1.5.155) approximates closely the equation
(1.5.145). Also as VP/Q and VQ/R approach the speed of light we have
VP/Q + VQ/R
lim VP/Q VQ/R
=c (1.5.156)
VP/Q →C
VQ/R →C
1+ c c
which agrees with Einstein’s hypothesis that the speed of light is an invariant.
Let us return now to the viewpoint of what gravitation is. Einstein thought of space and time as being
related and viewed the motion of the planets as being that of geodesic paths in a spacetime continuum.
Recall the equations of geodesics are given by
d2 xi i dxj dxk
+ = 0, (1.5.157)
ds2 jk ds ds
where s is arc length. These equations are to be associated with a 4dimensional spacetime metric gij
where the indices i, j take on the values 1, 2, 3, 4 and the xi are generalized coordinates. Einstein asked
the question, ”Can one introduce a spacetime metric gij such that the equations (1.5.157) can somehow
d2 ρ
~ GM
reproduce the law of gravitational attraction dt2 + ρ3 ρ
~ = 0?” Then the motion of the planets can be
viewed as optimized motion in a spacetime continuum where the metrices of the space simulate the law of
gravitational attraction. Einstein thought that this motion should be related to the curvature of the space
which can be obtained from the RiemannChristoffel tensor Rijkl . The metric we desire gij , i, j = 1, 2, 3, 4
has 16 components. The conjugate metric tensor g ij is defined such that g ij gjk = δki and an element of
arc length squared is given by ds2 = gij dxi dxj . Einstein thought that the metrices should come from the
RiemannChristoffel curvature tensor which, for n = 4 has 256 components, but only 20 of these are linearly
independent. This seems like a large number of equations from which to obtain the law of gravitational
attraction and so Einstein considered the contracted tensor
∂ n ∂ n m n m n
Gij = Rtijt = − + − . (1.5.158)
∂xj i n ∂xn i j in mj ij mn
where g11 = −1, g22 = −ρ2 , g33 = −ρ2 sin2 θ, g44 = c2 and gij = 0 for i 6= j. The negative signs are
2
introduced so that ds
dt = c2 − v 2 is positive when v < c and the velocity is not greater than c. However,
this metric will not work since the curvature tensor vanishes. The spherical symmetry of the problem suggest
that g11 and g44 change while g22 and g33 remain fixed. Let (x1 , x2 , x3 , x4 ) = (ρ, θ, φ, t) and assume
where u and v are unknown functions of ρ to be determined. This gives the conjugate metric tensor
−1 −1
g 11 = −e−u , g 22 = , g 33 = , g 44 = e−v (1.5.160)
ρ2 ρ2 sin2 θ
The equation (1.5.158) is used to calculate the nonzero Gij and we find that
2
1 d2 v 1 dv 1 du dv 1 du
G11 = 2
+ − −
2 dρ 4 dρ 4 dρ dρ ρ dρ
1 dv 1 du
G22 =e−u 1 + ρ − ρ − eu
2 dρ 2 dρ
(1.5.163)
1 dv 1 du
G33 =e−u 1 + ρ − ρ − eu sin2 θ
2 dρ 2 dρ
2 !
2
1 d v 1 du dv 1 dv 1 dv
G44 = − ev−u − + +
2 dρ2 4 dρ dρ 4 dρ ρ dρ
and Gij = 0 for i 6= j. The assumption that Gij = 0 for all i, j leads to the differential equations
2
d2 v 1 dv
1 du dv 2 du
+ − −
=0
dρ2 2 dρ
2 dρ dρ ρ dρ
1 dv 1 du
1+ ρ − ρ − eu =0 (1.5.164)
2 dρ 2 dρ
2
d2 v 1 dv 1 du dv 2 dv
2
+ − + =0.
dρ 2 dρ 2 dρ dρ ρ dρ
158
du dv
+ =0 or u + v = c1 = constant. (1.5.165)
dρ dρ
du
ρ = 1 − eu (1.5.166)
dρ
Separate the variables in equation (1.5.166) and integrate to obtain the result
1
eu = (1.5.167)
1 − cρ2
The constant c1 is selected such that g44 approaches c2 as ρ increases without bound. This produces the
metrices
−1 c2
g11 = , g22 = −ρ2 , g33 = −ρ2 sin2 θ, g44 = c2 (1 − ) (1.5.169)
1 − cρ2 ρ
where c2 is a constant still to be determined. The metrices given by equation (1.5.169) are now used to
expand the equations (1.5.157) representing the geodesics in this four dimensional space. The differential
equations representing the geodesics are found to be
2 2 2 2
d2 ρ 1 du dρ dθ dφ 1 v−u dv dt
+ − ρe−u − ρe−u sin2 θ + e =0 (1.5.170)
ds2 2 dρ ds ds ds 2 dρ ds
2
d2 θ 2 dθ dρ dφ
+ − sin θ cos θ =0 (1.5.171)
ds2 ρ ds ds ds
d2 φ 2 dφ dρ cos θ dφ dθ
+ +2 =0 (1.5.172)
ds2 ρ ds ds sin θ ds ds
d2 t dv dt dρ
+ = 0. (1.5.173)
ds2 dρ ds ds
π
The equation (1.5.171) is identically satisfied if we examine planar orbits where θ = 2 is a constant. This
value of θ also simplifies the equations (1.5.170) and (1.5.172). The equation (1.5.172) becomes an exact
differential equation
d dφ dφ
ρ2 =0 or ρ2 = c4 , (1.5.174)
ds ds ds
and the equation (1.5.173) also becomes an exact differential
d dt v dt v
e =0 or e = c5 , (1.5.175)
ds ds ds
where c4 and c5 are constants of integration. This leaves the equation (1.5.170) which determines ρ. Substi
tuting the results from equations (1.5.174) and (1.5.175), together with the relation (1.5.161), the equation
(1.5.170) reduces to
d2 ρ c2 c2 c24 c2 c24
+ + − (1 − ) = 0. (1.5.176)
ds2 2ρ2 2ρ4 ρ ρ3
159
d2 u c2 3
+ u − 2 = c2 u 2 . (1.5.178)
dφ2 2c4 2
du
Multiply the equation (1.5.178) by 2 dφ and integrate with respect to φ to obtain
2
du c2
+ u2 − u = c2 u 3 + c 6 . (1.5.179)
dφ c24
where c6 is a constant of integration. To determine the constant c6 we write the equation (1.5.161) in the
π
special case θ = 2 and use the substitutions from the equations (1.5.174) and (1.5.175) to obtain
2 2 2 2
dρ dρ dφ 2 dφ dt
e u
=e u
=1−ρ +e v
ds dφ ds ds ds
or 2
dρ c2 c2 c2 ρ 4
+ 1− ρ2 + 1 − − 52 = 0. (1.5.180)
dφ ρ ρ c c24
1
The substitution ρ = u reduces the equation (1.5.180) to the form
2
du 1 c2 c2
+ u 2 − c2 u 3 + 2 − 2 u − 2 5 2 = 0. (1.5.181)
dφ c4 c4 c c4
Now we can compare our relativistic equation (1.5.182) with our Newtonian equation (1.5.143). In order
that the two equations almost agree we select the constants c2 , c4 , c5 so that
c2
c2 2GM 1 − c52 E
2 = and 2 = 2. (1.5.183)
c4 h2 c4 h
The equations (1.5.183) are only two equations in three unknowns and so we use the additional equation
dφ dφ ds
lim ρ2 = lim ρ2 =h (1.5.184)
ρ→∞ dt ρ→∞ ds dt
160
which is obtained from equation (1.5.141). Substituting equations (1.5.174) and (1.5.175) into equation
(1.5.184), rearranging terms and taking the limit we find that
c4 c2
= h. (1.5.185)
c5
These values substituted into equation (1.5.181) produce the differential equation
2
du 22GM E 2GM 1
+u − 2
u+ 2 = u3 . (1.5.187)
dφ h h c2 1 + E/c2
c2 2GM 2GM 1
Let α = c24
= h2 and β = c2 = c2 ( 1+E/c2 ) then the differential equation (1.5.178) can be written as
d2 u α 3
+ u − = βu2 . (1.5.188)
dφ2 2 2
1
u= = A(1 + cos(φ − φ0 )) (1.5.189)
ρ
and so we assume a solution to equation (1.5.188) of this same general form. We know that A is small and so
we make the assumption that the solution of equation (1.5.188) given by equation (1.5.189) is such that φ0 is
approximately constant and varies slowly as a function of Aφ. Observe that if φ0 = φ0 (Aφ), then dφ0
dφ = φ00 A
d 2 φ0
and dφ2 = φ000 A2 , where primes denote differentiation with respect to the argument of the function. (i.e.
Aφ for this problem.) The derivatives of equation (1.5.189) produce
du
= − A sin(φ − φ0 )(1 − φ00 A)
dφ
d2 u
=A3 sin(φ − φ0 )φ000 − A cos(φ − φ0 )(1 − 2Aφ00 + A2 (φ00 )2 )
dφ2
= − A cos(φ − φ0 ) + 2A2 φ00 cos(φ − φ0 ) + O(A3 ).
Substituting these derivatives into the differential equation (1.5.188) produces the equations
α 3β
2A2 φ00 cos(φ − φ0 ) + A − = A2 + 2A2 cos(φ − φ0 ) + 2 A2 cos2 (φ − φ0 ) + O(A3 ).
2 2
Now A is small so that terms O(A3 ) can be neglected. Equating the constant terms and the coefficient of
the cos(φ − φ0 ) terms we obtain the equations
α 3β 2 3β 2 2
A− = A 2A2 φ00 = 3βA2 + A cos(φ − φ0 ).
2 2 2
Treating φ0 as essentially constant, the above system has the approximate solutions
α 3β 3β
A≈ φ0 ≈ Aφ + A sin(φ − φ0 ) (1.5.190)
2 2 4
161
The solutions given by equations (1.5.190) tells us that φ0 varies slowly with time. For less than 1, the
elliptical motion is affected by this change in φ0 . It causes the semimajor axis of the ellipse to slowly rotate
dφ0
at a rate given by dt . Using the following values for the planet Mercury
This slow variation in Mercury’s semimajor axis has been observed and measured and is in agreement with
the above value. Newtonian mechanics could not account for the changes in Mercury’s semimajor axis, but
Einstein’s theory of relativity does give this prediction. The resulting solution of equation (1.5.188) can be
viewed as being caused by the curvature of the spacetime continuum.
The contracted curvature tensor Gij set equal to zero is just one of many conditions that can be assumed
in order to arrive at a metric for the spacetime continuum. Any assumption on the value of Gij relates to
imposing some kind of curvature on the space. Within the large expanse of our universe only our imaginations
limit us as to how space, time and matter interact. You can also imagine the existence of other tensor metrics
in higher dimensional spaces where the geodesics within the spacetime continuum give rise to the motion
of other physical quantities.
This short introduction to relativity is concluded with a quote from the NASA News@hg.nasa.gov news
release, spring 1998, Release:9851. “An international team of NASA and university researchers has found
the first direct evidence of a phenomenon predicted 80 years ago using Einstein’s theory of general relativity–
that the Earth is dragging space and time around itself as it rotates.”The news release explains that the
effect is known as frame dragging and goes on to say “Frame dragging is like what happens if a bowling
ball spins in a thick fluid such as molasses. As the ball spins, it pulls the molasses around itself. Anything
stuck in the molasses will also move around the ball. Similarly, as the Earth rotates it pulls spacetime in
its vicinity around itself. This will shift the orbits of satellites near the Earth.”This research is reported in
the journal Science.
162
EXERCISE 1.5
~
~ and τ = δN~ · B.
I 1. Let κ = ·N
δT
δs δs
~ Assume in turn that each of the intrinsic derivatives of T~ , N
~,B
~ are
some linear combination of T~ , N
~,B ~ and hence derive the FrenetSerret formulas of differential geometry.
I 2. Determine the given surfaces. Describe and sketch the curvilinear coordinates upon each surface.
2uv 2 2u2 v
e1 + v b
(a) ~r(u, v) = u b e2 (b) ~r(u, v) = u cos v b
e1 + u sin v b
e2 (c) ~r(u, v) = b
e 1 + b
e2 .
u2 + v 2 u2 + v 2
I 3. Determine the given surfaces and describe the curvilinear coordinates upon the surface. Use some
graphics package to plot the surface and illustrate the coordinate curves on the surface. Find element of
area dS in terms of u and v.
e1 + b sin u sin v b
(a) ~r(u, v) = a sin u cos v b e2 + c cos u b
e3 a, b, c constants 0 ≤ u, v ≤ 2π
u u u
(b) ~r(u, v) = (4 + v sin ) cos u b e1 + (4 + v sin ) sin u b e2 + v cos b e3 − 1 ≤ v ≤ 1, 0 ≤ u ≤ 2π
2 2 2
e1 + bu sin v b
(c) ~r(u, v) = au cos v b e2 + cu be3
(d) ~r(u, v) = u cos v b
e1 + u sin v b
e2 + αv b
e3 α constant
e1 + b sin v b
(e) ~r(u, v) = a cos v b e2 + u b
e3 a, b constant
2
e2 + u b
e1 + u sin v b
(f ) ~r(u, v) = u cos v b e3
E F
I 4. Consider a two dimensional space with metric tensor (aαβ ) = . Assume that the surface is
F G
described by equations of the form y i = y i (u, v) and that any point on the surface is given by the position
vector ~r = ~r(u, v) = y i b
ei . Show that the metrices E, F, G are functions of the parameters u, v and are given
by
∂~r ∂~r
E = ~ru · ~ru , F = ~ru · ~rv , G = ~rv · ~rv where ~ru = and ~rv = .
∂u ∂v
I 5. For the metric given in problem 4 show that the Christoffel symbols of the first kind are given by
[1 1, 1] = ~ru · ~ruu [1 2, 1] = [2 1, 1] = ~ru · ~ruv [2 2, 1] = ~ru · ~rvv
[1 1, 2] = ~rv · ~ruu [1 2, 2] = [2 1, 2] = ~rv · ~ruv [2 2, 2] = ~rv · ~rvv
∂ 2~r ∂~r
which can be represented [α β, γ] = · , α, β, γ = 1, 2.
∂u ∂u ∂uγ
α β
I 6. Show that the results in problem 5 can also be written in the form
1 1 1
[1 1, 1] = Eu Ev
[1 2, 1] = [2 1, 1] = [2 2, 1] = Fv − Gu
2 2 2
1 1 1
[1 1, 2] = Fu − Ev [1 2, 2] = [2 1, 2] = Gu [2 2, 2] = Gv
2 2 2
where the subscripts indicate partial differentiation.
I 7. For the metricgivenin problem 4, show that the Christoffel symbols of the second kind can be
γ
expressed in the form = aγδ [α β, δ], α, β, γ = 1, 2 and produce the results
αβ
1 GEu − 2F Fu + F Ev 1 1 GEv − F Gu 2 2EFu − EEv − F Eu
= = = =
11 2(EG − F 2 ) 12 21 2(EG − F 2 ) 11 2(EG − F 2 )
1 2GFv − GGu − F Gv 2 2 EGu − F Ev 2 EGv − 2F Fv + F Gu
= 2
= = 2
=
22 2(EG − F ) 12 21 2(EG − F ) 22 2(EG − F 2 )
where the subscripts indicate partial differentiation.
163
b,
~ruu = c1~ru + c2~rv + c3 n b,
~ruv = c4~ru + c5~rv + c6 n b
~rvv = c7~ru + c8~rv + c9 n
b u = c1~ru + c2~rv
n ~ru = c∗1 n
bu + c∗2 n
bv
and
b v = c3~ru + c4~rv
n ~rv = c∗3 n
bu + c∗4 n
bv
and show
f F − eG gF − f G f F − gE f G − gF
c1 = c3 = c∗1 = c∗3 =
EG − F 2 EG − F 2 eg − f 2 eg − f 2
eF − f E f F − gE f E − eF f F − eG
c2 = c4 = c∗2 = c∗4 =
EG − F 2 EG − F 2 eg − f 2 eg − f 2
The constants in the above equations are determined in a manner similar to that suggested in problem 8.
Show that the Weingarten equations can be written in the form
∂b
n ∂~r
= −bβα β .
∂uα ∂u
~ru × ~rv
I 10. b= √
Using n , the results from exercise 1.1, problem 9(a), and the results from problem 5,
EG − F 2
verify that
p
2
(~ru × ~ruu ) · n
b= EG − F 2
11 p
p 1
2 (~rv × ~ruv ) · n
b=− EG − F 2
(~ru × ~ruv ) · n
b= EG − F 2 21
12 p
p 1
1 (~rv × ~rvv ) · nb=− EG − F 2
(~rv × ~ruu ) · n
b=− EG − F 2 22
11 p
p b = EG − F 2
(~ru × ~rv ) · n
2
(~ru × ~rvv ) · n
b= EG − F 2
22
and then derive the formula for the geodesic curvature given by equation (1.5.48).
dT~ dT~ α
n × T~ ) ·
Hint:(b = (T~ × )·n
b and aαδ ]β γ, δ] = .
ds ds βγ
164
I 11. Verify the equation (1.5.39) which shows that the normal curvature directions are orthogonal. i.e.
verify that Gλ1 λ2 + F (λ1 + λ2 ) + E = 0.
I 12. βγ ωα
Verify that δστ δλν Rωαβγ = 4Rλνστ .
I 13. Find the first fundamental form and unit normal to the surface defined by z = f (x, y).
I 14. Verify
Ai,jk − Ai,kj = Aσ R.ijk
σ
where
∂ σ ∂ σ n σ n σ
σ
R.ijk = − + − .
∂xj ik ∂xk ij ik nj ij nk
which is sometimes written
∂ s
∂ s
Rinjk
= ∂x j ∂x k
+
n j n k
[nj, k] [nk, i] [ij, s] [ik, s]
I 15. σ
For Rijkl = giσ R.jkl show
∂ ∂ s s
Rinjk = [nk, i] − [nj, i] + [ik, s] − [ij, s]
∂xj ∂xk nj nk
I 16. Show
1 ∂ 2 gil ∂ 2 gjl ∂ 2 gik ∂ 2 gjk
Rijkl = − − + + g αβ ([jk, β][il, α] − [jl, β][ik, α]) .
2 ∂xj ∂xk ∂xi ∂xk ∂xj ∂xl ∂xi ∂xl
Hence, the tensor Rijkl is skewsymmetric in the indices i, j and k, l. Also the tensor Rijkl is symmetric with
respect to the (ij) and (kl) pair of indices.
I 18. Verify the following cyclic properties of the Riemann Christoffel symbol:
I 19. By employing the results from the previous problems, show all components of the form:
Riijk , Rinjj , Riijj , Riiii , (no summation on i or j) must be zero.
165
I 20. Find the number of independent components associated with the Riemann Christoffel tensor
Rijkm , i, j, k, m = 1, 2, . . . , N. There are N 4 components to examine in an N −dimensional space. Many of
these components are zero and many of the nonzero components are related to one another by symmetries
or the cyclic properties. Verify the following cases:
CASE I We examine components of the form Rinin , i 6= n with no summation of i or n. The first index
can be chosen in N ways and therefore with i 6= n the second index can be chosen in N − 1 ways. Observe
that Rinin = Rnini , (no summation on i or n) and so one half of the total combinations are repeated. This
1
leaves M1 = 2 N (N − 1) components of the form Rinin . The quantity M1 can also be thought of as the
number of distinct pairs of indices (i, n).
CASE II We next examine components of the form Rinji , i 6= n 6= j where there is no summation on
the index i. We have previously shown that the first pair of indices can be chosen in M1 ways. Therefore,
the third index can be selected in N − 2 ways and consequently there are M2 = 12 N (N − 1)(N − 2) distinct
components of the form Rinji with i 6= n 6= j.
CASE III Next examine components of the form Rinjk where i 6= n 6= j 6= k. From CASE I the first pairs
of indices (i, n) can be chosen in M1 ways. Taking into account symmetries, it can be shown that the second
pair of indices can be chosen in 12 (N − 2)(N − 3) ways. This implies that there are 14 N (N − 1)(N − 2)(N − 3)
ways of choosing the indices i, n, j and k with i 6= n 6= j 6= k. By symmetry the pairs (i, n) and (j, k) can be
interchanged and therefore only one half of these combinations are distinct. This leaves
1
N (N − 1)(N − 2)(N − 3)
8
distinct pairs of indices. Also from the cyclic relations we find that only two thirds of the above components
are distinct. This produces
N (N − 1)(N − 2)(N − 3)
M3 =
12
distinct components of the form Rinjk with i 6= n 6= j 6= k.
Adding the above components from each case we find there are
N 2 (N 2 − 1)
M4 = M1 + M2 + M3 =
12
distinct and independent components.
Verify the entries in the following table:
Dimension of space N 1 2 3 4 5
Number of components N 4 1 16 81 256 625
M4 = Independent components of Rijkm 0 1 6 20 50
Note 1: A one dimensional space can not be curved and all one dimensional spaces are Euclidean. (i.e. if we have
an element of arc length squared given by ds2 = f (x)(dx)2 , we can make the coordinate transformation
p
f (x)dx = du and reduce the arc length squared to the form ds2 = du2 .)
Note 2: In a two dimensional space, the indices can only take on the values 1 and 2. In this special case there
are 16 possible components. It can be shown that the only nonvanishing components are:
For these nonvanishing components only one independent component exists. By convention, the com
ponent R1212 is selected as the single independent component and all other nonzero components are
expressed in terms of this component.
Find the nonvanishing independent components Rijkl for i, j, k, l = 1, 2, 3, 4 and show that
I 24. By employing the results from the previous problem show that in the case N = 2 we have
2R1221
(a) R12 = R21 = 0 (c) R=
g11 g22
(b) R11 g22 = R22 g11 = R1221 1
(d) Rij = Rgij , where R = g ij Rij
2
The scalar invariant R is known as the Einstein curvature of the surface and the tensor Gij = Rji − 12 δji R is
known as the Einstein tensor.
I 26. For N = 3 show that R1212 , R1313 , R2323 , R1213 , R2123 , R3132 are independent components of the
Riemann Christoffel tensor.
167
a11 0
I 27. For N = 2 and aαβ = show that
0 a22
R1212 1 ∂ 1 ∂a22 ∂ 1 ∂a11
K= =− √ √ + √ .
a 2 a ∂u1 a ∂u1 ∂u2 a ∂u2
a11 a12
I 28. For N = 2 and aαβ = show that
a21 a22
1 ∂ a12 ∂a11 1 ∂a22 ∂ 2 ∂a12 1 ∂a11 a12 ∂a11
K= √ √ −√ + √ −√ − √ .
2 a ∂u1 a11 a ∂u2 a ∂u1 ∂u2 a ∂u1 a ∂u2 a11 a ∂u1
Check your results by setting a12 = a21 = 0 and comparing this answer with that given in the problem 27.
I 29. Write out the FrenetSerret formulas (1.5.112)(1.5.113) for surface curves in terms of Christoffel
symbols of the second kind.
I 30.
(a) Use the fact that for n = 2 we have R1212 = R2121 = −R2112 = −R1221 together with eαβ , eαβ the two
dimensional alternating tensors to show that the equation (1.5.110) can be written as
√ 1
Rαβγδ = Kαβ γδ where αβ = aeαβ and αβ = √ eαβ
a
x = u, y = v, z = f (u, v).
I 34. Geodesics on a sphere Let (θ, φ) denote the surface coordinates of the sphere of radius ρ defined
by the parametric equations
Consider also a plane which passes through the origin with normal having the direction numbers (n1 , n2 , n3 ).
This plane is represented by n1 x + n2 y + n3 z = 0 and intersects the sphere in a great circle which is described
by the relation
n1 sin θ cos φ + n2 sin θ sin φ + n3 cos θ = 0. (2)
This is an implicit relation between the surface coordinates θ, φ which describes the great circle lying on the
sphere. We can write this later equation in the form
−n3
n1 cos φ + n2 sin φ = (3)
tan θ
168
and in the special case where n1 = cos β, n2 = sin β,n3 = − tan α is expressible in the form
tan α tan α
cos(φ − β) = or φ − β = cos−1 . (4)
tan θ tan θ
The above equation defines an explicit relationship between the surface coordinates which defines a great
circle on the sphere. The arc length squared relation satisfied by the surface coordinates together with the
equation obtained by differentiating equation (4) with respect to arc length s gives the relations
dφ tan α dθ
sin2 θ =q (5)
ds tan2 α ds
1 − tan2 θ
ds2 = ρ2 dθ2 + ρ2 sin2 θ dφ2 (6)
The above equations (1)(6) are needed to consider the following problem.
(a) Show that the differential equations defining the geodesics on the surface of a sphere (equations (1.5.51))
are
2
d2 θ dφ
− sin θ cos θ =0 (7)
ds2 ds
d2 φ dθ dφ
+ 2 cot θ =0 (8)
ds2 ds ds
dφ
sin2 θ = c1 (9)
ds
dθ
(c) Multiply equation (7) by ds and use the result of equation (9) to show that an integration produces
2
dθ −c21
= + c22 (10)
ds sin2 θ
sin α
(d) Use the equations (5)(6) to show that c2 = 1/ρ and c1 = ρ .
(e) Show that equations (9) and (10) imply that
dφ tan α sec2 θ
= q
dθ tan2 θ 1 − tan2 α
tan2 θ
tan α
and making the substitution u = tan θ this equation can be integrated to obtain the equation (4). We
can now expand the equation (4) and express the results in terms of x, y, z to obtain the equation (3).
This produces a plane which intersects the sphere in a great circle. Consequently, the geodesics on a
sphere are great circles.
169
I 35. Find the differential equations defining the geodesics on the surface of a cylinder.
I 36. Find the differential equations defining the geodesics on the surface of a torus. (See problem 13,
Exercise 1.3)
I 37. Find the differential equations defining the geodesics on the surface of revolution
Note the curve z = f (x) gives a profile of the surface. The curves r = Constant are the parallels, while the
curves φ = Constant are the meridians of the surface and
I 38. Find the unit normal and tangent plane to an arbitrary point on the right circular cone
This is a surface of revolution with r = u sin α and f (r) = r cot α with α constant.
I 39. Let s denote arc length and assume the position vector ~r(s) is analytic about a point s0 . Show that
h2 h3
the Taylor series ~r(s) = ~r(s0 ) + h~r 0 (s0 ) + ~r 00 (s0 ) + ~r 000 (s0 ) + · · · about the point s0 , with h = s − s0 is
2! 3!
given by ~r(s) = ~r(s0 ) + hT~ + 12 κh2 N ~ + 1 h3 (−κ2 T~ + κ0 N
6
~ + κτ B) ~ + · · · which is obtained by differentiating
the Frenet formulas.
I 40.
(a) Show that the circular helix defined by x = a cos t, y = a sin t, z = bt with a, b constants, has the
property that any tangent to the curve makes a constant angle with the line defining the zaxis.
(i.e. T~ · b
e3 = cos α = constant.)
(b) Show also that N e3 = 0 and consequently b
~ · b e3 is parallel to the rectifying plane, which implies that
b
e3 = T~ cos α + B
~ sin α.
(c) Differentiate the result in part (b) and show that κ/τ = tan α is a constant.
I 41. Consider a space curve xi = xi (s) in Cartesian coordinates.
dT~ p
(a) Show that κ = = x0i x0i
ds
1
(b) Show that τ = eijk x0i x00j x000 r 0 · ~r 00 × ~r 000
k . Hint: Consider ~
κ2
I 42.
(a) Find the direction cosines of a normal to a surface z = f (x, y).
(b) Find the direction cosines of a normal to a surface F (x, y, z) = 0.
(c) Find the direction cosines of a normal to a surface x = x(u, v), y = y(u, v), z = z(u, v).
I 43. Show that for a smooth surface z = f (x, y) the Gaussian curvature at a point on the surface is given
by
2
fxx fyy − fxy
K= .
(fx2 + fy2 + 1)2
170
I 44. Show that for a smooth surface z = f (x, y) the mean curvature at a point on the surface is given by
I 45. Express the FrenetSerret formulas (1.5.13) in terms of Christoffel symbols of the second kind.
I 46. Verify the relation (1.5.106).
I 47. In Vn assume that Rij = ρgij and show that ρ = R
n where R = g ij Rij . This result is known as
Einstein’s gravitational equation at points where matter is present. It is analogous to the Poisson equation
∇2 V = ρ from the Newtonian theory of gravitation.
I 48. In Vn assume that Rijkl = K(gik gjl − gil gjk ) and show that R = Kn(1 − n). (Hint: See problem 23.)
I 49. Assume gij = 0 for i 6= j and verify the following.
(a) Rhijk = 0 for h 6= i 6= j 6= k
2√ √ √ √ √
√ ∂ gii ∂ gii ∂ log ghh ∂ gii ∂ log gkk
(b) Rhiik = gii − − for h, i, k unequal.
∂xh∂xk ∂xh ∂xk ∂xk ∂xh
√ √ Xn √ √
√ √ ∂ 1 ∂ gii ∂ 1 ∂ ghh ∂ gii ∂ ghh
(c) Rhiih = gii ghh h √ + i √ + where h 6= i.
∂x ghh ∂xh ∂x gii ∂xi m=1
∂xm ∂xm
m6=h m6=i
I 50. Consider a surface of revolution where x = r cos θ, y = r sin θ and z = f (r) is a given function of r.
(a) Show in this V2 we have ds2 = (1 + (f 0 )2 )dr2 + r2 dθ2 where 0 = d
ds .
(b) Show the geodesic equations in this V2 are
2 2
d2 r f 0 f 00 dr r dθ
2
+ 0 2
− 0 2
=0
ds 1 + (f ) ds 1 + (f ) ds
d2 θ 2 dθ dr
+ =0
ds2 r ds ds
dθ a
(c) Solve the second equation in part (b) to obtain = 2 . Substitute this result for ds in part (a) to show
p ds r
a 1 + (f 0 )2
dθ = ± √ dr which theoretically can be integrated.
r r 2 − a2
171
In the following sections we develop some applications of tensor calculus in the areas of dynamics,
elasticity, fluids and electricity and magnetism. We begin by first developing generalized expressions for the
vector operations of gradient, divergence, and curl. Also generalized expressions for other vector operators
are considered in order that tensor equations can be converted to vector equations. We construct a table to
aid in the translating of generalized tensor equations to vector form and vice versa.
The basic equations of continuum mechanics are developed in the later sections. These equations are
developed in both Cartesian and generalized tensor form and then converted to vector form.
We consider the tensor representation of some vector expressions. Our goal is to develop the ability to
convert vector equations to tensor form as well as being able to represent tensor equations in vector form.
In this section the basic equations of continuum mechanics are represented using both a vector notation and
the indicial notation which focuses attention on the tensor components. In order to move back and forth
between these notations, the representation of vector quantities in tensor form is now considered.
Gradient
Note, if C i = g im Φ,m , i = 1, 2, 3 are the tensor components of the gradient then in an orthogonal coordinate
system we will have
C 1 = g 11 Φ,1 , C 2 = g 22 Φ,2 , C 3 = g 33 Φ,3 .
We note that in an orthogonal coordinate system that g ii = 1/h2i , (no sum on i), i = 1, 2, 3 and hence
replacing the tensor components by their equivalent physical components there results the equations
1 ∂Φ 1 ∂Φ 1 ∂Φ
C(1) = , C(2) = , C(3) = .
h1 ∂x1 h2 ∂x2 h3 ∂x3
These results are only valid when the coordinate system is orthogonal and gij = 0 for i 6= j and gii = h2i ,
with i = 1, 2, 3, and where i is not summed.
172
Divergence
The divergence of a contravariant tensor Ar is obtained by taking the covariant derivative with respect
to xk and then performing a contraction. This produces
Still another form for the divergence is obtained by simplifying the expression (2.1.3). The covariant deriva
tive can be represented
∂Ar r
Ar,k = + Am .
∂xk mk
Upon contracting the indices r and k and using the result from Exercise 1.4, problem 13, we obtain
√
∂Ar 1 ∂( g) m
Ar,r = + √ A
∂xr g ∂xm
√
1 √ ∂Ar r∂ g
Ar,r = √ g r +A (2.1.4)
g ∂x ∂xr
1 ∂ √ r
Ar,r = √ ( gA ) .
g ∂xr
EXAMPLE 2.11. (Divergence) Find the representation of the divergence of a vector Ar in spherical
coordinates (ρ, θ, φ). Solution: In spherical coordinates we have
for the physical components, the divergence can be expressed in either of the forms:
1 ∂ 2 ∂ 2 Aθ ∂ 2 Aφ
div Ar = (ρ sin θA ρ ) + (ρ sin θ ) + (ρ sin θ ) or
ρ2 sin θ ∂ρ ∂θ ρ ∂φ ρ sin θ
1 ∂ 1 ∂ 1 ∂Aφ
div Ar = 2 (ρ2 Aρ ) + (sin θAθ ) + .
ρ ∂ρ ρ sin θ ∂θ ρ sin θ ∂φ
173
Curl
~ = curl A
The contravariant components of the vector C ~ are represented
√
The determinant of gij is g = gij  = ρ4 sin2 θ with g = ρ2 sin θ. The relations (2.1.6) are tensor equations
representing the components of the vector curl A. ~ To find the components of curl A~ in spherical components
we write the equations (2.1.6) in terms of their physical components. These equations take on the form:
C(1) 1 ∂ ∂
=√ (h3 A(3)) − (h2 A(2))
h1 g ∂θ ∂φ
C(2) 1 ∂ ∂
=√ (h1 A(1)) − (h3 A(3)) (2.1.7)
h2 g ∂φ ∂ρ
C(3) 1 ∂ ∂
=√ (h2 A(2)) − (h1 A(1)) .
h3 g ∂ρ ∂θ
~ in spherical coordinates,
to denote the physical components, and find the components of the vector curl A,
are expressible in the form:
1 ∂ ∂
Cρ = 2 (ρ sin θAφ ) − (ρAθ )
ρ sin θ ∂θ ∂φ
1 ∂ ∂
Cθ = (Aρ ) − (ρ sin θAφ ) (2.1.8)
ρ sin θ ∂φ ∂ρ
1 ∂ ∂
Cφ = (ρAθ ) − (Aρ ) .
ρ ∂ρ ∂θ
174
Laplacian
The table 1 gives the vector and tensor representation for various quantities of interest.
175
~
A Ai or Ai Ai
~·B
A ~ Ai Bi = gij Ai B j = Ai B i Ai Bi
Ai Bi = g ij Ai Bj
1
~ =A
C ~×B
~ C i = √ eijk Aj Bk Ci = eijk Aj Bk
g
∂Φ
∇ Φ = grad Φ g im Φ,m Φ,i =
∂xi
1 ∂ √ r ∂Ai
~ = div A
∇·A ~ g mn Am,n = Ar,r = √ ( gA ) Ai,i =
g ∂xr ∂xi
∂Ak
~=C
∇×A ~ = curl A
~ C i = ijk Ak,j Ci = eijk
∂xj
1 ∂ √ ij ∂U ∂ ∂U
∇2 U g mn U ,mn = √ gg
g ∂xj ∂xi ∂xi ∂xi
∂Bi
C ~ · ∇)B
~ = (A ~ C i = Am B i,m Ci = Am
∂xm
∂Bm
~ = A(∇
C ~ ~
· B) C i = Ai B j,j Ci = Ai
∂xm
~ = ∇2 A ∂ ∂Ai
C ~ C i = g jm Ai ,mj or Ci = g jm Ai,mj Ci =
∂xm ∂xm
A~·∇ φ g im Ai φ ,m Ai φ,i
∂ 2 Ar
~
∇ ∇·A g im Ar,r ,m ∂xi ∂xr
∂ 2 Aj ∂ 2 Ai
~
∇× ∇×A ijk g jm kst At,s ,m
−
∂xj ∂xi ∂xj ∂xj
EXAMPLE 2.14. (Maxwell’s equations) In the study of electrodynamics there arises the following
vectors and scalars:
~ =Electric force vector, [E]
E ~ = Newton/coulomb
~ = Weber/m2
~ =Magnetic force vector, [B]
B
~ = coulomb/m2
~ =Displacement vector, [D]
D
~ =Auxilary magnetic force vector, [H]
H ~ = ampere/m
~ = ampere/m2
J~ =Free current density, [J]
% =free charge density, [%] = coulomb/m3
The above quantities arise in the representation of the following laws:
Faraday’s Law This law states the line integral of the electromagnetic force around a loop is proportional
to the rate of flux of magnetic induction through the loop. This gives rise to the first electromagnetic field
equation:
~
∂B ∂B i
~ =−
∇×E or ijk Ek,j = − . (2.1.15)
∂t ∂t
Ampere’s Law This law states the line integral of the magnetic force vector around a closed loop is
proportional to the sum of the current through the loop and the rate of flux of the displacement vector
through the loop. This produces the second electromagnetic field equation:
~
∂D ∂Di
~ = J~ +
∇×H or ijk Hk,j = J i + . (2.1.16)
∂t ∂t
Gauss’s Law for Electricity This law states that the flux of the electric force vector through a closed
surface is proportional to the total charge enclosed by the surface. This results in the third electromagnetic
field equation:
1 ∂ √ i
~ =%
∇·D or √ gD = %. (2.1.17)
g ∂xi
Gauss’s Law for Magnetism This law states the magnetic flux through any closed volume is zero. This
produces the fourth electromagnetic field equation:
1 ∂ √ i
~ =0
∇·B or √ gB = 0. (2.1.18)
g ∂xi
The four electromagnetic field equations are referred to as Maxwell’s equations. These equations arise
in the study of electrodynamics and can be represented in other forms. These other forms will depend upon
such things as the material assumptions and units of measurements used. Note that the tensor equations
(2.1.15) through (2.1.18) are representations of Maxwell’s equations in a form which is independent of the
coordinate system chosen.
In applications, the tensor quantities must be expressed in terms of their physical components. In a
general orthogonal curvilinear coordinate system we will have
the Maxwell equations can be represented by the equations in table 2. The tables 3, 4 and 5 are the
representation of Maxwell’s equations in rectangular, cylindrical, and spherical coordinates. These latter
tables are special cases associated with the more general table 2.
1 ∂ ∂ 1 ∂B(1)
2
(h3 E(3)) − 3 (h2 E(2)) = −
h1 h2 h3 ∂x ∂x h1 ∂t
1 ∂ ∂ 1 ∂B(2)
3
(h1 E(1)) − 1 (h3 E(3)) = −
h1 h2 h3 ∂x ∂x h2 ∂t
1 ∂ ∂ 1 ∂B(3)
1
(h2 E(2)) − 2 (h1 E(1)) = −
h1 h2 h3 ∂x ∂x h3 ∂t
1 ∂ ∂ J(1) 1 ∂D(1)
2
(h3 H(3)) − 3 (h2 H(2)) = +
h1 h2 h3 ∂x ∂x h1 h1 ∂t
1 ∂ ∂ J(2) 1 ∂D(2)
(h1 H(1)) − 1 (h3 H(3)) = +
h1 h2 h3 ∂x3 ∂x h2 h2 ∂t
1 ∂ ∂ J(3) 1 ∂D(3)
(h2 H(2)) − 2 (h1 H(1)) = +
h1 h2 h3 ∂x1 ∂x h3 h3 ∂t
1 ∂ D(1) ∂ D(2) ∂ D(3)
h1 h2 h3 + 2 h1 h2 h3 + 3 h1 h2 h3 =%
h1 h2 h3 ∂x1 h1 ∂x h2 ∂x h3
1 ∂ B(1) ∂ B(2) ∂ B(3)
h1 h2 h3 + 2 h1 h2 h3 + 3 h1 h2 h3 =0
h1 h2 h3 ∂x1 h1 ∂x h2 ∂x h3
with x1 = x, x2 = y, x3 = z, h1 = h 2 = h 3 = 1
with x1 = r, x2 = θ, x3 = z, h1 = 1, h2 = r, h3 = 1.
1 ∂ ∂Eθ ∂Bρ 1 ∂ ∂Hθ ∂Dρ
(sin θEφ ) − =− (sin θHφ ) − = Jρ +
ρ sin θ ∂θ ∂φ ∂t ρ sin θ ∂θ ∂φ ∂t
1 ∂Eρ 1 ∂ ∂Bθ 1 ∂Hρ 1 ∂ ∂Dθ
− (ρEφ ) = − − (ρHφ ) = Jθ +
ρ sin θ ∂φ ρ ∂ρ ∂t ρ sin θ ∂φ ρ ∂ρ ∂t
1 ∂ 1 ∂Eρ ∂Bφ 1 ∂ 1 ∂Hρ ∂Dφ
(ρEθ ) − =− (ρHθ ) − = Jφ +
ρ ∂ρ ρ ∂θ ∂t ρ ∂ρ ρ ∂θ ∂t
1 ∂ 2 1 ∂ 1 ∂Dφ
2
(ρ Dρ ) + (sin θDθ ) + =%
ρ ∂ρ ρ sin θ ∂θ ρ sin θ ∂φ
1 ∂ 2 1 ∂ 1 ∂Bφ
(ρ Bρ ) + (sin θBθ ) + =0
ρ2 ∂ρ ρ sin θ ∂θ ρ sin θ ∂φ
with x1 = ρ, x2 = θ, x3 = φ, h1 = 1, h2 = ρ, h3 = ρ sin θ
where Tij = Tji is symmetric, Ai are the components of a vector and λ is a scalar. Any nonzero solution
Ai of equation (2.1.19) is called an eigenvector of the tensor Tij and the associated scalar λ is called an
eigenvalue. When expanded these equations have the form
The condition for equation (2.1.19) to have a nonzero solution Ai is that the characteristic equation
should be zero. This equation is found from the determinant equation
T11 − λ T12 T13
f (λ) = T21 T22 − λ T23 = 0, (2.1.20)
T31 T32 T33 − λ
180
I1 = Tii
1 1
I2 = Tii Tjj − Tij Tij (2.1.22)
2 2
I3 = eijk Ti1 Tj2 Tk3 .
When Tij is subjected to an orthogonal transformation, where T̄mn = Tij `im `jn , then
`im `jn (Tmn − λ δmn ) = T̄ij − λ δij and det (Tmn − λ δmn ) = det T̄ij − λ δij .
Hence, the eigenvalues of a second order tensor remain invariant under an orthogonal transformation.
If Tij is real and symmetric then
Proof: To show a quantity is real we show that the conjugate of the quantity equals the given quantity. If
(2.1.19) is satisfied, we multiply by the conjugate Ai and obtain
The right hand side of this equation has the inner product Ai Ai which is real. It remains to show the left
hand side of equation (2.1.25) is also real. Consider the conjugate of this left hand side and write
Consequently, the left hand side of equation (2.1.25) is real and the eigenvalue λ can be represented as the
ratio of two real quantities.
Assume that λ(1) and λ(2) are two distinct eigenvalues which produce the unit eigenvectors L̂1 and L̂2
with components `i1 and `i2 , i = 1, 2, 3 respectively. We then have
Tij `j1 = λ(1) `i1 and Tij `j2 = λ(2) `i2 . (2.1.26)
By hypothesis, λ(1) is different from λ(2) and consequently the inner product `i1 `i2 must be zero. Therefore,
the eigenvectors corresponding to distinct eigenvalues are orthogonal.
181
Therefore, associated with distinct eigenvalues λ(i) , i = 1, 2, 3 there are unit eigenvectors
Tij `j1 = λ(1) `i1 Tij `j2 = λ(2) `i2 Tij `j3 = λ(3) `i3
xi = `ij xj or xm = `mj xj
which represents a rotation of axes, where `ij are the direction cosines from the eigenvectors of Tij . This is a
linear transformation where the `ij satisfy equation (2.1.23). Such a transformation is called an orthogonal
transformation. In the new x coordinate system, called principal axes, we have
∂xi ∂xj
T mn = Tij = Tij `im `jn = λ(n) `in `im = λ(n) δmn (no sum on n). (2.1.24)
∂xm ∂xn
This equation shows that in the barred coordinate system there are the components
λ(1) 0 0
T mn = 0 λ(2) 0 .
0 0 λ(3)
That is, along the principal axes the tensor components Tij are transformed to the components T ij where
T ij = 0 for i 6= j. The elements T (i)(i) , i not summed, represent the eigenvalues of the transformation
(2.1.19).
182
EXERCISE 2.1
~ = A(r,
I 3. In cylindrical coordinates (r, θ, z) for A ~ θ, z) find curl A.
~
~ = A(ρ,
I 7. In spherical coordinates (ρ, θ, φ) for A ~ θ, φ) find curl A.
~
I 9. Let ~r = x ê1 + y ê2 + z ê3 denote the position vector of a variable point (x, y, z) in Cartesian coordinates.
Let r = ~r denote the distance of this point from the origin. Find in terms of ~r and r:
1
(a) grad (r) (b) grad (rm ) (c) grad ( ) (d) grad (ln r) (e) grad (φ)
r
I 10. Let ~r = x ê1 +y ê2 +z ê3 denote the position vector of a variable point (x, y, z) in Cartesian coordinates.
Let r = ~r denote the distance of this point from the origin. Find:
(a) div (~r) (b) div (rm~r) (c) div (r−3 ~r) (d) div (φ ~r)
I 11. Let ~r = x ê1 + y ê2 + z ê3 denote the position vector of a variable point (x, y, z) in Cartesian
coordinates. Let r = ~r denote the distance of this point from the origin. Find: (a) curl ~r (b) curl (φ ~r)
where φ = φ(r) is an arbitrary function of r.
~
I 12. Expand and simplify the representation for curl (curl A).
I 13. Show that the curl of the gradient is zero in generalized coordinates.
I 14. Write out the physical components associated with the gradient of φ = φ(x1 , x2 , x3 ).
I 22. Write out the equations in problem 21 for a generalized orthogonal coordinate system in terms of
physical components.
I 25. Use equation (2.1.4) to represent the divergence in parabolic cylindrical coordinates (ξ, η, z).
I 26. Use equation (2.1.4) to represent the divergence in parabolic coordinates (ξ, η, φ).
I 27. Use equation (2.1.4) to represent the divergence in elliptic cylindrical coordinates (ξ, η, z).
I 28. ~ = ~v ∇ · A
B ~ + (∇ · ~v ) A
~
~ ~ ~
I 29.
d ~ ~ ~ = dA · (B
[A · (B × C)] ~ × C) ~ · ( dB × C)
~ +A ~ × dC )
~ · (B
~ +A
dt dt dt dt
d~v ∂~v
I 30. = + (~v · ∇)~v
dt ∂t
1 ∂H~
I 31. = −curl E ~
c ∂t
dB~
I 32. − (B~ · ∇)~v + B(∇
~ · ~v ) = ~0
dt
184
Z Z
I 37. The moment of inertia of an area or second moment of area is defined by Iij = (ym ym δij −yi yj ) dA
A
where dA is an element of area. Calculate
1 the3 moment of inertia
Iij , i, j = 1, 2 for the triangle illustrated in
1 2 2
12 bh − 24 b h
the figure 2.11 and show that Iij = 1 2 2 1 3 .
− 24 b h 12 b h
I 38. Use the results from problem 37 and rotate the axes in figure 2.11 through an angle θ to a barred
system of coordinates.
(a) Show that in the barred system of coordinates
I11 + I22 I11 − I22
I 11 = + cos 2θ + I12 sin 2θ
2 2
I11 − I22
I 12 = I 21 =− sin 2θ + I12 cos 2θ
2
I11 + I22 I11 − I22
I 22 = − cos 2θ − I12 sin 2θ
2 2
I 39. Otto Mohr1 gave the following physical interpretation to the results obtained in problem 38:
• Plot the points A(I11 , I12 ) and B(I22 , −I12 ) as illustrated in the figure 2.12
• Draw the line AB and calculate the point C where this line intersects the I axes. Show the point C
has the coordinates
I11 + I22
( , 0)
2
• Calculate the radius of the circle with center at the point C and with diagonal AB and show this
radius is s
2
I11 − I22 2
r= + I12
2
• Show the maximum and minimum values of I occur where the constructed circle intersects the I axes.
I11 + I22 I11 + I22
Show that Imax = I 11 = +r Imin = I 22 = − r.
2 2
I11 I12
I 40. Show directly that the eigenvalues of the symmetric matrix Iij = are λ1 = Imax and
I21 I22
λ2 = Imin where Imax and Imin are given in problem 39.
I 41. Find the principal axes and moments of inertia for the triangle given in problem 37 and summarize
your results from problems 37,38,39, and 40.
or
h1 ê1 h2 ê2 h3 ê3
1
~=
∇×A ∂ ∂ ∂ .
h1 h2 h3 ∂x1 ∂x2 ∂x3
h1 A(1) h2 A(2) h3 A(3)
1
Christian Otto Mohr (18351918) German civil engineer.
186
I 50. The harmonic and biharmonic functions or potential functions occur in the mathematical modeling
of many physical problems. Any solution of Laplace’s equation ∇2 Φ = 0 is called a harmonic function and
any solution of the biharmonic equation ∇4 Φ = 0 is called a biharmonic function.
(a) Expand the Laplace equation in Cartesian, cylindrical and spherical coordinates.
(b) Expand the biharmonic equation in two dimensional Cartesian and polar coordinates.
Hint: Consider ∇4 Φ = ∇2 (∇2 Φ). In Cartesian coordinates ∇2 Φ = Φ,ii and ∇4 Φ = Φ,iijj .
187
§2.2 DYNAMICS
Dynamics is concerned with studying the motion of particles and rigid bodies. By studying the motion
of a single hypothetical particle, one can discern the motion of a system of particles. This in turn leads to
the study of the motion of individual points in a continuous deformable medium.
Particle Movement
The trajectory of a particle in a generalized coordinate system is described by the parametric equations
xi = xi (t), i = 1, . . . , N (2.2.1)
where t is a time parameter. If the coordinates are changed to a barred system by introducing a coordinate
transformation
xi = xi (x1 , x2 , . . . , xN ), i = 1, . . . , N
dxi
vi = , i = 1, . . . , N. (2.2.3)
dt
By the chain rule differentiation of the transformation equations (2.2.2) one can verify that the velocity in
the barred system is
dxr ∂xr dxj ∂xr j
vr = = j
= v , r = 1, . . . , N. (2.2.4)
dt ∂x dt ∂xj
Consequently, the generalized velocity v i is a first order contravariant tensor. The speed of the particle is
obtained from the magnitude of the velocity and is
v 2 = gij v i v j .
The generalized acceleration f i of the particle is defined as the intrinsic derivative of the generalized velocity.
The generalized acceleration has the form
m n
iδv i i dx
n
dv i i m n d2 xi i dx dx
f = = v,n = + v v = + (2.2.5)
δt dt dt mn dt2 m n dt dt
FrenetSerret Formulas
The parametric equations (2.2.1) describe a curve in our generalized space. With reference to the figure
2.21 we wish to define at each point P of the curve the following orthogonal unit vectors:
These vectors define the osculating, normal and rectifying planes illustrated in the figure 2.21.
In the generalized coordinates the arc length squared is
dxi
Define T i = ds as the tangent vector to the parametric curve defined by equation (2.2.1). This vector is a
unit tangent vector because if we write the element of arc length squared in the form
dxi dxj
1 = gij = gij T i T j , (2.2.6)
ds ds
we obtain the generalized dot product for T i . This generalized dot product implies that the tangent vector
is a unit vector. Differentiating the equation (2.2.6) intrinsically with respect to arc length s along the curve
produces
δT m n δT n
gmn T + gmn T m = 0,
δs δs
which simplifies to
δT m
gmn T n = 0. (2.2.7)
δs
189
δT m
The equation (2.2.7) is a statement that the vector δs is orthogonal to the vector T m . The unit normal
vector is defined as
1 δT i 1 δTi
Ni = or Ni = , (2.2.8)
κ δs κ δs
where κ is a scalar called the curvature and is chosen such that the magnitude of N i is unity. The reciprocal
1
of the curvature is R = κ, which is called the radius of curvature. The curvature of a straight line is zero
while the curvature of a circle is a constant. The curvature measures the rate of change of the tangent vector
as the arc length varies.
The equation (2.2.7) can be expressed in the form
gij T i N j = 0. (2.2.9)
Taking the intrinsic derivative of equation (2.2.9) with respect to the arc length s produces
δN j δT i j
gij T i + gij N =0
δs δs
or
δN j δT i j
gij T i = −gij N = −κgij N i N j = −κ. (2.2.10)
δs δs
The generalized dot product can be written
gij T i T j = 1,
δN j
gij N i = 0.
δs
This in turn can be expressed in the form
i δN j
gij N + κT j = 0.
δs
This form of the equation implies that the vector represented in equation (2.2.12) is also orthogonal to the
unit normal N i . We define the unit binormal vector as
i 1 δN i i 1 δNi
B = + κT or Bi = + κTi (2.2.13)
τ δs τ δs
where τ is a scalar called the torsion. The torsion is chosen such that the binormal vector is a unit vector.
The torsion measures the rate of change of the osculating plane and consequently, the torsion τ is a measure
190
of the twisting of the curve out of a plane. The value τ = 0 corresponds to a plane curve. The vectors
T i , N i , B i , i = 1, 2, 3 satisfy the cross product relation
B i = ijk Tj Nk .
= τ ijk Tj Bk = −τ ikj Bk Tj = −τ N i .
The relations (2.2.8),(2.2.13) and (2.2.14) are now summarized and written
δT i
= κN i
δs
δN i
= τ B i − κT i (2.2.15)
δs
δB i
= −τ N i .
δs
These equations are known as the FrenetSerret formulas of differential geometry.
dxi dxi ds
vi = = = T i v, (2.2.16)
dt ds dt
ds
where v = dt is the speed of the particle and is the magnitude of v i . The vector T i is the unit tangent vector
to the trajectory curve at the time t. The equation (2.2.16) is a statement of the fact that the velocity of a
particle is always in the direction of the tangent vector to the curve and has the speed v.
By chain rule differentiation, the generalized acceleration is expressible in the form
δv r dv r δT r
fr = = T +v
δt dt δt
dv r δT r ds
= T +v (2.2.17)
dt δs dt
dv r
= T + κv 2 N r .
dt
The equation (2.2.17) states that the acceleration lies in the osculating plane. Further, the equation (2.2.17)
dv
indicates that the tangential component of the acceleration is dt , while the normal component of the accel
2
eration is κv .
191
Define M as the constant mass of the particle as it moves along the curve defined by equation (2.2.1).
Also let Qr denote the components of a force vector (in appropriate units of measurements) which acts upon
the particle. Newton’s second law of motion can then be expressed in the form
Qr = M f r or Qr = M fr . (2.2.18)
The work done W in moving a particle from a point P0 to a point P1 along a curve xr = xr (t), r = 1, 2, 3,
with parameter t, is represented by a summation of the tangential components of the forces acting along the
path and is defined as the line integral
Z P1 Z P1 Z t1 Z t1
dxr r dxr
W = Qr ds = Qr dx = Qr dt = Qr v r dt (2.2.19)
P0 ds P0 t0 dt t0
where Qr = grs Qs is the covariant form of the force vector, t is the time parameter and s is arc length along
the curve.
Conservative Systems
If the force vector is conservative it means that the force is derivable from a scalar potential function
∂V
V = V (x1 , x2 , . . . , xN ) such that Qr = −V ,r = − , r = 1, . . . , N. (2.2.20)
∂xr
In this case the equation (2.2.19) can be integrated and we find that to within an additive constant we will
have V = −W. The potential function V is called the potential energy of the particle and the work done
becomes the change in potential energy between the starting and end points and is independent of the path
connecting the points.
The kinetic energy T of the particle is defined as one half the mass times the velocity squared and can
be expressed in any of the forms
2
1 ds 1 1 1
T = M = M v 2 = M gmn v m v n = M gmn ẋm ẋn , (2.2.21)
2 dt 2 2 2
where the dot notation denotes differentiation with respect to time. It is an easy exercise to calculate the
derivatives
∂T
r
= M grmẋm
∂ ẋ
d ∂T m ∂grm n m
= M g rm ẍ + ẋ ẋ (2.2.22)
dt ∂ ẋr ∂xn
∂T 1 ∂gmn m n
r
= M ẋ ẋ ,
∂x 2 ∂xr
and thereby verify the relation
d ∂T ∂T
− = M fr = Qr , r = 1, . . . , N. (2.2.23)
dt ∂ ẋr ∂xr
192
EXAMPLE 2.21. (Equations of motion in spherical coordinates) Find the Lagrange’s form of
the equations of motion in spherical coordinates.
Solution: Let x1 = ρ, x2 = θ, x3 = φ then the element of arc length squared in spherical coordinates has
the form
ds2 = (dρ)2 + ρ2 (dθ)2 + ρ2 sin2 θ(dφ)2 .
The element of arc length squared can be used to construct the kinetic energy. For example,
2
1 ds 1 h i
T = M = M (ρ̇)2 + ρ2 (θ̇)2 + ρ2 sin2 θ(φ̇)2 .
2 dt 2
The Lagrange form of the equations of motion of a particle are found from the relations (2.2.23) and are
calculated to be:
h i
d ∂T ∂T
M f1 = Q1 = − = M ρ̈ − ρ(θ̇)2 − ρ sin2 θ(φ̇)2
dt ∂ ρ̇ ∂ρ
d ∂T ∂T d
M f2 = Q2 =