2. EVALUACIÓN DE LOS EOM POR LOS CEIM
2.4. Evaluación ética
Notice that the definition of a tensor does not mention components of the vectors. A tensor must be a rule [that] gives the same real number independently of the reference frame in which the vectors’ components are calculated.
Bernard Schutz, §3.2
3.1 Exercises
3.2 Prove that the set of all one-forms is a vector space.
Solution: While most instructors will be satisfied with a much more succinct answer, we give a detailed response to help clarify the concepts. One-forms were introduced in Schutz Eqs. (3.6a,b):
˜s = ˜p + ˜q,
˜r = α ˜p,
Schutz Eq. (3.6a) (3.1)
where ˜s and ˜r above must obey the following for all arguments A:
˜s( A)= ˜p( A)+ ˜q( A),
˜r( A)= α ˜p( A).
Schutz Eq. (3.6b) (3.2) To prove that the set of all one-forms is a vector space, we must show that this set meets axioms (1) and (2) given in Schutz Appendix A, p. 374. Axiom (1) states that
“[The set] V is an abelian group with operation + (A + B = B + A ∈ V ) and identity 0 (A+ 0 = A).” For a brief account of group theory see (e.g. Schutz,1980).
Very briefly, a group is a set of objects that can be combined with a binary operation, here “+” but it can be very general, to form new elements that are also in the set.
There are four axioms of group theory that must be met: (i) The closure property, A+ B ∈ V , is the one we just mentioned that the result of the binary operation (here “+”) between two elements results in an element of the set, (ii) Associativity property, A+ (B + C) = (A + B) + C, (iii) There must be an identity element, i.e.
an element that does not alter any elements of the set, A+ 0 = A for all A ∈ V , (iv) For every element A ∈ V there must be an inverse element B that combines to produce the identity element, A+ B = 0 for any A ∈ V . So applying this to a vector space, Axiom (1) tells us that the binary operation is vector addition and the identity 47
element is the zero vector 0. Some books will emphasize that 0 in this context is a vector by writing 0. We are asked to apply this to one-forms.
The sum of two one-forms must also be a one-form (so closure property (i) met) which is satisfied byeqn.(3.1),˜s = ˜p+ ˜q. We require that the order of summation does not matter (this is the “Abelian” property), which is satisfied byeqn.(3.2) because a one-form acting on a vector evaluates to a real and the sum of two reals does not depend upon the order. Similarly the property (ii) of associativity is also met. We also require a zero (there must be an identity element (iii)). The zero one-form gives zero for any vector. So say ˜q is the zero one-form. Then assumingeqn.(3.1) and by eqn.(3.2)
˜s( A)= ˜p( A)+ ˜q( A)= ˜p( A)+ 0 = ˜p( A), (3.3) so ˜p + 0 = ˜p and we have a zero. Finally setting α = −1 ineqns.(3.1) and (3.2) we see that for each ˜p we can always construct the inverse element ˜r = −1 · ˜p that sums to give the zero element, ˜p + ˜r = ˜p − 1 · ˜p = 0. So the set of one-forms with addition as defined ineqns.(3.1,3.2) satisfies Axiom (1).
Axiom (2) of Appendix A requires that multiplication of an element of the vector space by a real number gives another element of the vector space, with four requirements:
(i) a(A+ B) = aA + aB, (ii) (a+ b)A = aA + bA,
(iii) (ab)A= a(bA), (iv) 1(A)= A, (3.4)
with A, B elements of the vector space, a, b reals. Although it is not made explicit in eqns.(3.1) and (3.2), it was clear from the context that α ∈ R. Byeqns.(3.1and3.2) it is immediately clear that multiplication of a one-form by a real scalar meets all the requirements of Axiom 2. For instance Axiom 2(i) requires
α(˜p + ˜q) = α ˜p + α ˜q. (3.5)
On the LHS ofeqn. (3.5) we have a one-form, say ˜s = α( ˜p + ˜q), where ˜s is the one-form such that
˜s( V )= α · ( ˜p + ˜q)( V )= α ·
˜p( V )+ ˜q( V )
usedeqn.(3.2)
= α ˜p( V )+ α ˜q( V ). because ˜p( V )and ˜q( V )are reals (3.6) On the RHS ofeqn.(3.5) we have a one-form, say˜s = α ˜p + α ˜q, such that
˜s ( V )= (α ˜p + α ˜q)( V )= α ˜p( V )+ α ˜q( V ) usedeqn.(3.2)
= ˜s( V ). usedeqn.(3.6) (3.7)
So because ˜s and ˜s are one-forms that evaluate to the same real number when they operate on the same vector V, we conclude ˜s = ˜s andeqn. (3.5) holds, i.e.
multiplication by a scalar α is distributive over addition of one-forms. The other three properties in Axiom 2 follow similarly.
49 Exercises
3.4 Given the following vectors inO:
A→
O (2, 1, 1, 0), B→
O (1, 2, 0, 0), C →
O (0, 0, 1, 1), D →
O (−3, 2, 0, 0), (b) find components of ˜p if
˜p( A)= 1, ˜p( B)= −1, ˜p( C)= −1, ˜p( D)= 0.
Solution: Using the expression
˜p( A)= pαAα Schutz Eq. (3.8) (3.8)
we can write a linear system in the four unknown components:
⎛
Note the matrix is written with the rows given by the vectors (so that one-forms appeared as columns) but this choice was arbitrary. This was solved in the accompanying MapleTMworksheet, giving ˜p →O (−2, −3, 15, −23)/8.
(d) determine whether the one-forms ˜p, ˜q, ˜r, ˜s are linearly independent if
˜q( A)= 0, ˜q( B)= 0, ˜q( C)= 1, ˜q( D)= −1,
˜r( A)= 2, ˜r( B)= 0, ˜r( C)= 0, ˜r( D)= 0,
˜s( A)= −1, ˜s( B)= −1, ˜s( C)= 0, ˜s( D)= 0. (3.10)
Solution: Given the values of the four one-forms, ˜p, ˜q, ˜r, ˜s applied to the four known vectors A, B, C, Dwe can, in principle, find all components of all four one-forms, repeating the procedure we did in Exercise 3.4(b). And then one could write a matrix M where the columns of M are taken from the one-form components. If the determinant of M is zero the one-forms are linearly dependent. But that is a lot of work.
There is a simpler way to test for linear dependence. If the one-forms are linearly dependent, then there are nontrivial real numbers a, b, c, d such that
a˜p + b ˜q + c˜r + d ˜s = ˜t = 0
But then
The latter can only be true if the determinant is zero, but
so the one-forms must be linearly independent.
3.5 Justify steps from Schutz Eq. (3.10a) to Eq. (3.10d), where
A¯αp¯α= (Λ¯αβAβ)(Λμ¯αpμ), Schutz Eq. (3.10a)
= (Λμ¯αΛ¯αβ) (Aβpμ), Schutz Eq. (3.10b)
= δμβAβpμ, Schutz Eq. (3.10c)
= Aβpβ. Schutz Eq. (3.10d) (3.15)
Solution: We start with the respective Lorentz transformations of the vector and one-form, and then use the inverse property of the two Lorentz transformations. Summing over the dummy index results in the desired expression. In particular,
A¯αp¯α= (Λ¯αβAβ)(Λμ¯αpμ), by Schutz Eqs. (2.7) and (3.9) respectively
= (Λμ¯αΛ¯αβ) (Aβpμ), just rearranged the terms
= δμβ(Aβpμ), byeqn.(3.23)
= Aβpβ. sum over μ, use properties of the Kronecker delta
51 Exercises
3.6 Consider the basis{eα} of a frame O and a basis {˜λ0, ˜λ1, ˜λ2, ˜λ3} for the space of one-forms, with
˜λ0→
O (1, 1, 0, 0), ˜λ1→
O (1,−1, 0, 0),
˜λ2→
O (0, 0, 1,−1), ˜λ3→
O (0, 0, 1, 1). (3.16) Note that{˜λβ} is not the basis dual to {eα}.
(a) Show that ˜p ˜p(eα)˜λαfor arbitrary ˜p.
Solution: Applying a one-form to a basis vector results in the corresponding compo-nent of the one-form,
pα = ˜p(eα), Schutz Eq. (3.7) (3.17)
but this component “belongs to” the basis one-form dual to the basis vector eα. So using these components with a different basis such as ˜λα leads to a different one-form. That is, ˜p = pα˜ωα, when and only when ˜ωα is dual toeα so thateqn.(3.91) applies. Indeed some texts use the same symbol for both the basis vectors and their corresponding dual basis one-forms, (e.g. Hobson et al., 2006, Eq. (3.2)), which emphasizes this correspondence but de-emphasizes the distinction between one-forms and vectors.
More formally, consider an arbitrary one-form ˜p and vector A.
˜p(eα)˜λα( A)= pα˜λα( A)= pα˜λα(Aβeβ)= pαAβ ˜λα(eβ)
= pαAα iff ˜λα(eβ)= δαβ. (3.18) But it is clear that ˜λα(eβ) δαβby inspection of the given basis. For example, δ01= 0 but
˜λ0(e1)=
1 1 0 0
⎛
⎜⎜
⎝ 0 1 0 0
⎞
⎟⎟
⎠ = 1 · 0 + 1 · 1 + 0 · 0 + 0 · 0 = 1. usedeqn.(3.8)
3.7 Prove that the basis one-forms transform under a change of basis as follows:
˜ω¯α = Λ¯αβ˜ωβ. Schutz Eq. (3.13) (3.19)
Solution: We were told just before Schutz Eq. (3.13) that its derivation is analogous to the corresponding relation for basis vectors, see derivation of Schutz Eq. (2.13) in
§2.2. Imagine that ˜p is an arbitrary one-form. Let ˜ωαand ˜ω¯αbe the sets of basis one-forms in framesO and O respectively. ˜p can be expressed in terms of either basis set
and we want this to be the same geometrical object,
˜p = pα˜ωα = p¯α˜ω¯α.
Note the analogy with Schutz Eq. (2.12), which specifies that vectors are frame independent. It is important to realize that in general p0 p¯0. Part of the subtlety arises here because the notation, while very standard, is actually misleading. In fact a few textbooks (e.g. Hobson et al.,2006; Weinberg, 1972) use a different symbol for the components of ˜p in the O frame, namely ¯pα. Here we will continue with the standard notation used by Schutz and most others
pα˜ωα = p¯α˜ω¯α same geometrical object
= Λβ¯αpβ˜ω¯α. usedeqn.(3.87) (3.20) At this point we may relabel dummy indices and replace β with α,
pα˜ωα = Λα¯αpα˜ω¯α. relabeled dummy index (3.21) Because the equality above must hold for arbitrary ˜p it is clear that we require
˜ωα = Λα¯α ˜ω¯α.
If you are uncomfortable with this last step, see SP3.5below.
3.10 (a) Given a frameO whose coordinates are {xα}, show that
∂xα/∂xβ = δαβ.
Solution: Let’s use an asterix to denote an arbitrary but fixed index, like α∗. When α∗ β∗, then xα∗and xβ∗ are independent variables and of course ∂xα∗/∂xβ∗ = 0. But when α∗ = β∗ then xα∗ = xβ∗ are the very same variable and of course
∂xα∗/∂xα∗= 1. This completes the proof, which is valid in any coordinate system.
To be more concrete, consider pseudo-Cartesian coordinates wherein x0= t, x1= x, x2= y, x3= z. When α β we have terms like, say,
∂x0
∂x1 = ∂t
∂x = 0,
because t and x are independent variables. But when α= β we have terms like, say,
∂x3
∂x3 =∂z
∂z = 1.
3.10 (b) For any two frames, we have:
∂xβ
∂x¯α = Λβ¯α. Schutz Eq. (3.18) (3.22)
53 Exercises
Show that (a) and the chain rule imply
Λβ¯αΛ¯αμ= δβμ. (3.23) This is the inverse property again.
Solution: We start with the result from Exercise 3.10(a), and apply a Lorentz transformation as follows:
∂
∂xμxβ = δβμ from Exercise 3.10(a)
∂
∂xμ
Λβ¯αx¯α
= δβμ sub coordinate transform
Λβ¯α ∂x¯α
∂xμ = δβμ transform is a constant Λβ¯αΛ¯αμ= δβμ. from Schutz Eq. (3.18)
Eqn. (3.22) reveals that the Lorentz transformations are a type of coordinate transformation with the special property that they apply globally. Later in the text, when one studies GR, the idea of so-called general covariance will be of funda-mental importance and general coordinate transformations xα(x¯α)will be considered wherein ∂xα/∂x¯αis not constant throughout spacetime; however the inverse property eqn.(3.23) will still apply locally.
3.12 Let S be the two-dimensional plane x = 0 in three-dimensional Euclidean space. Let
˜n 0 be a normal one-form to S.
(a) Show that if V is a vector that is not tangent to S, then˜n( V ) 0.
Solution: Applying the general rule for the contraction of V and ˜n to this three-dimensional Euclidean space, we have
˜n( V )= nxVx+ nyVy+ nzVz. usedeqn.(3.8) (3.24) A normal one-form must produce zero when contracted with any vector tangent to the surface. It follows that for the x= 0 surface in 3D space,
˜n →O (nx, 0, 0), (3.25)
where any nx 0 will give us a non-zero, normal one-form. So the contraction eqn.(3.24) reduces to
˜n( V )= nxVx.
Because V is not tangent to surface S it must have a component in theex-direction, Vx 0. Thus the contraction is non-zero.
(c) Show that any normal to S is a multiple of ˜n.
Solution: Any normal to the surface S must have ny= 0 and nz= 0, as we argued in Exercise 3.12(a) in derivingeqn.(3.25). To be non-zero it requires nx 0. So any
˜n →O a(nx, 0, 0), (3.26)
where a 0 and a ∈ R will serve also as a non-zero, normal one-form.
3.13 Prove, by geometric or algebraic arguments, that the gradient of f , denoted by the tensor ˜df , is normal to surfaces of constant f .
Solution: Consider an arbitrary point p = (t, x, y, z) where f (t, x, y, z) = fp. Now imagine taking an infinitesimal step (t, x, y, z) such that the change in value of f ,
f = ∂f
∂tt+∂f
∂xx+∂f
∂yy+∂f
∂zz= 0. (3.27)
This ensures we do not leave the surface of constant f . So a tangent vector to the surface of constant f is obtained from an arbitrary multiple of such a step:
A→
O a(t, x, y, z), (3.28)
where a∈ R and a 0. The gradient one-form applied to such a tangent vector is
˜df ( A)= a
∂f
∂tt+∂f
∂xx+∂f
∂yy+∂f
∂zz
used Schutz Eq. (3.15)
= 0. usedeqn.(3.27) (3.29)
3.15 Supply the reasoning leading from
f= fαβ˜ωαβ Schutz Eqs. (3.23) to
˜ωαβ(eμ,eν)= δαμδβν. Schutz Eq. (3.24)
Solution: This exercise provides a good opportunity to deepen one’s understand of tensor algebra. We proceed step-by-step:
f= fαβ˜ωαβ, what we mean by a basis fμν = f(eμ,eν), what we mean by components
fμν = fαβ ˜ωαβ(eμ,eν), sub first line into second line (3.30)
˜ωαβ(eμ,eν)= δαμδβν. solving above (3.31)
55 Exercises
The last step deserves a few words of explanation. While fμν = fαβδαμδβν is obviously a solution, how do we know this is the solution? Here is one way to convince yourself:
fμν = fαβAαβμν, re-writeeqn.(3.30) with Aαβμνas unknown
fσ γδσμδγν = fαβAαβμν, re-write LHS
fαβδαμδβν = fαβAαβμν, relabel dummy indices, see SP3.5 0= fαβ
δαμδβν− Aαβμν
. rearrange
But fαβ is arbitrary so the only way this last line above can always be true is for δαμδβν = Aαβμν.
3.17 (a) Suppose that h is a0
2
tensor with the property that, for any two vectors Aand B(where B 0),
h( , A)= α h( , B),
where α is a number that may depend on Aand B. Show that there exist one-forms ˜p and ˜q such that
h= ˜p ⊗ ˜q.
Solution: In general,
h( C, A)= hγβ CγAβ.
Treat C as an arbitrary vector. We were given that for arbitrary vectors Aand B, but with B 0
h( , A)= αh( , B). (3.32)
Suppose Cγ →
O (1, 0, 0, 0). Then
h( C, A)= αh( C, B)
h0βAβ = αh0μBμ (3.33)
˜q( A)= α ˜q( B). (3.34)
The LHS and RHS ofeqn.(3.33) have the form of a one-form contracted with vectors Aand Brespectively, so we wrote that explicitly ineqn.(3.34), defining qβ ≡ h0β. So far there is no restriction on ˜q; we simply choose
α= ˜q( A)
˜q( B), (3.35)
and we note the stipulation that B 0.
Now suppose Cγ →
O (1, 1, 0, 0). Then h( C, A)= αh( C, B)
h0βAβ + h1βAβ = α(h0μBμ+ h1μBμ) (3.36) h1βAβ = α(h1μBμ). subtractedeqn.(3.33) (3.37) Now α is no longer a free variable, being set byeqn.(3.33) or equivalentlyeqn.(3.35).
Botheqns.(3.33) and (3.37) can be satisfied if
a h0β = h1β (3.38)
for an arbitrary a since theneqn.(3.37) is simplyeqn.(3.33) multiplied on both sides by a. And this must be the unique solution as is obvious from considering the case where the vectors have only a single component, e.g. A →
O (0, 1, 0, 0) and B → (1, 0, 0, 0). For theneqns.(3.33) and (3.37) reduce to simple algebraic equations, e.g.O
h01 = αh00
h11 = αh10
⇒h01 h11 = h00
h10 = a−1. (3.39)
Repeating the argument above for different C, we find a different constant for each first index of h. In particular, with Cγ →
O (1, 0, 1, 0) we conclude that bh0β = h2β, for some constant b. With Cγ →
O (1, 0, 0, 1) we find ch0β= h3β. That is, if the tensor h is written as a matrix, the rows are arbitrary scalar constants pμ= (1, a, b, c) of the first row (h0ν)≡ qν. And so
hμβ= pμqβ (3.40)
and so the tensor h has the form of an outer product,
h= ˜p ⊗ ˜q. (3.41)
tensor provides a map of vectors to vectors and one-forms to one-forms.
Solution: You can think of a1
1
tensor as a machine with two input slots, a slot for one-forms and another for vectors. Inserting a one-form and a vector into their respective slots produces a real number. So once you’ve filled the one-form slot, this machine takes one more input, a vector, to produce a real. But a machine that takes a vector as input to produce a real is, by definition, a one-form. Now let’s go back to the1
1
tensor, the machine with two empty slots. We’ve just seen that the act of inserting the one-form converts the machine to one-form; so we can say a1
1
tensor is a map from one-forms to one-forms. By a similar argument, a1
1
tensor is also a map from vectors to vectors.
57 Exercises
It’s also satisfying to see how this plays out using the mathematical symbols. Write T acting on the vector alone:
T( ;v) = T( ; vβeβ) usedeqn.(2.4)
= Tνμvβ eν⊗ ˜ωμ( ;eβ) used Schutz Eq. (3.61)
= Tνμvβ ˜ωμ(eβ)eν applied basis to argument
= Tνμvβδμβ eν usedeqn.(3.91)
= Tνμvμ
eν. summed over β (3.42)
And now we are done because the RHS of eqn.(3.42) is the product of four reals
Tνμvμ
, i.e. one for each value of ν, times the four basis vectors,eν. This is a vector, cf.eqn.(2.4).
And similarly we can write T acting on the one-form alone:
T(˜p; ) = T(pα˜ωα; ) used Schutz Eq. (3.11)
= Tνμpαeν⊗ ˜ωμ(˜ωα; ) used Schutz Eq. (3.61)
= Tνμpα ˜ωμδαν usedeqn.(3.91)
= Tνμpν
˜ωμ. summed over α (3.43)
Again we are done here because the RHS is clearly a one-form, i.e. it is the product of four reals
Tνμpν
, one for each value of μ, and the set of four one-form bases ˜ωμ. So a1
1
tensor is a map from vectors to vectors in the sense that when you contract it with a vector the result is a vector. Likewise a1
1
tensor is a map from one-forms to one-forms because contracting it with a one-form results in a one-form.
3.19 (b) Derive the equation for the dot product of two one-forms:
˜p · ˜q = −p0q0+ p1q1+ p2q2+ p3q3. Schutz Eq. (3.53) (3.44)
Solution: To derive the formula for the inner product of one-forms in terms of components,eqn.(3.44), we start with the definition:
˜p · ˜q = 1
2[( ˜p + ˜q)2− ˜p2− ˜q2]. Schutz Eq. (3.52) (3.45) This involves only addition and squares of one-forms. The square of a one-form has been defined in component form ineqn.(3.83). The addition of one-forms was defined abstractly ineqns.(3.1,3.2), and in SP3.3we show that this implies that we just add the components; if˜s = ˜p + ˜q then sα = pα+ qα. So the square of the sum will be:
(˜p + ˜q)2= ˜s2= ηαβsαsβ, byeqn.(3.83)
= ηαβ(pα+ qα)(pβ+ qβ), component-wise addition
= ηαβ(pαpβ+ qαqβ+ pαqβ+ pβqα). components are just reals (3.46)
Substitutingeqn.(3.46) intoeqn.(3.45) and using{ηαβ} from Schutz Eq. (3.44):
˜p · ˜q = 1
2[( ˜p + ˜q)2− ˜p2− ˜q2]
= 1
2[ηαβ(pαpβ+ qαqβ+ pαqβ+ pβqα)− ηαβpαpβ− ηαβqαqβ],
= 1
2[ηαβ(pαqβ+ pβqα)] cancelled terms
= ηαβpαqβ ηαβsymmetric
= −p0q0+ p1q1+ p2q2+ p3q3, (3.47)
which iseqn.(3.44).
3.20 In Euclidean three-space in Cartesian coordinates, we do not normally distinguish between vectors and one-forms, because their components transform identically.
Prove this in two steps. Suppose we are in Euclidean three-space in Cartesian coordinates.
(a) Show that
A¯α = Λ¯αβAβ and
P¯β= Λα¯βPα (3.48)
are the same transformation if the matrix (Λ¯αβ)is equal to the transpose of its inverse. Such a matrix is said to be orthogonal.
Hint: Solution available in (Schutz,1985, Appendix B).
(b) The metric of such a space has components{δij, i, j = 1, . . . , 3}. Prove that a transformation from one Cartesian coordinate system to another must obey
δ¯i ¯j= Λk¯iΛl¯jδkl (3.49)
and that this implies (Λk¯i)is an orthogonal matrix. See Exercise 3.32 for the analog of this in SR.
Solution: All we are given is that the metric tensor for Cartesian three-space is {δij, i, j = 1, 2, 3}. The metric tensor is used in forming the inner product of vectors, which we know must be frame invariant. So write the inner product between two three-space vectors in two different frames as
59 Exercises
δ¯i ¯jA¯iB¯j= A· B= δklAkBl
= δkl(A¯iΛk¯i)(B¯jΛl¯j), (3.50) and so upon “cancelling” the A¯iB¯jon either side (allowed because these are arbitrary vectors) and rearranging we see that
δ¯i ¯j= Λk¯iΛl¯jδkl, (3.51)
as required. If you were uncomfortable with cancelling the A¯iB¯jon either side, see SP3.5.
Now to show that this implies an orthogonal transformation matrix, sum over l:
δ¯i ¯j= Λk¯iΛl¯jδkl, fromeqn.(3.51)
= Λk¯iΛk¯j. after summing over l (3.52) The RHS of eqn.(3.52) is the product of a matrix by its transpose, and for this to equal the identity matrix (i.e. the LHS), we require the matrix to be orthogonal.
And now you see clearly why we never learned about one-forms when working in Cartesians coordinates in Euclidean space, and why we called the gradient of a scalar field a vector.
3.23 (a) Prove that the set of allM
N
tensors for fixed M and N forms a vector space.
(You must define addition of such tensors and their multiplication by numbers.) Solution: This is like Exercise 3.2, but now we need to define what we mean by the addition of twoM
N
tensors and the multiplication of anM
N
tensor by a scalar. So we are guided byeqns.(3.5) and (3.6) above. That is, we note thatM
N
tensors produce real numbers that can be added like real numbers, so the generalization ofeqns.(3.1) and (3.2) is trivial. The tensor S where
S= P + Q (3.53)
is defined to be that which gives the sum of the two values obtained by applying the input to P and Q. That is,
S(˜a1,˜a2, . . . ,˜aM; b1, b2, . . . , bN)=
P(˜a1,˜a2, . . . ,˜aM; b1, b2, . . . , bN)+ Q(˜a1,˜a2, . . . ,˜aM; b1, b2, . . . , bN), (3.54) where the notation (˜a1,˜a2, . . . ,˜aM; b1, b2, . . . , bN)was used to represent the M one-form inputs and N vector inputs. The choice of one-one-forms first, we will see later, gives the basis in the order Schutz gave in Exercise 3.23(b). We have followed the convention that superscript integers are used as indices of different one-forms. That is, ˜a1 and ˜a2 are two different one-forms, not components of the same one-form.
Similarly, subscripts are used to denote different vectors. Some authors will use a different notation to avoid this ambiguity; Carroll (2004) uses parentheses around the indices, like˜a(1), to distinguish them from components.
In analogy with eqn. (3.2) we can define multiplication of an M
N
tensor by a scalar α
R= αP (3.55)
to be the tensor that, for a given input, gives just α times the real number produced by supplying the input to P:
R(˜a1,˜a2, . . . ,˜aM; b1, b2, . . . , bN)= αP(˜a1,˜a2, . . . ,˜aM; b1, b2, . . . , bN). (3.56) The set of M
N
tensors for fixed M and N forms a vector space by the same argument as given for Exercise 3.2. What do we mean by the zeroM
N
tensor? This is the tensor that gives zero for any input,
0(˜a1,˜a2, . . . ,˜aM; b1, b2, . . . , bN)= 0.
The set ofM
N
tensors, with addition defined byeqns. (3.53) and (3.54) and scalar multiplication defined byeqns.(3.55) and (3.56) then meets axiom (1) in Appendix A:
M
N
tensors form an abelian group with the operation of addition. Similarly the requirements of Axiom (2) in Appendix A are clearly met.
3.23 (b) Prove that a basis for the vector space formed from the set of allM
N
tensors for fixed M and N is the set:
{eα⊗ eβ⊗ · · · ⊗ eγ ⊗ ˜ωμ⊗ ˜ων⊗ · · · ⊗ ˜ωλ} (3.57) with M vectors labeled with α . . . γ and N one-forms labeled μ . . . λ.
Solution: This is a nice question because it forces us to think about what we mean by a basis. The answer is a straightforward generalization of the argument for the basis of the0
2
tensors starting after Schutz Eq. (3.22) and ending with Eq. (3.26) of §3.4.
The notation is cumbersome because one needs to refer to M superscripts and N subscripts where M and N are arbitrary positive integers. In defining the basis eqn. (3.57) above Schutz has used a series of Greek letters like α . . . γ . Here we put subscript indices on the Greek letters α1, α2, . . . αM to be explicit about how many there are. Remember that each Greek letter index can take on four values, e.g.
α1∈ {0, 1, 2, 3} corresponding to the four dimensions.
As in Schutz Eq. (3.23) we write theM
N
tensor as a sum of components times the basis that we seek:
R= Rα1,α2,...,αMβ1,β2,...,βN ˜ωα1,α2,...,αMβ1,β2,...,βN. (3.58) And furthermore, the components correspond to the real values produced by applying the tensor to arguments that are the basis one-forms and basis vectors. So,
61 Exercises
Rα1,α2,...,αMβ
1,β2,...,βN = R( ˜ωα1,˜ωα2, . . . ,˜ωαM;eβ1,eβ2, . . . ,eβN), (3.59) which is the generalization of the formula given between Schutz Eq. (3.23) and Eq. (3.24). Now, we simply substitute the tensoreqn.(3.58) intoeqn.(3.59) to obtain:
Rμ1,...,μMν
1,...,νN =Rα1,...,αM β1,...,βN˜ωα1,...,αM β1,...,βN(˜ωμ1, . . . ,˜ωμM;eν1, . . . ,eνN).
(3.60) This implies the analogue to Schutz Eq. (3.24),
˜ωα1,...,αMβ1,...,βN(˜ωμ1, . . . ,˜ωμM;eν1, . . . ,eνN)= δαμ11. . . δαμM
M δβν1
1. . . δβνN
N. (3.61) Usingeqn.(3.91) we identify
δβν11 = ˜ωβ1(eν1), δβν22 = ˜ωβ2(eν2), . . . δβνN
N = ˜ωβN(eνN). (3.62) Based upon the dualism between vectors and one-forms, we identify:
δαμ1
1 = eα1(˜ωμ1), δαμ2
2 = eα2(˜ωμ2), . . . δαμM
M = eαM(˜ωμM).
So focusing on just the tensor, i.e. dropping the arguments, we are left with the basis that is the analogue to Schutz Eq. (3.25),
So focusing on just the tensor, i.e. dropping the arguments, we are left with the basis that is the analogue to Schutz Eq. (3.25),