Tensor Analysis-Chapter 1

Chapter 1
Tensor s in Rectilinear Coor dinates in Two

Dimensions
1. The Usefulness of Tensor s
T
he r udiments of vector analysis ar e familiar to near ly ever yone in
the physical and engineering sciences. Tensor analysis, however, still verges on the
esoteric. This circumstance is unfortunate, for tensors provide a mathematical tool of
great simplicity, power, and elegance. This circumstance is likewise unnecessary, for a
proper exposition of vectors may be extended almost effortlessly to include tensors. At
the same time, tensor notation enhances the utility of vector analysis and divests the
latter from all of the ritual abracadabra which cannot fail to mystify the student who has
not yet succumbed to habit.
In some branches of physics and mathematics, tensors become indispensable. The
general relativity theory would be virtually impossible to formulate without tensor
theory. Indeed, so much has been made of the intimate relation between general
relativity and tensor analysis that it is widely supposed that tensors have little
application save in that theory. However, the fact is that all of classical and relativistic
dynamics and electromagnetism are best formulated in tensor notation, and differential
geometry is the mathematical application par excellence. The common denominator of
these disciplines is their essentially geometric nature.
An indication of the variety and utility of tensors may be had from a mere sampling
of quantities representable as tensors of various orders: potential functions, lengths,
curvatures, velocities, forces, stresses, the electromagnetic field, infinitesimal volumes,
and so forth. The versatility of tensors may be judged by the fact that they are not
limited to Cartesian coordinate systems nor to two or three dimensions. The power of
a tensor equation is that it is true for all coordinate systems if it is true for one. The
virtues of a knowledge of tensor analysis are clearly manifold. Let us develop the
essentials of the subject by proceeding from simple and concrete considerations to the
more general and abstract formulation.
2. Vector s in a Car tesian Coor dinate System
A
plane vector is familiar ly defined as a di r ected l i ne segment. It has a
prescribed length and a prescribed direction. Different vectors have either different
lengths or different directions. To state a length and a direction, therefore, is to specify
a simple plane vector. More commonly, a vector is defined implicitly by stating the
signed projections which it has on the axes of a rectangular Cartesian coordinate system
(see Fig. 1). Such numbers are called the components of the vector.
Chapt er 1: Tensor s i n Rect i l i near Coor di nat es i n Two Di mensi ons 4
Figure 1
In Fig. 1, the horizontal axis of the Cartesian system is indicated as the vertical
axis as The Cartesian coordinates measured parallel to these axes are designated
as and respectively. The superscripts 1 and 2 are thus identifying indices, not to
be confused with exponents. (Should it be necessary to consider the square of for
example, this will be indicated as The index notation is an almost universal
convention in tensor analysis because of its generality, its ready extension to any
number of dimensions, and its special convenience in indicating sums over
indexed quantities. If we employ the index notation, we may then write the
coordi nates of poi nt as or, for short,
where is understood to have one of the values 1 or 2.
Similarly, the point B has coordinates Hence the directed
projections upon the axes equal to and line segment AB has
These equations may be written
Hence we will refer to the vector as
It is clear that the values of the components of a plane vector are independent of a
translation of the origin of any Cartesian coordinate system to which they are referred (see
Fig. 2). That is to say, the directed line segment may be placed anywhere in the plane
without changing the values of its projections upon the coordinate axes. This means that
if the coordinates themselves undergo a transformation of the form
(2.1)
from the original system to a new system it is still true that
2. Vect or s i n a Cart esi an Coordi nat e Syst em 5
Figure 2
where are the components referred to the new axes. Fig. 2 shows the
effect of the transformation (2.1); the origin of coordinates at O is removed to the new
origin of coordinates whose coordinates in the original system are We say,
therefore, that the Cartesian components of a plane vector are invariant to a translation
of the origin.
We have shown thus far that a plane vector referred to a Cartesian coordinate
system is defined by its components which are such that (1) each component
is equal to the difference of the corresponding Cartesian coordinates of its end-points and
(2) the components of the vector are invariant to a translation of the origin. It should be
here remarked that neither of these results is true in general. The difference of the
coordinates of two points is usually not a vector in more general coordinate systems, and
in such systems a relocation of the origin will entail a change in the components of a
vector. The reasons for this will become clear as we refine and generalize the rather
unsophisticated definition of a vector which we have thus far used.
Ex. (2.1) On a map showing wind velocities as directed line segments having
in various localities the local direction of the wind and having respective
lengths equal to the wind speed on an appropriate scale, we superpose a
Cartesian axis system with positive axis to the east and positive axis
to the north. In such a reference system, find the components of a wind
velocity if the wind is (a) out of the SW at 10 miles per hour; (b) from the NNE
at 25 miles per hour; (c) from N of W at 20 miles per hour.
Chapt er 1: Tensors i n Rect i l i near Coordi nat es i n Two Di mensi ons 6
Ans.
(Hint: )
Ex. (2.2) (a) What relation have the vectors and
where a is any positive number? (b) What is the relation
between and if a is an arbitrary negative number?
Ans. and are parallel. and are antiparallel.
3. The Effect of a Rotation of Axes
A
s we have seen, the Car tesian components of a plane vector ar e invar iant to
a translation of the origin. Without loss of generality, therefore, we may place the
origin of Cartesian coordinates at the initial point of any vector. The projections of the
vector upon the coordinate axes through this origin are then the components of the
vector. Clearly, they are also the Cartesian coordinates of the end point of the vector.
Such a vector is thus called a position vector inasmuch as it locates the end point of
the vector.
Consider a position vector, such as the directed line segment in Fig. 3. In a
Cartesian coordinate system with axes and it has components
If, however, we generate a new Cartesian coordinate system by rotating the axes about
the origin , then the new components become the projections of
upon the rotated axes and These are obviously not equal to the
components except in the special instances when being any positive
or negative integer or zero. In general, therefore, the Cartesian components of a plane
vector are not invariant to a rotation of coordinate axes; the components change in a
determinate way. We may express this by saying that the vector is covariant with the
coordinates during a rotation of axes.
Let us consider more explicitly the relations in Figure 3 between the components
in the old system and the components in the new
system. In both coordinate systems, the coordinates of the origin are
(0 , 0). On the other hand, the coordinates of the point P are in the
original system, in the new system. Since the components of the vector
in each system are equal to the signed differences of the coordinates of the end
points, the vector will have components which are numerically equal to
whereas in the new coordinate system it becomes the vector whose
components are numerically equal to Hence transforms into
in exactly the same manner as transforms into and
conversely. Let us see what this transformation is.
3. The Ef f ect of a Rot at i on of Axes 7
In general, we may write,
(3.1)
Calling and noting that
we have
(3.2)
We may therefore conclude that the components of the vector transform according to
the equations
(3.3)
when the coordinates are transformed according to equations (3.2).
We may note in passing that a second rotation, by an angle will recover the
original coordinates and components. Hence
(3.4)
as may be verified by solving equations (3.3) for and
Figure 3
Ex. (3.1) Using equation (3.2), determine how Cartesian coordinates transform
under a rotation of (a) (b) (c) (d)
Ans.
Ex. (3.2) ( ) What do the coordinates of the point (3, 2) become under the
transformations of exercise (3.1)?
Ans.
( ) What do the components of the vector (3, 2) become under the same
transformation?
Ans. Same as in ( ).
Ex (3.3) Determine the inverse of equations ( 3.3 ) by solving them for and
Ans. Equations (3.4).
Ex (3.4) Find the transformations inverse to those of Ex. (3.1).
Ans. ;
Ex. (3.5) The determinant
is called the Jacobi an of the transformation (a) What is the
Jacobian of the transformation (3.2)? (b) Of its inverse? (c) Of the transformation
(3.1)?
4. Cont ravari ant Obl i que Rect i l i near Coordi nat e Syst ems 9
Figure 4
Ans.
We will have frequent occasion to use the Jacobian in some form in what follows.
We have now seen that in Car tesian coor dinates the components of a
plane vector are invariant to a translation of the origin but covariant with the
coordinates under a rotation of the axes. We may describe these results in another way
by saying that to define a plane vector one must both give the components of the
vector and, aside from the location of the origin, identify the coordinate system in
which these are the components. Thus a vector may be equally well defined in any one
coordinate system as in any other; a vector transcends definition in any particular
coordinate system. In the sense that every given Cartesian vector may be defined in
any Cartesian coordinate system, a vector is independent of coordinate system.
Relations between vectors such as equality, proportionality, etc. therefore
express coordinate-free relations.
4. Contr avar iant Oblique Rectilinear Coor dinate Systems
W
e have consider ed the effect upon the components of a vector of a r otation
of Cartesian axes. Now let us generalize our coordinate systems to include
oblique rectilinear coordinates. There are two ways to carry out the proposed
generalization. We will consider one of these ways in this section, the second way in
the next section, and the relation between them in the section after that.
Let the straight lines and intersect at , not necessarily
perpendicularly, as in Fig. 4. Consider the point . Through , draw lines parallel to
the axes, intersecting at a distance from and at a distance from
. Then are the oblique rectilinear contr avar iant coor dinates of the
point . We define the contr avar iant components of a vector to be the
difference of the corresponding coordinates of the endpoints and . The position
vector evidently has contravariant components which are identical
with the oblique contravariant coordinates of .
Figure 5
Let us consider how the oblique contravariant coordinates of a point transform on
going to a new pair of oblique axes through the same origin. In Figure 5, a point has
contravariant coordinates and in an oblique rectilinear coordinate system whose
axes are and In a new coordinate system whose axes are and
the coordinate of is the length of the line segment . From the point drop the
perpendicular onto Since the angle the length of is
Now through draw perpendicular to and through draw
parallel to and hence perpendicular to
In the triangle , the angle is and the
side is Hence Further, the angle is equal to
and the side is Hence Since is a
rectangle by construction, . Therefore
(4.1)
where
(4.2)*
The quantities which appear in equations (4.1) and (4.2) are denoted collectively as
*
The significance of the dots is solely that of position indicators, showing that
in this particular instance the superscript is the f i r st index and the subscript is the second. We will
have occasions elsewhere to consider a set of quantities in which the superscript is evidently the
Observe that when equations (4.1) and (4.3) reduce to equations
(3.2) for rotated rectangular coordinate systems.
Ex. (4.1) Show that Eqs. (4.1) and (4.3) reduce to Eq. (3.2) for simple
rotations of Cartesian axes. (Hint: take

Ex. (4.2) (a) Show that the transformation (4.1) may also be written as
(4.3)
(b) Show that when both sets of axes are orthogonal, this equation reduces to
the first of equations (3.2).
Ex. (4.3) (a) Show that the transformation (4.4) may also be written as
(4.4)
(Hint: Use the relations
(b) Show that when both sets of axes are orthogonal, this equation reduces to
the second of equations (3.2).
Ex. (4.4) Show that when equations (4.7) and (4.8) reduce to
equations (3.1).
Ex. (4.5) Determine the transformations (4.1) and (4.4) when
Ans.
second index and the subscript the f i r st. This is not a mere picayune notational device with which to
burden the novice. It is found to be a necessary means for distinguishing order of indices when that
order is significant. For example, it is virtually universal practice to display the in a square array
called a matr i x. By convention, the f i r st i ndex denotes r ow, the second col umn. In the matrices
and the first index again denotes row, the second column, but in this case the indices themselves
serve as position indicators and no dots are needed. Later, as familiarity crystallizes into habit, the
dots may be omitted.
Chapter 1: Tensors i n Recti l i near Coordi nates i n Two Di mensi ons 12
Figure 6
Hence
Note: Though the final expression is identical with that given for in Ex.
(3.1), it must be remembered that there the coordinates were Cartesian, here
oblique. The content of the two expressions is therefore altogether different.
Ex. (4.6) (a) Use the transformations of Ex. (4.5) to find the coordinates
and for the point (b) Construct and label the axes of
the two coordinate systems of Ex. (4.5). Mark the point of part (a), using for
this purpose the coordinates (c) By the appropriate geometric
construction, determine the coordinates Do they appear to have the
values computed in (a)?
4. Contravariant Obl ique Rectilinear Coordinate Systems 13
Ans. (a) For (b), (c), see
Figure 6.
Ex. (4.7) Show that the inverse transformations to equations (4.1) and (4.3)
are
(4.5)
(4.6)
(Hint: Consider the axes and to be first given and axes and
to be gotten by rotations through the angles and )
Ex. (4.8) Determine the transformations (4.7) and (4. 8) between the axis
systems of Ex. (4.5).
Ans.
Ex. (4.9) Use the transformations of the preceding Exercise to find the
coordinates and for the point whose coordinates are
Are your results those of Ex. (4.6)?
Ex. (4.10) By solving equations (4.1) and (4.3) simultaneously for and
show that the inverse transformations may also be written as
(4.7)
(4.8)
Show that equations (4.11) and (4.12) are equivalent to (4.9) and (4.10),
respectively. (Hint: Show that
by replacing by and
by Then
Ex. (4.11) (a) Show that the Jacobian of the transformation defined by
equations (4.1) and (4.3) is
(Hint: Use the identity derived in the preceding Exercise.) (b) From part (a),
determine what values of and are impermissible for any
geometrically meaningful transformation of coordinates. (c) What is the
condition for a pure rotation of axes, whether oblique or rectangular? What,
then, is the value of the Jacobian for a purely rotational transformation?
Ans.
Ex. (4.12) Construct graphically the contravariant vectors whose components
with respect to axes making an angle are
Equations (4.2) and (4.5) offer occasion to intr oduce a notational convention
which represents a characteristic feature of tensorial notation and, in addition, a most
advantageous economy. Using the index notation, the two equations may be written
collectively as
(4.9)
where it is understood that has the range of values since we are
considering two dimensions. We abbreviate this equation to
(4.10)
by the familiar summation symbol The summed-over index j is called a dummy
index, umbral index, or summation index to distinguish it from the free index . It
is evident that both summation index and free index have the range 1 to 2 . In
dimensions they would, of course, range from 1 to . Clearly, a summation index
may be replaced by any other summation index without affecting the expression; just
as clearly, this is not in general true of a free index.
We now introduce the very useful summation convention, whereby any r epeated
index appear ing once as a subscr ipt and once as a super scr ipt is to be summed
over fr om 1 to , wher e is the number of dimensions. Thus we may write
equation (4.10) in the still simpler form
(4.11)
the summation over being tacitly understood since it appears as a subscript in
and in the same term as a superscript on the Admittedly, the notational economy
in this simple instance is not particularly great. It becomes substantial, however, in
terms where several dummy indices appear and in spaces of several or many
dimensions. We shall make use of the summation convention hereafter. To avoid
possible confusion, it is to be understood that no summation index may appear more
than once as a subscript nor more than once as a superscript in any term.
Ex. (4.13) Write out in full the explicit equations expressed in index notation
as: (a) (b) Assume that the range of each index
is
Ans.
Ex. (4.14) Repeat Ex. (4.13) assuming a range (1, 2, 3) for each index. (Is the
utility of the summation convention more apparent now?)
Ex. (4.15) Find the coefficients when
Ans.
Ex. (4.16) Evaluate the Jacobian of the transformation in Ex. (4.15). Use
Ex. (4.11) to verify your answer.
Ans.
Ex. (4.17) Find the transformation of contravariant coordinates from
Cartesian coordinates to oblique coordinates where the axis
coincides with the axis and the axis makes an angle with the
axis.
Ans.
Ex. (4.18) Consider a vector whose magnitude is and which makes an angle
with the axis. (a) Show that its contravariant Cartesian components
are (b) Show that under the transformation of
Ex. (4.17) its components become Does this suggest cir-
cumstances when oblique coordinates may be advantageous?
We may now say how the components of a vector tr ansfor m under a
transformation from one oblique contravariant rectilinear coordinate system to a
new system We need only recall that transforms into in exactly the same
manner as transforms into From equation (4.11), therefore, it is evident that
(4.12)
where, as before, the are given by (4.3) and (4.6).
Ex. (4.19) Write out explicitly the matrix of coefficients in equation
(4.16). Observe the convention of the footnote on p. 13.
Ans.
(4.13)
5. Covar iant Oblique Rectilinear Coor dinate Systems
W
e have seen how, by equations (4.12), we may compute fr om the
components of a vector in one rectilinear coordinate system its components
in another coordinate system when the coordinate transformation between the two
rectilinear systems is known. Implicit in this fundamental procedure is the very method
of defining coordinates: it has been done, as in Fig. 4, by drawing through any point
straight lines parallel to the axes and then measuring their
intercepts on the axes. There is, however, a different way of defining the coordinates of
, a way which has an equal claim upon our consideration.
Again, consider coordinate axes and as in Fig. 7. Let be any point
whose coordinates are to be determined. From drop perpendiculars and onto
and Their intercepts, and , are the coordinates and of the
Figure 7
5. Covari ant Obl i que Rect i l i near Coordi nat e Syst ems 17
point P. To distinguish the coordinates so defined from those defined in the previous
way, they have been given inferior indices. They are called covar iant coor dinates
and (The former, designated by superior indices, have been called the
contr avar iant coor dinates and
The relation between the covariant and contravariant coordinates of any point P is
easily derived. Thus, in Fig. 7, let PC be drawn parallel to and PD parallel to
These line segments are then equal to and respectively. Hence it is
obvious from Fig. 7 that
(5.1)
By solving this set of equations for and we obtain the inverse transformation
(5.2)
Ex. (5.1) Find the covariant coordinates of the point whose contravariant
coordinates are when (Hint: apply equations
(5.1)). Illustrate by a graphical construction.
Ans.
Ex. (5.2) Find the contravariant coordinates of the point whose covariant
coordinates are when
Ans.
Figure 8
Let us now der ive the equations of transfor mation by which the covar iant
coordinates in a new rectilinear coordinate system may be found from the covariant
coordinates in a given system. Thus, in Fig. 8, the point P has coordinates and
relative to the original coordinate axes and through the origin
In a new coordinate system in which is the new axis, however, the covariant
coordinate of P is Hence PC is perpendicular to PB is perpendicular
to and PA is perpendicular to Now let PF be an extension of CP; draw AF
parallel to and therefore perpendicular to CF. From A drop the perpendicular AE
It is
onto the perpendicular AH onto PB, and the perpendicular AD onto
then clear that the angle APB is equal to and the angle APF is equal
to Now since AECF and AHBD are rectangles, it is evident that
(5.3)
and
But HA =AP sin and AF =AP so that
AP = or HA
Substituting for HA and AF from equations (5.3), we have
However, from Fig. 8 it is clear that so that
Therefore, we obtain at last the desired relation
(5.4)
Similarly, we can show (Appendix 1.2) that
(5.5)
Ex. (5.3) Show that the transformations given in equations (5.4) and (5.5) may
also be written as
(5.6)
(5.7)
(Hint: Use the relations and
5. Covari ant Obl i que Rect i l i near Coordi nat e Syst ems 19
Ex. (5.4) With the aid of the Table below, determine the transformation of
coordinates for the system of axes as shown in Fig. 9.
Ans.
Ex. (5.5) Using only the values for the entries in the Table below,
verify the remaining entries without referring to a table of trigonometric
functions.
Angle Sine
3/5
12/13
63/65
8/17
817/1105
220/221
Figure 9
Ex. (5.6) Show by solving equations (5.4) and (5.5) simultaneously for and
and by making use of the results of Ex. (4.11) that the inverse of the
transformation of covariant coordinates is
(5.8) ,
(5.9) .
Ex. (5.7) Show that the Jacobian of the transformation expressed in
equations (5.4) and (5.5) is
where is the same as in Ex. (4.11).
Ex. (5.8) Write out explicitly the matrix of coefficients in equation (5.11)
below. (Use the first index to indicate row, second index to indicate column, as
noted in 4).
Ans.
(5.10)
Ex. (5.9) Evaluate the Jacobian of the transformation derived in Ex. (5.4). Verify
your answer by making use of Ex. (5.7).
Ans.
Ex. (5.10) Find the transformation of covariant coordinates from Cartesian
coordinates to oblique covariant coordinates where the axis
coincides with the axis and the axis makes an angle with the
axis.
Ans.
Ex. (5.11) Consider the vector of Ex. (4.18). (a) Show that its covariant
components in the Cartesian coordinate system are
(b) Show that under the transformation of Ex. (5.10) its covariant components
become
Equations (5.4) and (5.5) ar e the equations of tr ansfor mation of covar iant
rectilinear coordinates. We have seen, however, that a position vector with components
transform exactly as the rectilinear covariant coordinates.
6. The Rel ati on Between Contravari ant and Covari ant Transformati ons 21
Therefore
(5.11)
is the equation of transformation of a covar iant vector when the rectilinear
covariant coordinates are transformed from to The coefficients are those
given in equations (5.4) and (5.5) (or in Ex. (5.8)). Equation (5.11) is the covariant
analogue of equation (4.16) for contravariant vectors and equations (5.4) and (5.5) are
analogous to equations (4.1) and (4.4).
The difference between contravariant and covariant vectors in the plane may be
delineated to some degree by observing that if a component of a contravariant vector
is zero, this shows that the vector is parallel to the other axis; on the other hand, if
a covariant component is zero, this shows that there is no perpendicular projection
upon that axis, i.e., that the vector is perpendicular to that axis.
6. The Relation Between Contr avar iant and Covar iant Tr ansfor mations
W
e have seen that in any plane r ectilinear coor dinate system we may give to
any point P a set of contravariant or covariant coordinates. These coordinates are
at the same time the components of the position vector from the origin to the point
P. If one and the same vector can be represented in two ways, there must exist some
close relation between the two representations and between their respective laws of
transformation. Let us see what this is.
We begin by forming the quantity Using equations (4.16) and (5.11), we
find that
A substitution of the values for and from equations (4.17) and (5.10) yields
the results
(6.1)
and
(6.2)
From the answer to Ex. (4.11) we see that
(6.3)
Combining this with equation (6.1), we summarize these results in the formula
(6.4)
We see, therefore, that by the definition of inverse, the covariant transformation is the
inverse of the contravariant transformation.
With the results given in equation (6.4), we can now observe that the quantity
is
Figure 10
Since it has the same value in the new coordinate system as in the old, and indeed
has the same value in every rectilinear coordinate system, it is called an invar iant.
We may readily interpret it when is the position vector of the point
Then
In an orthogonal Cartesian coordinates system
where is the distance of P from the origin O. Since this invariant is the same in every
coordinate system, its value and its meaning are the same in all.
We thus see that is the length, or magnitude, of the position vector whose
rectilinear contravariant components are We write this as Similarly,
is the length or magnitude of the vector whose contravariant components are
Therefore we have shown that
where the covariant components of the vector whose contravariant components are
are to be found by an equation analogous to (5.1), namely
(6.5)
Clearly, when i.e., when the axes are orthogonal, there is no difference
in the values of the respective covariant and contravariant vector components.
6. The Rel at i on Bet ween Cont ravari ant and Covari ant Transf ormat i ons 23
Ex. (6.1) Show by direct geometric proof that
(Hint: Use Fig. 10 to calculate as the square of the hypotenuse of triangles
and .
Thus
Therefore
and
Hence from triangle we have
whereas from triangle we have
Adding the two expressions for gives
Therefore, as required,
Ex. (6.2) By solving equations (6.5), show that
(6.6)
Ex. (6.3) Find the magnitudes of the following vectors:
Ans. (a) 17; (b) 25. (Hint: see Ex. (6.4)); (c) 7 (Hint: use Eq. (6.5)).
Ex. (6.4) Show that
(6.7)
(Hint: use equation (6.6)).
7. Changes of Scale
W
e have defined a plane vector to be a dir ected line segment. The pr ojections
of the directed line segment upon rectilinear axes are the components of the vector.*
If the components are orthogonal projections, the vector is a covariant vector. The
transformations which the components undergo when the directed line segment is
projected upon a new set of rectilinear (but not necessarily rectangular!) axes have been
found in terms of the orientation of the new axes with respect to the old. At the same time,
the vector is invariant to a translation of the origin.
In addition to arbitrary change of origin and independent rotation of axes, one other
kind of transformation is possible, namely, a change of scale. Thus, suppose that
new units along the axis are equal to 1 former unit along the axis.
(Note that the index on the scale factor is parenthesized to indicate that it is not a
vector index.) Then the coordinate of a point on the axis will be times as great
in the new coordinates as in the old. To express this most conveniently, we first set
(7.1)
(Clearly, the index does not satisfy the conditions of the summation convention.) We
then have new coordinates which are given by
(7.2)
Written out, this is simply
A similar transformation
(7.3)
may be applied to covariant rectilinear coordinates.
Our language reflects the fact that we think not only of the line segment, but also of the components
*
by means of which we identify the line segment, as constituting the vector.
7. Changes of Scal e 25
For fullest generality, let us determine the combined effect of a change of axes
followed by a change of scale. The change of axes is effected by equation (4.14) as
(4.14)
The so found are substituted into equation (7.2), yielding
(7.4)
We see that a single transformation
(7.5)
can effect both changes simultaneously; written out, the coefficients of the
transformation (7.5) are
(7.6)
where they have been arranged as a matrix.
Let us note that the order of operations is not immaterial. Had we first applied a scale
change, then a change of axes, the result would have been
(7.7)
The transformation now has the matrix
(7.8)
This is clearly not the same as equation (7.6). In equation (7.6), each row has a common
whereas in equation (7.8) each column has a common
The transformations (7.6) and (7.8) are now quite general. By them, we may effect
any change of axes and any change of scale in whatever order may be desired.
Ex. (7.1) (a) Show that
(b) Show that
Ex. (7.2) (a) Derive the transformation matrix for a transformation of
covariant coordinates with changes of scale following rotation of axes.
(b) Derive the transformation matrix for a transformation of covariant
coordinates with change of scale preceding rotation of axes.
Ans.
Ex. (7.3) Show that
(Hint: use the fact that covariant and contravariant transformations with
respect to the same sets of axes are mutual inverses, their Jacobians therefore
reciprocals.)
8. Inter pr etation of an Ar bitr ar y Tr ansfor mation Matr ix
I
t is of quite as much inter est to consider the conver se pr oblem: Given a
reversible transformation
to what change of axes and changes of scale does it correspond? Since the coefficients
of the transformation do not, by themselves, indicate whether it is covariant or
contravariant, we shall have to specify. For the purposes of illustration, let us consider
a contravariant transformation of coordinates.
From equation (7.6) it is clear that the ratio
is independent of the scale factors In other words, we need not know or
in order to derive the result that
(8.1)
8. Int erpret at i on of an Arbi t rary Transf ormat i on Mat r i x 27
In a like manner, we can readily show that
(8.2)
(Note that and depend upon as well as the )
Since it is assumed that the and are known, the angles and may be
found. The only ambiguity is in the quadrants of and However, any particular
choice of quadrants may be tested, for and
This determines the quadrant of a similar procedure determines the quadrant of
We complete the analysis by noting from equation (7.6) that
or
(8.3)
Ex. (8.1) Determine the effect of the transformation
upon the contravariant rectilinear coordinates if Assume that
scale changes follow rotation of axes. (Hint: use equations (8.1), (8.2), and
(8.3).)
Ans.
Ex. (8.2) From Ex. (7.2a) show that for a covariant transformation of
coordinates
(8.4a)
(8.4b)
when the scale changes follow the rotation of axes. For a geometric derivation of
similar results, see Appendix 1.3.
Ex. (8.3) From equation (7.8) show that
etc. Hence show that it is necessary to know both the matrix of coefficients and
the angle between the new axes, in order to interpret the transformation when the
change of scale precedes the rotation of axes.
Ex. (8.4) From Ex. (7.2b) show that
etc.
We may summar ize our r esults thus far by obser ving that for a given change
of axes followed by a change of scale, the general contravariant and covariant
transformations may be written may be written as
(8.5) and
We therefore see that the matrix whose elements are the product of the
matrices (8.5), is
From equations (6.4) we see that this is simply
When no changes of scale occur, i.e., when the right hand side is
simply To achieve this result more generally, we associate to any contravariant
transformation that covariant transformation which is its inverse. Then we must have
(8.6)
Conversely, to any covar iant tr ansfor mation ther e is associated a contr avar iant
tr ansfor mation which is its inver se.
Ex. (8.5) Determine the relation between contravariant and covariant
coordinates following a change of scale. (Hint: apply a scale change to
equations (5.1) and (5.2) and make use of equations (8.6).)
8. Int erpret at i on of an Arbi t rary Transf ormat i on Mat r i x 29
Ans.
(8.7)
(8.8)
Ex. (8.6) From equation (8.1) show that
(8.9)
(Hint: Use the relation and the formula for the tangent of
the sum of two angles.) See Appendix 1.3 for a geometric derivation of the
same result.
Ex. (8.7) From equation (8.2), show that
(8.10)
(Hint: use a method similar to that in Ex. (8.6).) For a geometric derivation, see
Appendix 1.3.
Ex. (8.8) Show that the covariant and contravariant components of a vector are
related by the equations
(8.11)
(8.12)
(Hint: use equations (8.7), (8.8), and the definition of a vector.)
Ex. (8.9) Show that the equation implies that (Hint: apply
equations (8.11) and (8.12) to the original vectors.)
9. The Definition of a Vector in Rectilinear Coor dinates
L
et us now summar ize our pr evious analysis by saying that a plane vector
may be defined in a particular rectilinear coordinate system as a pair
of components which relate to the components of the same vector in any other
rectilinear coordinate system by the same linear homogeneous transformation with
constant coefficients as the transformation of coordinates themselves. Suppose that
the latter is
Then the components of the vector in the coordinate system become
in terms of the components in the coordinate system. The intimate relation of
the transformation of vectors to the transformation of coordinates may be displayed
explicitly by noting that
(9.1)
We may then write the vector transformation as
(9.2)
Thi s i s by def i ni ti on both the necessar y and the suf f i ci ent condi ti on that two pai r s
of number s consti tute the components of a contr avar i ant vector i n two coor di nate
systems and Hence equation (9.2) may be used to determine what the
components will be in any other rectilinear coordinate system or it may be used to test
whether or not two number-pairs can be the components of the same vector in two
given coordinate systems. The definition is more often used in the former mode than
in the latter.
The transformation law for covariant vectors may be similarly obtained. Since the
covariant transformation of rectilinear coordinates is the inverse of the contravariant
transformation, and since
(9.3)
the law of transformation for covariant vectors must be
(9.4)
The equations (9.2) and (9.4) in effect define contravariant and covariant vectors,
respectively. Vector s ar e evi dentl y or der ed sets of number s i .e., components
whi ch obey the appr opr i ate tr ansf or mati on l aw under r ever si bl e tr ansf or mati ons
of coor di nates. The general forms (9.2) and (9.4) clearly apply not only when the
transformation of coordinates is linear, as between rectilinear systems, but also when
10. The Addi t i on of Vect ors 31
more general transformations are considered. It is therefore on account of this future
usefulness rather than any present necessity that the equations (9.2) and (9.4) are
adopted as the defining relations for vectors.
Note as well that equations (9.2) and (9.4) also permit non-homogeneous linear
transformations (i.e., linear transformations which include additive constants). In other
words, in rectilinear coordinates, the vector components are independent of location.
Ex. (9.1) Are ( -5, 8) the components of the vector in a coordinate
system
?
Ex. (9.2) What are the components of the vector ( -5, 8) in a new coordinate system
related to the original by the transformation
?
Ans.
Ex. (9.3) (a) What are the components of the vector in a new
coordinate system? (b) Of the vector ?(c) Of the vector
? (d) Of the vector ?
Ans.
10. The Addition of Vector s
C
onsider two vector s and . Since they ar e vector s, they satisfy the
relations
and
Adding their left and right sides, respectively, we have
Figure 11
Clearly, the sum
satisfies the vector transformation law and is therefore itself a vector. This sum has
a simple interpretation. Thus, let
From Fig. 11 it is obvious that is the diagonal of the parallelogram whose adjacent
sides are the vectors , lying along and of length , and lying
the axis and of length .
along
Since a rectilinear transformation could change the axes and therewith the
components but would not alter the magnitudes or directions of or it is
clear that this interpretation of the vector sum is entirely general. It is known as the
par allelogr am law: The sum of two contravariant vectors is the diagonal of the
parallelogram whose two adjacent sides are the given vectors.
Since in rectilinear coordinates, the location of the vector is arbitrary, it is
common practice to add vectors by a geometric construction whereby the vector
is translated so as to lie along AC. Any number of vectors may then be added in this
head-to-tail fashion, the resultant being the single vector extending from the initial
point of the first to the end point of the last. It should be clearly understood, however,
that such a procedure is defined only in rectilinear systems and does not hold in the
more general curvilinear coordinate systems to be considered later.
We note that the difference is simply the sum of and
As in ordinary algebraic subtraction, vector subtraction is thus
a special case of addition.
10. The Addi t i on of Vect ors 33
Figure 12
Figure 13
The parallelogram law also holds for covariant vectors, as may be seen in Fig. 12,
where the vectors are and Change of axes
will not alter the geometric relationships.
Ex. (10.1) (a) A man can paddle a canoe 3 miles per hour in still water. If he
paddles directly across a stream whose current is 4 miles per hour, what are
the components of the canoes velocity vector? Take downstream to be in the
direction of the positive axis and across the stream to be in the direction
of the positive axis. (b) What is the magnitude of the velocity? (c) The
direction? (d) Draw the appropriate vector diagram.
Ans. (a) (b) 5 miles per hour (c)
Ex. (10.2) If the man wished to arrive at a point directly across from his
starting point, at what speed and in what direction would he need to paddle?
Ans. His still-water velocity must be such that when added to the stream
velocity it yields a resultant velocity vector with zero
component. Thus
Hence his speed (the magnitude of his velocity vector) must be
and his direction relative to the axis must be at an angle
11. The Inner Pr oduct of Vector s
C
onsider the vectors and They transform according to the respective
laws
and
Then the sum indicated by
(11.1)
evidently has the same value in both coordinate systems and is therefore an invariant.
It is called the inner pr oduct of and
Since its meaning and value do not depend upon the coordinate system used, we
will choose the coordinate system which best serves our convenience. For this
purpose, consider a Cartesian system as in Fig. 14 with axis lying along the
direction of the vector The components of must therefore be where
is the magnitude of On the other hand, the components of are evidently
where is the magnitude of and is the angle between
and Hence the inner product has the value
(11.2)
11. The Inner Product of Vect ors 35
Figure 14
Figure 15
Since each of the factors on the right is independent of scale changes or rotations of
the axes, we may state as a general rule that the inner product of a covariant and a
contravariant vector is the product of their magnitudes times the cosine of the angle
between them.
When and are the same vector, and the result is evidently the
square of the magnitude of the vector. We have already seen this in the special case
of the position vector. On the other hand, when Hence if
we can infer that the vectors are orthogonal.
Ex. (11.1) (a) Show from equation (11.2) that
(b) Prove the same result by using equations (6.5) and (6.6).
Ex. (11.2) Derive the cosine law of plane trigonometry.
Ans. Let and be the sides of a plane triangle , as in Figure 15.
Then evidently, because the triangle is a closed figure, and by virtue of the
parallelogram rule for the addition of plane vectors,
Hence, by Ex. (8.9),
Multiplying the respective left hand sides and the corresponding right hand sides,
we have
Then, since
this becomes
the law of cosines.
Ex. (11.3) Show that is posi ti ve def i ni te, i.e., that the
equality holding only if
Ans. Since
where is the angle between the axis and the axis, and since we must
have it follows that, if and are of the same sign,
provided However, if then
The parallel argument if and are of opposite sign is obvious. Only if
therefore, is
12. Poi nt s and Li nes 37
Figure 16
Ex. (11.4) In a Cartesian coordinates system, find the angle between the following
vectors, using equation (11.2):
Ans.
Ex. (11.5) Determine the component of so that will be
perpendicular to in a Cartesian coordinate system.
Ex. (11.6) Prove that is orthogonal to
12. Points and Lines
O
ur r esults thus far may be given a simple but useful application. In
Fig.16, let and be rectilinear axes and let L be any line in the
plane. The covariant vector is orthogonal to L and the position vector locates
any point P on L. Consider the invariant
where d is the magnitude of and is the magnitude of It is clear that
Hence
(12.1)
For fixed this is the equation of the line . Therefore, is identified by the
components of which are thus known as the line var iables, just as the are
known as the point var iables. The meaning of is obvious: it is the distance of the
origin from the nearest point on the line . The proof is trivially simple if one
takes the point to coincide with the point . Then
The line may be readily identified by locating its intercepts on the and
axes. The exceptional case arises when passes through the origin. Then the
equation becomes
Ex. (12.1) Equation (12.1) represents the line through the end point of and
perpendicular to it. Show that if is replaced by where is any finite
real number different from zero, then equation (12.1) will become(12.2)
(What interpretation has ? See Ex. (2.2).)
Ex. (12.2) From equation (12.1) show that the vector is orthogonal
to the vector (Hint: replace by )
Ex. (12.3) From equation (12.2) show that if is a unit vector
then represents the perpendicular distance of the line from the origin. How
does one interpret the case ?
Ex. (12.4) Interpret the equation of the straight line in
a Cartesian coordinate system.
Ans. The straight line of slope whose minimum distance from the origin
is 2 units.
Ex. (12.5) Interpret the equation of the straight line
in a rectilinear system where
Ans. Same as in Ex. (12.4).
Ex. (12.6) Interpret the equation of the straight line in
a rectilinear system where
Ans. Same as in Ex. (12.4).
13. The Def i ni t i on of Tensors 39
Figure 17
Ex. (12.7) Draw a single figure showing a common axis, the several
axes of Exs. (12.4, 5, 6), and the line given in each. Place by each
intercept the value of the non-zero coordinate.
Ans. Fig. 17.
Ex. (12.8) Express the condition for parallelism of two lines in terms of their
respective vectors and
Ans. where is a real number. Ex. (12.9) Show that the
distance between two parallel lines is
The symmetry between point coor dinates and line coor dinates may be
exhibited further by considering once more the equation When is
fixed and the are allowed to assume all consistent sets of values, this is the
equation of the line L, the different values of identifying the different points on
the line. On the other hand, if the values of are held fixed and the allowed to
assume all consistent sets of values, this is the equation of the point P, the different
values of identifying all lines through the point.
13. The Definition of Tensor s
W
e have seen that if we ar e given two vector s and their sum and
difference are also vectors. The same may be said of their covariant forms. Let
us now consider the result of multiplying the vectors and Since
and
the several products of the form
(13.1)
transform in a manner which is an obvious and direct generalization of the
transformation law for contravariant vectors. The number of such products is four in
two dimensions; more generally, it is in dimensions. A set of quantities
which transforms according to equation (13.1) when the variables undergo a
transformation is called a contr avar iant tensor of the second r ank. Lest the example
be misleading, let it be clearly understood that such a set of quantities constitutes
a tensor whether or not they are the product of two contravariant vectors such as
and Satisfaction of the condition
is by definition both the necessary and the sufficient condition that the constitute
a second rank contravariant tensor.
In analogous fashion, one may consider the product to define a prototype
second rank covariant tensor By such a procedure we may be led to the general
definition that a covariant tensor of the second rank is a set of 4 (or ) quantities
whose transformation under a transformation of coordinates is according to the
rule
(13.2)
Again, this is both a necessary and a sufficient condition. It may be applied as a recipe
for determining the components in a new system of coordinates when the components
in the old system are given; or it may be used as a test to see whether sets of
components in old and new coordinate systems are those of the same tensor.
A third possibility, one which did not arise with the sum and difference of two
vectors, is that of defining a mixed tensor of the second rank, once covariant, once
contravariant. By taking as a prototype, one can show in a similar
straightforward fashion that the equation of transformation for a mixed tensor of the
second rank is
(13.3)
One particularly useful and interesting mixed tensor of the second rank is the
Kronecker delta, for its transformed values are
It is thus one of the few tensors whose components have identically the same values
in different coordinate systems.
The definitions of tensors of the second rank bring out for the first time thus far
the fact that the rank is equal to the number of free indices. This therefore implies that
a vector may be considered to be a tensor of the first rank, an invariant a tensor of the
zeroth rank.
It should now be apparent that tensors of higher rank may also be defined. Thus
a tensor of rank having components in two dimensions (or components in
dimensions), is a set of quantities which transform according to the rule that
(13.4)
where the total number of indices of all kinds is . If of the indices are superscripts
and of the indices subscripts the tensor is said to be -fold
contravariant and -fold covariant. Evidently there may be types of tensors of
rank , from wholly contravariant to wholly covariant
We need only the definition (13.4) to prove an important theorem about tensors
known as the Quotient Law. The quotient law states that: If the pr oduct of any
indexed system with a tensor whose components ar e ar bitr ar y is known to be a
tensor , then the or iginal system must be a tensor of the appr opr iate r ank and
type.
To prove this theorem, we start with the fact that a system times
an arbitrary tensor is known to be a tensor That is,
(13.5)
Since this is true in a new coordinate system, we have also that
But since
we have
If the components of are entirely arbitrary and if the transformation of
coordinates is reversible, then this equation can be true if and only if
But this is precisely the equation of transformation of a tensor Satisfying the
tensor transformation is both the necessary and sufficient condition that a system be
a tensor. Hence is a tensor of the kind and rank indicated by the full set of
indices. Note that the product in equation (13.5) may include contracted products.
Note further that the tensor must have arbitrary components, else the argument
either breaks down or is qualified in some way. We shall make a number of important
applications of the Quotient Law in future developments.
Tensors of the second and higher rank offer a possibility not afforded by vectors
and invariants, namely symmetries or antisymmetries with respect to indices. Thus,
consider a second rank covariant tensor of which it is true that
(13.6)
Such a tensor is said to be a tensor symmetric in and . One of the most important
features of such symmetry is the fact that it is preserved under transformation to any
new coordinate system. Thus
Symmetry is therefore independent of coordinate system. Similarly, symmetry may be
defined for a second rank contravariant tensor Tensors of higher rank may
possess similar symmetries in some or all of their contravariant or covariant indices,
respectively.
It should be noted, however, that symmetry between covariant and contravariant
indices is not defined, for it would not be independent of coordinate systems.
By analogy with the definition of a symmetric tensor, we may define an
antisymmetr ic tensor as one for which it is true that
(13.7) or
Again, antisymmetry is a property independent of coordinate system. The proof is
exactly the same as for symmetry.
It should be recognized that symmetry or antisymmetry reduces the number of
independent components of a tensor. Thus, a symmetric covariant tensor of the second
rank has only three distinct components: and in two
dimensions. By comparison, an antisymmetric second rank covariant tensor has only
one non-zero component, apart from sign, namely The same is true
of contravariant second rank tensors. Clearly, tensors antisymmetric in more than two
indices will be identically zero in two dimensions since at least two indices cannot fail
to be the same.
Ex. (13.1) (a) How can one tell by inspection that is an invariant? (b)
That is a tensor of second rank? (Hint: how many free indices does each
have?)
Ex. (13.2) (a) If and are arbitrary vectors and is known to be
an invariant, show that must be a covariant tensor of the second rank. (b)
If is an arbitrary vector and is known to be an invariant, show
that is a covariant tensor of the second rank. Hence show that
is a second rank covariant tensor if it is symmetric. (Hint: Write out all terms
in two coordinate systems and assign to the successive sets of values
and
Ex. (13.3) Determine the conditions under which is a tensor if: (a)
is an invariant, arbitrary; (b) is an invariant,
and arbitrary; (c) is an invariant, and arbitrary.
Ans. (a) completely symmetric in , , ; (b) symmetric in and ;
(c) independent of any symmetry in , , and .
We have now defined tensor s in general, pr oved the impor tant Quotient Law
for them, and recognized the possibility of symmetry and antisymmetry for tensors of
rank two or higher. It is time to confront the inevitable and persistent question: What
is a tensor?
A literally correct but unedifying answer is that a tensor is a set of quantities which
obey the appropriate law of tensor transformation. In the end, this answer will have
to suffice, though it is somewhat disappointing that whereas tensors of rank zero and
one can be visualized, this is no longer possible for tensors of higher rank. Two
remarks may offer some slight comfort in this respect, however. First, examples of
quantities represented by second rank tensors include stress, strain, viscosity, angular
velocity, and electromagnetic field. To the extent that these are familiar physical
quantities, so are tensors.
Second, a tensor of the second rank may be regarded from a mathematical
viewpoint as a linear operator. Thus a tensor will transform a vector into the
vector This operation is not difficult to picture in the minds eye. Since
this is a linear operation, the tensor becomes a linear operator. This is true not only
of mixed tensors but of covariant and contravariant tensors as well. By extension,
tensors of higher rank may also be thought of as linear operators. The higher the rank,
however, the more remote is the operation from any direct pictorial apprehension. In
this matter, therefore, one should expect to rely more upon the familiarity which
comes with long acquaintance and abundant usage than on ready mental imagery.
Beyond tensors of the second rank, one must in general reluctantly forsake geometric
habiliments.
Ex. (13.4) Show that the tensor regarded as an operator, has upon the
effects: (a) rotates by an angle
(13.8)
and multiplies it by a factor
(13.9)
(b) rotates by an angle
(13.10)
and multiplies it by a factor
(13.11)
Here is the angle between the coordinate axes. (Hint: use equations (8.1),
(8.2), and (8.3), regarding as the coefficients of a linear transformation of
coordinates.)
Ex. (13.5) Let and be four definite vectors and let a mixed
tensor of rank two, be expressible (see Ex. (1.15.12)) as
Then show that the effect of the operator upon any contravariant vector
is to transform it into
Similarly, show that the effect of the operator upon any covariant vector
is to transform it into
Thus, by choosing and to be vectors along the and axes,
respectively, and by choosing and to be vectors perpendicular to the
and axes, respectively, interpret the components of the tensor
14. The Fundament al Tensor and t he Li ne El ement 45
14. The Fundamental Tensor and the Line Element
C
onsider the tr ansfor mation fr om given var iables to new
variables If we take differentials of both sides, we get
(14.1)
From this equation it is clear that the contravariant coordinate differentials transform
as contravariant vectors. They are differ ential vector s.
It may seem arbitrary that we have chosen the differentials of the contravariant
coordinates to consider. Could we not have chosen the differentials of the covariant
coordinates as well? True, this could have been done, but as we have seen in 12, it
is the contravariant coordinates which are the point variables, the covariant
coordinates which are the line variables (or surface variables in three dimensions,
hyper-surface variables in four or more dimensions). Since we customarily desire to
think of a set of variables as a set of point coordinates, we thereby tacitly elect to
regard them as contravariant.
Let us further note, in anticipation of developments to come, that only in
rectilinear coordinate systems is it true that
Therefore it is only in such coordinate systems that the coordinates are themselves
the components of a position vector. In contrast, equation (14.1) remains valid in all
coordinate systems, rectilinear or not. Hence the differentials of point coordinates are
unexceptionably contravariant vectors.
Let us now form from the differential vector a differential invariant of
particular interest. For this purpose we define a covariant tensor whose components
in a Cartesian coordinate system are
(14.2)
(Beware of confusing this a symmetric second rank covariant tensor whose
components are 0 and 1 only in Cartesian coordinate systems, with a second rank
mixed tensor without symmetry whose components are 0 and 1 in all coordinate
systems whatever.) In any coordinate system the components of this tensor are,
by definition,
(14.3)
With the aid of the tensor we form the differential invariant
(14.4)
The expression on the right, written out in fully in Cartesian coordinates, is
the square of the line element between adjacent points in the plane. Since by equation
(14.4) the tensor is so intimately associated with such a fundamental quantity as
the square of the differential of the line element, it is called the fundamental tensor .
Since by equation (14.2) it is symmetric in its indices in a Cartesian coordinate
system, it is symmetric in its indices in any coordinate system, i.e.,
(14.5)
The fundamental tensor defines lengths in an invariant way by means of equation
(14.4). This equation therefore expresses what is also known as a congr uence
r elation. That is to say, it sets up a standard independent of location or direction by
which lengths of vectors may be determined (and thereby compared). Since it defines
the standard of measure at each point and in every coordinate system, it is also called
the metr ic tensor . In rectilinear coordinate systems, where the contravariant
coordinates may also be taken as the components of a position vector, the equation
(14.6)
analogous to equation (14.4), amounts to a statement in a general rectilinear
coordinate system of the cosine law (see Ex. (11.2)), which reduces to the Pythagorean
Theorem in Cartesian coordinates. In non-rectilinear coordinate systems, we shall find
that only the coordinate differentials (rather than the coordinates themselves) are
contravariant vectors and only for these is there any analogue of the cosine law or
Pythagorean Theorem. Therefore equation (14.4) is entirely general whereas equation
(14.6) is not.
Had we formed the covariant differentials by taking perpendicular projections
of ds upon the coordinate axes rather than parallel projections, we could as well have
formed the invariant By evaluating and interpreting this invariant in a
Cartesian system, we could readily show that this too is Consequently we
have
This can be true for arbitrary only if
(14.7)
It is clear that for coordinate differentials, at any rate, the fundamental tensors
relate the covariant and contravariant forms of the infinitesimal displacement vector.
Equation (14.7) must therefore be the relation between the covariant and contravariant
forms of any vector. That is,
(14.8)
for any vector The fundamental nature of the tensor is thus further
demonstrated by its providing the direct link between the contravariant and covariant
forms of any vector
14. The Fundament al Tensor and t he Li ne El ement 47
It is obvious, however, that to complete the symmetry of the relation between the
contravariant and the covariant forms of any vector, we must also define a
contravariant tensor of the second rank, designated such that
(14.9)
Then from equation (14.8) we have that
(14.10)
Thus, if the contravariant components of a vector are known, its covariant
components are at once obtainable from equation (14.8). Conversely, if the covariant
components of the vector are known, its contravariant components are at once
obtainable from equation (14.10). The compatibility of the relations between covariant
and contravariant components is guaranteed by the equation (14.9). The tensor
defined by equation (14.9) is therefore called the contr avar iant fundamental tensor
to distinguish it from the covariant fundamental tensor The and are
mutual inverses. Either may be gotten from the other by equation (14.9).
From equations (14.8) and (14.10) together it is clear that
(14.11)
Either one of the forms of equation (14.11) therefore suffices to define the square of
the magnitude of any vector or
Looking at equation (14.8) from a purely mechanical standpoint, it is clear that
from a superscripted quantity we have generated an associated subscripted quantity
We may say, therefore, that by equation (14.8) we have lowered the index.
Indices may thus be raised or lowered at will with the aid of the covariant and
contravariant fundamental tensors. This holds true not only for vectors but for any or
all indices in tensors of any order. For example, equation (14.9) may be thought of as
the result of raising one index of Hence from equation (14.9) it will be seen that
the Kronecker delta is the mixed form of the fundamental tensor. Similarly,
One final result concerning the fundamental tensors should be noted here. If we
apply equation (14.9) in a Cartesian coordinate system, the equation becomes
whose solution is
Thus in a Cartesian system, the contravariant fundamental tensor is symmetric. It is
therefore symmetric in all coordinate systems.
Ex. (14.1) (a) Given Cartesian coordinates what are the components of
the fundamental tensor in a coordinate system where
?
(b) How do you account for the fact that the components of are the same
as those of ? (Hint: see Ex. (3.1a).)
Ans. (a)
(b) The transformation is a pure rotation.
Ex. (14.2) Given that the Cartesian coordinates are
what are the components of ? What do the
components become in the coordinate system of Ex. (14.1)?
Ans.
Ex. (14.3) If the fundamental tensor has components
in a rectilinear coordinate system what do the components become in the
coordinate system of Ex. (14.2)?
Ans.
Ex. (14.4) Compare equation (14.8) term-by-term with the results of Ex. (8.8).
Since the vector is wholly arbitrary, what do you conclude about the
meaning of the several components of ?
Ans.
Ex. (14.5) Compare equation (14.10) term-by-term with the results of Ex.
(8.8). Since the vector is wholly arbitrary , what do you conclude about the
interpretation of the several components of ?
Ans.
Ex. (14.6) Prove that
Ex. (14.7) Show by writing out in full, that if then
15. Uni t Vect or s and t he Basi s Syst em 49
Figure 18
15. Unit Vector s and the Basis System
C
onsider two adjacent points in the plane, such as and with
coordinates and as in Fig. 18. The differential vector has the
magnitude
Therefore it must be true that the vector
(15.1)
has unit magnitude, since
(15.2)
Any vector having unit magnitude is called a unit vector . Thus any vector is a
unit vector.
Consider the case when lies parallel to the axis for example, suppose
that and both lie on the axis. Then denoting such a vector as
Since this is a unit vector,
Similarly, a unit vector parallel to the axis can be shown to have components
(15.4)
Now since
it is apparent that the coordinate differential and the length are in the ratio
or * Evidently is the scale factor between coordinate
differential and invariant length differential along the axis. In the same fashion,
we may show that is the scale factor between coordinate differential and
invariant length along the axis.
The contravariant unit vectors
have a form which is at the same time so simple and so suggestive that it at once
prompts the query as to whether there are analogous covariant unit vectors and what
interpretation they have. Clearly, there is nothing to prevent our considering a vector
of the form
(15.5)
(We have here chosen the negative square root, for reasons given below.)
Analogously, we may define the unit vector
(15.6)
Note that since is not a perfect differential of a point function, we may not take partial differentials;
*
i.e., we may not set
Figure 19
To give a concrete meaning, we note that by Ex. (11.6)
In other words, is the unit normal to the axis and is the unit normal to
the axis. By reference to Fig. 19, it is clear that the first component of the
projection of on the axis, is negative; this is the reason for choosing the
negative square root in equation (15.5). Otherwise, would be the negative normal
to the axis.
Now consider the inner product of and We call that the inner product
of two vectors is the product of their magnitudes times the cosine of the angle
between them. Since these particular vectors are unit vectors, their respective
magnitudes are both 1, and since they are respectively parallel to the coordinate axes,
the angle between them is Therefore
where the middle term results from the actual substitution of equations (15.3) and
(15.4) into the left-hand expression, and where the right-hand term is the product of
the magnitudes times the cosine of the angle between the vectors. We see, then, that
in complete generality
(15.7)
Since we have already determined that we can also say that
(15.8)
Summing up, we say that
where and are the same scale factors as in equation (8.5). Moreover, the
result (15.8) may be derived directly (see Appendix 1.4). In general,
(15.9)
Ex. (15.1) Show that in general
(no summation over or ). Hence determine that
(15.10)
Now consider any vector with magnitude Clearly we may define
from it a unit vector
which has the direction of In other words, for any vector it is possible to write
it as
the product of its scalar (invariant) magnitude times a unit vector in the same direction
as the vector. Since is a unit vector,
Constructing a unit vector from any vector is called normalizing
Consider a Cartesian coordinate system in which are defined two vectors
(15.11) and
with associated covariant vectors
(15.12) and
Given any vector it is clear that
(15.13) and whence
(15.14)
We say that the and associated form a basis system. Clearly, any vector may
be expressed as a sum of scalar multiples of the basis vectors.*
Our particular illustration of basis vectors has been misleading in two respects; let
us hasten to forestall the possibility of misunderstanding. First, the vectors and
are unit vectors in a Cartesian coordinate system. This need not be so generally, where
unit vectors parallel to the axes will, by equations (15.3) and (15.4), have the form
(15.15)
(no summation over ); unit normals to the axes have the form
(15.16)
where the minus sign applies when the plus sign when Hence in general
the basis vectors have the form
(15.17)
and are not unit vectors.
Secondly, in a Cartesian system This is true in orthogonal
coordinate systems but not in general. Hence, though the basis vectors bear a close
relation to the unit vectors and the distinction between them must not be
overlooked.
Ex. (15.2) Write down the fundamental tensor for a coordinate system in
which
This statement, implied by equations (15.13) and (15.14), is sometimes interpreted to mean that the
*
numbers which we have called the components ought rather to be regarded as invariants, the invariant
projections of a directed line segment in a particular coordinate system. According to this interpretation,
it is only the basis vectors which undergo change when subjected to a transformation of coordinates.
Though such a point of view is logically permissible, it is of no utility for our purposes; we will adhere to
our understanding of the meaning of the components of a vector.
(Hint: use equation (15.9).)
Ans.
Ex. (15.3) How may one identify orthogonal coordinate systems by inspection
of the components of the fundamental tensor?
Ans. If
Ex. (15.4) Find the components of the contravariant fundamental tensor
corresponding to the coordinate systems of Ex. (15.2).
Ans.
Ex. (15.5) Show that
Use this result to check your answer to Ex. (15.4).
Now consider mor e gener ally two unit vector s and not necessar ily
in a Cartesian coordinate system. To insure that these vectors are distinct, we require that
where is the angle between them. Let us now formally represent an arbitrary vector
as
(15.18)
Are there values of and which make this equation true? If so, the and
constitute a basis system. To determine and , we first form the inner
products of with and Thus,
Solving for and and using the fact that and are unit vectors, we find
(15.19)
Clearly, when and are perpendicular, as in a Cartesian system, the results
(15.18) and (15.19) agree with those previously found. The present results are entirely
general, however, in a space of two dimensions. This is sometimes expressed by saying
that for arbitrary and the vectors span the two-space
of the coordinate plane.
It might be asked at this point how we can be assured that a representation such as
equation (15.18) is invariably possible. The answer is that we have assumed the number
of independent basis vectors is two. If for some value(s) of and
were true as well as the conditions on the the
vector would by definition exist in a higher number of dimensions than two. We will
not here concern ourselves further with this matter.*
Ex. (15.6) Show that Hence interpret for fixed
as the contravariant components of the unit normal to the non-j axis.
Ex. (15.7) Normalize the vectors Assume
Cartesian coordinates. ) Express the vector in terms of the
two preceding vectors.
Simply by lower ing indices in equation (15.18) we can der ive the anticipated
result that
(15.20)
the are given, as before, by equation (15.19) and the unit covariant vectors
are the covariant forms of the considered previously.
We have seen that there is some advantage in the use of basis vectors which are
orthogonal. It is a relatively simple matter to construct from a given pair of basis vectors
and an orthogonal pair. First, choose Then define
(15.21)
The point of these comments is irrelevant, in any case, when we adhere to the notation which has been
*
our practice, for by assuming that i has the range (1, 2) we have thereby specified that the number of
dimensions is two. The question of dimensionality is an important matter, however, particularly for the
representation of generalized vectors in function space. This then calls for tests of the completeness of any
basis system.
It is clear that
whence and are orthogonal. We then normalize by requiring that
The pair then forms an orthogonal basis system of unit vectors or an
or thonor mal system.
The expansion of a given vector in terms of other vectors suggests an inquiry as to
whether tensors in general may be expanded as products of the appropriate number and
kind of vectors. The answer is contained in a general Pr oduct Theor em that any
tensor may be r epr esented as the sum of pr oducts of vector s.
To prove that this is so, we need only provide a general construction by which such
a representation might be achieved in one particular coordinate system. Since the sum
of tensors is a tensor and since a product of vectors is a tensor, it then follows that the
sum of the products of vectors is a tensor. It suffices, therefore, if the constructed tensor
has the given components in the given coordinate system.
Suppose that the tensor is of order q, with components Let these
components be arranged dictionary-wise, i.e., arranged in order of the values of the first
contravariant index, then the succeeding contravariant indices, and finally the several
covariant indices. Now define vectors, enumerated dictionary-wise according to
(j, ... k, l, ...),
whose components are respectively, in the given coordinate system; here the
parentheses are meant to indicate that the indices are purely enumerative and do not
possess tensorial character. The components of the tensor have thus been
distributed over the components of the vectors.
The next step in the construction is to form the sum of products
where the various and are basis vectors. The several sums are the components
of the tensor whose components are To be convinced of this, consider one sum
at a time; thus
etc., until finally all parentheses are removed. One may also simply observe that the
contracted product with summed over j simply substitutes m for j. The end result is
again the same.
The theorem we have just proven is important not so much because it prescribes to
us how actually to resolve given tensors into sums of products of vectors, but because
it amounts to an existence theorem, guaranteeing that some such resolution is possible.
Clearly, it is in some respects a converse of the Quotient Theorem, which it
complements in a useful way. We will have occasion in what follows to appeal to this
Pr oduct Theor em.
Ex. (15.8) Form an orthogonal system from the Cartesian vectors
Ans.
Ex. (15.9) (a) Show that (b) Show that
where
Ex. (15.10) Show that if a tensor is the simple product of two
vectors, then it is necessary that the determinant (b) Can the
fundamental tensor be so represented? (c) How many vectors are required for
the representation in Ex. (15.9)?
Ans. (a) (b) Since by Ex. (14.4)
(none of nor can be
zero), the fundamental tensor cannot be represented as the simple product of
vectors. (c) Four.
Ex. (15.11) (a) Show that any vector may be written as where
is the tensor
and being unit basis vectors, the angle between them. (Hint:
combine equations (15.11) and (15.18).) (b) Show that Such a
tensor is called a projection operator. (c) Hence show that the Kronecker
delta is a projection operator, and that its resolution in terms of the basis
vectors is that given in (a).
Ex. (15.12) Show that in two dimensions any mixed tensor can be expressed as
for suitably chosen vectors
Ex. (15.13) (a) Given the transformation of Ex. (14.2) and the resultant
fundamental tensor, what are the scale factors along the new axes and the
angle between the axes? (b) What are the unit vectors along the new axes in both
the old and new coordinate systems? (c) Calculate the inner products of the unit
vectors along the and axes with the unit vector along the axis. Since
these are the cosines of and determine these angles. (d) The angle
between the axes of the coordinates is Use the result of
(a) to check your answer. (e) What is the inverse of the transformation of Ex.
(14.2)? (f) Use the coefficients together with the equation
from Appendix 1.3 to check your determination of
Ans.
16. Anti symmetri c Tensors and the General i zed Kronecker Del ta 59
16. Antisymmetr ic Tensor s and the Gener alized Kr onecker Delta
W
e have noted that a tensor of or der two or higher may be antisymmetr ic in
any pair of covariant indices or any pair of contravariant indices. This means, for
example, that is antisymmetric in and if
In two dimensions, a tensor of higher order may be antisymmetric in at most pairs of
subscripts or pairs of superscripts or both, since antisymmetry in more than two subscripts,
for example, would require every component to be equal to its own negative, hence zero.
Aside from sign, therefore, a completely antisymmetric second order tensor has but
one non-zero component, as The only other non-vanishing component is
Consider now the tensor which is antisymmetric in and as well as in and
. Let its non-zero components be (a) +1 if and are both distinct pairs and both even
or both odd permutations of 12, (b) 1 if and are both distinct pairs and one is even
and the other an odd permutation of 12, and (c) zero otherwise. This tensor is called the
gener alized Kr onecker delta, of rank four.
The generalized Kronecker delta may be simply expressed in terms of the usual
Kronecker delta, namely as
(16.1)
Such a decomposition is oftentimes useful in reducing inner products of tensors with
generalized Kronecker deltas. So also are such relations as
(16.2)
Since the components of have the same values in all coordinate systems, by
equation (16.1) so also do the This is an exceptional property.
Consider now a tensor of rank two whose components in a Cartesian coordinate
system are
(16.3)
Its components in some other coordinate system will be
This is clearly antisymmetric in and . Hence it is zero unless The only
non-vanishing component is numerically equal to
This may be given a more familiar and more useful form. To this end, we write
(16.4)
Therefore
(16.5)
In equation (16.5) we have chosen the positive square root on the assumption that
the coordinate system is right-handed. This means that unit vectors along the two
coordinate curves are in the same relation as thumb and forefinger of the right hand
when the pointer finger is parallel to the axis. More generally, it means that the
positive axis makes an angle less than with the positive axis. If the
coordinate system were left-handed, we should need to choose the negative square root
of , whence equation (16.5) would be replaced by
(16.6)
By a similar development, we may readily define a contravariant antisymmetric
tensor of the second rank which is given by
(16.7)
where in a Cartesian coordinate system. The relation between the and
the is very simple, namely
(16.8)
as may be verified most simply by considering the values of the separate components.
The generalized Kronecker delta is often useful for the purpose of forming or
isolating the antisymmetric part of any second rank tensor. Thus, if a tensor is the
sum of a symmetric part and an antisymmetric part we have that
But
16. Ant i symmet r i c Tensor s and t he Gener al i zed Kr onecker Del t a 61
Figure 20
and
since it is zero if
Therefore
(16.9)
A similar procedure is clearly possible for contravariant tensors
The usefulness of the antisymmetric tensor may be illustrated by giving it a
representative application. Thus, we form from and two arbitrary vectors
and the invariant
Because this is an invariant, its value and meaning in one coordinate system is the same
as in any other. We therefore choose as a most convenient coordinate system a set of
Cartesian axes, of which the axis lies along (see Fig. 20). This implies that
while Then, since ,
(16.10)
This is clearly the area of the parallelogram AOBC. The result is as true when and
are differential vectors along the coordinate axes; this result will be helpful in computing
areas by integration, especially in non-rectilinear coordinates.
Ex. (16.1) Show that the condition that two unit vectors and be
distinct is that either
Ex. (16.2) Derive vectorially the law of sines of plane trigonometry.
Ans. Using the figure and notation of Ex. (11.2), we have
Multiplying both sides of this equation by gives
This and the analogous equations which follow by multiplying by or
instead, are the statement of the law of sines.
Ex. (16.3) Show that is orthogonal to (Hint: take the inner
product of with
17. Isotropy
I
t may sometimes happen that a physical or geometric quantity is independent
of direction. When this is so, the quantity is said to be isotropic. Often, isotropy
characterizes invariant (or scalar) quantities, as a radially symmetric distribution, for
example. Isotropy may also be a property of tensor quantities, however. For instance,
we have seen that the Kronecker delta has components which are unchanged under
any reversible transformation. That is to say, in every admissible rectilinear system of
coordinates, the components of are the same in spite of the fact that ordinarily such
quantities as vectors and tensors change their values under a change of coordinate
system. In other words, the directions of the axes are implicitly reflected in the
components of the vectors and tensors when the components are given with reference
to those axes. Therefore, if a quantity does not change with a change of axes, it
must be independent of dir ection, or isotropic.
Can a vector be isotropic? If it were, its components in a new coordinate system
would be the same as its components in the old; i.e., But if it is a vector, it
must also satisfy the condition that
Hence
17. Isot ropy 63
These equations have no real non-zero solutions unless the determinant of
coefficients is
for every reversible rectilinear transformation. Hence we cannot attach a meaning to the
isotropy of a vector.
Consider, however, the tensor
(17.1)
where is some invariant function of the coordinates. In a new coordinate system
it has the components
It is therefore an isotropic tensor. Its contravariant and covariant forms are had by the
straightforward process of raising or lowering an index and are clearly of the form
(17.2a)
(17.2b)
We may infer that isotropic tensors are always of even order, for as we have seen,
any tensor may be represented as a sum of products of vectors. Hence the same sort of
argument by which isotropy was ruled out for vectors will likewise rule out isotropy for
all tensors of odd order.
On the other hand, we may at the same time conclude that all isotropic tensors, when
expressed as tensors which are covariant and contravariant to the same rank, will be
expressible as functions of the Kronecker deltas and the fundamental tensors, of which
the Kronecker deltas are the mixed form. For example, consider the isotropic fourth
order tensor symmetric in m, n and in r, s. We may give its general form
as
where and are two invariant functions. On the other hand, had been
antisymmetric in r and s, this would have to be modified to
etc. Note that it is also antisymmetric in m and n. Evidently, the symmetry or
antisymmetry must match that in the superscripts, and vice versa, for isotropic tensors.
Ex. (17.1) Determine the form of an isotropic tensor which is
antisymmetric in and as well as in and but symmetric to interchange
of with .
Ans. where is an invariant.
Ex. (17.2) Show that the tensor of Ex. (17.1) is the wholly covariant form
of the tensor
18. Eigenvector s
C
onsider the isotropic tensor Because it is isotropic, we have for any vector
Evidently an isotropic mixed tensor may be regarded as a linear operator which
transforms any vector into a multiple of itself. This is obviously not true of second order
mixed tensors in general. Is it true under any circumstances?
Let us examine the conditions under which
(18.1)
The only solution to such an equation which is not the solution will be one
for values of for which
or
The roots and of this deter minantal equation are called the eigenvalues
of the tensor The solution of equation (18.1) for these values of are, aside from
a scale factor, the eigenvector s of Since both and
are invariants, as is plain from the notation, the
eigenvalues and must also be invariants.
From the determinantal equation we see that the roots and will be real and
distinct provided
Expanding and simplifying, this becomes
which will always be the case provided and are of the same sign (or both are zero
and
There is one case of particular interest and importance when this condition is
automatically fulfilled. That is the case when or i. e., when
and are symmetric. In such an event, we may also show that the eigenvectors are
orthogonal, for
18. Ei genvect ors 65
Therefore
and since it follows that
whence the two eigenvectors are orthogonal.
Let us now compare and contrast an isotropic tensor with a general symmetric
tensor In their covariant or contravariant forms, both are symmetric. In their
mixed forms, they satisfy similar equations
and
respectively. However, the first of these equations is true for any vector the
secondonly for two orthogonal eigenvectors and determinate in direction
though not in magnitude. These various properties are invariant under reversible
transformations of the coordinates. The directions of the eigenvectors of are called
the invar iant dir ections. For an isotropic tensor, all directions are invariant
directions.
Ex. (18.1) Show that the determinantal equation may be written generally as
Ex. (18.2) (a) What are the eigenvalues and eigenvectors of the tensor whose
components in a Cartesian system are
?
(b) Are the eigenvectors orthogonal, and if not, why not?
Ans. (a)
where and are arbitrary. (b) No, because is not symmetric.
components in a rectilinear system are those of Ex. (18.2) but in which the
fundamental tensor is
(b) Are the eigenvectors orthogonal, and if so, why? (c) What is the angle
between the axes and what are the scale factors?
Ans. (a) Same as in Ex. (18.2a) . (b) Yes, because is symmetric.
(c)
components in a Cartesian system are
(b) How is this tensor related to that in Ex. (18.2)? (c) How are the
eigenvalues to those in Ex. (18.2)? (d) How are the eigenvectors related to
those in Ex. (18.2)?
Ans. where
and are arbitrary non-zero real numbers. (b)
(c) (d) Identically the same.
We must now consider one r esult of gr eat inter est concer ning the eigen-
vectors and invariant directions of a symmetric tensor We have seen that the
invariant directions of are orthogonal. Let us therefore adopt these invariant
directions as those of a set of orthogonal axes. In such a system, the eigenvectors are
evidently and Then, since
we have, on substituting the components of
with analogous results for the substitution of We therefore see that the
components of in this coordinate system are
We say that a transformation to these axes diagonalizes the tensor Its symmetric
covariant form now becomes
if we remember that in an orthogonal coordinate system when It
therefore differs from the fundamental tensor only in that is multiplied by
19. Deri vat i ves of Vect or s and Tensors i n Rect i l i near Coordi nat es 67
and by But where is the scale factor along the Hence,
at least when and are both positive, operates as would if we were to
multiply the scale factor by the scale factor by
Let us now consider all position vectors for which
where is a constant. In the orthogonal coordinate system based on the invariant
directions, this evidently reduces to
Then, if
where
It is therefore the equation of an ellipse with the center at the origin and axes and
if and are of the same sign. If is opposite in sign to and
, we can then set and have
which is the equation of a hyperbola with axes and If the sign of is
opposite the sign of and then no position vector satisfies the equation.
If then the locus becomes a pair of straight lines, provided and
are of opposite signs; otherwise the locus degenerates to the single point at the origin.
19. Der ivatives of Vectors and Tensor s in Rectilinear Coor dinates
W
e have seen that the gener al vector and tensor transfor mation laws per mit
non-homogeneous linear reversible transformations and that this implies an
arbitrariness of the origin of coordinates, hence of position of the vector in the plane.
There is therefore no difficulty in comparing vectors at different points, for they may be
treated as though they were at a common point; this is a most significant property, for
it is in general not true in non-rectilinear coordinates. In rectilinear coordinates,
however, we are thereby guaranteed that we may form differences of neighboring
vectors, hence derivatives.
Thus, consider a plane curve, , as in Fig. 21. On it, choose two points and
corresponding to invariant curve parameter values and where
is small. If a vector is defined such that each component is a function of , we then
define the derivative of as the vector whose components are
Figure 21
Actually, there is somewhat more to this definition than meets the eye, for the
coordinates along the curve may also be functions of the parameter .
Therefore, in greater generality, we must allow for a functional dependence of the vector
components in the form
Then
For a vector rooted to a fixed point, like a weather vane, the second term on the right
drops out. On the other hand, if the vector changed only with displacement along the
curve C, the first term on the right side would vanish. Thus, the hands of a clock on the
wall define vectors of the first sort, whereas the velocity of a bird in flight is a vector of
the second sort; the hands on a clock in a plane in flight would illustrate the completely
general case. In each instance, however, the left hand side is a vector, the derivative of
a vector with respect to a scalar.
Ex. (19.1) (a) In a Cartesian coordinate system with origin at the center of a
clock face, positive axis in the 12 oclock direction and positive axis
in the 3 oclock direction, what are the components at time of the position
vector representing a minute hand of length ? Take time to be the
moment when the hand is in the 12 oclock position. Measure time in minutes.
(b) What is the velocity of the tip of an hour hand of length ?
Ans.

Ex. (19.2) If a ball is projected horizontally with velocity v and allowed to fall
freely under the action of gravity, what are the components of the position
vector from its starting point? Take the axis to be vertical, the axis
horizontal, the starting time to be , the acceleration of gravity to be ,
the origin at the point of projection. (b) What is the balls velocity vector?
Ans.

Ex. (19.3) (a) Suppose that instead of a ball as in Ex. (19.2), a clock is
projected horizontally, etc. What are the components of the position vector of
the tip of the minute hand? (b) What is the velocity vector of the minute hand?
Ans.
(19.4) (a) Suppose that the vector depends upon position and time
according to the equations
What is the time derivative of the vector ? (b) What is the time derivative
of along the curve
Ans.
Let us consider , then, the case when Now is a
differential vector since by definition
hence
Therefore is a non-differential vector. But the curve is entirely arbitrary, so
that in turn the vector may be chosen at our pleasure. We now invoke the Quotient
Theorem: since is known to be a vector and since is an arbitrary vector, the
quantity is a mixed tensor of the second rank.
Figure 22
The same may be seen to be true of covariant vectors. Therefore, in rectilinear
coordinates the partial derivatives of a contravariant vector constitute a tensor of second
rank. The same may be seen to be true of covariant vectors. Moreover, since we have
shown that any tensor may be resolved into the sum of products of vectors, we can say
further that in a rectilinear system the partial derivatives of any tensor form a tensor of
one greater covariant rank. (Beware of incautiously extrapolating these results to
non-rectilinear coordinate systems, where they are not true!)
A curve such as C (Fig. 22) is a one-dimensional manifold of points. When,
therefore, a vector is defined at every point along a length of arc of a curve, the set of
vectors so defined is called a vector field with par ameter . The vectors themselves
are two-dimensional though the field is a one-parameter field. However, we may also
elect to define each component of the vector as a function of position, as
The vectors so defined constitute a vector field over the entire region of definition of
the functions Henceforth, we tacitly assume that any vector or tensor is,
more accurately, a vector or tensor field with respect to the coordinate variables or
some relevant parameters.
Let us now illustrate the differentiation of vectors and tensors with a few simple
applications. Consider the curve C. Let arc length from some definite point be
chosen as the curve parameter. Then the differential of arc length is, as we have seen,
(19.1)
Now consider the vector , where is a point on the curve and
is a neighboring point on the curve. It is a unit vector since
(19.2)
Its direction must be tangent to the curve C, for the vector is the difference of the
coordinates of two neighboring points on the curve, hence is directed along the arc
. Therefore is the unit tangent to the curve C.
If we differentiate the equation (19.2), we get
remembering that the are constants in a rectilinear coordinate system. Evidently the
vector is orthogonal to It is therefore called thenor mal to the curve.
Though is a unit vector, of course need not be. In fact, its magnitude would
be calculated, as for any vector, as
(19.3)
We then define the unit nor mal as
(19.4)
To interpret consider the case when the curve C is a circle of radius about the
origin. Then in Cartesian coordinates
where is the arc length measured from the axis. Hence
Hence is a unit tangent, therefore perpendicular to the radius vector; is a unit
normal, perpendicular to the unit tangent and directed inward along the radius of the
circle. The quantity is called the cur vatur e of the curve at the given point
and its reciprocal is ther adius of cur vatur e of the curve at the given point.
Though a circle is obviously a very special case, it shows how we may interpret
our results in general. At any point of a continuous, sufficiently differentiable curve,
the vector is a unit tangent. The vector is the unit normal, directed inward along
a radius of the osculating cir cle to the curve at the given point. The radius of the
osculating circle is Though we have found it convenient to derive these
results in a Cartesian coordinate system, tangency and orthogonality are invariant
relations and is an invariant quantity. Therefore our results are true in all rectilinear
coordinate systems.
Ex. (19.5) Given the curve
in a Cartesian coordinate system, find the unit tangent and unit normal.
Ans. Since
we have
One kind of cur ve of special note is, of cour se, the str aight line. If we define
it as a curve of zero curvature, we see at once that its differential equations are, from
equation (19.4),
Appendi x 1.1 Transf or mat i on of Obl i que Rect i l i near Cont ravari ant Coordi nat es 73
Figure 23
On the other hand, we may also define it as a curve whose tangent is always parallel to
itself. In rectilinear coordinates this means that
where and are the coordinates of any two points and on the line. In other
words, is a constant vector, whence
the same equation as was found previously.
Appendix 1.1 Tr ansfor mation of Oblique Rectilinear Contr avar iant
Coor dinates
In Fig. 23 a point has coordinates and in a system whose axes
are and In a new system whose axes are and the coordinate
of is . Now drop the perpendicular onto . Then
Next, draw through the line perpendicular to and
therefore parallel to . In the triangle the angle is and the side
Hence
Further, in the triangle the angle is equal to and the side
so that
Figure 24
Now is a rectangle by construction. Therefore
or
(4.4)
(4.5)
where
Appendix 1.2 - Tr ansfor mation of Oblique Rectilinear Covar iant Coor dinates
In Fig. 24 the point P has coor dinates and r elative to the
original axes and through the origin . In a new coordinate system in
is a new axis, however, the coordinate of is . Now which
Appendix 1.3 - Interpretation of an Arbitrary Transformation in Rectilinear Coordinates 75
Figure 25
from which it follows that But
.
and
Hence
Since and since
we have finally that
(5.5)
Appendix 1.3 - Inter pr etation of an Ar bitr ar y Tr ansfor mation in Rectilinear
Coor dinates
Assume in the fir st instance that the change of scale follows the change of axes.
Now a point (in Fig. 25) at unit distance on the axis will have coordinates
Therefore the coordinates of this same point must satisfy the relations
Figure 26
of which the solution is
But from Fig. 25 it is apparent that
Therefore
Since the angle between the original axes may be assumed known, only the coefficients
and need be known to determine the angle between the original axis
and the new axis. Clearly, the ratio is independent of subsequent scale
changes if scale changes follow axis transformation.
In a similar fashion, we may determine that the point on the axis whose
coordinates are will have coordinates
Appendix 1.3 - Interpretation of an Arbitrary Transformation in Rectilinear Coordinates 77
Figure 27
Hence from Fig. 26 it is apparent that
and
Therefore
(8.9)
Again it is clear that this result is independent of subsequent changes of scale, for the
ratio will be unaffected by them.
The remaining steps are straightforward. From the relations
we may calculate the from equations (4.3) and (4.6). Then
(8.4) and
Let us now consider how a covariant transformation may be interpreted. Given the
covariant transformation let us apply it to the point P (Fig. 27) whose
covariant coordinates are Then the coordinates relative to new axes
must be
Figure 28
Now from Fig. 27 we can also see that when the scale factors are unity,
Substituting the values of and we see that
(A1.4.1)
Similarly, we see from Fig. 28 that the point whose coordinates are
must have new coordinates while at the same
time
Therefore we must have that
(A1.3.2)
Appendi x 1.4 - Int erpret at i on of t he Component s of t he Fundament al Tensor 79
It will be more useful if we put the second of equations (A1.3.1) into the form
Substituting from equation (A1.3.2) the value of gives
Therefore
In the same fashion, since from equation (A1.3.1)
we can show that
whence
It is clear from these equations, together with the results of Ex. (7.2a) that the change
of axes is not affected by the changes of scale if the scale changes follow the change of
axes. The scale changes may then be found from equation (7.6) as
(A1.3.3)
where the are given by equations (5.4) and (5.5).
Appendix 1.4 - Inter pr etation of the Components of the Fundamental Tensor
W
e wish to find the values of the fundamental tensor components after
transformation from a Cartesian coordinate system with axes and to an
oblique coordinate system with axes and (see Fig. 29). For this purpose we
use the tensor transformation
.
The quantities are given, by equations (5.4), (5.5), and (7.8) as
Figure 29
Since we find then for the expression
whence Similarly,
And finally,
as was to be shown.
Not es Chapt er 1 81
Notes Chapter 1
U
pon this chapter mor e than any of the r est falls the bur den of defining and
applying fundamental terms, concepts and usages. As so often happens, an
appreciation of their finer points usually cannot be achieved except in that retrospect
which is possible only after having extensively penetrated the subject in an
unsophisticated or brute-force fashion. It seems desirable, however, to call attention to
some of these matters in passing, in the hope that the anticipation of them may obviate
difficulties for the especially acute reader or illuminate the minutiae for the unwary.
1.21.5. For example, the terms invariant, covariant and contravariant are
introduced in Sections 1.2-1.5. The first two of these terms, at least, have shades of
meaning which must be judged from the context of their usage. Thus, invariant literally
means unchanging. It implies that whatever is invariant remains fixed despite changes
or differences in kindred or related quantities. In this sense, a plane vector is invariant
to a change of origin, for neither its length nor its direction will have been altered
thereby. In a somewhat more sophisticated sense, we may even say that the vector is
invariant to a rotation of axes, for the rotation induces no change in the length or
direction of the vector; it is this kind of invariance which is the key to the power of
vector equations.
Beware of understanding invariant thusly, however. Most often, the term invariant
is intended to be taken as an abbreviation for invariant function, a single-valued point
function. Such might be the density distribution of matter, the temperature distribution
within a star, or the length of a particular line segment. Plainly, no mere change of
coordinates will affect these quantities.
In practice, however, there is seldom occasion to decry this ambiguity, for the
invariance of a vector to change of axes may be as well expressed in the covariance of
its components. Literally, covariance means changing with. As applied to vectors, it
expresses the fact that a vectors components are so altered by a transformation of
coordinates that the length and direction of the vector remain the same. Thus,
covariant may refer figuratively to any vector-like change of components which is
induced by a transformation of coordinates, or it may refer to the specific type of
transformation considered in 1.5. Which sense is intended must be inferred from the
context of its usage.
The term contravariant (varying oppositely) is intended to contrast to covariant in
its latter meaning, and is therefore used to identify a specific mode of transformation
of components under a change of coordinates. It is only superficially paradoxical,
therefore, to note that it would be proper to say that a contravariant vector is covariant
in the more general sense though not in the specific sense.
Let it be noted again that the invariance of a vector is embodied in the covariance
of its components because the latter refers implicitly to a particular set of coordinate
axes. The role of the coordinate axes in injecting the distinction between a vector and
its components seems like an intrusion only because it has so often been allowed to be
a tacit role. To compensate, we have labored the point somewhat.
1.2 The index notation is an integral part of tensor analysis. Much of the mystery of
tensor analysis resides in this very simple device. The beginner would be well advised
to practice translating equations in index notation into their explicit expanded forms
until he is thoroughly at ease with the compact notation and can think in terms of it.
In doing this, it would be advantageous to include equations which also employ the
summation convention, introduced in 1.4; such are Exercises 4.13 and 4.14. At the risk
of seeming tedious, it can be said that facility with the index notation is a virtual sine
Figure 30
qua non for mastery of tensor analysis. Fortunately, such facility can be come by with
very little effort. The rewards are disproportionate, for the indices will be found to
compress a great deal of information that can be read at a glance.
The index notation may be contrasted with the index-free notation, which depends
upon the use of bold face or other type fonts to designate vector or tensor quantities. The
commonly used Gibbs notation for vectors is an example; thus a plane vector might be
written as
where is the vector, is the length of its projection upon the axis, is the
length of its projection upon the axis, is a unit vector along the axis, and is
a unit vector along the axis. If we were to limit our consideration to vectors in
two-dimensional Cartesian coordinate systems, there might be little to choose between
the notations. However, in non-Cartesian systems, in tensors of rank higher than one,
and in more than two dimensions, the index notation greatly facilitates the expression
of symmetries and antisymmetries and renders almost trivially transparent the invariance
of inner and triple products, the orthogonality of cross products, etc.; the index-free
notation posits these relations in what can only appear a pragmatic but cabalistic manner.
For other comments upon index-free notation, consult Laugwitz (8), p. 94.
1.4 The summation convention, closely related to the index notation, is another
notational device which for reasons inscrutable seems to present a difficulty to many
beginners. Perhaps its very simplicity is a clue to the difficulty, for the tyro is prone to
seek a deep significance which is not there; he is misled because he cannot credit the
notion that such a luminously simple convention could by itself be as supremely
economical and functional as experience will prove it to be. It might be well, therefore,
for him to juxtapose equations (4.13), (4.14) and (4.15) thusly
and contemplate them at such length as will secure forevermore his automatic and
unhesitating translation of repeated subscript and superscript into implied summation.
Let these equations be read both forward and backward. Let the same be done for
Exercises 1.4.13 and 1.4.14.
Not es Chapt er 1 83
1.4, 1.5 For an optional discussion of the geometrical meaning of covariance and
contravariance, consult Kreyszig (5), pp. 101-105, after reading Section 14 of this
chapter.
An interesting alternative to the method of defining rectilinear coordinates as
intercepts upon the axes is given in Wells (21), p. 12. The coordinates are instead taken
as the distances from the axes. For example, consider Fig. 30. It is evident from the
figure that
It follows, then, that
Clearly, in Cartesian coordinates, with the coordinates are identical with
the usual contravariant coordinates but not otherwise.
1.13 (a) An antisymmetric second order tensor is an example of a quantity whose
indices have a significant order. Later, as in 3.5, for example, we will find it convenient
to consider closely related quantities of the form or whose index order is
conspicuously preserved by the insertion of dots where necessary. (b) Any covariant or
contravariant tensor of rank two may be resolved uniquely into a sum of a symmetric
and an antisymmetric part. As much cannot be said for tensors of higher rank. We make
use of this characteristic resolution in 16. In any case, the generalized Kronecker delta
may be employed to isolate the completely antisymmetric portion of any set of indexed
quantities.
1.15 (a) It is shown in Adler, Bazin and Schiffer (1), pp. 25-28, that in general
is the minimum number of products of vectors into the sums of products of which a
tensor of rank may be decomposed in dimensions. Since each vector has
components, this provides a total of components. Thus in two dimensions
a second rank tensor requires products of two vectors,
such as for example. However, in three dimensions a
second rank tensor requires 3 products, as etc. (b)
Projection operators arise not only in ordinary geometry but also in the geometry of
function space. It can easily be shown that if and are projection operators, then
(i) is a projection operator if , i.e., if and commute; (ii) is
a projection operator if , i.e., if and anticommute; (iii) is
a projection operator if , as in (ii), or if is the identity operator . The
latter case is the one which arises in 4.10; the operator is called the
complement of .
1.17 It would be well to note two points respecting the definition of isotropy here
given. First, the condition of isotropy may be achieved at an isolated point. However,
unless the condition is satisfied throughout some connected region, it is of very little
significance or interest either physically or mathematically. We are therefore really
discussing isotropic tensor fields (see 1.19). Second, we exclude vectors whose
components are not real. Isotropic vectors with complex components are defined in
spinor theor y.

Tensor Analysis-Chapter 1

Uploaded by

Copyright:

Available Formats

Tensor Analysis-Chapter 1

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Tensor Analysis-Chapter 1

Uploaded by

Copyright:

Available Formats

Chapter 1

Tensor s in Rectilinear Coor dinates in Two

You might also like