Notes on General Relativity

Maurı́cio O. Calvão 1
Institute of Physics
Federal University of Rio de Janeiro

August 13, 2008

1 Email : orca@if.ufrj.br.
2
Contents

Introduction 1

I Geometrical foundations 3
1 Tensor algebra 5
1.1 Vector (linear) spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.2 Dual spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
1.3 Tensor product spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
1.4 Relative tensors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
1.5 Additional operations and results . . . . . . . . . . . . . . . . . . . . . . . . . . 20
1.5.1 Contraction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
1.5.2 Symmetrization and anti-symmetrization . . . . . . . . . . . . . . . . . . 21
1.5.3 Quotient rules . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22

2 Manifolds 25
2.1 Differentiable manifolds . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
2.2 The tangent space and tensor fields . . . . . . . . . . . . . . . . . . . . . . . . . 26
2.3 Differentiation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
2.3.1 Derivation in general . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
2.4 Symmetries and Lie derivative . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30

i
ii CONTENTS
List of Figures

2.1 Lie derivative. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30

iii
iv LIST OF FIGURES
List of Tables

v
vi LIST OF TABLES
Introduction

General relativity, that is, Einstein’s gravitational theory, is in a particularly blossoming epoch.
This is due mainly to new experiments and observations, either already carried out, or taking
place right now or being planned.
binary pulsar
LIGO, GEO, Gravity Probe B
LISA, STEP, GAIA, LAGEOS, etc
Global positioning system
A relatividade geral, ou seja, a teoria de gravitação de Einstein, está passando por uma época
particularmente efervescente. Isso deve-se, principalmente, a novos experimentos e observações,
quer já realizados, quer em curso, quer em planejamento. Valem citar (i) as observações do pul-
sar binário PSR1913+16, que conferiram o prêmio Nobel a Hulse e Taylor, em 1993, com a usual
interpretação de detecção, mesmo indireta, de ondas gravitacionais; (ii) a construção, em curso,
de uma série de detetores de ondas gravitacionais de nova geração [LIGO (Laser Interferometer
Gravitational wave Observatory), Schömberg,. . . ] ; (iii) o planejamento de detetores assom-
brosamente ambiciosos, como o LISA (Laser Interferometer Space Antenna), um interferômetro
espacial, com braço de 5 milhões de quilômetros (≃ 3,3 UA); (iv) planejamento de experimen-
tos a serem realizados em satélites para teste da universalidade da queda livre ou princı́pio de
equivalência fraco [STEP (Satellite Teste of Equivalence Principle)], e da precessão geodésica ou
arrasto do referencial inercial local ou efeito de Sitter ou acoplamento spin-gravitoeletricidade,
assim como do efeito Lense-Thirring ou acoplamento spin-gravitomagnetismo [Gravity Probe B,
LAGEOS (Laser Geodynamic Sattelite Experiment)].
These notes are based on a graduate level course of general relativity, which I delivered on
several occasion in the Institue of Physics of the Federal University of Rio de Janeiro, during
the period 1991-2008. Our intention is to present a reasonably honest and deep exposition,
in an intermediate level (= Weinberg; Anderson; Carroll; Hobson; > Rindler; Shutz; foster &
Nightingale; Hughston & Tod; d’Inverno; Hartle; < Wald; Straumann; Sachs & Wu; de Felice &
Clarke). Honest means we shall try to make clear the validity and certainty of the arguments
and results, at least from our point of view; deep, in the intermediate level means that we will
not take profit of mathematically or physically advanced arguments (algebraic geometry, fiber
bundles; quantum gravity) but we sure will use the potential of the so-called “classical tensor
calculus” in all its richness.
The presentation is divided into four parts, besides this Introduction. In Part I, “Geometrical
foundations”, we deal first with tensor algebra, and then with tensor analysis (manifolds and
their geometries). In Part II, “Physical foundations”, we deal with the very important notions of
instantaneous observers, observers, and frames of reference, then with some “non-gravitational”
physics, particularly, continuum physics (dynamical and thermal aspects) and electromagnetism

1
2 LIST OF TABLES

in arbitrary curved spacetimes and frames of reference. In Part III, “Einstein’s general relativ-
ity”, we start with a very critical analysis of the historico-heuristical basis of general relativity,
explicitly the general covariance principle, the general relativity principle, the equivalence prin-
ciole(s), Mach’s principle and the minimal coupling principle. We present then Einstein’s field
equations and its possibly most famous solution, the Schwarzschild one, together with the so-
called classical tests: gravitational spectral shift, light deflection, perihelion precession, and radar
echo delay. In Part IV, the last one, entitled “Further developments”, we study gravitational
collapse, black holes and the linear perturbations around Minkowski spacetime, which allow a
simple study of gravitational waves.
 Part I

Geometrical foundations

3
Chapter 1

Tensor algebra

1.1 Vector (linear) spaces

Of course, the precise and plain concept of vector is the one provided by linear algebra; in
particular, it is obvious that, rigorously, a vector: (i) is neither a “quantity characterized by a
magnitude, and a direction”, (ii) nor a “numeric (real, complex, rational) triple (or n-tuple)”.
In fact, in a plain (“raw”) vector space, there does not exist the notion of magnitude (norm)
for a vector, and the numerical triple (n-tuple) merely represents the abstract vector (by means
of its components) in a given basis (taken for granted from the context)1 . Granted this, we
should hasten to immediately recognize the very fundamental, and possibly most important,
from the heuristic point of view, role of the primitive association of vectors with displacements
in the physical three-dimensional space; by the way, in general, new concepts and theories arise
from practice, enshrouded in a series of superfluous superstructures, which are only evidenced
by future logical development and purging.
In the specific case of displacements, it urges to model the physical space as an affine space
in the mathematical sense[1, 2, 6]. Here, we would just like to remind that only within this
structure can we assign some sense to the notion that the physical space, at least in the context
of Euclidean geometry, does not have a privileged origin; the most primitive concept is that of
point, wherefrom the concept of displacement is built and, finally, as a consequence, the concept
of vector2 . Phrasing it another way, in a plain vector space, there is no notion of point at all.
A vector (or linear) space T (over K) is a triple {T, +, ·, K}, where T is a (non-empty) set of
elements called vectors, +, · are two composition laws:
+ : T × T → T (addition)
· : K × T → T (multiplication by a scalar)
and K := (K, ⊕, ¯) is a field, whose elements of the base set, K, are called, in this context,
scalars (for instance, the real, or complex numbers). Furthermore, such compositions must
satisfy, for any u, v, w ∈ T and a, b ∈ K, the following other axioms3 :
1
In a gross analogy, I, Maurı́cio Ortiz Calvão, am “represented”, in Brazil, by a given identity card with a
certain general registration number, whereas, in the United States, I have another card with a distinct number;
am I, ipso facto, two distinct persons?
2
It pays to ponder upon three concepts of vector traditionally introduced in the elementary physics literature:
free, sliding and bound; think about the notion of equipolence in Euclidean geometry[10].
3
Unfortunately, by a notational abuse, it is usual to denote both operations · and ¯ simply by juxtaposition,

5
6 CHAPTER 1. TENSOR ALGEBRA

1. v + w = w + v (commutativity of addition)

2. u + (v + w) = (u + v) + w (associativity of addition)

3. ∃ 0 | v + 0 = v (existence of a neutral element for addition)

4. ∃ −v ∈ T | v + (−v) = 0, ∀ v (existence of inverse elements for addition)

5. a · (v + w) = a · v + a · w

6. (a ⊕ b) · v = a · v + b · v

7. a · (b · v) = (a ¯ b) · v

8. 1·v = v.

Axioms (1) to (4) render (T, +) an Abelian (or commutative) group. Axioms (2) and (7), with
the notational abuse mentioned in footnote 3, allow the elimination of parentheses in certain
expressions; e.g., u+v+w=(u+v)+w and abv = (ab)v.

Exercise 1.1 Prove the following immediate consequences of the axioms:

1. the neutral element, 0, for addition is unique.

2. for all v ∈ T , 0v = 0.

3. the inverse elements for addition are unique.

4. if a ∈ K, v ∈ T , and av = 0, then either a = 0 or v = 0.

d times
z }| {
d
Exercise 1.2 Let T := R := R × · · · × R, where R is the set of real numbers. Define

(u1 , . . . , ud ) + (v1 , . . . , vd ) = (u1 + v1 , . . . , ud + vd ),

a · (v1 , . . . , vd ) = (av1 , . . . , avd ), ∀a ∈ R.

Prove that (Rd , +, ·) is, then, a vector space (over the real field).

Exercise 1.3 Let R+ be the set of positive real numbers. Define the “sum” of two elementes of
R+ as the product in the usual sense (p + q := pq), and multiplication by a scalar of R as

· : R × R+
(r, p) 7→ r · p := pr .

With such operations, show that (R+ , +, ·) is a vector space over R.

and the operations + and ⊕ by the same symbol +. You have been warned!
1.1. VECTOR (LINEAR) SPACES 7

If U and V are two vector spaces over the same field of scalars, then we build a new vector
space, called the direct sum of U and V and denoted by U + V, in the following manner: the new
set of vectors is U × V , the new addition and multiplication by a scalar are defined by (with an
evident notational abuse)

(u, v) + (u′ , v′ ) := (u + u′ , v + v′ )
a · (u, v) := (a · u, a · v).

Exercise 1.4 Show that, in Exercise 1.2 above, (Rd , +, ·) is the direct sum of R with itself.

Following [4], it is simplest to start with an example of matrix equation:

u = Av.

Here v is a column matrix (“vector”), of order, say, N × 1; A is a matrix of order M × N ;

and u is thus a column matrix (“vector”), of order M × 1. This matrix equation tells us that
each individual element of u is determined from the individual elements of v via A. To write
down explicitly such an expression, we introduce a notation for the elements (“components”) of
u and v, as well as for the elements of A: let us agree that v a stands for the a-th element of v
(a = 1, 2, . . . , N ), uα the α-th element of u (α = 1, 2, . . . , M ), and Aα a the element of the α-th
row and a-th column of A. The matrix equation above is, thus, equivalent to the M equations
N
X
α
u = Aα a v a .
a=1

The range convention arises from the observation that it is not necessary to spell out, in each
occurrence of such a set of equations, that M equations are involved and that the validity of each
one of them is stated. This can be perceived from the presence of the index α in each member
of the equation: for α is a free index, as opposed to a, which is subject to the summation sign.
On the other hand, the summation convention follows from the observation that, whenever a
summation occurs in such an expression, it is a summation over an index (a in that case) which
occurs precisely twice in the expression to be summed. Thus, a summation occurs when there
is a repeated index; and when an index is repeated,PN a summation is almost always implied.
Under these circumnstances, the summation sign a=1 does not play any useful role, since the
summation may be recognized by the repetition of an index; the sign can, therefore, be omitted.
Thus the “component” or “element” equation above is written, when the domain and sum-
mation conventions hold, in the simple form

uα = Aα a v a .

The presence of the repeated index a in the right hand side implies a summation over its allowed
domain of values (1, 2, . . . , N ) by virute of the summation convention; whereas the presence of
a free index α, in both sides of the equation, implies an equality for each value 1, 2, . . . , M that
it may assume, by virtue of the domain convention.
In general, the range and summation conventions work in the following way. If, in an equation
involving indexed quantities, there exist free (non-repeated) indices, then the equation holds for
8 CHAPTER 1. TENSOR ALGEBRA

all values in the domains of all free indices, such domains having been declared before: that is the
range convention. In an expression involving indexed quantities, wherever an index is repeated,
a summation over all possible values of the doamin of that index is implied, the domain again
having been declared previously: that is the summation convention.
The way these conventions work is relatively straightforward. One or two rules–often em-
ployed for an interactive verification of the correction of a calculation–should be mentioned. The
number of free indices in the two members of an equation must be the same; and, naturally,
each different free index in an expression must be represented by a different letter (symbol). Re-
peated indices in an expression may only occur in pairs. The replacement of a letter representing
an index by another letter is allowed, provided all occurrences of the letter are altered at the
same time and in the same way, and provided it is understood that the new letter has the same
range of values as the one it replaces. The most convenient pratice to adopt, when indices with
different ranges are involved in the same calculation (expression), is to reserve a small section of
a particular alphabet to stand for the indices of a given range. Thereby, in the case discussed
above, we might take a, b, c to range and sum from 1 to N , and α, β, γ to range and sum from
1 to M ; then uβ = Aβ c v c would mean exactly the same as uα = Aα a v a .
Two points should be emphasized regarding the way those conventions are used in these
notes. First, we arrange things in such a way that the pair of repeated indices implying a
summation will occur (almost always) with an index in the upper position and the other in
the lower position. This is already apparent from the way we chose to write down the matrix
equation above, when something like uα = Aαa va might be expected. This point is related to
the importance of distinguishing between a vector space and its dual (column vectors versus row
vectors), which will be dealt with later on.
The second point worth being mentioned is that an expression such as (xc ) is often used to
represent (x1 , x2 , . . . , xN ). Additionally, the value of a function, say f , of N variables, say xc ,
will be denoted by f (xc ). In this situation, the index c is subject neither to the summation nor
the the range conventions. In this particular context, (xc ) should be generally thought of as the
set of coordinates of a “point” in a given “space”.
Let T be a vector space. A finite set of vectors, say {v1 , . . . , vr }, is said to be linearly
dependent if there are scalars a1 , . . . , ar , not all of them zero, such that P ai vi = 0 (here, it is
understood, from the range and summation conventions, that we mean ri=1 ai vi = 0.). An
infinite set is linearly dependent if some finite subset is linearly dependent. A set of vectors is
linearly independent if it is not linearly dependent.
A (finite) sum of the form ai vi , where vi ∈ T and ai ∈ K, is called a linear combination of
v1 , . . . , vr .
As a simple consequence, we notice that two vectors are linearly dependent if one of them
is a multiple of the other; we cannot state that each one is a multiple of the other, since one of
them might be 0. If a set S includes 0, then it is linearly dependent whatever the other elements
(if any).
Exercise 1.5 Prove these two statements.
The maximum number of linearly independent vectors in a vector space T is called the
dimension of T and is denoted by dim T. Naturally, there might not be a finite maximum,
in which case we write dim T = ∞; this means that, for every positive n, there is a linearly
independent set of T having n elements.
1.1. VECTOR (LINEAR) SPACES 9

A basis of T is a linearly independent subset S of T such that every vector is a linear

combination of elements of S.4

Exercise 1.6 Prove that, if S is a basis, then a linear combination that expresses v ∈ T in
terms of elements of S is unique, except for the order of the summands.

If S is a basis of T, then, for each v ∈ T , the unique scalars that occur as coefficients in
the linear combination of elements of S that expresses v are called the components of v in (or
with respect to) the basis S. We consider that a component of v is assigned to each element of
S; however only a finite number of components is nonvanishing.5

Exercise 1.7 Prove that all bases have the same number of elements, the dimension of T.

Let {e1 , . . . , eN } (or, simply, {eα }) be a basis of an N -dimensional vector space T, such that
any v ∈ T may be written as v = v α eα , for convenient scalars v α . The scalars v α are the
components of v with respect to the basis {eα }.
We now want to see how the components of a vector transform when a new basis is introduced.
′
Let {eα′ } be the new basis of T, and v α the components of v with respect to this new basis.
Then
′
v = v α eα′ . (1.1)
The new basis vectors can, as any vectors, be expressed as a linear combination of the old ones:

eα′ = Xαβ′ eβ , (1.2)

and, inversely, the old ones as a linear combination of the new ones:
′
eγ = Xγα eα′ . (1.3)

(Though we are using the same kernel letter X, the N 2 numbers Xαβ′ are different of the N 2
′
numbers Xβα , the position of the primes indicating the difference.) Replacing, now, eα′ , from
(1.2), in 1.3, there comes
eγ = Xγα Xαβ′ eβ .
′
(1.4)
Due to the linear independence of {eβ }, we have then

Xγα Xαβ′ = δγβ ,

′
(1.5)

where δγβ is the Kronecker delta, defined by

½
0, se β 6= γ,
δγβ := (1.6)
1, se β = γ.
4
We mention, without proof, that a basis always exists. This is obvious if dim T is finite, but, otherwise,
demands transfinite induction, or the axiom of choice, or Zorn’s lemma.
5
In plain vector spaces, only linear combinations of a finite number of terms are defined, since no meaning can
be attributed to limits or convergence. Vector spaces where a notion of limit is defined and that satisfy certain
additional (pun intended) relations are called topological vector spaces. When such an additional structure derives
from a definite positive inner product, the space is called a Hilbert space.
10 CHAPTER 1. TENSOR ALGEBRA

(Notice that we cannot assert that δββ = 1, since β is present both as a superindex and as a
subindex and, according to the summation convention, a sum is implicit; in fact, δββ = N , the
dimension of T.) Analogously, we can show that
′ ′
Xαβ′ Xβγ = δαγ ′ (= δαγ ). (1.7)

Exercise 1.8 Derive this.

The replacement of eα′ , from equation (1.2), in (1.1), furnishes

v = v α Xαβ′ eβ ,
′
(1.8)

and, due to the linear independence of eβ ,

v β = Xαβ′ v α .
′
(1.9)

Consequently, ′
Xαγ v α = Xαγ Xβα′ v β = δβγ ′ v β = v β .
′ ′ ′ ′ ′
(1.10)
Summing up, if the primed and unprimed bases are related by

eα′ = Xαβ′ eβ ,
′
eα = Xαβ eβ ′ , (1.11)
then the components of a given vector v are related by
′ ′ ′
v α = Xβα v β , v α = Xβα′ v β , (1.12)
and

Xβα Xγβ′ = δγα′

′ ′ ′
Xβα′ Xγβ = δγα , (1.13)
hold

1.2 Dual spaces

Though it is usually suggested that it may be useful to visualize vectors of a given vector space
as a set of arrows starting from a given origin, in a certain sense, this image may be somewhat
troublesome, since several sets of objects with no similarity to arrows constitute vector spaces
under adequate definitions of addition and multiplication by a scalar. Among such of them, we
have functions in a given set; would you imagine one of them as an arrow?
Let us restrict our discussion to real functions defined on a real vector space of vector set
T . Mathematically, such a function f is represented by f : T → R, indicating that it applies a
vector of T to a real number. We may endow the set of all these functions with the structure of
a vector space, by defining:
1. the sum f + g of two function f and g as

(f + g)(v) = f (v) + g(v), for all v ∈ T ;

1.2. DUAL SPACES 11

2. the product af of the scalar a by the function f as

(af )(v) = a(f (v)), for all v ∈ T ;

3. the function zero 0 as

0(v) = 0, for all v ∈ T
(where, in the left, 0 stands for a function, whereas, in the right, it is the real number
zero);
4. the inverse function −f of function f as
(−f )(v) = −(f (v)), for all v ∈ T.

Exercise 1.9 Prove that, under these operations, the set of functions f constitutes a vector
space. What is its dimension?

The set of all real functions over the vector space T is too big for our purposes: we will thus
restrict ourselves to those functions which are linear; that is, functions satisfying
f (au + bv) = af (u) + bf (v), (1.14)
for all a, b ∈ R and all u, v ∈ T . Real linear functions over a real vector space are generally
called linear functionals or forms. It is easy to verify that the sum of two linear functionals is
also a linear functional, and that the product by a scalar provides a linear functional as well.
These observations guarantee that the set of all linear functionals over a vector space T is also
a vector space. This space is called the dual space of T and is denoted by T∗ .
Exercise 1.10 Prove the last statements.
Since the linear functionals or forms are vectors as well, we will henceforth use boldface for
them. Thus, if v ∈ T and f ∈ T ∗ , then f (v) ∈ R, that is, it is a scalar, despited the boldface.
We now have two kinds of vectors, those in T and those in T ∗ . To distinguish them, those in
T will be called contravariant or primal vectors, whereas those in T ∗ will be called covariant or
dual vectors. As an additional distinctive feature, the basis vectors of T∗ will carry superindices
and the components of vectors in T ∗ will carry subindices. Thus, if {eα } is a basis of T∗ , then
g ∈ T ∗ has a unique expression g = gα eα , in terms of components. Indeed, the reason for the
choice of the expressions contravariant and covariant will be elucidated presently.
The use of the lowercase letter α in the implicit sum above suggests, according to our range
convention, that the range of the summation is from 1 to N , the dimension of T, i.e., that T∗
has the same dimension as T, so that these spaces are isomorphic.6 This, indeed, is the case,
as we shall prove now, by showing that a given basis {eα } of T induces, in a natural manner, a
dual basis {eα } in T∗ possessing N elements which satisfy eα (eβ ) = δβα .
We start by defining eα as a function which takes each vector v ∈ T into the real number
which is its α-th component v α with respect to {eα }, i.e., eα (v) = v α , for all v ∈ T . This gives
us N real functions which clearly satisfy eα (eβ ) = δβα ; it remains to show that they are linear
and that they constitute a basis of T∗ .
6
However, if dim T is infinite, then it can be shown that dim T∗ > dim T, provided the usual meaning is given
to the inequality between different orders of (infinite) cardinality.
12 CHAPTER 1. TENSOR ALGEBRA

Exercise 1.11 Check that the functions eα are, indeed, linear.

In order to show that they constitute a basis, we proceed in the following way.
For any g ∈ T ∗ , we may define N real numbers gα by g(eα ) =: gα . Thus, for any v ∈ T ,

g(v) = g(v α eα ) = v α g(eα ) (pela linearidade de g)

= v α gα = gα eα (v).

Then, for any g ∈ T ∗ , we have g = gα eα , showing that {eα } spans T ∗ and the issue of the
linear independence of {eα } is the only one that remains. This can be settled by noting that a
relationship xα eα = 0, where xα ∈ R and 0 is the functional zero, implies

0 = xα eα (eβ ) = xα δβα = xβ , for all β.

Thereby, we see that, given a basis {eα } of T , the components gα of g ∈ T ∗ with respect to the
dual basis {eα } are given by gα = g(eα ).
A change of basis (1.11) in T induces a change of dual basis. Let us denote the dual of the
′ ′ ′ ′ ′ ′
primed basis {eα′ } by eα , so that, by definition, eα (eβ ′ ) = δβα′ , and eα = Yβα eβ , for some Yβα .
Then,

δβα′ = eα (eβ ′ ) = Yγα eγ (Xβµ′ eµ )

′ ′ ′

= Yγα Xβµ′ eγ (eµ ) (by linearity of eγ )

′

= Yγα Xβµ′ δµγ = Yγα Xβγ′ .

′ ′

′ ′
which means Yγα = Xγα .
Exercise 1.12 Prove this last statement.
Thereby, through a change of basis of T given by (1.11), the dual bases of T ∗ transform as
′ ′ ′
eα = Xβα eβ , eα = Xβα′ eβ . (1.15)

It is immediately shown that the components of g ∈ T ∗ , with respect to the dual bases, transform
as
gα′ = Xαβ′ gβ , gα = Xαβ gβ ′ .
′
(1.16)

Exercise 1.13 Prove this.

′
Then, the same matrix [Xβα ] and its inverse [Xβα′ ] are involved, but their roles with respect to
the basis vectors and the components are changed and this finally gives some justification for
the terminology contravariant and covariant.
Given T and one of its bases {eα }, we have just seen how to build its dual space T∗ with dual
basis {eα } satisfying eα (eβ ) = δβα . We may apply this process again to get the dual space T∗∗ of
T∗ , with dual basis {fα }, say, satisfying fα (eβ ) = δαβ , and the vectors h ∈ T ∗∗ may be expressed
in terms of components as h = hα fα . Under a change of basis of T, the components of vectors
′ ′
in T transform according to v α = X α β v β . This induces a change of dual basis of T∗ , under
which the components of the vectors in T ∗ transform according to gα′ = Xαβ′ gβ . On its turn, this
1.3. TENSOR PRODUCT SPACES 13

induces a change of basis of T∗∗ , under which it is readily seen that the components of vectors
′ ′
of T ∗∗ transform according to hα = Xβα hβ (because the inverse of the inverse of a matrix is
the original matrix itself). That is, the components of vectors in T ∗∗ transform exactly in the
same manner as the components of vectors in T . This means that, if we establish a one-to-one
correspondence between vectors of T and of T ∗∗ , making v α eα in T correspond to v α fα in T ∗∗ ,
where {fα } is the dual of the dual of {eα }, then this correspondence will be independent of basis.
Exercise 1.14 Convince yourself of that!
A one-to-one basis-independent linear correspondence between two vector spaces is called a
natural isomorphism and, of course, naturally isomorphic vector spaces are generally identified,
by identifying the corresponding vectors. Consequently, we will identify T∗∗ and T.

1.3 Tensor product spaces

Given a vector space T, we saw how to create a new vector space, viz. its dual T∗ , but the process
is over here (when we identify T∗∗ with T). However, it is possible to generate a new vector
space from two vector spaces, by forming what is called a tensor product. As a preliminary to
this, we need to define bilinear functionals in a pair of vector spaces.
Let T and U be two real finite-dimensional vector spaces. The Cartesian product T × U is
the set of all (ordered) pairs of the form (v, w), where v ∈ T and w ∈ U . A bilinear functional
f over T × U is a real function f : T × U → R, which is bilinear, that is, satisfies
f (au + bv, w) = af (u, w) + bf (v, w),
for all a, b ∈ R, u, v ∈ T and w ∈ U,
and
f (v, cw + ex) = cf (v, w) + ef (v, x),
for all c, e ∈ R, v ∈ T and w, x ∈ U.
With definitions of addition, multiplication by a scalar, the zero function and inverses anal-
ogous to those given for linear functionals, it is immediate to show that the set of bilinear
functionals over T × U is a vector space and, henceforth, we will use boldface for the bilinear
functionals.
Exercise 1.15 Demonstrate the statement above.
We are now in a position to define the tensor product T ⊗ U of T and U as the vector space
of all bilinear functionals over T ∗ × U ∗ . Notice that, in this definition, we use the base sets T ∗
and U ∗ of the dual spaces, not the spaces T and U themselves.
There naturally arises the question of the dimension of T ⊗ U. It is indeed N M , where N
and M are the dimensions of T and U, respectively; we prove this by showing that, from given
bases of T and U, we can define N M elements of T ⊗ U, which constitute a basis for it.
Let {eα }, α = 1, . . . , N and {f a }, a = 1, . . . , M , be bases of T∗ and U∗ , dual to the bases {eα }
and {fa } of T and U, respectively (Note that we used two different alphabets for the suffixes
which possess different ranges.). Let us define N M functions eαa : T ∗ × U ∗ → R as
eαa (g, h) = gα ha , (1.17)
14 CHAPTER 1. TENSOR ALGEBRA

where gα are the components of g ∈ T ∗ relatively to {eα } and ha those of h ∈ U ∗ relatively to

{f a }. In particular,

eαa (eβ , f b ) = δαβ δab . (1.18)

It is simple to show that eαa are bilinear and thus belong to T ⊗ U. To show that they constitute
a basis we must prove that they span T ⊗ U and that they are linearly independent.

Exercise 1.16 Following a reasoning similar to one in Section 1.2, show that {eαa } (i) spans
T ⊗ U, and (ii) is linearly independent.

With the preceding exercise, we proved that the dimension of T ⊗ U is N M , the product of
the dimensions of T and U, and that, in a natural manner, bases {eα } of T and {fa } of U induce
a basis {eαa } of T ⊗ U, the components τ αa of any τ ∈ T ⊗ U, relatively to this basis, being
given, in terms of the dual bases, by τ αa =τ (eα , f a ).
Let us investigate now how the components τ αa and the basis vectors eαa transform when
new bases are introduced in T and U. Let us suppose the bases of T and U transform according
to
eα′ = Xαβ′ eβ , fa′ = Yab′ fb . (1.19)
This induces a new basis {eα′ a′ } of T ⊗ U, and, for any (g, h) ∈ T ∗ × U ∗ ,

eα′ a′ (g, h) = gα′ ha′ = Xαβ′ Yab′ gβ hb

= Xαβ′ Yab′ eβb (g, h).

Therefore,
eα′ a′ = Xαβ′ Yab′ eβb . (1.20)
Analogously, for the components, we obtain
′ ′ ′ ′
τ α a = Xβα Yba τ βb . (1.21)

Exercise 1.17 Prove this.

A vector which is an element of the tensor product of two (or more) spaces (cf. below) is
called a tensor. The tensor product, as defined above, is a product of spaces. It is possible to
define a tensor which is the tensor product g ⊗ h of individual vectors g ∈ T and h ∈ U, by
demanding that
g ⊗ h = g α ha eαa , (1.22)
where g α and ha are the components of g and h relatively to bases of T and U which induce a
basis {eαa } of T ⊗ U. Although this definition is given by means of bases, it is, in fact, basis
independent.

Exercise 1.18 Prove this.

In particular we have
eα ⊗ fa = eαa . (1.23)
1.3. TENSOR PRODUCT SPACES 15

The tensor product g ⊗ h belongs to T ⊗ U, but not all tensors of T ⊗ U have this form. Those
which do have are called decomposable or simple tensors.
Having established the basic idea of the tensor product of two vector spaces, we can now
extend it to three or more spaces. However, given three vector spaces T, U, V, we can form their
tensor product in several ways: (T ⊗ U) ⊗ V or T ⊗ (U ⊗ V). These two spaces clearly have
the same dimension and are, in fact, naturally isomorphic, in the sense that we can establish
a bijective basis-independent linear correspondence between their elements, just as we have
done with T and T ∗∗ . This is achieved by choosing bases {eα }, {fa } and {gA } in T, U e V,
respectively (three ranges, therefore three “alphabets”), and letting τ αaA eα ⊗(fa ⊗gA ) correspond
to (τ αaA eα ⊗fa )⊗gA , and showing, then, that such a correspondence is basis-independent. Due to
this natural isomorphism, we identify these spaces, and the notation T⊗U⊗V is not ambiguous.
Exercise 1.19 Prove the existence of the mentioned natural isomorphism.
An alternative way to define T⊗U⊗V is as the space of trilinear functionals over T ∗ ×U ∗ ×V ∗ .
This leads to a space which is naturally isomorphic to those of the preceding paragraph, all of
them being identified.
Exercise 1.20 Convince yourself of this.
There are other natural isomorphisms, for example between T⊗U and U⊗T, or between (T⊗U)∗
and T∗ ⊗ U∗ , and, whenever they do exist, the vector spaces are identified.
Exercise 1.21 Convince yourself of this.
From now on, we will restrict ourselves to tensor produt spaces obtained by taking repeated
tensor products of a unique vector space T and its dual T∗ . We introduce the following notation:
r times
z }| {
T ⊗ T ⊗ · · · ⊗ T =: Tr = T r (this last notation, by abuse),
s times
z }| {
T∗ ⊗ T∗ ⊗ · · · ⊗ T∗ =: Ts = Ts (this last notation, by abuse),
T r ⊗ Ts =: Tsr .
In particular T = T 1 and T ∗ = T1 .
An element of T r is a (totally) contravariant tensor of rank r, an element of Ts is a (totally)
covariant tensor of rank s, whereas an element of Tsr is a (mixed) tensor of rank (r, s). Notice
that this naming labels contravariant and covariant vectors as tensor of rank (1, 0) and (0, 1),
respectively. Scalar can be included in the general scheme by considering them as tensors of
rank (0, 0).
A basis {eα } of T (of dimension N ) induces a dual basis {eα } of T ∗ and these, together,
induce a basis {eβα11···β r r
···αr } of Ts . Each tensor τ ∈ Ts has N
s r+s
unique components unı́vocas
relatively to the induced dual basis:
τ = τ α1 ···αr β1 ···βs eβα11···β
···αr .
s
(1.24)
A change of basis of T induces a change of basis of Tsr , under which the components transform
according to:
′ ′ ′ ′
τ α1 ···αr β1′ ···βs′ = Xµα11 · · · Xµαrr Xβν1′ · · · Xβνss′ τ µ1 ···µr ν1 ···νs . (1.25)
1
16 CHAPTER 1. TENSOR ALGEBRA

For instance, for a tensor τ ∈ T21 ,

′ ′
τ α β ′ γ ′ = Xρα Xβσ′ Xγλ′ τ ρ σλ .

It is usual to define tensors as objects having components which transform accoding to the
equations (1.25). Such a way of conceiving tensors is justified noticing that if to each basis of
T are associated N r+s real numbers, which, under a change of basis given by equations (1.11),
transform as (1.25), then these numbers are the components of a tensor τ of rank (r, s), in
accordance with the way we defined such an object; simply make

τ = τ α1 ···αr β1 ···βs eβα11···β

···αr .
s

1.4 Relative tensors

We will show now that “there are more things in heaven and earth than are dreamt of in [our]
philosophy”, that is, there are still more general (and equally useful) geometric objects than
the tensors we defined above, whose components transform not simply with factors of the basis
transformation matrix, but also with a factor depending on the determinant of such a matrix.
To that end, let us remind something about determinants
Let there be given an N × N square matrix[Z α β ], where, as usual, we will suppose the
superindex α indicates a row and the subindex β indicates a column, that is,
 
Z 11 Z 12 · · · Z 1N
 Z 21 Z 22 · · · Z 2N 
α  
[Z β ] =  .. .. . . ..  . (1.26)
 . . . . 
ZN 1 ZN 2 · · · ZN N

The determinant of this matrix, det[Z], or simply Z, can be defined through the general
Cramer’s rule or, in a more geometrical fashion, through the study of the notion of volume or
extension in an N -dimensional space. Here we want you to convince yourself that it may also
be written as:

det[Z] = ²α1 α2 ...αN Z α1 1 Z α2 2 · · · Z αN N ,

= ²α1 α2 ...αN Z 1 α1 Z 2 α2 · · · Z N αN , (1.27)

where we introduced the so called Levi-Civita permutation symbols:


 1, if (α1 , α2 , . . . , αN ) for even permutation of (1, 2, . . . , N );
²α1 α2 ...αN := ²α1 α2 ...αN := −1, if (α1 , α2 , . . . , αN ) for odd permutation of (1, 2, . . . , N );

0, otherwise, i.e., if there are repeated indices.
(1.28)
Observations:

1. Notice that we follow the convention that ²12...N = ²12...N = 1 and not that ²12...N =
−²12...N = 1, sometimes adopted by certain authors. Watch out!
1.4. RELATIVE TENSORS 17

2. Notice that, with equation (1.27), it is indeed obvious that, under an exchange of two rows
or columns, the determinant changes sign, due to the anti-symmetry of the Levi-Civita
symbols.

3. As a consequence of the former item, or right from equation (1.27) itself, the determinant
of a matrix with two proportional rows or columns turns out to be vanishing.

4. To help accept the expression (1.27) for the determinant, remember (or prove now), from
basic vector calculus, the expression for (i) the vector product in terms of Cartesian com-
ponents:
 
x̂ ŷ ẑ
A × B = det  Ax Ay Az 
Bx By Bz
= ²ijk Ai Bj êk (implicit sums in i, j, k).

or (ii) the mixed product, also in terms of Cartesian components:

 
Ax Ay Az
A · (B × C) = det  Bx By Bz  ,
Cx Cy Cz

which also connects to the idea, mentioned above, of the determinant as a measure of
volume (in this case, of a parallelepiped with sides A, B e C).

Taking into account the anti-symmetry of the Levi-Civita symbols, we can recast (1.27) as

²α1 α2 ...αN Z = ²β1 β2 ...βN Z α1 β1 Z α2 β2 . . . Z αN βN , (1.29)

or
²α1 α2 ...αN Z = ²β1 β2 ...βN Z β1 α1 Z β2 α2 . . . Z βN αN , (1.30)
forms which will be useful later on.

Exercise 1.22 Convince yourself of the validity of these formulae.

It would be interesting if we could have a final expression for the determinant, Z, isolated in
a member of this mentioned expressioni as a function of the Levi-Civita symbols and the matrix
elements. To that end, it is expedient to introduce new objects, which are the generalized
Kronecker delta tensors, δβα11βα22···β
···αr
r
, defined by
 
δβα11 · · · δβαk1 · · · δβαr1
 .. ... .. ... .. 
 . . . 
 αk αk αk 
δβα11···β
···αk ···αr
k ···βr
:= det  δβ1 · · · δβ2 · · · δβr  (1.31)
 . .. .. .. .. 
 .. . . . . 
αr αr αr
δβ1 · · · δβk · · · δβr
18 CHAPTER 1. TENSOR ALGEBRA

Obviously, the (usual) Kronecker delta is a particular case of that definition, corresponding to
the value r = 1. With r = 2 in (1.31), we see that
αβ
δµν = δµα δνβ − δνα δµβ .

In general, δβα11βα22···β
···αr
r
is a sum of r! terms, each one of which is a product of r (usual) Kronecker
deltas. Since, as we have already seen, the (usual) Kronecker delta is a type (1,1) tensor, it
follows immediately that the generalized Kronecker delta is a tensor of rank (r, r). From its very
definition, it is easy to show that: (i) the generalized Kronecker delta is anti-symmetric in all its
upper indices and all its lower indices; (ii) if r > N , where N is the dimension of the underlying
(primal) space, then δβα11βα22···β···αr
r
≡ 0.

Exercise 1.23 Convince yourself of these statements.

In order to obtain other important properties of the generalized Kronecker delta, following [9],
we now return to (1.31) and restrict ourselves to the case r ≤ N . We can expand (1.31) in terms
of its elements of the last column, to obtain
α α ···α ···α α α ···α α α ···α α
r−1 r
δβ11β22···βkk···βr−1β r
= (−1)r+r δβαrr δβ11···βr−1
r−1
+ (−1)r+r−1 δβrr−1 δβ11···βr−2
r−2 r
βr−1 + · · ·
α ···α α α
· · · + (−1)r+k δβαrk δβ11···βk−1
k−1 k+1 r−1 r···α
βk βk+1 ···βr−1 + · · ·
α α ···α α α ···α α
· · · + (−1)r+2 δβαr2 δβ11β23β3 ···β
r−1 r
r−1
+ (−1)r+1 δβαr1 δβ12β2 ···β
r−1 r
r−1
. (1.32)

Contract now over the indices αr , βr , to derive

α α ···α ···α α
r−1 r α ···α
r−1 α ···α
r−2 r−1 α
δβ11β22···βkk···βr−1α r
= N δβ11···βr−1 − δβ11···βr−2 βr−1 + · · ·
α ···α α ···α α
· · · + (−1)r+k δβ11···βk−1
k−1 k+1 r−1 k
βk βk+1 ···βr−1 + · · ·
α α ···α α α ···α α
· · · + (−1)r+2 δβ11β23β3 ···β
r−1 2
r−1
+ (−1)r+1 δβ12β2 ···β
r−1 1
r−1
. (1.33)

It is easy to check that, in the right hand side of this expression, we have, besides the first term,
α ···αr−1 α ···αr−1
N δβ11···βr−1 , r − 1 others; furthermore these other terms are all equal to −δβ11···βr−1 .

Exercise 1.24 Prove these two last statements.

Thus, we get
α ···α r−1 r α r−1 α ···α
δβ11···βr−1 αr = (N − r + 1)δβ1 ···βr−1 .
1
(1.34)
We now contract, in this formula, αr−1 and βr−1 , applying the formula to itself for the case when
r is reduced to r − 1, obtaining
α ···α α
r−2 r−1 rα r−2 α ···α
δβ11···βr−2 αr−1 αr = (N − r + 2)(N − r + 1)δβ1 ···βr−2 .
1
(1.35)

Repeating this process r − s times, we get the following general identity

α ···α α ···α (N − s)! α1 ···αs

δβ11···βssαs+1
s+1 r
···αr = δ . (1.36)
(N − r)! β1 ···βs

Exercise 1.25 Prove this.

1.4. RELATIVE TENSORS 19

An equivalent form of (1.36) is

α ···α α ···α (N − r + t)! αt+1 ···αr

δα11···αttβt+1
t+1 r
···βr = δ . (1.37)
(N − r)! βt+1 ···βr

Exercise 1.26 Prove this.

In particular, when t = r, we have

N!
δαα11···α
···αr
= . (1.38)
r
(N − r)!

Now we want to establish a fundamental relation or identity among ²α1 α2 ···αN , ²β1 β2 ···βN and
δβα11βα22···β
···αN
N
:
²α1 α2 ···αN ²β1 β2 ···βN = δβα11βα22···β
···αN
N
. (1.39)

To that end, consider the quantity

Aαβ11βα22···β
···αN
N
:= ²β1 β2 ···βN ²α1 α2 ···αN − δβα11βα22···β
···αN
N
, (1.40)

which is obviously anti-symmetric in the subindices and the superindices. Therefore, the only
possible nonvanishing components of Aαβ11βα22···β
···αN
N
will occur when (α1 , α2 , . . . , αN ) and (β1 , β2 , . . . , βN )
are permutations (without repetition) of (1, 2, . . . , N ). However, from (1.28), (1.40) and (1.31),
we easily see that
A12···N
12···N = 0.

Thus, we have just shown that

Aαβ11βα22···β
···αN
N
≡ 0,
which, due to (1.40), establishes (1.39).

Exercise 1.27 Show, from (1.39), that, generically,

1 α1 ...αN −j γ1 ...γj α ...α −j

² ²β1 ...βN −j γ1 ...γj = δβ11...βNN−j . (1.41)
j!

Thereby, or from (1.39) itself, it comes, in particular,

²α1 α2 ···αN ²α1 α2 ···αN = N !

Exercise 1.28 Prove that!

Finally, we can then have the expression we looked for:

1 α1 α2 ···αN
Z := det[Z µ ν ] = ² ²β1 β2 ···βN Z β1 α1 Z β2 α2 · · · Z βN αN . (1.42)
N!
Exercise 1.29 Prove it!
20 CHAPTER 1. TENSOR ALGEBRA

Let us consider now, as a particular case for the matrix, [Z α β ], dealt with above a matrix for
the change of bases, in a certain vector space:
eα′ = Xαβ′ eβ . (1.43)

In this situation, the expression for the determinant of [Xαβ′ ] or its inverse [Xβα ], according to
′

(1.29) or (1.30), shows that, if we postulate, as is all too natural, that, regardless of basis, the
′ ′ ′
values of the (homonimous) components of the Levi-Civita symbols are the same (²α1 α2 ···αN =
²α1 α2 ···αN and ²α′1 α′2 ···α′N = ²α1 α2 ···αN ), then we conclude that the transformation laws for the
components of these symbols (invariant by postulation or definition) turn out to be:
′ ′ ′ α′ α′ α′
²α1 α2 ···αN = X −1 Xβ11 Xβ22 · · · XβNN
and
²α′1 α′2 ···α′N = XXαβ′1 Xαβ′2 · · · Xαβ′N ²β1 β2 ···βN .
1 2 N

Such laws are equal to those for tensors, except for the presence of a factor which is a power
of the determinant fo the matrix of basis change. This suggests the importance of dealing
with algebraic (geometric) objects whose components transform in a more general way. Hence,
deviating a little from our geometrical or basis-independent presentation up to now, we will say
that a set of N r+s numbers Λα1 ···αr β1 ···βs constitute the components of a relative tensor relativo of
rank (r, s) and weight w, if, under a basis change (1.43), these numbers (called the components
of the relative tensor) transform according to
′ ′ ′ ′
Λα1 ···αr β1′ ···βs′ = X −w Xµα11 · · · Xµαrr Xβν1′ · · · Xβνss′ Λµ1 ···µr ν1 ···νs . (1.44)
1

Notice the sign of the exponent of the determinant X. We can then notice that:
1. ²α1 ···αN and ²α1 ···αN constitute the components of relative tensors of weight 1 and −1,
respectively.
2. the tensors we have dealt with so far are relative tensors of weight 0; sometimes they are
called absolute tensors.

1.5 Additional operations and results

1.5.1 Contraction
So far, we have three basic (closed) operations with tensors (relative or absolute): addition of
relative tensors of the same rank and weight, multiplication of a tensor by a scalar, and formation
of the tensor product. There is a fourth operation, which is most easily explained in terms of
components. This operation is the contraction, which associates N r+s−2 numbers (components)
Rα1 ···αp−1 αp+1 ···αr β1 ···βq−1 βq+1 ···βs with N r+s numbers (components) Qα1 ···αr β1 ···βs , defined by
Rα1 ···αp−1 αp+1 ···αr β1 ···βq−1 βq+1 ···βs := Qα1 ···αp−1 γαp+1 ···αr β1 ···βq−1 γβq+1 ···βs . (1.45)
That is, rendering one subindex equal to a superindex and summing, as the summation conven-
tion implies. Of course there are rs ways to acomplish that, each one of which leading to an (in
principle) different contraction of the original set of numbers.
1.5. ADDITIONAL OPERATIONS AND RESULTS 21

The special meaning this operation has for (absolute or relative) tensors is that, if the original
numbers are components of a relative tensor of rank (r, s) and weight w, then its contractions
are the components of a relative tensor of rank (r − 1, s − 1) and same weight, w.

Exercise 1.30 Prove this!

1.5.2 Symmetrization and anti-symmetrization

Given a matrix [Mαβ ], we can always express it as a sum of two other matrices, [M(αβ) ] and
[M[αβ] ], such that
Mαβ = M(αβ) + M[αβ] , (1.46)
where
1
M(αβ) := (Mαβ + Mβα ) (1.47)
2
1
M[αβ] := (Mαβ − Mβα ). (1.48)
2
The matrix [M(αβ) ] is called the symmetric part of [Mαβ ] and the process shown in (1.47) is
called symmetrization of [Mαβ ], whereas the matrix [M[αβ] ] is called the anti-symmetric (or skew-
symmetric) part of [Mαβ ] and the process indicated in (1.48) is called the anti-symmetrization
of [Mαβ ]; such a terminology is justified by the fact that

M(αβ) = M(βα)

and
M[αβ] = −M[βα] .
Besides, if, indeed, Mαβ constitute the components¡ of a relative
¢ tensor of rank (2, 0) and weight
w, so will M(αβ) e M[αβ] , in contradistinction to M α β − M β α /2.

Exercise 1.31 Prove this!

In the more general case, the component M(α1 ···αr ) of the so called (totally) symmetric part of
Mα1 ···αr is obtained by summing all components obtained by all possible r! permutations of the
indices (α1 , . . . , αr ) and dividing the result by r! ; i.e., in the case of three indices, we would
have:
1
M(αβγ) := (Mαβγ + Mαγβ + Mβαγ + Mβγα + Mγαβ + Mγβα ) .
3!
Something analogous holds for the so called (totally) anti-symmetric part, but here the even per-
mutaions of the indices (α1 , . . . , αr ) are summed, whereas the odd permutations are subtracted,
that is:
1
M[αβγ] := (Mαβγ − Mαγβ − Mβαγ + Mβγα + Mγαβ − Mγβα ) .
3!
Naturally, all this can be extended to “covariant indices”, always preserving the (relative) tensor
character of the resulting objects (the symmetric and anti-symmetric parts). On the other hand,
the symmetrization or anti-symmetrization over indices in distinct “levels” do not yield (relative)
tensors.
22 CHAPTER 1. TENSOR ALGEBRA

Exercise 1.32 Prove this!

We can relate the anti-symmetrization operation to the previously defined generalized Kro-
αβγ
necker deltas. In fact, as an example, given the tensor Tαβγ , consider the tensor δρσλ Tαβγ . From
(1.32) (with r = 3), we have

αβγ αβ αγ βγ
δρσλ Tαβγ = δρσ Tαβλ − δρσ Tαλγ + δρσ Tλβγ
= 3!T[ρσλ] . (1.49)

Exercise 1.33 Show this.

For any integer r ≤ N , it is indeed easily seen that

δβα11···β
···αr
r
Tα1 ···αr = r!T[β1 ···βr ] . (1.50)

Exercise 1.34 Convince yourself of this.

1.5.3 Quotient rules

The quotient rules allow us to establish directly the (relative) tensor character of an object once
it is known that its product with an arbitrary (relative) tensor always yields a (relative) tensor.
Let us argue with a concrete example, again in terms of components.
Let there be given, in a certain basis, a set of numbers Y α βγ , which, when multiplied by the
components T γµ of an arbitrary tensor, we know always furnishes a tensor C α β µ , say; that is,

C α β µ = Y α βγ T γµ (1.51)

is a tensor for any tensor T γµ . Then, the quotient rule, in this case, states that Y α βγ will
constitute the components of a tensor as well, of rank (1,2), as suggested by the structure of its
indices.
To show that, we use the characteristic law of transformation of tensor components. We
imagine tha, in a new basis, there still hodls the equation (1.51), as if by definition of the new
1.5. ADDITIONAL OPERATIONS AND RESULTS 23

components of the object Y, whose character we wish to discover. Then,

′ ′ ′ ′ ′
C α β′ µ = Y α β′γ′ T γ µ
w
w
w
w (for C and T are tensors)
w
Ä
Xαα Xββ′ Xµµ C α β µ = Y α β ′ γ ′ Xγγ Xµµ T γµ
′ ′ ′ ′ ′

w
w
w
w replacing (1.51)
w
Ä
Xαα Xββ′ Xµµ Y α βγ T γµ =
′ ′ ′ ′ ′
Y α β ′ γ ′ Xγγ Xµµ T γµ
w
w
w
w
w since T is arbitrary
Ä
³ ′ ´
Y α β ′ γ ′ Xγγ − Xαα Xββ′ Y α βγ Xµµ =
′ ′ ′
0
w
w
w
w
w multiplying by Xµµ′
Ä
= Xαα Xββ′ Y α βγ
′ ′ ′
Y α β ′ γ ′ Xγγ
w
w
w
w multiplying by X γ′
w σ
Ä
Y α β ′ σ′ = Xαα Xββ′ Xσγ′ Y α βγ ,
′ ′

which is exactly what we wished to demonstrate. The “quotient rule” expression itself is ex-
plained by the manner in which Y α βγ appears in (1.51).

Exercise 1.35 How would you adapt the statement of this rule for the case of relative tensors?

Exercise 1.36 If, for a symmetric but otherwise arbitrary tensor, with components S αβ , the
result
C α = Y α βγ S βγ
is always a contravariant vector, what can you deduce about the character of Y α βγ or of any of
its parts?
Bibliography

[1] R. Annequin and J. Boutigny. Curso de ciencias fı́sicas: mecánica 1. Editorial Reverté,
1978.

[2] P. Bamberg and S. Sternberg. A course in mathematics for students of physics 1. Cambridge
University Press, 1988.

[3] R. L. Bishop and S. I. Goldberg. Tensor analysis on manifolds. Dover, 1980.

[4] M. Crampin and F. A. E. Pirani. Applicable differential geometry. Cambridge University

Press, 1986.

[5] R. d’Inverno. Introducing Einstein’s relativity. Oxford University Press, 1992.

[6] C. T. J. Dodson and T. Poston. Tensor geometry: the geometric viewpoint and its uses.
Springer, second edition, 1991.

[7] J. Foster and J. D. Nightingale. A short course in general relativity. Longman, 1979.

[8] L. P. Hughston and K. P. Tod. An introduction to general relativity. Cambridge University

Press, 1990.

[9] D. Lovelock and H. Rund. Tensors, differential forms, and variational principles. John
Wiley and Sons, 1975.

[10] L. Nachbin. Introdução à álgebra. Editora McGraw-Hill do Brasil, 1971.

[11] C. Nash and S. Sen. Topology and geometry for physicists. Academic Press, 1983.

[12] B. O’Neill. Semi-Riemannian geometry with applications to relativity. Academic Press,

1983.

[13] B. Schutz. Geometrical methods of mathematical physics. Cambridge University Press,

1980.

31