Documenti di Didattica
Documenti di Professioni
Documenti di Cultura
for
Manifolds, Tensors, and Forms
Paul Renteln
Department of Physics
California State University
San Bernardino, CA 92407
and
Department of Mathematics
California Institute of Technology
Pasadena, CA 91125
prenteln@csusb. edu
Contents
iii
1
Linear algebra
1.1 We have
But the ei ’s are a basis for V , so they are linearly independent, which implies
vi − vi = 0.
1.3 Let V = U ⊕ W , and let E := {ei }i=1 n
be a basis for U and F := { f j }mj=1 a
basis for W . Define a collection of vectors G := {gk }n+m k=1 where gi = ei for
1 ≤ i ≤ n and gn+i = f i for 1 ≤ i ≤ m. Then the claim follows if we can
show G is a basis for V . To that end, assume
n+m
n
m
0= ci gi = ci ei + ci f i .
i=1 i=1 i=1
The first sum in the rightmost expression lives in U and the second sum lives
in W , so by the uniqueness property of direct sums, each sum must vanish
by itself. But then by the linear independence of E and F, all the constants
ci must vanish. Therefore G is linearly independent. Moreover, every vector
v ∈ V is of the form v = u + w for some u ∈ U and w ∈ W , each of which
1
2 Linear algebra
can be written as a linear combination of the gi ’s. Hence the gi ’s form a basis
for V .
1.4 Let S be any linearly independent set of vectors with |S| < n. The claim is
that we can always find a vector v ∈ V so that S ∪ {v} is linearly independent.
If not, consider the sum
|S|
cv + ci si = 0,
i=1
T v1 = T v2 ⇒ T (v1 − v2 ) = 0 ⇒ v1 − v2 = 0 ⇒ v1 = v2 .
Assume the kernel of T consists only of the zero vector. Then for any two
vectors v1 and v2 , T (v1 − v2 ) = 0 implies v1 − v2 = 0, which is equivalent
to saying that T v1 = T v2 implies v1 = v2 , namely that T is injective. The
converse follows similarly.
1.8 Let V and W be two vector spaces of the same dimension, and choose a basis
{ei } for V and a basis { f i } for W . Let T : V → W be the map that sends ei to
f i , extended by linearity. Then the claim is that T is an isomorphism. Let v =
i ai ei be a vector in V . If v ∈ ker T , then 0 = T v = i ai T ei = i ai f i .
By linear independence, all the ai ’s vanish, which means that the kernel of
T consists only of the zero vector, and hence by Exercise 1.7, T is injective.
Also, if w = i ai f i , then w = i ai T ei = T i ai ei , which shows that T
is also surjective.
1.9 a. Let v ∈ V and define w := π(v) and u := (1 − π)(v). Then π(u) =
(π − π 2 )(v) = 0, so v = w + u with w ∈ im π and u ∈ ker π . Now
Linear algebra 3
Hence
(ST )i j = Sik Tk j = (ST )i j ,
k
1.17 The easiest way to see this is just to observe that the identity automorphism
I is represented by the identity matrix I (in any basis). Suppose T −1 is
represented by U in some basis. Then by the results of Exercise 1.16,
T T −1 → T U.
But T T −1 = I , so T U = I, which shows that U = T −1 .
1.18 Choose a basis {ei } for V . Then by definition,
T ej = Ti j ei .
i
Also,
g(v, u) = vi u i = u i vi = g(u, v).
i i
Assume g(u, v) = 0 for all v. Let v run through all the vectors v (i) =
(0, . . . , 1, . . . , 0), where the ‘1’ is in the i th place. Plugging into the defi-
nition of g gives u i = 0 for all i, so u = 0. Thus g is indeed an inner product.
The same proof works equally well for the Euclidean and Lorentzian inner
products.
Again consider the standard inner product on Cn . Then
g(u, u) = ui ui = |u i |2 ≥ 0,
i i
n−1
g(u, u) = −u 20 + u i2 ,
i=1
while
∗ ∗
e j , Aei = e j , Aki ek = Aki δ jk = A ji ,
k k
∗
so the matrix representing A is just the transpose of the matrix
representing A.
1.26 We have
† †
A e j , ei = (A )k j ek , ei = (A† )k j δki = (A† )i j ,
k k
while
∗ ∗
e j , Aei = e j , Aki ek = Aki δ jk = A ji ,
k k
which gives
(A† )i j = A ji .
1.27 Let w = i ai vi (where not all the ai ’s vanish) and suppose i ci vi + cw =
0. The latter equation may be solved by choosing c = 1 and ci = −ai , so the
set {v1 , . . . , vn , w} is linearly dependent. Conversely, suppose {v1 , . . . , vn , w}
is linearly dependent. Then the equations i ci vi + cw = 0 have a nontrivial
solution (c, c1 , . . . , cn ). We must have c = 0 else the set {vi } is not linearly
independent. But then w = − i (ci /c)vi .
1.28 Obviously, the monomials span V , so we need only check linear indepen-
dence. Assume
c0 + c1 x + c2 x 2 + c3 x 3 = 0.
Linear algebra 7
The zero on the right side represents the zero vector, namely the polynomial
that is zero for all values of x. In other words, this equation must hold for all
values of x. In particular, it must hold for x = 0. Plugging in gives c0 = 0.
Next let x = 1 and x = −1, giving c1 + c2 + c3 = 0 and −c1 + c2 −
c3 = 0. Adding and subtracting the latter two equations gives c2 = 0 and
c1 + c3 = 0. Finally, choose x = 2 to get 2c1 + 8c3 = 0. Combining this with
c1 + c3 = 0 gives c1 = c3 = 0.
1.29 We must show exactness at each space. Clearly the sequence is exact at ker T ,
because the inclusion map ι : ker T → V is injective, so only zero gets sent
to zero. By definition, the kernel of T is ker T , namely the image of ι, so
the sequence is exact at V . Let π : W → coker T be the projection map
onto the quotient W/ im T . Then by definition π kills everything in im T , so
the sequence is exact at W . Finally, π is surjective onto the quotient, so the
sequence is exact at coker T .
1.30 Write the exact sequence together with its maps
ϕ0 ϕ1 ϕn−1
0 −−−→ V0 −−−→ V1 −−−→ · · · −−−→ Vn −−−→ 0
and set ϕ−1 = ϕn = 0. By exactness, im ϕi−1 = ker ϕi . But the rank/nullity
theorem gives
Hence,
(−1)i dim Vi = (−1)i (dim ker ϕi + dim im ϕi )
i i
= (−1)i (dim im ϕi−1 + dim im ϕi )
i
= 0,
As each number from 1 to n appears precisely once among the set σ (1), σ (2),
. . . , σ (n), the product may be rewritten (after some rearrangement) as
Hence
−1
det A = (−1)σ Aσ −1 (1)1 Aσ −1 (2)2 . . . Aσ −1 (n)n . (4)
σ ∈Sn
det A. As there was nothing special about the choice i = 1, (1.57) is proved.
Equation (1.58) is proved similarly.
1.33 Suppose we begin with a matrix A and substitute for its i th row a new row of
elements labeled Bi j , where j runs from 1 to n. Now, the cofactors of the Bi j
in the new matrix are obviously the same as those of the Ai j in the old matrix,
so we may write the determinant of the new matrix as, for instance,
i1 + Bi2 A
Bi1 A i2 + · · · + Bin A
in . (1)
Again, a similar result holds for columns. (The cofactors appearing in (1) are
called alien cofactors, because they are the cofactors properly corresponding
to the elements Ai j , j = 1, . . . , n, of the i th row of A rather than the k th row.)
We may summarize (2) by saying that expansions in terms of alien cofactors
vanish identically.
Consider the ik th element of A(adj A):
n
n
[A(adj A)]ik = Ai j (adj A) jk = k j .
Ai j A
j=1 j=1
1.34 By (1.59),
and therefore
det A(i)
xi = .
det A
1.36 From (1.57),
∂ ∂ 11 + A12 A
12 + · · · ) = A
12 ,
(det A) = (A11 A
∂ A12 ∂ A12
because A12 only appears in the second term. A similar argument shows that,
in general,
∂ i j .
(det A) = A
∂ Ai j
But from (1.59), adj A = (det A) A−1 , so
i j = (adj A) ji = (det A)(A−1 ) ji .
A
1.37 a. If T is an automorphism then it is surjective. Hence its rank equals dim V .
b. If T is an automorphism then it is invertible. Suppose T −1 is represented
by the matrix S. Then I = T T −1 is represented by the matrix T S. But any
basis, the identity automorphism I is represented by the identity matrix I,
so T S = I, which shows that T is invertible, and hence nonsingular.
1.38 a. Suppose {vi } is an orthonormal basis. Then
g(Rvi , Rv j ) = g(vi , v j ) = δi j ,
whence we see that {Rvi } is again orthonormal. Conversely, if {T vi } is
orthonormal, then
g(T vi , T v j ) = δi j = g(vi , v j ).
If v = i ai vi and w = j b j v j then
g(T v, T w) = ai b j g(T vi , T v j ) = ai b j g(vi , v j ) = g(v, w),
ij ij
so T is orthogonal.
Linear algebra 11
and similarly,
Aej = ei Ai j ⇒ Ae = e A . (2)
i
1.46 a.
∞
(1 + t x j ) = (1 + t x1 )(1 + t x2 ) · · ·
j=1
b.
j
p j t j−1 = ( xi )t j−1
j j i
= xi (t xi ) j−1
i j
xi
= .
i
1 − t xi
c. We have
dE
= xj (1 + xk t),
dt j k= j
so
1 dE xj
= .
E dt j
1 + xjt
k
kek = (−1) j−1 ek− j p j .
j=1
14 Linear algebra
Now multiply the odd columns by −1 and the even rows by −1 to get
p1 1 0 · · · · · ·
p2 p1 2 0 · · ·
· · · · · · .
det A = p3 p2 p1
. .. .. ..
.. . . .
p pn−1 pn−2 · · · p1
n
and so
1
det A = (tr A)4 − 6(tr A)2 (tr A2 ) + 3(tr A2 )2
4!
+ 8(tr A)(tr A3 ) − 6 tr A4 .
1.47 This follows immediately from the results of Exercises 1.40, 1.43 and 1.44.
1.48 First assume A to be diagonalizable with eigenvectors e1 , e2 , . . . , en and
corresponding eigenvalues λ1 , λ2 , . . . , λn . Then
1 2 1 3
e ei = 1 + A + A + A + · · · ei ,
A
2 3!
1 2 1 3
= 1 + λi + λi + λi + · · · ei ,
2 3!
= eλi ei .
It follows that e A is diagonalizable with eigenvectors e1 , e2 , . . . , en and
corresponding eigenvalues eλ1 , eλ2 , . . . , eλn . [If we had not assumed diago-
nalizability, we could not say that we had gotten all the eigenvectors of det e A
this way.] The result now follows from Exercise 1.44, because
e i λi = eλi .
i
Next, suppose A is not necessarily diagonal. Because the trace and the
determinant are both similarity invariants we may assume, using Schur’s
theorem, that A = D + N, where D = diag(d1 , . . . , dn ). Observe that
A2 = ( D + N)2 = D2 + D N + N D + N 2 .
But D2 is diagonal and D N, N D, and N 2 are all strictly upper triangular, so
we can write
A2 = D2 + N ,
for some strictly upper triangular matrix N . By induction, it follows that
e A = e D + N ,
where N is some other strictly upper triangular matrix. The matrix on the
right is upper triangular, so by the Laplace expansion its determinant is just
the product of its diagonal elements. Thus,
det e A = edi .
i
But tr A = tr D, so
etr A = etr D = e i di
,
whereupon the claim follows.
16 Linear algebra
Finally,
∞
( f, f ) = f 2 (x) d x ≥ 0
−∞
c1 v1 + c2 v2 + . . . cn vn = 0. (1)
Take the inner product of (1) with each of the vectors vi to get
Regarding this as a set of linear equations for the constants c j , we see that the
Grammian must vanish.
Conversely, suppose the Grammian of {v1 , . . . , vn } is zero. Then the system
(2) has a nonzero solution. Multiplying the equations in (2) by each c j in
succession and then adding them all together gives
c1 v1 + c2 v2 + · · · + cm vm = 0
where v2 := g(v, v). Equation (1) now follows by virtue of the nondegen-
eracy of the inner product, so the vectors are linearly dependent.
1.52 Define vi = x i for i = 0, 1, 2, 3. Then
1
g(v0 , v0 ) = d x = 2,
−1
so
1
e0 := √ .
2
Next,
1
1 1 1 1
x 2
e1 = x − g x, √ ·√ =x− x dx = x − = x.
2 2 2 −1 4 −1
Thus,
1
1
x 3 2
g(e1 , e1 ) = g(x, x) = x dx = 2
= 3.
−1 3 −1
Hence
e1 3
e1 = = x.
g(e1 , e1 )1/2 2
Next we have
The last inner product vanishes by a simple parity argument, so we only need
to compute the second term, which is
1 1 2 1
e0 g(e0 , x ) =
2
x dx = .
2 −1 3
Thus,
1
e2 = x 2 − .
3
18 Linear algebra
Now we normalize.
The only inner product we haven’t done yet is the first, which is
1
2
g(x 2 , x 2 ) = x4 dx = .
−1 5
Hence
2 4 2 8
g(e2 , e2 ) = − + = ,
5 9 9 45
whereupon we obtain
e2 45 1
e2 = = x −
2
.
g(e2 , e2 )1/2 8 3
Lastly, we have
Again by parity we need only compute the third term on the right, which is
1
3 3
e1 g(e1 , x ) = x
3
x 4 d x = x.
2 −1 5
Thus,
3
e3 = x 3 − x.
5
The next step is to normalize. We have
6 9
g(e3 , e3 ) = g(x 3 −(3/5)x, x 3 −(3/5)x) = g(x 3 , x 3 )− g(x 3 , x)+ g(x, x).
5 25
Having done this many times by now, we can pretty much read off the answer:
2 12 6 8
g(e3 , e3 ) = − + = .
7 25 25 175
Hence
175 3
e3 = x3 − x .
8 5
1.53 For the first, we have, by the definition above,
f ∈ ker T ∗ ⇔ T ∗ f = 0 ⇔ f T = 0 ⇔ f ∈ Ann im T.
Linear algebra 19
Hence,
v fv = g(ei , ei ) f v,i ei = g(ei , ei )g(v, ei )ei = g(ei , ei )2 vi ei = v.
i i i
Also,
f v f (e j ) = g(v f , e j ) = g(ei , ei ) f i g(ei , e j ) = g(e j , e j )2 f j = f j ,
i
2.1 Let e and e be two bases related by the change of basis matrix A, so that
ei = ei Ai i .
i
Suppose Ti j = T ji . Then
T j i = Ai j A j i Ti j = Ai j A j i T ji = A j j Ai i Ti j = Ti j ,
ij ij ij
where in the penultimate step we changed the names of the dummy indices
from i and j to j and i, respectively. The antisymmetric case is similar and is
left to the reader.
2.2 We have
Ai j B i j = A ji B i j = − A ji B ji = − Ai j B i j = 0.
ij ij ij ij
20
Multilinear algebra 21
2.4 We have
a [i1 ...i p ] ei1 ∧ · · · ∧ ei p
i 1 ,...,i p
1
= (−1)σ a iσ (1) ...iσ ( p) ei1 ∧ · · · ∧ ei p (1)
p! i ,...,i
1 p σ ∈Sp
1 iσ (1) ...iσ ( p)
= a eiσ (1) ∧ · · · ∧ eiσ ( p) (2)
p! i ,...,iσ ∈Sp 1 p
1
= a iσ (1) ...iσ ( p) eiσ (1) ∧ · · · ∧ eiσ ( p) (3)
p!
σ ∈Sp i σ (1) ,...,i σ ( p)
1 i1 ...i p
= a ei1 ∧ · · · ∧ ei p (4)
p!
σ ∈Sp i 1 ,...,i p
= a i1 ...i p ei1 ∧ · · · ∧ ei p . (5)
i 1 ,...,i p
Equality (1) is just the definition (2.21), while in (2) we have used (2.39)
and flipped the order of summation. In (3) we have changed the name of the
dummy indices from i 1 , . . . , i p to i σ (1) , . . . , i σ ( p) , and then in (4) we have
changed the dummy indices back to i 1 , . . . , i p . Finally, (5) holds because S p
has p! elements.
2.5 By linearity is suffices to prove Property (3) for two monomials. So let λ =
v1 ∧· · ·∧v p and μ = w1 ∧· · ·∧wq . Then by moving the vectors vi successively
through the wi ’s (of which there are q), we get
μ ∧ λ = w1 ∧ · · · ∧ wq ∧ v1 ∧ · · · ∧ v p
= (−1)q v1 ∧ w1 ∧ · · · ∧ wq ∧ v2 ∧ · · · ∧ v p
= (−1)2q v1 ∧ v2 ∧ w1 ∧ · · · ∧ wq ∧ v3 ∧ · · · ∧ v p
..
= .
= (−1) pq v1 ∧ · · · ∧ v p ∧ w1 ∧ · · · ∧ wq
= (−1) pq λ ∧ μ.
2.6 By multilinearity is suffices to prove everything for monomials. So, let T =
e1 ⊗ · · · ⊗ e p , say, and define an action of an element σ ∈ S p on tensors
by T σ := eσ (1) ⊗ · · · ⊗ eσ ( p) , extended by linearity. Note that, by definition,
(T σ )τ = T τ σ . With this notation we have
1
alt(T ) = (−1)σ T σ .
p!
σ ∈Sp
22 Multilinear algebra
Therefore
2
1 σ σ 1
alt(alt(T )) = (−1) (alt(T )) = (−1)σ (−1)τ (T τ )σ
p! p!
σ ∈Sp σ ∈Sp τ ∈Sp
2 2
1 1
= (−1)σ τ T σ τ = (−1)π T π
p! p!
σ ∈Sp τ ∈Sp −1σ ∈Sp σ π∈Sp
1 2 π π 1
= (−1) T = (−1)π T π = alt(T ).
p! p!
σ ∈Sp π∈Sp π∈Sp
The key point is that S p+q naturally decomposes into pieces, and each term
in the sum vanishes on each of the pieces. Specifically, contained in S p+q is
a subgroup isomorphic to S p that permutes the first p numbers and leaves
the remaining q numbers fixed. For all such permutations, the right side of
(1) vanishes by the hypothesis alt(S) = 0 (because the sign of such a permu-
tation just equals the sign of the permutation of the first p numbers, and the
remaining q vectors pull out of the sum).
To show the rest of the terms vanish, we need a little bit of group theory. If
G is a group and H a subgroup, a right coset of H in G is any subset of the
form H g for g ∈ G. The right cosets partition G. (Proof. If x ∈ H g1 ∩ H g2
then x = h 1 g1 = h 2 g2 for some h 1 , h 2 ∈ H . So g1 = hg2 for h = h −1 1 h2 ∈
H , which shows that H g1 = H g2 . In other words, if two cosets are not
disjoint they coincide.)
Returning to our problem, let {H τ1 , H τ2 , . . . , H τk } be a partition of S p+q
into right cosets of H = S p , the subgroup permuting the first p numbers.
Multilinear algebra 23
Then
1 k
alt(S ⊗ T ) = (−1)σ vσ (1) ⊗ vσ (2) ⊗ · · · ⊗ vσ ( p+q)
( p + q)! i=1 σ ∈H τ
i
1
k
= (−1)τi (−1)π vπ τi (1) ⊗ vπ τi (2) ⊗ · · · ⊗ vπ τi ( p+q) .
( p + q)! i=1 π∈H
A moment’s thought shows that, for each i, the inner sum vanishes for pre-
cisely the same reason as before, because the only effect of τi is to renumber
the indices. A similar argument shows that T ∧ S = 0.
By multilinearity,
alt(R ∧ S − R ⊗ S) = 0.
R ∧ (S ∧ T ) = alt(R ⊗ S ⊗ T ),
whereupon we conclude that the wedge product defined by alt is indeed asso-
ciative. Wow. All that just to prove a fact that is obvious when viewed from
the axiomatic perspective. Well, chacun à son gout.
2.7 We have
3
T (e1 ∧ e2 ∧ e3 ) = T e1 ∧ T e2 ∧ T e3
= (e1 + 2e2 ) ∧ (3e2 + 2e3 ) ∧ (e1 + e3 )
= (e1 + 2e2 ) ∧ (3e2 ∧ e1 + 3e2 ∧ e3 + 2e3 ∧ e1 )
= 7e1 ∧ e2 ∧ e3 .
2.8 Pick a basis {e1 , . . . , en }, and suppose that T is represented by T in that basis.
Then (ignoring index placement, as it is irrelevant here),
( n T )e1 ∧ · · · ∧ en = T e1 ∧ · · · ∧ T en
= Ti1 1 · · · Tin n ei1 ∧ · · · ein
i 1 ,...,i n
= (−1)σ Tσ (1)1 · · · Tσ (n)n e1 ∧ · · · ∧ en ,
σ ∈Sn
because the only terms contributing to the sum are permutations of {1, . . . , n},
and by definition of the sign,
2.9 We have
2
( T )(ei ∧ e j ) = T ei ∧ T e j
= Tki ek ∧ Tj e
k
= Tki Tj (ek ∧ e )
k
= Tki Tj − Ti Tk j (ek ∧ e ).
k<
For example,
( 2 T )(e1 ∧ e2 ) = (T11 T22 − T21 T12 )(e1 ∧ e2 )
+ (T11 T32 − T31 T12 )(e1 ∧ e3 )
+ (T21 T32 − T31 T22 )(e2 ∧ e3 ).
Let T (2) denote the matrix representation of the operator 2 T , and arrange
the basis elements of 2 V in lexicographic order: e1 ∧ e2 , e1 ∧ e3 , e2 ∧ e3 .
Then similar calculations reveal that
⎛ ⎞
T11 T22 − T21 T12 T11 T23 − T21 T13 T12 T23 − T22 T13
⎝T11 T32 − T31 T12 T11 T33 − T31 T13 T12 T33 − T32 T13 ⎠ .
T21 T32 − T31 T22 T21 T33 − T31 T23 T22 T33 − T32 T23
2.10 With the setup of the hint, we have
det T = det T = ( n T )(e1 ∧ · · · ∧ en ) = T e1 ∧ · · · ∧ T en = v1 ∧ · · · ∧ vn .
It is now painfully obvious from the properties of wedge products that, (1)
swapping two columns of T flips the sign of the determinant, (2) setting two
Multilinear algebra 25
v1 ∧ · · · ∧ (vi + λv j ) ∧ · · · ∧ v j ∧ · · · ∧ vn
= v1 ∧ · · · ∧vi ∧ · · · ∧v j ∧ · · · ∧ vn +λv1 ∧ · · · ∧v j ∧· · · ∧v j ∧· · · ∧vn
= v1 ∧ · · · ∧ vi ∧ · · · ∧ v j ∧ · · · ∧ vn ,
and (4) multiplying a column vector by a scalar multiplies the entire deter-
minant by that scalar. The corresponding statements with the word ‘column’
replaced by the word ‘row’ follow by appealing to (1.56).
2.11 Assume the same setup as in the proof of (2.60). By linearity we may assume
η = e I . Now λ = K a K e K , but g(e I , e K ) = 0 unless K = I , so we may as
well assume λ = e I . Then using (2.63) and (2.65) we have
The other equality follows from the symmetry of the inner product.
2.12 Complete {v1 , . . . , vk } to a basis {v1 , . . . , vk , vk+1 , . . . , vn }. Then
n
wi = ai j v j ,
j=1
k
n
k
n
0= vi ⊗ ai j v j = ai j vi ⊗ v j .
i=1 j=1 i=1 j=1
It follows that
(A ⊗ B)k,i j = Aki Bj .
26 Multilinear algebra
But this is just ( A ⊗ B)k,i j , as one can see by unpacking the definition of the
Kronecker product.
2.14 Assume v1 , v2 , . . . , v p are linearly dependent. Then there exist constants, not
all zero, such that
c1 v1 + · · · c p v p = 0.
By renumbering the vectors if necessary, we may take c p = 0. Then
1
vp = − (c1 v1 + · · · + c p−1 v p−1 ).
cp
By the multilinearity and antisymmetry properties of the wedge product, the
expression v1 ∧ · · · ∧ v p is a sum of terms, each of which involves the wedge
product of two copies of the same vector, so it must vanish.
Conversely, suppose v1 , v2 , . . . , v p are linearly independent. Then they
form a basis for the p dimensional subspace W ⊆ V that they span. The
p-vector v1 ∧ · · · ∧ v p is a basis for the one dimensional space p W , and
therefore cannot vanish.
2.15 Following the hint, let {v1 , v2 , . . . , v p , v p+1 , . . . , vn } be a basis of V . Since
any vector can be expanded in terms of the basis, we can write
p
n
wi = Ai j v j + Bi j v j
j=1 j= p+1
p
n p
= Ai j (vi ∧ v j ) + Bi j (vi ∧ v j ).
i, j=1 i=1 j= p+1
Each term on the right side must vanish separately, because they involve
linearly independent bivectors. The first term can be written
(Ai j − A ji )(vi ∧ v j ),
1<i< j< p
from which it follows that Ai j = A ji . The only way the second term can
vanish is if Bi j = 0 for all i and j for which it is defined.
2.16 To compute the determinant of p A, we compute its eigenvalues and take
their product. The map p A acts on the vector space p V consisting of all
p forms of V . As A is assumed diagonal, we may choose a basis {e1 , . . . , en }
for V consisting of eigenvectors of A. Set Aei = λi ei . Then
Multilinear algebra 27
p
( A)(ei1 ∧ ei2 ∧ · · · ∧ ei p ) = Aei1 ∧ Aei2 ∧ · · · ∧ Aei p
= λi1 λi2 · · · λi p (ei1 ∧ ei2 ∧ · · · ∧ ei p ). (1)
p p
In other words, the eigenvectors of A acting on V are products of
eigenvalues of A whose indices range over all the p subsets of {1, 2, . . . , n}.
Multiplying all these eigenvalues of p A together, the only question is
how often each individual λ appears. Let’s do a simple example. Suppose
p = 3 and n = 5. Then the following subsets appear:
123, 124, 125, 134, 135, 145, 234, 235, 245, 345.
How many times does the ‘1’ appear? Well, to construct the subsets
contain-
ing 1, we first chose 1, and then the rest of the numbers in 3−1 = 42 = 6
5−1
ways. This holds in general, so each index appears in the product n−1
p−1
times,
which means that
det( p A) = (det A)( p−1) .
n−1
2.17 As in Example 2.4 we get (using the fact that d = 1 for a Euclidean signature)
ei = (−1)i−1 e1 ∧ · · · ∧ ei ∧ · · · ∧ en ,
where the hat means that ei is omitted. Thus
μ = (−1)i−1 ai e1 ∧ · · · ∧ ei ∧ · · · ∧ en .
i
b. Define
B := I + z A.
From the definition of the determinant,
n
B(e1 ∧ e2 ∧ · · · ∧ en ) = (det B)e1 ∧ e2 ∧ · · · ∧ en .
Thus,
n
tr B = det B,
because the trace just sums all the components of the linear operator that
map the basis elements to themselves. (Basically, the trace counts the
‘multiplicity’ of the fixed points of the action of the linear operator on
all the basis elements.) But, from the definition,
n
B(e1 ∧ e2 ∧ · · · ∧ en ) = Be1 ∧ Be2 ∧ · · · ∧ Ben
= (I +z A)e1 ∧ (I +z A)e2 ∧ · · · ∧ (I +z A)en .
Now we are finished, because the coefficient of z r consists of a sum
of all possible (ordered) wedge products of r Aei ’s and n − r ei ’s. More
precisely, each term in the sum involves the action of r A on all possible
r
basis elements of V . Adding up all these contributions yields exactly
r
tr A.
We can see this explicitly when n = 3:
n
B(e1 ∧ e2 ∧ e3 )
= (I + z A)e1 ∧ (I + z A)e2 ∧ (I + z A)e3
= e1 ∧ e2 ∧ e3
+ z[Ae1 ∧ e2 ∧ e3 + e1 ∧ Ae2 ∧ e3 + e1 ∧ e2 ∧ Ae3 ]
+ z 2 [Ae1 ∧ Ae2 ∧ e3 + Ae1 ∧ e2 ∧ Ae3 + e1 ∧ Ae2 ∧ Ae3 ]
Multilinear algebra 29
where the straight bracket symbol means that the term is omitted from
the action of the exterior algebra operator. Here we are using the fact
that, for example,
2
A(e1 ∧ e2 ) = α12 (e1 ∧ e2 ) + . . .
2
A(e1 ∧ e3 ) = α13 (e1 ∧ e3 ) + . . .
2
A(e2 ∧ e3 ) = α23 (e2 ∧ e3 ) + . . . ,
and so
2
tr A = α12 + α13 + α23 .
2.21 a. The question asks for the number of multisets of size p chosen from a
set of size n, where a multiset is just a set where repetitions are allowed.
For example, {1, 1, 1, 2, 2, 4} is a multiset of size 6 chosen from a set
of 4 objects, corresponding to the basis element e1 e1 e1 e2
e2 e4 . The slickest way to count these objects is to use a “stars and
bars” argument. We observe that we can represent the multiset above by
the following picture:
∗ ∗ ∗| ∗ ∗||∗,
so
tr Symk A = λi1 · · · λik = Hk (λ1 , . . . , λn ).
i 1 ≤i 2 <···≤i k
2.22 Let {vi } be a basis for H, so that {vi1 · · ·vi p } with 1 ≤ i 1 · · · ≤ i p ≤ dim V
is a basis for Sym p H and {vi1 ∧ · · · ∧ vi p } with 1 < i 1 · · · < i p < dim V is a
basis for p H.
(i).
a (u)
a (v)v1 · · · v p
p
=
a (u) (v, vi )v1 · · ·
vi · · · v p
i=1
p
p−1
= (v, vi ) (u, v j )v1 · · ·
vi · · ·
v j · · · vp.
i=1 j=1
j=i
a † (u)
a † (v)v1 · · · v p =
a † (u)v v1 · · · v p
= u v v1 · · · v p .
a (u)
a † (v)v1 · · · v p =
a (u)v v1 · · · v p
= (u, vi )v v1 · · ·
vi · · · v p ,
i
Multilinear algebra 31
where the prime indicates that the sum includes the case in which vi = v.
On the other hand,
a † (v)
a (u)v1 · · · v p =
a † (v) (u, vi )v1 · · ·
vi · · · v p
i
= (u, vi )v v1 · · ·
vi · · · v p .
i
Therefore,
[
a (u),
a † (v)]− v1 · · · v p = (u, v)v1 · · · v p ,
from which we conclude that
[ a † (v)]− = (u, v)
a (u), 1.
(iv).
b(u)
b(v)v1 ∧ · · · ∧ v p
=b(u) (−1)i (v, vi )v1 ∧ · · · ∧
vi ∧ · · · ∧ v p
i
= (v, vi )(u, v j )(−1)i+ j v1 ∧ · · · ∧
vj ∧ · · · ∧
vi · · · ∧ v p
j<i
+ (v, vi )(u, v j )(−1)i+ j−1 v1 ∧ · · · ∧
vi ∧ · · · ∧
v j · · · ∧ vp.
j>i
(1)
Now change the names of the dummy indices in both terms from i and j
to j and i, respectively, to get
b(u)
b(v)v1 ∧ · · · ∧ v p
= (v, v j )(u, vi )(−1)i+ j v1 ∧ · · · ∧
vi ∧ · · · ∧
v j · · · ∧ vp
i< j
+ (v, v j )(u, vi )(−1)i+ j−1 v1 ∧ · · · ∧
vj ∧ · · · ∧
vi · · · ∧ v p .
i> j
(2)
If we now flip u and v in (2) and add it to (1) we get zero because the
terms in the sums cancel pairwise. Hence, [
b(u),
b(v)]+ = 0.
(v).
b † (u)
b † (v)v1 ∧ · · · ∧ v p =
b † (u)v ∧ v1 ∧ · · · ∧ v p
= u ∧ v ∧ v1 ∧ · · · ∧ v p .
Obviously, [
b † (u),
b † (v)]+ = 0.
32 Multilinear algebra
(vi).
b(u)
b † (v)v1 ∧ · · · ∧ v p =
b(u)v ∧ v1 ∧ · · · ∧ v p
= (−1)i (u, vi )v ∧ v1 ∧ · · · ∧
vi ∧ · · · ∧ v p ,
i
(3)
where the prime means that the sum includes the case i = 0 correspond-
ing to vi = v. On the other hand,
b † (v)
b(u)v1 ∧ · · · ∧ v p =
b † (v) (−1)i (u, vi )v1 ∧ · · · ∧
vi ∧ · · · ∧v p
i
= (−1)i (u, vi )v ∧ v1 ∧ · · · ∧
vi ∧ · · ·∧ v p .
i
(4)
If we add (3) and (4) all the terms cancel except the i = 0 term in (3), so
[
b(u),
b † (v)]+ v1 ∧ · · · ∧ v p = (u, v)v1 ∧ · · · ∧ v p ,
whereupon we conclude that [ b † (v)]+ = (u, v)
b(u), 1.
3
Differentiation on manifolds
3.1 For every point y ∈ Y let U (y) ⊂ Y be the open set whose existence is guar-
anteed by the hypothesis. Then the claim is that Y = ∪ y U (y), which implies
that Y is open in X . Clearly, ∪ y U (y) ⊆ Y . But if y ∈ Y then y ∈ U (y), so
Y ⊆ ∪ y U (y).
3.2 The first two properties hold by virtue of de Morgan’s laws.
1. Let C be a collection of closed sets. Then
C= C,
C∈C C∈C
which is open in X .
2. Let C be a finite collection of closed sets. Then
C= C,
C∈C C∈C
which is open in X .
3. The empty set and X are both open, so they are also both closed.
3.3 Let U be a neighborhood of y. If U ∩ Y = ∅ then there is a closed set, namely
U , that contains Y but does not contain y, so y ∈ cl Y . Conversely, suppose
y ∈ cl Y , and let C be a closed set containing Y that does not contain y. Then
C is a neighborhood of y that misses Y .
3.4 Let Y be closed and suppose x is an accumulation point of Y . If x ∈ Y then
x ∈ cl(Y − {x}) = cl Y = Y , a contradiction. Conversely, suppose Y contains
all its accumulation points, and let x ∈ Y . Then x is not an accumulation
point, so x ∈ cl(Y − {x}) = cl Y . By the previous exercise, this means there
is an open neighborhood of x that does not meet Y , so Y must be open (being
the union of open sets). Therefore, Y is closed.
33
34 Differentiation on manifolds
3.13 We construct f as follows. Let [x] ∈ Y and choose some x ∈ [x]. Define
f ([x]) := g(x). This is well defined, because if x is any other point in [x]
then by hypothesis, g(x ) = g(x). Moreover, it is determined uniquely by g.
Lastly, if U is open in Z then by the continuity of g, g −1 (U ) = π −1 ◦ f −1 (U )
is open in X . So by the definition of the quotient topology, f −1 (U ) is open in
Y , so f is continuous.
3.14 It suffices to show that the Jacobian matrix equals the matrix ( f i j ) represent-
ing f itself relative to the standard bases. By linearity,
f (x) = f ( x jej) = ei f i j x j ,
j ij
so the i th coordinate function is f i (x) = j f i j x j . Taking partial derivatives
gives
∂fi
= f i j.
∂x j
3.15 Following the hint, we observe that f ◦ f −1 = id implies that (D f )(x) ·
(D f −1 )(y) = I , which shows that (D f −1 )(y) = (D f (x))−1 . Well, ok, that
skips a bunch a steps. To fill in a few of those steps, we have to show that the
chain rule applied to multivariate functions just gives the product of the Jaco-
bian matrices, and that the derivative of the identity map (as a multivariate
function) is the identity map (as a linear operator). For the first, we have
∂( f ◦ g)i ∂fi ∂g k
[(D( f ◦ g))(x)]i j = = k (g(x)) · (x)
∂x j ∂x ∂x j
= [(D f )(g(x))]ik [(Dg)(x)]k j = [(D f )(g(x)) · (Dg)(x)]i j .
(ϕ8 ◦ ϕ2−1 )(1/6, 1/2) = (7/6, 1/2). In general, (ϕ8 ◦ ϕ2−1 )(x, y) = (x + 1, y).
In fact, we see that all the transition functions are either the identity map or
else a translation through 1 in one of the coordinate directions. Therefore,
the Jacobians all equal +1, and the torus is seen to be a smooth orientable
manifold.
3.17 First we observe that every point of RPn is in some Ui , because the zero vector
is not in RPn , so at least one entry in [x 0 , x 1 , . . . , x n ] must be nonzero. We
must show that the charts are compatible. To do this, we show that ϕi ◦ ϕ −1 j
is a diffeomorphism of ϕi (Ui ∩ U j ) and ϕ j (Ui ∩ U j ). Let p ∈ Ui ∩ U j . By
construction, p = [x 0 , x 1 , . . . , x n ] where x i = 0 and x j = 0. We have
x0 x1 x i−1 x i+1 xn
ϕi ( p) = , , . . . , , . . . ,
xi xi xi xi xi
and
x0 x1 x j−1 x j+1 xn
ϕ j ( p) = , , . . . , , . . . , .
xj xj xj xj xj
ϕ −1
j (y , . . . , y ) = (y , . . . , !"#$
1 n 1
1 , . . . , yn )
j th position
Let’s verify that this is indeed an inverse. One direction is trivial. We have
(ϕ j ◦ ϕ −1
j )(y , . . . , y ) = ϕ j (y , . . . , !"#$
1 n 1
1 , . . . , yn )
j th position
= (y 1 , . . . , y n ).
(The caret means that the term is excised.) This is clearly a bijection from
φ j (Ui ∩ U j ) to φi (Ui ∩ U j ).
To show ϕi ◦ ϕ −1 −1
j is smooth, we proceed as follows. If i = j then ϕi ◦ ϕ j
is just the identity map, so there is nothing to prove. (The identity map is
always a diffeomorphism.) So we examine the cases in which i = j. For
concreteness, let us choose i = 0 and j = 1, as the other cases are similar.
Then, by the above discussion, we have (with x 0 = 0 and x 1 = 0)
0 2 1 2
−1 x x xn x x xn
(ϕ0 ◦ ϕ1 ) , ,..., 1 = , ,..., 0 .
x1 x1 x x0 x0 x
That is,
1 y2 yn
(ϕ0 ◦ ϕ1−1 )(y 1 , y ,..., y ) =
2 n
, , . . . , ,
y1 y1 y1
It has determinant −1/(y 1 )n+1 . If n is even, this varies with the sign of y 1 , so
no consistent orientation is possible. If n is odd, this is negative for every pair
of overlaps, so a consistent choice of orientation is possible. We conclude that
if n is odd, RPn is orientable, but if n is even we cannot say anything (because
we do not know if some other choice of coordinates would make the manifold
orientable). It turns out, in fact, that for n even, the manifold is not orientable.
38 Differentiation on manifolds
∂x ∂y ∂z
= sin θ cos φ = sin θ sin φ = cos θ
∂r ∂r ∂r
∂x ∂y ∂z
= r cos θ cos φ = r cos θ sin φ = −r sin θ
∂θ ∂θ ∂θ
∂x ∂y ∂z
= −r sin θ sin φ = r sin θ cos φ = 0,
∂φ ∂φ ∂φ
gives
∂ ∂ ∂ ∂
= sin θ cos φ + sin θ sin φ + cos θ
∂r ∂x ∂y ∂z
∂ ∂ ∂ ∂
= r cos θ cos φ +r cos θ sin φ −r sin θ
∂θ ∂x ∂y ∂z
∂ ∂ ∂
= −r sin θ sin φ +r sin θ cos φ .
∂φ ∂x ∂y
[X, Y ]( f g) = (X Y )( f g) − (Y X )( f g)
= X (gY f + f Y g) − Y (g X f + f Xg)
= (Y f )(Xg) + g(X Y f ) + (Y g)(X f ) + f (X Y g)
− (X f )(Y g) − g(Y X f ) − (Xg)(Y f ) − f (Y Xg)
= g[X, Y ] f + f [X, Y ]g,
because the terms with the mixed partial derivatives of f cancel. Hence
∂Y j ∂X j ∂
[X, Y ] = X i i − Y i i ,
∂x ∂x ∂x j
as desired.
c.
∂xk ∂y j
ai (x) = ak (x) = ak (x)δik = ai (x).
∂y j ∂xi
Similarly we obtain the identity b j (y) = b j (y) if we plug (3.38) into (3.37),
so the two formulae are inverses of one another.
40 Differentiation on manifolds
and
3.26
F = −E x dt ∧ d x − E y dt ∧ dy − E z dt ∧ dz
+ Bx dy ∧ dz + B y dz ∧ d x + Bz d x ∧ dy,
J = −ρ d x ∧ dy ∧ dz + Jx dt ∧ dy ∧ dz
+ Jy dt ∧ dz ∧ d x + Jz dt ∧ d x ∧ dy.
42 Differentiation on manifolds
we get
∂ρ ∂ Jx ∂ Jy ∂ Jz
dJ = − − − − dt ∧ d x ∧ dy ∧ dz.
∂t ∂x ∂y ∂z
Taking ‘d’ of the Maxwell’s equation with sources gives
0 = d 2 F = 4πd J,
whence we obtain the law of charge conservation.
3.30 a. Let X = X i (∂/∂ x i ) (implicit sum). Then, as
∂f
df = dxi ,
∂xi
property (2) of the interior product and the linearity of the dual pairing,
yield
∂f
i X (d f ) = d f (X ) = d f, X = X i= X f.
∂xi
b. The first thing to observe is that the dual pairing of a one form ω and a
vector field X is function linear:
ω, f X = f ω, X ,
which follows immediately from its definition. Hence, on one forms,
i f X ω = ω, f X = f ω, X = f i X ω
Now let η be a p − 1 form, so that λ = ω ∧ η is a p form. Then
i f Xλ = i f Xω ∧ η − ω ∧ i f Xη
= ( f i X ω) ∧ η − f ω ∧ i X η
= f i X λ,
where the second step follows by the induction hypothesis. As any p
form can be written as a sum of monomials, the claim is proved.
c. Let ω ∧ η be a p-form. Then by Property (iii) of the interior product,
i X i Y (ω ∧ η) = i X (i Y ω ∧ η + (−1)deg ω ω ∧ i Y η)
= i X i Y ω ∧ η + (−1)deg ω − 1 i Y ω ∧ i X η
+ (−1)deg ω [i X ω ∧ i Y η + (−1)deg ω ω ∧ i X i Y η]
= i X iY ω ∧ η + ω ∧ i X iY η
+ (−1)deg ω [i X ω ∧ i Y η − i Y ω ∧ i X η].
By induction on p, the first two terms of this expression are antisym-
metric under the interchange of X and Y . As the last term is manifestly
Differentiation on manifolds 43
for some constants ai , not all of which are zero. By multilinearity, we have
n−1
ω(X 1 , X 2 , . . . , X n−1 , X n ) = ai ω(X 1 , X 2 , . . . , X n−1 , X i ).
i=1
3.33 We compute
φ ∗ θ = −(u 3 cos v)(3u 2 sin v du + u 3 cos v dv) ∧ (2u du)
− (u 3 sin v)(2u du) ∧ (3u 2 cos v du − u 3 sin v dv)
+ 2(u 2 )(3u 2 cos v du − u 3 sin v dv) ∧ (3u 2 sin v du + u 3 cos v dv)
= (2u 7 cos2 v + 2u 7 sin2 v + 6u 7 cos2 v + 6u 7 sin2 v)du ∧ dv
= 8u 7 du ∧ dv.
3.34 We have
φ ∗ ω = 4v du + 2uv dv + 2u du + 2v dv = 2(u + 2v) du + 2v(u + 1) dv,
so
dφ ∗ ω = 4 dv ∧ du + 2v du ∧ dv = (2v − 4) du ∧ dv.
On the other hand,
dω = 4 dy ∧ d x + 2y d x ∧ dy = (2y − 4) d x ∧ dy,
so
φ ∗ dω = (2v − 4) du ∧ dv,
and dφ ∗ ω = φ ∗ dω.
3.35 Let X = ∂/∂ x i . Then
∗ ∂(g ◦ f ) ∂g ∂ f j ∂ y j ∂g
X p ( f g) = = = .
∂ xi p ∂ y j f ( p) ∂ x i p ∂ x i p ∂ y j f ( p)
Also,
∂
( f ∗ X p ) f ( p) (g) = f∗ i (g),
∂x f ( p)
so we conclude
∂ ∂ y j ∂
f∗ i = .
∂ x f ( p) ∂ x i p ∂ y j f ( p)
But recall that the components of the matrix representing the linear trans-
formation A : U → V relative to the bases {ei } of U and { f i } of V are
defined by
Aei = A ji f j .
It follows that the matrix representing f ∗ relative to the local bases ∂/∂ x i and
∂/∂ y j has components ( f ∗ ) ji = (∂ y j /∂ x i ) p , whence the conclusion follows.
Differentiation on manifolds 45
L X Y = X Y − Y X = [X, Y ].
[L X , LY ]Z = L X LY Z − LY L X Z
= L X [Y, Z ] − LY [X, Z ]
= [X, [Y, Z ]] − [Y, [X, Z ]
= [X, [Y, Z ]] + [Y, [Z , X ]]
= [[X, Y ], Z ]
= L[X,Y ] Z .
3.38 Bisect the circle by the line L = {(x, 0) ∈ R2 and let U = S 1 − {N } and
V = S 1 − {S}, where the ‘north pole’ is N = (0, 1) and the ‘south pole’ is
S = (0, −1). Let ϕ be the projection onto L from N and ψ be the projection
onto L from S. Let ϕ(P) = Q and let the coordinate function on U be u. The
parametric line from N to Q ∈ L intersects the circle at
We have
2
t 2 u 2 + (1 − t)2 = 1 ⇒ t= ,
1 + u2
so
2u 1 − u2
xP = and yP = − .
1 + u2 1 + u2
Projecting from the south pole instead gives
2v 1 − v2
xP = and yP = .
1 + v2 1 + v2
where v is the coordinate function on V . Equating x P and y P in both
coordinate systems gives
1
v(u) = (ψ ◦ ϕ −1 )(u) = .
u
This is clearly a diffeomorphism on the overlap (where u = 0), and the
Jacobian of the transformation is −1/u 2 < 0, so the circle is orientable.
Differentiation on manifolds 47
3.39 The patches Ui × V j clearly cover M × N , so we need only show that the
coordinate maps are compatible. First we observe that
(ϕi × ϕ j )−1 (x, y) = (ϕi−1 (x), ϕ −1
j (y)).
Also,
1 y 1
dφ = − 2 d x + dy
1 + (y/x)2 x x
1
= 2 (−y d x + x dy).
x + y2
Note that
* *
sin θ = 1 − cos2 θ = 1 − (z/r )2 .
− (x 2 + y 2 )(x dy ∧ dz + y dz ∧ d x)]
1
= 3 (x dy ∧ dz + y dz ∧ d x + z d x ∧ dy),
r
as advertised.
b. We first compute σU . To avoid drowning in superscripts, let a := u 1 and
b := u 2 . Then
2a 2b η−1
x= , y= , and z= ,
1+η 1+η η+1
with η = a 2 + b2 , so
2(1 + η) − (2a)(2a) (2a)(2b)
dx = da − db
(1 + η) 2 (1 + η)2
2
= [(1 + b2 − a 2 ) da − 2ab db],
(1 + η)2
(2b)(2a) 2(1 + η) − (2b)(2b)
dy = − da + db
(1 + η) 2 (1 + η)2
2
= [−2ab da + (1 + a 2 − b2 ) db],
(1 + η)2
and
2a(η + 1) − (η − 1)(2a) 2b(η + 1) − (η − 1)(2b)
dz = da + db
(1 + η) 2 (1 + η)2
4
= (a da + b db).
(1 + η)2
Differentiation on manifolds 49
Thus,
16a
x dy ∧ dz = [(−2ab)b − (1 + a 2 − b2 )a] da ∧ db
(1 + η)5
−16a 2
= da ∧ db.
(1 + η)4
Similarly, we find
−16b2
y dz ∧ d x = da ∧ db
(1 + η)4
and
4(η − 1)
z d x ∧ dy = [(1 + (b2 − a 2 ))(1 − (b2 − a 2 )) − 4a 2 b2 ] da ∧ db
(1 + η)5
4(η − 1)
= [(1 − (b2 − a 2 )2 ) − 4a 2 b2 ] da ∧ db
(1 + η)5
4(η − 1)
= (1 − b4 + 2a 2 b2 − a 4 − 4a 2 b2 ) da ∧ db
(1 + η)5
4(η − 1)
= (1 − η2 ) da ∧ db.
(1 + η)5
Adding everything together we get
−16η(1 + η) 4(1 − η − η2 + η3 ) 4
− =−
(1 + η)5 (1 + η)5 (1 + η)2
times da ∧ db, as advertised.
The calculation for σV is similar. Evidently σU = 4ωU and σV = 4ωV ,
where ωU and ωV were defined in Exercise 3.10.
3.41 Note that f : R2 → R, and is f −1 (0). The Jacobian matrix of f is (−3x 2 −
a 2y), so the map is rank deficient if y = 0 and 3x 2 + a = 0. But if y = 0
then x 3 + ax + b = 0, which means that x is a double root of x 3 + ax +
b. This can only happen if the discriminant vanishes. Thus, as long as the
discriminant is nonzero, is an embedded submanifold of the plane.
3.42 f is constant on X so the differential f ∗ must vanish everywhere on X .
Hence, T p X ⊆ ker f ∗ . Now, f ∗ maps T p M surjectively onto T p N , so by the
rank/nullity theorem dim ker f ∗ = m − n. But by the regular value theorem,
dim T p X = dim X = m − n as well, so T p X = ker f ∗ .
3.43 a. The determinant of a matrix is a polynomial in its entries, so it is a smooth
function from Mn (R) to R. The inverse image of 0 is closed in Mn (R), so
its complement is open. Any open subset of Euclidean space is a smooth
manifold (just take a single coordinate chart consisting of the open subset
itself). A single entry of a product of two matrices is a sum of products
50 Differentiation on manifolds
ϕi j := ϕ(Ai j ) = Aik A jk .
So
∂ϕi j
= δim δkn A jk + Aik δ jm δkn .
∂ Amn
This is the Jacobian matrix. Viewing B as a “vector” in the tangent space
to Mn (R) we get
∂ϕi j
Bmn = Bik A jk + Aik B jk ,
∂ Amn
or
ϕ∗A B = B A T + AB T .
1 = a 2 + b2 = c2 + d 2 ,
0 = ac + bd.
ad − bc = 1.
where a 2 + b2 = 1.
The circle S 1 is the locus of points (x, y) ∈ R2 where x 2 + y 2 = 1.
The map ϕ : R2 → R4 given by
x y
(x, y) →
−y x
52 Differentiation on manifolds
L g∗ [X, Y ] = [L g∗ X, L g∗ Y ] = [X, Y ].
In particular, η1 and η2 are both integral curves of X and they satisfy the
same initial condition
so they must be equal. Finally, we can extend γ (t) to all values of t using
the group law, by defining γ (t) := γ (t/n)n for some large n.
3.45 (ii) By definition, for any matrices A and B, L A B = AB (matrix multiplica-
tion). Thus,
so
L ∗A x = Ax.
and therefore
L ∗A x −1 = x −1 A−1 .
Hence,
L ∗A = L ∗A (x −1 d x) = (L ∗A x −1 ) d(L ∗A x)
= x −1 A−1 A d x = x −1 d x = .
(iii) First note that left multiplication by g is a transitive action on any group
G, meaning that, for any two points x and y in G, there exists a g such
that gx = y. (Just take g = yx −1 .) Thus, it suffices to show that, for any
A, L ∗A , X = , X , as that will show B X has the same value at any
point. But as pullback commutes with contraction,
L ∗A , X = L ∗A , L ∗A X = , X .
It follows that we may as well evaluate , X at the identity. But then
we get a map from an element of Te G L(n, R) = g(n, R) to a matrix in
Mn (R). It is obviously linear by the linearity of the dual pairing.
(iv) Applied to the current situation, (3.123) gives
d(X, Y ) = X (Y ) − Y (X ) − ([X, Y ]).
But we already showed that if X and Y are left invariant, (X ) and (Y )
are constant functions, so the first two terms on the right must vanish.
(v) Differentiating both sides of
x x −1 = 1
gives
d x · x −1 + x · d(x −1 ) = 0,
which shows that
d(x −1 ) = −x −1 d x x −1 .
Thus,
d = d(x −1 d x) = −x −1 d x x −1 ∧ d x = − ∧ .
(vi) We have
([X, Y ]) = B[X,Y ]
and
( ∧ )(X, Y ) = (X )(Y ) − (Y )(X ) = [B X , BY ],
so we conclude that X → B X is indeed a Lie algebra homomorphism.
54 Differentiation on manifolds
where γ (0) = I . Now iterate by repeatedly substituting the left side into the
right side to get
1
γ (t) = I + t A + (t A)2 + · · · = et A .
2!
But Exp t A is also an integral curve whose tangent vector at the identity is A,
so the conclusion follows from the uniqueness of integral curves.
3.48 Proceeding as in the solution of Exercise 3.46 we take a curve A(t) on SU (n)
through the identity and differentiate I = A A† to get
0 = Ȧ(0) + ( Ȧ(0))† ,
is automatic from the requirement that it be skew symmetric, but in the com-
plex case we need to work a little harder. By virtue of Exercise 3.47 every
element of SU (n) can be written in the form e Ȧ(0) for some element Ȧ(0)
in su(n). But by Exercise 1.48 det e Ȧ(0) = etr Ȧ(0) , so the condition that e Ȧ(0)
have determinant one is exactly the condition that Ȧ(0) be traceless.
3.49 a. ϕ(g) is a smooth bijection with smooth inverse ϕ(g)−1 = ϕ(g −1 ). There-
fore it is a diffeomorphism, and its derivative is a local isomorphism. Let
X and Y be left invariant vector fields on G. By (3.87),
ϕ(g)∗,e [X e , Ye ] = ϕ(g)∗,e [X, Y ]e = [ϕ(g)∗,e X, ϕ(g)∗,e Y ]e ,
so Ad g is a Lie algebra automorphism.
b. In terms of matrices, the conjugation map is
ϕ(A)B = AB A−1 .
This is linear in B, so its derivative equals itself.
3.50 We write
d d t X −t X
Ad et X (Y ) = e Ye
dt dt
d tX d −t X
= e Y e−t X + et X Y e
dt dt
= et X X Y e−t X − et X Y X e−t X
= et X [X, Y ]e−t X
= Ad et X [X, Y ].
But also (as ad X commutes with itself)
d t ad X
e (Y ) = et ad X (ad X )(Y ) = et ad X [X, Y ].
dt
Hence both Ad et X and et ad X satisfy the same differential equation with the
same initial condition (both reduce to the identity at t = 0), so by uniqueness
they must coincide.
3.51 Assuming (3.114), we get, by setting t = 1,
e A+B = e A e B e−(1/2)[A,B] ,
from which (3.111) follows, because [A, B] commutes with A and B.
Differentiating both sides of (3.111) with respect to t gives
(A + B)et (A+B) = Aet A f (t) + et A f (t),
so using (3.113) we get
(A + B)et A f (t) = Aet A f (t) + et A f (t),
56 Differentiation on manifolds
or, simplifying,
Bet A f (t) = et A f (t).
Multiplying both sides on the left by e−t A yields
e−t A Bet A f (t) = f (t),
so applying (3.110) we obtain
f (t) = (Ad e−t A B) f (t) = (e−t ad A B) f (t).
Using the fact that [A, [A, B]] = 0 we thus get
f (t) = (B − t[A, B]) f (t),
together with the initial condition f (0) = 1.
On the other hand, mindful of the fact that [B, [A, B]] = 0, we see that
d t B −(t 2 /2)[A,B]
= Bet B e−(t /2)[A,B] − et B t[A, B]e−(t /2)[A,B]
2 2
e e
dt
= (B − t[A, B])et B e−(t /2)[A,B] .
2
3.52 a. ad X is a linear map, so this follows from the cyclicity of the trace.
b. By definition, an automorphism T of g must carry brackets to brackets:
T [X, Y ] = [T X, T Y ].
Writing Z = T Y this becomes
T [X, T −1 Z ] = [T X, Z ]
which just says that ad T X = T ◦ ad X ◦ T −1 . Thus, by cyclicity of the
trace,
(T X, T Y ) = tr(ad T X ◦ ad T Y ) = tr(T ◦ ad X ◦ ad Y ◦ T −1 )
= tr(ad X ◦ ad Y ) = (X, Y ).
But according to Exercise 3.49a Ad g is an automorphism of g for any
g ∈ G.
c. By the Jacobi identity,
ad[X, Y ](Z ) = [[X, Y ], Z ] = [X, [Y, Z ]] + [Y, [Z , X ]]
= ad X ◦ ad Y (Z ) − ad Y ◦ ad X (Z )
= [ad X, ad Y ](Z ),
so by the cyclicity of the trace again,
Differentiation on manifolds 57
In particular, a and d are real, and b and c are complex conjugates of one
another. The condition that the matrix be traceless gives a + d = 0. Therefore
the most general element of su(2) is of the form
x y − iz
= xσ1 + yσ2 + zσ3 = x · σ ,
y + iz −x
where x, y, and z are real. It follows that the matrices τk := (−i/2)σk form a
basis for su(2).
To show that so(3) and su(2) are isomorphic it suffices to show that they
have the same structure constants. The commutation relations for the Pauli
matrices are well known and easy to prove by multiplying out the matrices.
Using the summation convention again we get
[σi , σ j ] = 2ii jk σk .
Hence,
[τi , τ j ] = i jk τk ,
and the claim is proved.
3.55 a. We want to show that
L X = i X d + di X (1)
when acting on p forms. The result holds for zero forms, because
(i X d + di X ) f = i X d f = X f = L X f.
Now suppose it holds for p − 1 forms, and let η ∧ λ be a p-form, with
deg η = r . Then by the induction hypothesis and the properties of the
Lie derivative,
L X (η ∧ λ) = L X η ∧ λ + η ∧ L X λ
= (i X d + di X )η ∧ λ + η ∧ (i X d + di X )λ.
But we also have, by the properties of the differential and the interior
product,
(i X d + di X )(η ∧ λ) = (i X d)(η ∧ λ) + (di X )(η ∧ λ)
= i X (dη ∧ λ + (−1)r η ∧ dλ)
+ d(i X η ∧ λ + (−1)r η ∧ i X λ)
= i X dη ∧ λ + (−1)r+1 dη ∧ i X λ
+ (−1)r [i X η ∧ dλ + (−1)r η ∧ i X dλ]
+ di X η ∧ λ + (−1)r−1 i X η ∧ dλ
+ (−1)r [dη ∧ i X λ + (−1)r η ∧ di X λ]
= (i X d + di X )η ∧ λ + η ∧ (i X d + di X )λ.
Differentiation on manifolds 59
(i λ ((
− iY L X η ∧ λ − ( ((L(X(
(−1) η∧
r
Y
−i(
( (∧(L(X(
Yη λ − (−1)r η ∧ i Y L X λ
= i L X Y η ∧ λ + (−1)r η ∧ i L X Y λ
= i L X Y (η ∧ λ).
c. We want to show
(L X 0 ω)(X 1 , . . . , X p ) = L X 0 (ω(X 1 , . . . , X p ))
p
− ω(X 1 , . . . , L X 0 X i , . . . , X p ). (3)
i=1
For p = 1, this is
(L X 0 ω)(X 1 ) = L X 0 (ω(X 1 )) − ω(L X 0 X 1 ),
or
i X 1 L X 0 ω = L X 0 i X 1 ω − i L X 0 X 1 ω,
which is just the case p = 1 of the formula proved in Part (b). Now
assume (3) holds for p − 1. Then, by induction and the result of Part (b)
again,
L X 0 (ω(X 1 , . . . , X p )) = L X 0 ((i X 1 ω)(X 2 , . . . , X p ))
= (L X 0 (i X 1 ω))(X 2 , . . . , X p )
p
+ (i X 1 ω)(X 2 , . . . , L X 0 X i , . . . , X p ).
i=2
60 Differentiation on manifolds
= (i X 1 L X 0 ω + i L X 0 X 1 ω)(X 2 , . . . , X p )
p
+ ω(X 1 , X 2 , . . . , L X 0 X i , . . . , X p ).
i=2
= (L X 0 ω)(X 1 , . . . , X p )
p
+ ω(X 1 , X 2 , . . . , L X 0 X i , . . . , X p ).
i=1
p
= (−1)i−1 L X i (ω(X 0 , X 1 , . . . ,
X i , . . . , X p ))
i=1
+ (−1)i+ j ω(X 0 , L X i X j , X 1 , . . . ,
Xi , . . . ,
X j , . . . , X p)
1≤i< j≤ p
p
= (−1)i−1 L X i (ω(X 0 , . . . ,
X i , . . . , X p ))
i=1
+ (−1)i+ j+1 ω(L X i X j , X 0 , X 1 , . . . ,
Xi , . . . ,
X j , . . . , X p ).
1≤i< j≤ p
(4)
Equations (1) and (3) yield
(i X 0 dω + di X 0 ω)(X 1 , . . . , X p )
= L X 0 (ω(X 1 , . . . , X p ))
p
− ω(X 1 , . . . , L X 0 X i , . . . , X p )
i=1
= L X 0 (ω(X 1 , . . . , X p ))
p
− (−1)i+ j−1 ω(L X 0 X j , X 1 , . . . , %
X j , . . . , X p ), (5)
i=0,1≤ j≤ p
Differentiation on manifolds 61
where the last equality follows by changing dummy indices and by the
antisymmetry of the contraction map. Combining (4) and (5) we get
dω(X 0 , X 1 , . . . , X p )
= L X 0 (ω(X 1 , . . . , X p ))
− (−1)i+ j−1 ω(L X 0 X j , X 1 , . . . , %
X j , . . . , X p)
i=0,1≤ j≤ p
p
− (−1)i−1 L X i (ω(X 0 , . . . ,
X i , . . . , X p ))
i=1
− (−1)i+ j+1 ω(L X i X j , X 0 , X 1 , . . . ,
Xi , . . . ,
X j , . . . , X p)
1≤i< j≤ p
p
= (−1)i L X i (ω(X 0 , . . . ,
X i , . . . , X p ))
i=0
+ (−1)i+ j ω(L X i X j , X 0 , X 1 , . . . ,
Xi , . . . ,
X j , . . . , X p ).
0≤i< j≤ p
The second formula in the problem follows because for any function f
and vector fields X and Y , L X f = X f and L X Y = [X, Y ].
3.56 By compatibility of the Lie derivative with the dual pairing,
X α, ∂i = L X α, ∂i + α, L X ∂i . (1)
From (3.125) we get
L X ∂i = −X k ,i ∂k . (2)
Combining this with (1) gives
X j αi, j = (L X α)i − α j X j ,i ,
which is (3.126). In particular, if α = d x j we get
L X d x j = X j ,k d x k . (3)
Now we have
T = T i1 ...ir j1 ... js ∂i1 ⊗ · · · ⊗ ∂ir ⊗ d x j1 ⊗ · · · ⊗ d x js ,
so applying the Leibniz rule and using (2) and (3) yields
L X T = X (T i1 ...ir j1 ... js )∂i1 ⊗ · · · ⊗ ∂ir ⊗ d x j1 ⊗ · · · ⊗ d x js
+ T i1 ...ir j1 ... js (−X k ,i1 )∂k ⊗ ∂i2 ⊗ · · · ⊗ ∂ir ⊗ d x j1 ⊗ · · · ⊗ d x js
+ ···
+ T i1 ...ir j1 ... js (−X k ,ir )∂i1 ⊗ · · · ⊗ ∂ir −1 ⊗ ∂k ⊗ d x j1 ⊗ · · · ⊗ d x js
+ T i1 ...ir j1 ... js (+X j1 ,k )∂i1 ⊗ · · · ⊗ ∂ir ⊗ d x k ⊗ d x j2 ⊗ · · · ⊗ d x js
+ ···
+ T i1 ...ir j1 ... js (+X js ,k )∂i1 ⊗ · · · ⊗ ∂ir ⊗ d x i1 ⊗ · · · ⊗ d x js−1 ⊗ d x k .
62 Differentiation on manifolds
65
66 Homotopy and de Rham cohomology
Hence,
1
∗
hF ω = A(t x, t y, t z) t dt (y dz − z dy)
0
1
+ B(t x, t y, t z) t dt (z d x − x dz)
0
1
+ C(t x, t y, t z) t dt (x dy − y d x).
0
so that
Then
∂(ez) ∂( f z) ∂(gx) ∂(gy)
dα = − − + + d x ∧ dy + cyclic
∂x ∂y ∂x ∂y
∂e ∂f ∂g ∂g
= − z− z+ x+g+ y + g d x ∧ dy + cyclic. (4)
∂x ∂y ∂x ∂y
Now add and subtract (∂g/∂z)z inside the parenthetical term in Equation
(4). Observe that
∂e ∂f ∂g
+ + = 0. (5)
∂x ∂y ∂z
This follows from
∂A ∂B ∂C
+ + = 0,
∂x ∂y ∂z
which holds because ω is closed. Also,
∂g ∂g ∂g
x+ y+ z = (x · ∇)g
∂x ∂y ∂z
1
= (x · ∇)C(t x, t y, t z) t dt (6)
0
Homotopy and de Rham cohomology 67
1
d
= C(t x, t y, t z) t 2 dt (7)
0 dt
( )1 1
= C(t x, t y, t z) t 2
−2 C(t x, t y, t z) t dt (8)
0 0
= C(x, y, z) − 2g. (9)
Equation (7) holds by virtue of the chain rule. The easiest way to see this
is to write
∂ ∂ ∂
∇=t , ,
∂(t x) ∂(t y) ∂(t z)
in Equation (6). Equation (8) follows by integration by parts, and Equation
(9) from definition (3). Combining (4), (5), and (9) gives
dα = C d x ∧ dy + cyclic = ω,
as was to be shown.
4.4 Following the hint, we consider the homology class [a] ∈ Hi (A). As ϕ is a
chain map and a ∈ Z (Ai ) is closed,
di ϕi a = ϕi+1 di a = 0,
so ϕi (a) ∈ Z (B i ). Hence it makes sense to consider its cohomology class.
So define h i ([a]) := [ϕi (a)]. We need only show that the map is independent
of class representative. So, suppose a ∈ [a]. Then a − a = di−1 γ for some
γ ∈ Ai−1 . But ϕ is a chain map, so
ϕi (a − a ) = ϕi di−1 γ = di−1 ϕi−1 γ .
This means that [ϕi (a − a )] is zero in Hi (B), which, by linearity, means that
[ϕi (a)] = [ϕi (a )].
4.5 Begin again with [c] ∈ H i (C). Choose an element c ∈ [c] different from c
and follow the same steps as before. This gives a b ∈ B i satisfying ψi b = c
and an a ∈ Ai+1 such that ϕi+1 a = di b . We must show that [a] = [a ].
As [c] = [c ], c − c = di−1 w for some w ∈ C i−1 . By exactness, there is
a v ∈ B i−1 such that ψi−1 v = w. Thus c − c = di−1 ψi−1 v = ψi di−1 v. It
follows that ψi (b − b ) = c − c = ψi di−1 v or ψi (b − b − di−1 v) = 0.
By exactness, there exists a u ∈ Ai such that b − b − di−1 v = ϕi u, so
ϕi+1 (a − a ) = di (b − b ) = di ϕi u = ϕi+1 di u, or ϕi+1 (a − a − di u) = 0. By
exactness, we conclude a − a = di u, so [a] = [a ].
4.6 We show that the sequence is exact at Hi (A), Hi (B), and Hi (C), in that order.
1. (im δi−1 ⊆ ker αi .) Using the same notation as in the text, we have
αi δi−1 [c] = αi [a] = [ϕi a] = [di−1 b] = 0.
68 Homotopy and de Rham cohomology
3. (im βi ⊆ ker δi .)
for k < n. Therefore, the bottom two rows of the Mayer-Vietoris sequence
look like this:
⎪
⎨α s+1 , if 0 ≤ t ≤ (s + 1)/4,
4t
⎪
F(s, t) = β(4t
− (s + 1)),
if (s + 1)/4 ≤ t ≤ (s + 2)/4,
⎪
⎪
⎩γ 4t−(s+2)
, if (s + 2)/4 ≤ t ≤ 1.
2−s
the base. But we can shrink the cylinder back down to the base, whereupon we
are left with just a cone, and this is contractible by Exercise 4.2. Similarly,
U ∩ V is just the cylinder over M (because both endpoints are removed),
which is homotopic to M. Hence, the Mayer-Vietoris sequence looks like
this:
Looking at the diagram, we see that for every k ≥ 1 we get an exact sequence
of the form
so that
H k (M) ∼
= H k+1 ( M). (1)
As for the lowest level, once again we have rk ψ0∗ = 1. By exactness and
dimension counting we get
H 0 ( M) = R. (2)
H 0 (M) = H 1 ( M) ⊕ R. (3)
It follows that
77
78 Elementary homology theory
not three dimensional. But the inequality implies that p and q cannot both
exceed 5. This leaves only {3, 3}, {4, 3}, {3, 4}, {5, 3}, and {3, 5}, namely
the Platonic solids.
5.7 From the figure, we count 9 vertices, 27 edges, and 18 2-faces. (Identified
vertices are labeled, and edges with the same endpoints are identified. All the
2-faces are distinct.) The Euler characteristic is therefore
χ = V − E + F = 9 − 27 + 18 = 0 .
5.8 a. The wedge sum of two tetrahedra has 7 vertices, 12 edges, and 8 2-faces,
so χ = 7 − 12 + 8 = 3.
b. We computed the homology of the tetrahedron in Exercise 5.3. It is R in
dimensions 0 and 2, and 0 in dimension 1. So the homology of the wedge
sum is R in dimension 0, 0 in dimension 1, and R ⊕ R in dimension 2. The
alternating sum of Betti numbers is therefore 1 − 0 + 2 = 3, as before.
5.9 a. We have
Hence, removing an open disk from a surface reduces its Euler character-
istic by 1. We show χ (T #k ) = 2 − 2k by induction on k. For k = 0 we get
the 2-sphere, whose Euler characteristic we already know to be 2. By the
definition of connected sum, the additivity of the Euler characteristic, and
induction, we get
5.10 Following the hint, we consider the homology class [αi ] ∈ Hi (A). As ψ is a
chain map and αi is a cycle in Ai ,
∂i ψi (αi ) = ψi−1 ∂i (αi ) = 0,
so ψi (αi ) is a cycle in Bi . Hence it makes sense to consider its homology
class. So define h i ([αi ]) := [ψi (αi )]. We need only show that the map is
independent of class representative. So, suppose [αi ] = [βi ]. Then αi − βi =
∂i+1 (γi+1 ) for some γi+1 ∈ Ai+1 . But ψ is a chain map, so
ψi (αi − βi ) = ψi ∂i+1 (γi+1 ) = ∂i+1 ψi+1 (γi+1 ).
This means that [ψi (αi − βi )] is zero in Hi (B), which, by linearity, means
that [ψi (αi )] = [ψi (βi )].
5.11 From the result of Exercise 1.30 and the given Mayer-Vietoris sequence we
get
0= dim H0 (U ∪ V ) − dim H0 (U ) − dim H0 (V ) + dim H0 (U ∩ V )
− dim H1 (U ∪ V ) + dim H1 (U ) + dim H1 (V ) − dim H1 (U ∩ V )
+··· ,
from which we conclude
χ (U ∪ V ) − χ (U ) − χ(V ) + χ(U ∩ V ) = 0.
5.12 The sequence
ι ∂i
0 −−−→ Z i −−−→ Ci −−−→ Bi−1 −−−→ 0
is trivially exact at Z i because Z i ⊆ Ci , and at Bi−1 by definition. It is exact
at Ci because, by definition, i-cycles are annihilated by the boundary operator
∂i . It follows that
tr(ψi , Ci ) = tr(ψi , Z i ) + tr(ψi , s(Bi−1 )),
where ∂i ◦ s = 1. But ψ is a chain map, so if bi−1 ∈ Bi−1 then
∂i ψi (s(bi−1 )) = ψi−1 ∂i (s(bi−1 )) = ψi−1 (bi−1 ).
Hence, ∂i ψi s = ψi−1 restricted to Bi−1 . In particular,
tr(ψi , s(Bi−1 )) = tr(ψi−1 , Bi−1 ).
The sequence
ι πi
0 −−−→ Bi −−−→ Z i −−−→ Hi −−−→ 0
is exact because Hi = Z i /Bi . It follows that
tr(ψi , Z i ) = tr(ψi , Bi ) + tr(ψi , Hi ).
Elementary homology theory 83
6.1 We have
so
1
∗
1
6
ω= γ ω=6 t dt = t = 1.
5
γ I 0 0
6.2 We have
and
Thus
84
Integration on manifolds 85
and
∗
β= ϕ β= B(ϕ(u, v)) · n(ϕ(u, v)) du ∧ dv = B · d S.
c c
We have
ω = −yz d x + x z dy ⇒ dω = 2z d x ∧ dy − x dy ∧ dz − y dz ∧ d x.
Hence
ϕ ∗ dω = 2(u 2 )(3u 2 cos v du − u 3 sin v dv) ∧ (3u 2 sin v du + u 3 cos v dv)
− (u 3 cos v)(3u 2 sin v du + u 3 cos v dv) ∧ (2u du)
− (u 3 sin v)(2u du) ∧ (3u 2 cos v du − u 3 sin v dv)
= 8u 7 du ∧ dv.
This gives
2
2π 2 8
dω = dv 8u du = 2πu = 510π.
7
0 1 1
We have
ϕ ∗ ω = −(u 5 sin v)(3u 2 cos v du − u 3 sin v dv)
+ (u 5 cos v)(3u 2 sin v du + u 3 cos v dv)
= u 8 dv.
The boundary of U is a square, so we need to compute the line integral around
the square. This gives
/
ω= u 8 dv = 28 (2π − 0) + 18 (0 − 2π) = 510π,
∂ ∂U
6.5 Let ω be a k-form and c a (k + 2)-chain. Then Stokes theorem applied twice
gives
d ω=
2
dω = ω.
c ∂c ∂2c
By de Rham’s
0 first theorem, ω is exact, so [ω] = 0, whence we see that
[ω] → [ ω] is injective.
Integration on manifolds 87
6.10 We choose the orientation to be the one in which the variables are ordered
(u 2 , u 1 ) and (v 1 , v 2 ), so that the Jacobian is positive on the overlap. As we
see from Exercise 3.40, this is equivalent to choosing the area element on the
sphere to be + sin θ dθ ∧ dφ. (Equivalently, we choose the ‘outward pointing
normal’.) Integrating over all of U gives
du 2 ∧ du 1
ωU =
U0 (1 + η)
2
U
2π ∞
r dr dθ
= dθ dr
(1 + r 2 )2
0 0
∞
1/2
= −2π · = π.
1 + r 2 0
Alternatively,
ω= ωU + ωV ,
S2 U0 V0
6.12 a. We see immediately that the integral over the base must vanish, because
dz = 0 there. A natural parameterization for the rest of the cone surface is
ϕ(u, v) = (u cos v, u sin v, uh/a), 0 ≤ u ≤ a, 0 ≤ v ≤ 2π.
This gives
ϕ ∗ σ = (u cos v)(sin v du + u cos v dv) ∧ ((h/a)du)
+ (u sin v)((h/a)du) ∧ (cos v du − u sin v dv)
= −(h/a)u 2 du ∧ dv.
Thus,
2π a
h 2π 2
σ =− dv u 2 du = − ha .
∂V a 0 0 3
The negative sign just means that our parameterization naturally selects
the inward pointing normal rather than the outward pointing normal. To
see this, look back at Exercise 6.2. With our parameterization,
n = (cos v, sin v, h/a) × (−u sin v, u cos v, 0)
= (−(uh/a) cos v, −(uh/a) sin v, u),
which definitely points inwards. This is easily remedied, though, simply by
flipping the sign of the integral.
b. By Stokes’ theorem,
σ = dσ = 2 d x ∧ dy ∧ dz,
∂V V V
which is twice the volume. Hence the volume of the cone is π ha 2 /3.
6.13 The disk meets the sphere in a circle ∂ of radius a/2. We parameterize
the circle by
√
γ (t) = (a/2)(cos t, sin t, 3), 0 ≤ t ≤ 2π,
so
√
γ ∗ ω = (a/2)2 [− sin2 t + 3 cos t] dt
and
2π
∗
ω= γ ω = −(a/2) 2
sin2 t dt = −a 2 π/4.
∂ I 0
Also,
dω = dy ∧ d x + dz ∧ dy + d x ∧ dz,
so
σ ∗ dω = (sin v du + u cos v dv) ∧ (cos v du − u sin v dv)
= −u du ∧ dv.
Note that the unit normal to the disk with this parameterization points
upwards, which is consistent with the choice of direction for ∂. Integrating
over the disk gives
2π a/2
∗
dω = σ dω = − dv u du dv = −πa 2 /4.
R 0 0
7.1 By definition,
(ϕV ◦ ϕU−1 )( p, yU ) = ( p, yV ) = ( p, gV U ( p)yU ).
If U = V then ϕV = ϕU , so we get
( p, yU ) = ( p, gUU ( p)yU ).
for all yU , which gives gUU ( p) = id.
Similarly,
(ϕU ◦ ϕV−1 )( p, yV ) = ( p, gU V ( p)yV ) = ( p, gU V gV U yU ).
But
(ϕU ◦ ϕV−1 )( p, yV ) = (ϕU ◦ ϕV−1 )(ϕV ◦ ϕU−1 )( p, yU ) = ( p, yU ),
whereupon we conclude
gU V ( p)gV U ( p) = 1.
Lastly, if U , V , and W are three overlapping neighborhoods,
(ϕV ◦ ϕU−1 )( p, yU ) = ( p, yV ),
(ϕW ◦ ϕV−1 )( p, yV ) = ( p, yW ),
−1
(ϕU ◦ ϕW )( p, yW ) = ( p, yU ),
where yV = gV U yU , yW = gW V yV and yU = gU W yW . Combining these
equations together with the ones above yields
gU W gW V gV U = 1.
7.2 Let ϕ : E → M × Y be a global trivialization and choose ϕU = ϕV = ϕ.
This immediately implies gU V = id.
90
Vector bundles 91
∇gY ∇ f X h Z = g∇Y ( f ∇ X (h Z ))
= g∇Y ( f X (h)Z + f h∇ X Z )
= gY ( f X (h))Z + f g X (h)∇Y Z
+ gY ( f h)∇ X Z + f gh∇Y ∇ X Z . (3)
+ gY ( f )X (h)Z + gY ( f )h∇ X Z
! "# $ ! "# $
a b
− f g([X, Y ]h)Z − f gh∇[X,Y ] Z . (5)
! "# $ ! "# $
a d
Close examination of this expression reveals that all the terms labeled ‘a’, ‘b’,
and ‘c’ cancel, while the terms labeled ‘d’ combine to yield the desired result:
R( f X, gY )h Z = f gh R(X, Y )Z .
7.5 A point of the unit tangent bundle T1 (S 2 ) is a pair (x, y), where x ∈ S 2
and y ∈ Tx S 2 . Such a pair determines a unique orthonormal triple (x, y, z),
where z = x × y. Take these to be the column vectors of a matrix R. By
orthonormality, R R T = 1, so R ∈ S O(3). Conversely, given an element R ∈
S O(3) (viewed as a subgroup of G L(3, R)) the columns form an orthonormal
triple, and therefore we may choose the first and second entries to be a point
of T1 (S 2 ). This shows that the map T1 (S 2 ) → S O(3) given by (x, y) → R
is bijective, and continuity is clear.
7.6 The three vector fields are mutually orthogonal. (Indeed, on S 4 they are
orthonormal.) In particular, they are clearly linearly independent. Moreover,
they are nowhere vanishing on S 4 , because (0, 0, 0, 0) ∈ S 4 . Smoothness
is obvious, so there is a global frame field and the tangent bundle is indeed
trivial.
7.7 The projection map π just sends (E ⊕ F) p to p. If ϕ is a local trivializa-
tion of E → M and ψ is a local trivialization of F, then (ϕ, ψ) is a local
trivialization of E ⊕ F according to
(It is worth noting that this also shows that an arbitrary linear combination of
connections is not necessarily a connnection.)
7.9 a. This follows immediately from the cyclicity of the trace and the prop-
erties of matrix multiplication. We have = A−1 A, so 2 =
(A−1 A)(A−1 A) = A−1 2 A. Inductively, we have k = A−1 k A.
By the cyclicity of the trace,
b. We have
where the middle step follows from the antisymmetry properties of the
wedge. (Each element i1 i2 is a two form, so it commutes with all other
forms.) Every other term in (2) can be brought to the same form, and there
are k of them. Equation (4) follows from the Bianchi identity. Finally,
Equation (5) uses the same idea that we used to obtain Equation (3).
94 Vector bundles
gives
tr(η ∧ dk−1
t ) = tr(η ∧ {(t ∧ ωt − ωt ∧ t ) ∧ t ∧ · · · ∧ t
+ t ∧ (t ∧ ωt − ωt ∧ t ) ∧ t ∧ · · · ∧ t
+ t ∧ · · · ∧ t ∧ (t ∧ ωt − ωt ∧ t )})
= tr(−η ∧ ωt ∧ k−1
t + η ∧ k−1
t ∧ ωt )
= − tr({η ∧ ωt + ωt ∧ η} ∧ k−1
t ).
(The second equality follows because the sum is telescoping, meaning that
one term in each line cancels with one term in the next line. The last
equality follows by commuting forms and using the cyclicity of the trace.)
Hence
dα = tr({dη + η ∧ ωt + ωt ∧ η} ∧ k−1
t )
= tr({dη + η ∧ ω + ω ∧ η + 2tη ∧ η} ∧ k−1
t )
1 d
= tr kt .
k dt
Di (gψ) = ∂i (gψ) + Ai gψ
= (∂i g)ψ + g∂i ψ + gg −1 Ai gψ
= g g −1 ∂i g + g −1 Ai g ψ + g∂i ψ
= g Di ψ.
F = dA + A ∧ A
= ∂ j Ai d x j ∧ d x i + Ai A j d x i ∧ d x j
= ∂i A j + Ai A j d x i ∧ d x j
1
= ∂i A j − ∂ j Ai + [Ai , A j ] d x i ∧ d x j .
2
96 Vector bundles
Also,
[Di , D j ]ψ = [∂i + Ai , ∂ j + A j ]ψ
= ([∂i , ∂ j ] + [∂i , A j ] + [Ai , ∂ j ] + [Ai , A j ])ψ
= ∂i (A j ψ) − A j ∂i ψ + Ai ∂ j ψ − ∂ j (Ai ψ) + [Ai , A j ]ψ
= (∂i A j − ∂ j Ai + [Ai , A j ])ψ.
c. The Bianchi identity (7.20) reads
d F − F ∧ A + A ∧ F = 0.
In local coordinates (after multiplying by 2), this becomes
∂i F jk − Fi j Ak + Ai F jk d x i ∧ d x j ∧ d x k = 0.
Note that, by virtue of the antisymmetry of the wedge products,
Bi jk d x i ∧ d x j ∧ d x k = 0
holds if and only if B[i jk] , the totally antisymmetric part of B, vanishes.
Using the fact that Fi j is already antisymmetric, we get (dropping a factor
of 3),
0 = ∂[i F jk] − F[i j Ak] + A[i F jk]
= ∂i F jk + ∂ j Fki + ∂k Fi j − Fi j Ak − F jk Ai − Fki A j
+ Ai F jk + A j Fki + Ak Fi j
= ∂i F jk + [Ai , F jk ] + cyclic.
But
[Di , F jk ]ψ = [∂i + Ai , F jk ]ψ
= ∂i (F jk ψ) − F jk ∂i ψ + [Ai , F jk ]ψ
= (∂i F jk + [Ai , F jk ])ψ,
so the conclusion follows.
8
Geometric manifolds
8.1
τ ( f X, gY ) = ∇ f X (gY ) − ∇gY ( f X ) − [ f X, gY ]
= f (Xg)Y + f g∇ X Y − g(Y f )X − g f ∇Y X
− f (Xg)Y − f g X Y + g(Y f )X + g f Y X
= f gτ (X, Y ).
8.2 From the definition,
τ a (eb , ec ) = θ a (τ (eb , ec ))
= θ a (∇eb ec − ∇ec eb − [eb , ec ])
= a bc − a cb − θ a ([eb , ec ])
= ωa c (eb ) − ωa b (ec ) − θ a ([eb , ec ])
= (ωa d ∧ θ d )(eb , ec ) + dθ a (eb , ec ).
The last line follows from Exercise 3.15 and Equation 3.122. Specifically,
(ωa d ∧ θ d )(eb , ec ) = ωa d (eb )θ d (ec ) − ωa d (ec )θ d (eb )
= ωa d (eb )δcd − ωa d (ec )δbd
= ωa c (eb ) − ωa b (ec ),
and
dθ a (eb , ec ) = eb θ a (ec ) − ec θ a (eb ) − θ a ([eb , ec ])
= eb (δca ) − ec (δba ) − θ a ([eb , ec ])
= −θ a ([eb , ec ]).
8.3 By Properties C4 and C5 of Section 7.2,
0 = ∇ea eb , θ c = ∇ea eb , θ c + eb , ∇ea θ c = c ab + eb , ∇ea θ c ,
so that ∇ea θ c = − c ab θ b .
97
98 Geometric manifolds
8.4 Under a coordinate transformation x i → y i with ∂i := ∂/∂ x i and ∂i :=
∂/∂ y i we have
∂i = J i i ∂i and ∂i = (J −1 )i i ∂i ,
where
∂xi −1 i ∂ yi
J i
i = i and (J ) i = .
∂y ∂xi
Hence
k i j ∂k = ∇∂i ∂ j = J i i ∇∂i (J j j ∂ j ) = J i i J j j ∇∂i ∂ j + J i i (∂i J j j )∂ j
= J i i J j j k i j ∂k + J i i ((J −1 ) i ∂ J j j )(J −1 )k j ∂k
= J i i J j j k i j (J −1 )k k + δi (∂ J j j )(J −1 )k j ∂k ,
whereupon we conclude
k ∂ yk ∂ x i ∂ x j k ∂ yk ∂ 2 x j
i j = i j + .
∂ x k ∂ yi ∂ y j ∂ x j ∂ yi ∂ y j
8.5 The Christoffel symbols are given by
m i j = g mk ki j ,
where
1
ki j = ∂i g jk + ∂ j gik − ∂k gi j
2
and g mk are the components of the inverse metric.
Let the index ‘1’ be ‘θ’ and the index ‘2’ be ‘φ’. Then, written as a matrix,
the metric tensor components are
1 0
gi j = ,
0 sin2 θ
so the inverse metric components are
1 0
g =
ij
.
0 csc2 θ
Now we compute, using the fact that the off diagonal terms of the metric
vanish,
1
θθθ = (∂θ gθθ ) = 0,
2
1
θθφ = (∂φ gθθ ) = 0,
2
Geometric manifolds 99
θφθ = θθφ = 0,
1
θφφ = − (∂θ gφφ ) = − sin θ cos θ,
2
1
φθθ = − (∂φ gθθ ) = 0,
2
1
φθφ = (∂θ gφφ ) = sin θ cos θ,
2
φφθ = φθφ = sin θ cos θ,
1
φφφ = (∂φ gφφ ) = 0.
2
Raising indices with the inverse metric tensor we get
θ θθ = g θθ θθθ + g θφ φθθ = 0,
θ θφ = θ φθ = g θθ θθφ + g θφ φθφ = 0,
θ φφ = g θθ θφφ + g θφ φφφ = − sin θ cos θ,
φ θθ = g φθ θθθ + g φφ φθθ = 0,
φ θφ = φ φθ = g φθ θθφ + g φφ φθφ = cot θ,
φ φφ = g φθ θφφ + g φφ φφφ = 0.
Hence
n̂ m̂ = dωn̂ m̂ + ωn̂â ∧ ωb̂m̂ g â b̂
= −dωm̂ n̂ + ωâ n̂ ∧ ωm̂ b̂ g â b̂
= −dωm̂ n̂ − ωm̂ b̂ ∧ ωâ n̂ g â b̂
= −dωm̂ n̂ − ωm̂ â ∧ ωâ n̂
= −m̂ n̂ .
8.7 We have d A = Aω and d A T = (d A)T = ω T A T = −ω A T , so
d(A A T ) = d A · A T + A · d A T = Aω A T − Aω A T = 0.
Hence A A T = constant. But A A T ( p) = I , so A A T = I everywhere, i.e., A
is orthogonal.
8.8 Equation (7.34) says that if s = eσ is a section, then
R(X, Y ) ◦ s = e(X, Y )σ.
Let X = ec , Y = ed , and s = Z = eb Z b . Then
Z b R(ec , ed )eb = ea a b (ec , ed )Z b ⇒ a b (ec , ed ) = R a bcd ,
from which the result follows. (Okay, maybe this isn’t immediately obvious.
The point is that a b is a matrix of two forms, so it can be written
a b = a be f θ e ∧ θ f
for some coefficients a be f which are necessarily antisymmetric in the last
two indices. Then taking inner products gives
a b (ec , ed ) = (a be f θ e ∧θ f )(ec , ed ) = a be f (θ e (ec )θ f (ed ) − θ e (ed )θ f (ec ))
f
= a be f (δce δd − δde δcf ) = 2a bcd .
Now perhaps the conclusion is a bit more apparent.)
8.9 Equation (7.36) gives
R(X, Y )Z = ∇ X ∇Y Z − ∇Y ∇ X Z − ∇[X,Y ] Z .
Choose X = ∂k , Y = ∂ , and Z = ∂ j . Then [X, Y ] = 0, so we get
R(∂k , ∂ )∂ j = ∇∂k ∇∂ ∂ j − (k ↔ )
= ∇∂k ( m j ∂m ) − (k ↔ )
= m j,k ∂m + m j n km ∂n − (k ↔ )
= i j,k + i km m j ∂i − (k ↔ ),
which yields the desired result.
Geometric manifolds 101
8.10 Start with the following four equations, each of which follows from (8.49):
Adding the first two and subtracting the second two and applying (8.47) and
(8.48) liberally gives
:a :b
0 =g(W, R(X, Y )Z ) + g(W,
R(Y,Z )X ) + g(W,
R(Z ,
X )Y )
:
:c
a
+ g(X,
R(Y, Z )W ) + g(X, R(Z , W )Y ) + g(X,
R(W, Y )Z )
:d
:b
− g(Y, R(Z , W )X ) − g(Y,
R(W, X )Z ) − g(Y,
R(X, Z )W )
:d
:c
W
− g(Z
,
R(W, X )Y ) − g(Z , R(X, Y )W ) − g(Z
,
R(Y, )X ),
where terms with the same label cancel. This leaves
as desired.
8.11 Begin with (8.49) and take inner products to get
g(∂i , R(∂k , ∂ )∂ j ) = Ri jk ,
Ri jk + Rikj + Rijk = 0. (1)
Ri[ jk] = 0 (2)
we get
or
and
which shows that the definition of σ can be consistently extended across over-
laps to all of M. (We assumed J > 0, which we may do by virtue of the fact
that M is orientable.)
8.26 On U write d x i = Ai j θ j for some matrix A of smooth functions. Then at any
point of U ,
g i j = g(d x i , d x j ) = Ai k A j g(θ k , θ ) = Ai k A j δ k = (A A T )i j .
Taking determinants of both sides gives G −1 = (det A)2 . Therefore
√
σ = G dx1 ∧ · · · ∧ dxn
= (det A)−1 (det A)θ 1 ∧ · · · ∧ θ n
= θ1 ∧ · · · ∧ θn.
8.27 This follows immediately from the fact that 2 = ±1 and d 2 = 0.
8.28 If f is a harmonic function it must be closed, so d f = 0, which means f
is constant on connected components. Now let ω = f σ be a harmonic top
dimensional form. Assuming = we get 0 = ω = f (because
σ = 1), so f is harmonic and therefore constant. To prove the assumption,
let η be a k-form. Then using the fact that 2 = (−1)k(n−k) on a k-form (in
positive definite signature), we find
(dδ + δd)η = (−1)nk+n+1 d d η + (−1)nk+2n+1 (−1)k(n−k) d dη
2 +1
= (−1)nk+n+1 d d η + (−1)k d dη
and
(dδ + δd)η = (−1)n(n−k)+n+1 (−1)k(n−k) d dη + (−1)n(n−k+1)+n+1 d d η
2 +1
= (−1)k d dη + (−1)nk+n+1 d d η.
8.29 Let ω = η + dμ and π = λ + dν. Then (recalling that η and λ are closed),
([ω], [π]) = (η + dμ) ∧ (λ + dν) = η∧λ+η∧dν +dμ ∧ λ+dμ∧dν
M M
Thus, from (8.33), (8.36), and Property (C6) of the covariant derivative
operator,
∇k T = ∂k (T i j )∂i ⊗ d x j + T i j ∇k (∂i ) ⊗ d x j + T i j ∂i ⊗ ∇k (d x j )
= T i j,k ∂i ⊗ d x j + T i j ki ∂ ⊗ d x j − T i j ∂i ⊗ j km d x m
= T i j,k + i k T j − k j T i ∂i ⊗ d x j .
The general case is similar.
8.32
T i1 ...ir j1 ... js ;k X k
− T ki2 ...ir j1 ... js X i1 ;k − T i1 k...ir j1 ... js X i2 ;k − · · · − T i1 ...ir −1 k j1 ... js X ir ;k
+ T i1 ...ir k j2 ... js X k ; j1 + T i1 ...ir j1 k... js X k ; j2 + · · · + T i1 ...ir j1 ... js−1 k X k ; js
= (L X T )i1 ...ir j1 ... js
+ i1 k T i2 ...ir j1 ... js + i2 k T i1 ...ir j1 ... js + · · · + ir k T i1 ...ir−1 j1 ... js
8.34 There are a few ways to do this. One way would be to use (8.103) to write out
the covariant derivatives on both sides, then multiply on the left by the metric
and verify explicitly that both sides are the same. An easier way is just to
note that the covariant derivative obeys a tensorial Leibniz rule and respects
the dual pairing, so
Tm j; = (gim T i j ); = gim; T i j + gim T i j; .
But gim; = 0 by covariant constancy of the metric, so the result follows.
8.35 In a coordinate basis, the Bianchi identity (7.20) can be written
di j − i p ∧ ω p j + ωi p ∧ p j = 0.
Substituting into this equation using (8.35) and (8.45) (and dropping an
irrelevant factor of 1/2) gives
0 = d(R i jk d x k ∧ d x )
− (R i pmn d x m ∧ d x n ) ∧ p q j d x q + i q p d x q ∧ (R p jn d x n ∧ d x ),
= R i jk,m − R i pmk p j + R p jk i mp d x m ∧ d x k ∧ d x ,
= R i jk;m d x m ∧ d x k ∧ d x ,
from which R i j[k;m] = 0 follows. The last equation above holds by virtue of
the symmetry of the lower two Christoffel indices. Specifically, (8.103) yields
R i jk;m = R i jk,m + i mp R p jk − p m j R i pk − p mk R i j p − p m R i jkp .
Antisymmetrizing this expression on k, , and m kills the last two terms and
leaves
R i j[k;m] = R i j[k,m] + i [m| p R p j|k] − p [m| j R i p|k] ,
where the indices between the straight brackets are not antisymmetrized. But
p [| j R i p|mk] = p [m| j R i p|k]
by cyclic permutation of indices.
The other expression is equivalent, because
1 2
∇m {[∇k , ∇ ]∂i } = ∇m R i jk ∂i = R i jk;m ∂i ,
and
[∇m , [∇k , ∇ ]] + cyclic = 0 ⇔ ∇[m [∇k , ∇] ] = 0.
8.36 We have
∇ ab = ab , + a p pb + b p ap .
Geometric manifolds 109
The individual pieces are not tensors, but the whole thing is a tensor (that’s
the raison d’etre for the covariant derivative), so (8.103) gives
∇k (∇ ab ) = [∇ ab ],k − m k (∇m ab ) + a km (∇ mb ) + b km (∇ am )
= ab ,k + a p,k pb + a p pb ,k + b p,k ap + b p ap ,k
− m k [ ab ,m + a mp pb + b mp ap ]
+ a km [ mb , + m p pb + b p mp ]
+ b km [ am , + a p pm + m p ap ].
Now subtract the same thing with k and interchanged. Any term above that
is symmetric in k and will cancel. This means we can throw away the entire
term multiplying m k , because of the symmetry of the Christoffel symbols.
Moreover, the terms involving a derivative of all disappear as well, because
ab ,k + a p pb ,k + b p ap ,k + a km mb , + b km am ,
is symmetric under k ↔ . Finally, the terms
a km b p mp + b km a p pm
also disappear for the same reason. We are left with
( a p,k + a km m p ) pb + ( b p,k + b km m p ) ap
minus the same term with k and interchanged, which gives
R a pk pb + R b pk ap .
8.37 a. The Levi-Civita connection is torsion-free and metric compatible, so
∇ X Y − ∇Y X = [X, Y ] and Xg(Y, Z ) = g(∇ X Y, Z ) + g(Y, ∇ X Z ).
Hence
0 = (L X g)(Y, Z ) = Xg(Y, Z ) − g([X, Y ], Z ) − g(Y, [X, Z ])
= g(∇ X Y, Z ) + g(Y, ∇ X Z )
− g(∇ X Y − ∇Y X, Z ) − g(Y, ∇ X Z − ∇ Z X )
= g(∇Y X, Z ) + g(Y, ∇ Z X ).
The claim now follows.
b. Let X = X k ∂k , Y = ∂i , and Z = ∂ j . Then the result follows immediately
from the result of Part (a) and the covariant constancy of the metric.
0 = g(∇i (X k ∂k ), ∂ j ) + g(∂i , ∇ j (X k ∂k )),
= g(X k ;i ∂k , ∂ j ) + g(∂i , X k ; j ∂k ),
= X k ;i gk j + X k ; j gik = X j;i + X i; j .
110 Geometric manifolds
c. By metric compatibility,
Y g(X, Y ) = g(∇Y X, Y ) + g(X, ∇Y Y ).
But the curve is a geodesic, so ∇Y Y = 0. Also, X is Killing, so
g(∇Y X, Y ) = −g(Y, ∇Y X ) = 0,
whereupon we conclude Y g(X, Y ) = 0.
d. By our previous results,
∇∂i ∂ j = k i j ∂k = g k i j ∂k ,
where
1
i j = (gi, j + gj,i − gi j, ).
2
Choosing coordinates so that X = ∂1 , Y = Y j ∂ j , and Z = Z k ∂k gives
g(∇Y X, Z ) + g(Y, ∇ Z X )
= g(Y j m j1 ∂m , Z k ∂k ) + g(Y j ∂ j , Z k m k1 ∂m )
= Y j Z k (k j1 + jk1 )
= Y j Z k g jk,1 .
As Y and Z are arbitrary, it follows that g jk,1 = 0, as claimed.
e. Let φt be the flow corresponding to X . Then
t
d
φ−t∗ gφt p = g p + (φ−s∗ gφs p ) ds,
ds
0 t
d
= gp + (φ−(s+x)∗ gφ(s+x) p ) ds,
0 dx
t
x=0
d
= gp + φ−s∗ (φ−x∗ gφ(s+x) p ) ds,
0 dx x=0
t
= gp + φ−s∗ (L X g)φs p ds,
0
= gp,
which is precisely the statement that φt is an isometry.
f. It suffices to show that Killing fields are closed under the Lie bracket. But
this follows immediately from (3.102) applied to the metric. If X and Y are
Killing fields, then
L[X,Y ] g = L X LY g − LY L X g = 0,
so [X, Y ] is Killing as well.
Geometric manifolds 111
ξi, j = αi j
ξi = αi j x j + ai
ẋ 1 = x 2
ẋ 2 = −x 1
ẋ 3 = 0.
ẍ 1 = ẋ 2 = −x 1 x 1 = A cos(t + δ).
x 2 = −A sin(t + δ) + B,
but this is only compatible with the first equation if B = 0. Therefore the
integral curves are circles, as promised.
c. This is an elementary exercise in change of variables, but here is one
solution anyway. Recall that in spherical polar coordinates
112 Geometric manifolds
Lξ F = Fi j,k ξ k + Fk j ξ k ,i + Fik ξ k , j d x i ∧ d x j
For ξ (3) we have
0 = Lξ (3) F = Fi j,φ d x i ∧ d x j ,
which just yields
Fi j,φ = 0. (1)
Next, for ξ (1) we have
0 = Lξ (1) F = [Fi j,θ (− sin φ) + Fi j,φ (− cot θ cos φ)
+ Fθ j (− sin φ),i + Fφ j (− cot θ cos φ),i
+ Fiθ (− sin φ), j + Fiφ (− cot θ cos φ), j ] d x i ∧ d x j ,
which yields the following equations (using (1)):
Fτρ,θ = 0, (2)
Fτ θ,θ − Fτ φ csc2 θ cot φ = 0, (3)
Fτ φ,θ + Fτ θ cot φ − Fτ φ cot θ = 0, (4)
Fρθ,θ − Fρφ csc θ cot φ = 0,
2
(5)
Fρφ,θ + Fρθ cot φ − Fρφ cot θ = 0, (6)
Fθφ,θ − Fθφ cot θ = 0. (7)
Likewise,
0 = Lξ (2) F = [Fi j,θ (cos φ) + Fi j,φ (− cot θ sin φ)
+ Fθ j (cos φ),i + Fφ j (− cot θ sin φ),i
+ Fiθ (cos φ), j + Fiφ (− cot θ sin φ), j ] d x i ∧ d x j ,
which yields the following equations (using (1) again):
114 Geometric manifolds
Fτρ,θ = 0, (8)
Fτ θ,θ + Fτ φ csc θ tan φ = 0,
2
(9)
Fτ φ,θ − Fτ θ tan φ − Fτ φ cot θ = 0, (10)
Fρθ,θ + Fρφ csc θ tan φ = 0,
2
(11)
Fρφ,θ − Fρθ tan φ − Fρφ cot θ = 0, (12)
Fθφ,θ − Fθφ cot θ = 0. (13)
where B(τ, ρ) is some arbitrary function. By (1) and (14) Fτρ = A(τ, ρ)
for some arbitrary function A. Thus
as promised.
b. From (8.91) we get
*
(dτ ∧ dρ) ∧ (dθ ∧ dφ) = g((dτ ∧dρ), dθ ∧dφ) |G| dτ ∧dρ ∧dθ ∧dφ
*
= α |G|g(dθ ∧dφ, dθ ∧dφ) dτ ∧dρ ∧dθ ∧dφ,
so
*
α |G|g(dθ ∧ dφ, dθ ∧ dφ) = 1.
θ 0̂ = a dτ,
θ 1̂ = b dρ,
θ 2̂ = r dθ,
θ 3̂ = r sin θ dφ.
ds 2 = gâ b̂ θ â ⊗ θ b̂
where gâ b̂ is the Minkowski metric.
b. By the antisymmetry of the connection matrices with downstairs indices,
we have (with Greek indices running from 0 to 3 and Latin indices running
from 1 to 3):
ω0̂ 0̂ = g 0̂μ̂ ωμ̂0̂ = g 0̂0̂ ω0̂0̂ = 0
ω0̂ î = g 0̂μ̂ ωμ̂î = g 0̂0̂ ω0̂î = −ω0̂î
ωî 0̂ = g î μ̂ ωμ̂0̂ = g î î ωî 0̂ = ωî 0̂ = −ω0̂î = ω0̂ î
ωî ĵ = g î μ̂ ωμ̂ ĵ = g î î ωî ĵ = ωî ĵ = −ω ĵ î = −ω ĵ î .
c. We have
1 −1/2 2m −1/2 m
dθ = !
0̂
dr ∧ dt = −! θ 0̂ ∧ θ 1̂
2 r2 r2
= −ω0̂ â ∧ θ â
= −ω0̂ 0̂ ∧ θ 0̂ − ω0̂ 1̂ ∧ θ 1̂ − ω0̂ 2̂ ∧ θ 2̂ − ω0̂ 3̂ ∧ θ 3̂
= −ω0̂ 1̂ ∧ θ 1̂ .
Now we guess that
m m
ω0̂ 1̂ = ω1̂ 0̂ = !−1/2 dt, θ 0̂ =
r2 r2
which certainly satisfies the structure equation. (The reason this is a guess
is because ω0̂ 1̂ could have a term proportional to dr . It turns out that the
other structure equations rule out this possibility, but we would have to
write them all out in order to verify this.)
d. We have
0̂ 1̂ = dω0̂ 1̂ + ω0̂ μ̂ ∧ ωμ̂ 1̂
= dω0̂ 1̂
2m
= 3 dt ∧ dr
r
2m 0̂
= 3 θ ∧ θ 1̂ ,
r
Geometric manifolds 117
and
1̂ 2̂ = dω1̂ 2̂ + ω1̂ μ̂ ∧ ωμ̂ 2̂
= dω1̂ 2̂ + ω1̂ 3̂ ∧ ω3̂ 2̂
1 −1/2 2m
=− ! − 2 dr ∧ dθ
2 r
m 1̂
= 3 θ ∧ θ 2̂ .
r
8.41 a. Begin with the parallel transport equation (8.63)
dγ i ∂Y k
+ ijYk j
= 0.
dt ∂xi
We must choose a parameterization for γ . As the result is parameterization
independent we choose the simplest one, namely
θ(t) = θ0 and φ(t) = t.
Then θ̇ = 0 and φ̇ = 1, so (8.63) reduces to
Y k ,φ + k φθ Y θ + k φφ Y φ = 0.
Plugging in the known values of the Christoffel symbols gives
Y θ ,φ − sin θ0 cos θ0 Y φ = 0
and
Y φ ,φ + cot θ0 Y θ = 0.
Now differentiate the second equation and substitute the first to get
Y φ ,φφ + cos θ0 Y φ = 0,
whose solution is
Y φ = A cos(φ cos θ0 ) + B sin(φ cos θ0 ).
Plug this back into the differential equation for Y φ and solve for Y θ . This
gives
1
Yθ = − Y φ ,φ
cot θ0
= sin θ0 [A sin(φ cos θ0 ) − B cos(φ cos θ0 )].
From the initial conditions we get
a = Y θ (θ0 , 0) = −B sin θ0
118 Geometric manifolds
and
b = Y φ (θ0 , 0) = A,
so the general solutions are
θ
Y cos(φ cos θ0 ) sin θ0 sin(φ cos θ0 ) a
= .
Yφ − sin(φ cos θ0 )/ sin θ0 cos(φ cos θ0 ) b
b. The angle between the parallel transported vector Y and the original vector
Y0 is
g(Y, Y0 )
cos ψ = √ .
g(Y, Y )g(Y0 , Y0 )
We have
g(Y, Y ) = gθθ (Y θ )2 + gφφ (Y φ )2
= [a cos(φ cos θ0 ) + b sin θ0 sin(φ cos θ0 )]2
+ sin2 θ0 [−a sin(φ cos θ0 )/ sin θ0 + b cos(φ cos θ0 )]2
= a 2 + b2 sin2 θ0
= g(Y0 , Y0 ),
which just confirms that parallel transport preserves the length of vectors.
Also
g(Y, Y0 ) = gθθ Y θ a + gφφ Y φ b
= a[a cos(φ cos θ0 ) + b sin θ0 sin(φ cos θ0 )]
+ b sin2 θ0 [−a sin(φ cos θ0 )/ sin θ0 + b cos(φ cos θ0 )]
= (a 2 + b2 sin2 θ0 ) cos(φ cos θ0 ),
from which it follows that
cos ψ = cos(φ cos θ0 ),
or
ψ = φ cos θ0 .
As we walk with an attitude around a latitude, φ turns through a full 2π,
so the vector turns through an angle
ψ = 2π cos θ0 .
8.42 Rewrite (8.114) as
φ̈/φ̇ = −2(cot θ)θ̇ .
Geometric manifolds 119
J sec2 (at + c)
= ,
1 + (J/a)2 tan2 (at + c)
120 Geometric manifolds
or
J sec2 (at + c)
φ= dt.
1 + (J/a)2 tan2 (at + c)
Set x = (J/a) tan(at + c) so d x = J sec2 (at + c) dt to get
dx
φ= = arctan x + φ0 = arctan((J/a) tan(at + c)) + φ0 ,
1 + x2
or
which is (8.118).
For brevity, set β := J/a, α 2 = 1 − β 2 , and b := at + c. Then
* 5
sin θ = 1 − cos2 θ = cos2 b + β 2 sin2 b,
so
−α sin b
cot θ = *
cos b + β 2 sin2 b
2
α
=− 2
β + cot2 b
α/β
=−
1 + cot2 (φ − φ0 )
= C sin(φ − φ0 ),
where
α *
C := − = − (a/J )2 − 1.
β
This yields (8.119).
Now multiply both sides of (8.119) by sin θ and expand the right hand side
to get
z = Ay − Bx,
Equivalently,
Ri jk = K p (gik g j − gi g jk ).
Then plugging into (8.120) gives K (") = K p for any two plane ".
Conversely, assume K (") = K p for any two plane. By definition,
The same proof works backwards, so (8.122) implies (8.121). (You have
to change coordinates at the last moment from an orthonormal basis to a
general coordinate basis.)
e. The torsion free condition is dθ + ω ∧ θ = 0, so
dâ b̂ = d K ∧ θ â ∧ θ b̂ + K dθ â ∧ θ b̂ − K θ â ∧ dθ b̂
= d K ∧ θ â ∧ θ b̂ − K ωâ ĉ ∧ θ ĉ ∧ θ b̂ + K θ â ∧ ωb̂ ĉ ∧ θ ĉ
= d K ∧ θ â ∧ θ b̂ − K ωâ ĉ ∧ θ ĉ ∧ θ b̂ − K θ â ∧ θ ĉ ∧ ωb̂ ĉ
= d K ∧ θ â ∧ θ b̂ − ωâ ĉ ∧ ĉb̂ − â ĉ ∧ ωb̂ ĉ
= d K ∧ θ â ∧ θ b̂ − ωâ ĉ ∧ ĉb̂ + â ĉ ∧ ωĉ b̂ .
d K ∧ θ â ∧ θ b̂ = 0.
R = g i j Ri j = g i j R k ik j = g i j g k Rik j
= 2g 11 g 22 R1212 + 2(g 12 )2 R1221
= 2g 11 g 22 − 2(g 12 )2 R1221
= 2G −1 R1212 ,
where G is the determinant of the metric tensor. On the other hand, the
Gaussian curvature in two dimensions is, with X = ∂1 and Y = ∂2 ,
R1212
K = ,
g11 g22 − g12
2
Integrating gives
ln ẋ = 2 ln y + c ⇒ ẋ = cy 2 .
Using the hint we can write the second geodesic equation as
y ÿ − ẏ 2 ẋ 2
f˙ = = − = −c ẋ.
y2 y2
There are two cases. If c = 0 we have ẋ = 0 and f = ẏ/y = constant.
The solutions to these equations are
x = constant and y = y0 ebt ,
for some constant b. These are straight vertical lines (with a funny
parameterization).
If c = 0 we integrate to get
f = −cx + e
for some other constant e. Multiply through by y 2 = ẋ/c to get
y ẏ + x ẋ = a ẋ,
where a := e/c. Integrating both sides gives
1 2
(y + x 2 ) = ax + q
2
for some other constant q. This can be rewritten as
y 2 + (x − a)2 = r 2 ,
*
which is the equation of a circle of radius r = 2q + a 2 centered at a.
Using y 2 = ẋ/c again gives
ẋ + c(x − a)2 = cr 2 ,
which separates to
dx
= c dt.
r2 − (x − a)2
Changing variables to z = (x − a)/r gives
dz c ct
= dt ⇒ tanh−1 z = + h,
1−z 2 r r
and thus
x = r tanh(ct/r + h) + a.
126 Geometric manifolds
Rx yx y = gx x R x yx y
1
= 2 ( x yy,x − x x y,y + x x x x yy + x x y y yy
y
− x yx x x y − x yy y x y )
1
= − 4.
y
As the denominator is y −4 , we conclude that K = −1.
8.46 a. Let
ai z + bi
TAi (z) =
ci z + di
for i = 1, 2. Then
a2 (a1 z + b1 )/(c1 z + d1 ) + b2
TA2 (TA1 (z)) =
c2 (a1 z + b1 )/(c1 z + d1 ) + d2
a2 (a1 z + b1 ) + b2 (c1 z + d1 )
=
c2 (a1 z + b1 ) + d2 (c1 z + d1 )
(a2 a1 + b2 c1 )z + (a2 b1 + b2 d1 )
= .
(c2 a1 + d2 c1 )z + (c2 b1 + d2 d1 )
But, observe that
a2 b2 a1 b1 a a + b2 c1 a2 b1 + b2 d1
= 2 1 .
c2 d2 c1 d1 c2 a1 + d2 c1 c2 b1 + d2 d1
It follows that
T A2 ◦ T A1 = T A2 A1 , (1)
Geometric manifolds 127
TA ◦ TA−1 = TA−1 ◦ TA = TI ,
(Note that the inverse exists because A ∈ G L(2, C).) Therefore all the
group axioms are satisfied.
b. Equation (1) together with TI (z) = z shows that the map A → T A is a
group homomorphism. The kernel consists of those matrices A that also
map to TI . In other words,
az + b
= z.
cz + d
Cross multiplying gives
cz 2 + (d − a)z − b = 0,
(For the second map we used the fact that ad − bc = 1.) Observe that the
denominator appearing in these two expressions is always positive because
c and d cannot vanish simultaneously without violating the determinant
condition. It follows that the map carries H2+ to H2+ and is everywhere
differentiable. (The differentiability of the inverse map follows similarly
by sending A to A−1 .)
d. This one is a chore in either real or complex coordinates. A better way to
do it is to decompose an arbitrary Möbius transformation into a product of a
translation, an inversion, and a dilation and prove that each transformation
is an isometry, but as we did not discuss such decompositions we will do
the problem the long way instead.
First note that
dz = d x + idy
and
d z̄ = d x − idy,
so
dz d z̄ = d x 2 + dy 2 .
Also,
z − z̄
y=
2i
so
d x 2 + dy 2 dz d z̄
ds 2 = = −4 .
y 2 (z − z̄)2
In order to avoid confusing the differential dz with the product of d and
z we will denote the differential dz by η (and its complex conjugate by
η := d z̄). Thus,
aη(cz + d) − (az + b)(cη)
TA∗ (η) = dT A∗ z =
(cz + d)2
η(ad − bc) η
= = .
(cz + d)2 (cz + d)2
Similarly, because A is real,
η
TA∗ (η) = dTA∗ z̄ = dT A∗ z = .
(cz̄ + d)2
Thus
ηη
TA∗ (ηη) = .
|cz + d|4
Geometric manifolds 129
Lastly,
az + b a z̄ + b
TA∗ (z − z̄) = −
cz + d cz̄ + d
(az + b)(cz̄ + d) − (a z̄ + b)(cz + d)
=
|cz + d|2
(z − z̄)
= ,
|cz + d|2
so
(z − z̄)2
TA∗ (z − z̄)2 = .
|cz + d|4
Putting it all together gives
TA∗ (η)TA∗ (η)
TA∗ ds 2 = −4
T ∗ (z − z̄)2
A
ηη |cz + d|4
= −4
|cz + d|4 (z − z̄)2
ηη
= −4 = ds 2 .
(z − z̄)2
8.47 a. Following the hint we get
h = σ ∗ g = σ ∗ d x ⊗ σ ∗ d x + σ ∗ dy ⊗ σ ∗ dy + σ ∗ dz ⊗ σ ∗ dz.
Now
∂x ∂x
σ ∗ d x = dσ ∗ x = d(x ◦ σ ) = du + dv,
∂u ∂v
and similarly for σ ∗ dy and σ ∗ dz, so
∂x ∂x ∂x ∂x
h= du + dv ⊗ du + dv + · · ·
∂u ∂v ∂u ∂v
= g(σu , σu ) du 2 + 2g(σu , σv ) dudv + g(σv , σv ) dv 2 .
b. We have
σθ = (cos θ cos φ, cos θ sin φ, − sin θ),
σφ = (− sin θ sin φ, sin θ cos φ, 0),
so
E = g(σθ , σθ ) = 1,
F = g(σθ , σφ ) = 0,
G = g(σφ , σφ ) = sin2 θ,
130 Geometric manifolds
and thus
g = dθ 2 + sin2 dφ 2 ,
as expected.
8.48 We have
σu = (−a cosh v sin u, a cosh v cos u, 0),
σv = (a sinh v cos u, a sinh v sin u, a).
The metric tensor components are therefore
E = g(σu , σu ) = a 2 cosh2 v,
F = g(σu , σv ) = 0,
G = g(σv , σv ) = a 2 cosh2 v.
By inspection,
θ 1̂ = a cosh v du,
θ 2̂ = a cosh v dv.
Thus,
sinh v 1̂
dθ 1̂ = a sinh v dv ∧ du = − θ ∧ θ 2̂ ,
a cosh2 v
dθ 2̂ = 0.
There is only one independent connection form, namely ω1̂ 2̂ . Using the
torsion free condition dθ = −ω ∧ θ we guess
sinh v 1̂
ω1̂ 2̂ = θ = tanh v du.
cosh2 v
(The guess is consistent, because if we had added a term of the form
f (u, v) dv to ω1̂ 2̂ it would have contradicted the equation for dθ 2̂ .) The
corresponding independent curvature two form component is
1̂2̂ = 1̂ 2̂ = dω1̂ 2̂ + ω1̂ 2̂ ∧ ω2̂ 1̂
= sech2 v dv ∧ du
= −a −2 sech4 v θ 1̂ ∧ θ 2̂ ,
whereupon we conclude that K = −1/a 2 cosh4 v.
8.49 We have
σu = (−(b + a cos v) sin u, (b + a cos v) cos u, 0),
σv = (−a sin v cos u, −a sin v sin u, a cos v).
Geometric manifolds 131
By inspection,
θ 1̂ = (b + a cos v) du,
θ 2̂ = a dv.
Thus,
dθ 1̂ = −a sin v dv ∧ du,
dθ 2̂ = 0.
There is only one independent connection form, namely ω1̂ 2̂ . Using the
torsion free condition dθ = −ω ∧ θ we guess
(Rg∗−1 m)(X h , Yh )h
= m(Rg−1 ∗ X h , Rg−1 ∗ Yh )hg−1 (Equation (3.90)),
= B(L gh −1 ∗ Rg−1 ∗ X h , L gh −1 ∗ Rg−1 ∗ Yh ) (Equation 8.124),
= B(Rg−1 ∗ L gh −1 ∗ X h , Rg−1 ∗ L gh −1 ∗ Yh ) (Ra and L b commute),
= B(Rg−1 ∗ L g∗ L h −1 ∗ X h , Rg−1 ∗ L g∗ L h −1 ∗ Yh ) (chain rule),
= B(L h −1 ∗ X h , L h −1 ∗ Yh ) (Ad invariance of B),
= m(X h , Yh ) (Equation 8.124).
132 Geometric manifolds
b. Let X , Y , and Z be left invariant vector fields. The Koszul formula reads
(ad Z (X ), Y ) + (ad Z (Y ), X ) = 0.
(X g , Yg ) = (L g−1 ∗ X g , L g−1 ∗ Yg )e = (X e , Ye )
so (X, Y ) is constant and all terms of the form X (Y, Z ) vanish. As Z was
arbitrary (G is parallelizable, so we can always find a basis of left invariant
vector fields at any point), we conclude that
2∇ X Y = [X, Y ].
R(X, Y )Z = ∇ X ∇Y Z − ∇Y ∇ X Z − ∇[X,Y ] Z
1 1
= ([X, [Y, Z ]] − [Y, [X, Z ]]) − [[X, Y ], Z ]
4 2
1
= − [[X, Y ], Z ].
4
d. From Part (c) and Exercise 3.52c,
1 1
(X, R(X, Y )Y ) = − (X, [[X, Y ], Y ]) = ([X, Y ], [X, Y ]).
4 4
Now apply (8.120).
e. According to (8.53), the Ricci tensor is given by
8.51 a. We have
dθ 1̂ = 0,
f 1̂
dθ 2̂ = f dr ∧ dθ = θ ∧ θ 2̂ ,
f
f 1̂ 1
dθ 3̂ = f sin θ dr ∧ dφ + f cos θ dθ ∧ dφ = θ ∧ θ 3̂ + cot θ θ 2̂ ∧θ 3̂ .
f f
f 1 − f 2
− =K = .
f f2
If K = −1 the first equation gives f − f = 0 which has solution f =
a sinh r + b cosh r . Applying the boundary condition requires b = 0. But
the second equation requires f 2 − f 2 = 1, so a = ±1 and f = ± sinh r .
If K = 0 the first equation gives f = ar + b, while the second requires
134 Geometric manifolds
e. We have
εi1 ...in d x i1 ∧ · · · ∧ d x in = n! d x 1 ∧ · · · ∧ d x n ,
so
1 *
i1 ...in d x i1 ∧ · · · ∧ d x in = |G| d x 1 ∧ · · · ∧ d x n = σ.
n!
f. Using Part (a) again gives
∂ y i1 ∂ y in i1 ...in
· · · ε = J εi1 ...in ,
∂x i 1 ∂x i n
so η = J −1 .
g. Again by Part (a),
i1 ...in = g i1 j1 · · · g in jn j1 ... jn
*
= |G|g i1 j1 · · · g in jn ε j1 ... jn
*
= |G|G −1 ε j1 ... jn
= ±(sgn G) j1 ... jn .
because εa1 ...an = −1, 0, 1 and the determinant of the metric is built from
the metric components.
8.53 a. εi1 ...in ε j1 ... jn and δ ij11...i n
... jn have the same symmetry properties, namely anti-
symmetry under the interchange of any pair of i or j indices. This is true
for the Levi-Civita symbols by definition, and true for the determinant,
because swapping two i indices swaps two columns, while swapping two
j indices swaps to rows. Moreover, if i s = js = s then both sides equal 1.
Hence two two quantities must be equal.
To prove the identity in general, proceed by reverse induction, assuming it
holds for k + 1 and proving it for k. By hypothesis,
i ...i
εi1 ...ik+1 ik+2 ...in ε j1 ... jk+1 ik+2 ...in = (n − k − 1)!δ j11 ... jk+1
k+1
.
⎜ . .. .. .. ⎟
⎜ .. . . . ⎟
= det ⎜ ⎜ ik ik ik ⎟
⎟
⎝ δ j1 ··· δ jk δik+1 ⎠
i i i k+1
δ jk+1
1
··· δ jk+1k
δik+1
1 î1 i2 ...ik+1 i î ...i
= (−1)k δiik+1 δ j1 ... jk − δiik+1 2
δ j11 ...2 jk k+1
i ...î i ik+1 i 1 ...i k
· · · + (−1)k−1 δiik+1 k
δ j11 ... jkk k+1 + (−1)k δik+1 δ j1 ... jk
= (−1)k δ îj11i...2 ...i
jk
k i1
− δ ij11î...2 ...i
jk
k i2
i 1 ...i k
· · · + (−1)k−1 δ ij11...i ... jk + (−1) nδ j1 ... jk
k k
i1 ...ik
= −1! − 1
"#· · · − 1
$ +n δ j1 ... jk
k terms
= (n − k)δ ij11...i
... jk .
k
In the third line the caret indicates the absence of that index. The signs are
derived by permuting the indices in the δ symbols back to their canonical
forms.
Actually, there is a much easier way to prove the identity in general. Begin
again from result for k = n (base case of the induction). Next, observe
that for each fixed sequence of indices (i k+1 , . . . , i n ), the remaining indices
(i 1 , . . . , i k ) and ( j1 , . . . , jk ) must be permutations of one another, and so
the base case implies that, for any fixed collection (i k+1 , . . . , i n ) we have
(In this formula repeated indices are not summed.) Now we just observe
that there are (n − k)! choices for the index set (i k+1 , . . . , i n ).
b. We have
where we must choose the plus sign in Riemannian case, and the minus
sign in the Lorentzian case.
8.54 Recall the definition (8.91) of the Hodge dual:
α ∧ β = g(α, β)σ,
Hence
1 ···i n
(a ∗ ) jk+1 ,..., jn δ jk+1 σ = (a ∗ )ik+1 ...in σ,
i
g(α, β)σ = k+1 ··· jn
(n − k)!
and the claim is proved.
8.55 From Exercises 8.52, 8.53 and 8.54 we have
1
α ∧ β = ai ...i b∗ d x i1 ∧ · · · ∧ d x in
k!(n − k)! 1 k ik+1 ...in
1
= ai ...i b j1 ... jk j1 ... jk ik+1 ...in d x i1 ∧ · · · ∧ d x in
(k!) (n − k)! 1 k
2
1 j1 ... jk
*
= a i ...i b |G|ε j1 ... jk ik+1 ...in εi1 ...in d x 1 ∧ · · · ∧ d x n
(k!)2 (n − k)! 1 k
138 Geometric manifolds
1
= ai ...i b j1 ... jk δ ij11...i
... jk σ
k
(k!)2 1 k
1
= ai1 ...ik b j1 ... jk σ.
k!
8.56 a. According to Exercise 1.36,
∂G
= Gg i j .
∂gi j
By the chain rule
∂G ∂G ∂gi j
= = Gg i j gi j,k .
∂xk ∂gi j ∂ x k
Recall that
1
ki j = (g jk,i + gik, j − gi j,k ),
2
and observe that
g ik g jk,i = g ik g ji,k .
Thus,
1 ik 1 ∂G
i i j = g ik ki j = g gik, j = G −1 j .
2 2 ∂x
It follows that
∂G 1/2 1 ∂G
= G −1/2 j = G 1/2 i i j ,
∂x j 2 ∂x
and therefore
b. We have
L X σ = L X (G 1/2 ) d x 1 ∧ · · · ∧ d x n + G 1/2 L X (d x 1 ∧ · · · ∧ d x n ).
so
n
L X (d x ∧ · · · ∧ d x ) =
1 n
d x 1 ∧ · · · ∧ L X (d x j ) ∧ · · · ∧ d x n
j=1
= X j , j d x1 ∧ · · · ∧ d xn.
Adding together the two terms and using the result of Part (a) gives
because mixed partials commute and the Christoffel symbols are symmet-
ric in the lower indices.
8.57 a. This is just a change of variables problem. By definition,
∂xi ∂x j
gi j = gi j ,
∂ yi ∂ y j
where the primed coordinates are the spherical polar ones and gi j = δi j .
We have
∂x ∂y ∂z
= sin θ cos φ = sin θ sin φ = cos θ
∂r ∂r ∂r
∂x ∂y ∂z
= r cos θ cos φ = r cos θ sin φ = −r sin θ
∂θ ∂θ ∂θ
∂x ∂y ∂z
= −r sin θ sin φ = r sin θ cos φ = 0,
∂φ ∂φ ∂φ
so the diagonal elements are
2 2 2
∂x ∂y ∂z
grr = + + = 1,
∂r ∂r ∂r
2 2 2
∂x ∂y ∂z
gθθ = + + = r 2,
∂θ ∂θ ∂θ
2 2 2
∂x ∂y ∂z
gφφ = + + = r 2 sin2 θ,
∂φ ∂φ ∂φ
and all the off diagonal terms vanish.
140 Geometric manifolds
by virtue of the symmetry of the Christoffel symbols in the last two indices.
Now
(X r ),θ = (X r̂ ),θ ,
(X r ),φ = (X r̂ ),φ ,
(X θ ),r = r X θ̂ ,r ,
(X θ ),φ = r X θ̂ ,φ ,
(X φ ),r = sin θ r X φ̂ ,r ,
(X φ ),θ = r sin θ X φ̂ ,θ .
and note that ai1 ...ik is already totally antisymmetric. We can split up the
sum over Sk+1 by fixing a value for σ (1) and then summing over all k!
permutations of the remaining indices. The sign of such a permutation is
just the product of the sign of the identity permutation with one element
moved to the front times the sign of the rest of the permutation. For exam-
ple, the sign of the permutation 2143 is −1, the sign of the permutation
143 is −1, and their product is +1. Reasoning in this way, we get
1
k+1
∇[i1 ai2 ...ik+1 ] = (−1)r −1 (−1)π ∇ir aiπ(1) ...iπ(k) ,
(k + 1)! r =1
π∈S(r)
where S(r ) stands for the permutation subgroup that fixes the first element.
Summing over the elements in each subgroup gives
1 k+1
∇[i1 ai2 ...ik+1 ] = (−1)r−1 ∇ir ai1 ...îr ...ik+1 .
(k + 1) r =1
The second term vanishes, while the first term becomes, after permuting
some indices,
Therefore
δα = (−1)nk+n+1 d α
1
=− g p1 j1 ∇ p1 (a j1 jn−k+2 ... jn )
(k − 1)!
× d x jn−k+2 ∧ · · · ∧ d x jn .
c. Define
1
β = δα = βi2 ...ik d x i2 ∧ · · · ∧ d x ik ,
(k − 1)!
where
Then
dδα = dβ
1
k
= (−1)r−1 ∇ir βi1 ...îr ...ik d x i1 ∧ · · · ∧ d x ik
k! r =1
1
k
= (−1)r ∇ir ∇ j a ji1 ...îr ...ik d x i1 ∧ · · · ∧ d x ik ,
k! r =1
Next, if
1
β= bi ...i d x i1 ∧ · · · ∧ d x ik+1 ,
(k + 1)! 1 k+1
then
1 j
δβ = −
∇ b ji1 ...ik d x i1 ∧ · · · ∧ d x ik .
k!
Defining β := dα we have
k+1
bi1 ...ik+1 = (−1)r −1 ∇ir ai1 ...îr ...ik+1
r =1
k+1
= ∇i1 ai2 ...ik+1 + (−1)r −1 ∇ir ai1 ...îr ...ik+1 ,
r =2
and thus
3 4
1 k
δdα = − ∇ 2 ai1 ...ik + (−1)r ∇ j ∇ir a ji1 ...îr ...ik d x i1 ∧ · · · ∧ d x ik .
k! r =1
(a)i1 ...ik = −∇ 2 ai1 ...ik − (−1)r ∇ j ∇ir − ∇ir ∇ j a ji1 ...îr ...ik
r =1
k
= −∇ ai1 ...ik +
2
∇ j ∇ir − ∇ir ∇ j ai1 ...ir−1 jir+1 ...ik ,
r =1
where we permuted the j using the antisymmetry of the a’s. Now we have
to employ Ricci’s identity (8.109), then massage this expression until it
takes the desired form. To this end, we write
−[∇ j , ∇ir ]ai1 ...ir −1 jir+1 ...ik
r −1
= ai1 ...is−1 pis+1 ...ir −1 jir+1 ...ik R p is j ir
s=1
+ ai1 ...ir −1 pir +1 ...ik R p j j ir
k
+ ai1 ...ir−1 jir +1 ...is−1 pis+1 ...ik R p is j ir .
s=r +1
Also,
R p is j ir + R pj ir is + R p ir is j = 0,
so antisymmetrizing on p and j gives
1 1
R [ p is j] ir = − R pj ir is = R pj is ir .
2 2
Hence
[∇ j , ∇ir ]ai1 ...ir −1 jir +1 ...ik
= ai1 ...ir−1 pir+1 ...ik R p ir
1
k
− ai ...i ji ...i pi ...i R j p ir is .
2 s=r 1 r−1 r+1 s−1 s+1 k
k
= a ji1 ...îr ...ik aii1 ...îr ...ik R i j
r =1
= ka ji2 ...ik aii2 ......ik R i j .
Similarly,
1 i1 ...ik
a ai1 ...ir−1 iir +1 ...is−1 jis+1 ...ik R i j ir is
2 r =s
1 i1 ...ir ...is ...ik
= a ai1 ...ir−1 iir +1 ...is−1 jis+1 ...ik R i j ir is
2 r=s
1 pqi3 ...ik
= a ai ji3 ...ik R i j pq
2 r=s
1
= k(k − 1)a pqi3 ...ik ai ji3 ...ik R i j pq .
2
Hence the claim is proved.
b. By Weitzenböck, for any function f ,
f = −∇ 2 f.
so
Also,
a pqi3 ...ik ai ji3 ...ik R i j pq = K a pqi3 ...ik ai ji3 ...ik (δ ip δqj − δqi δ pj )
= 2K k!α2 .
Hence
dσ = (n + 1)ω,
so
An = σ =c τ =c dσ = (n + 1)c ω
Sn ∂ B n+1 B n+1 B n+1
= (n + 1)cVn+1 = c An ,
and therefore c = 1.
c. We follow the proof in ([5], p. 411). The domain of integration for the
(n + 1)-ball is
(x 1 )2 + · · · + (x n+1 )2 ≤ 1,
which is equivalent to
(x 3 )2 + · · · + (x n+1 )2 ≤ 1 − (x 1 )2 + (x 2 )2 and (x 1 )2 + (x 2 )2 ≤ 1.
Geometric manifolds 149
Thus,
Vn+1 = ω
x2 ≤1
= 1
dx dx 2
d x 3 . . . d x n+1 .
(x 1 )2 +(x 2 )2 ≤1 (x 3 )2 +···+(x n+1 )2
≤1−(x 1 )2 +(x 2 )2
9.1 We just extend the commutative diagram used to define ‘Deg’ to get
g∗ f∗
H n (P) −−−→ H n (N ) −−−→ H n (M)
⏐ ⏐ ⏐
0 ⏐ 0 ⏐ 0 ⏐
P ; N ; M;
(1)
Deg(g) Deg( f )
R −−−→ R −−−→ R.
The claim then follows from the fact that (g ◦ f )∗ = f ∗ ◦ g ∗ .
9.2 The form d x is globally defined on the torus. It is closed but not exact,
because x is not a single valued function on the torus. Nevertheless, in the
neighborhood of a point x is a well-defined function, so f ∗ (d x) = d f ∗ x
makes sense locally. Specifically, f ∗ (d x) is a closed one-form on the sphere
(because ‘d’ is a local operation). Let λ := f ∗ (d x) and η := f ∗ (dy). Every
closed one-form on the 2-sphere is exact (because H 1 (S 2 ) = 0), so λ = dg
and η = dh for some (globally defined) functions g and h on the 2-sphere.
But then
dg ∧ dh = d(g ∧ dh) = g ∧ dh = 0,
M M ∂M
151
152 The degree of a smooth map
9.5 The map F(t, x) is clearly a smooth homotopy between f and the antipodal
map, provided it exists. But the denominator vanishes only if
= n! pf(Q) σ.
1
λ= Qi j θ i ∧ θ j
2
1
= Q i j Ai k A j θ k ∧ θ
2
1
= (A T Q A)k θ k ∧ θ
2
1
= Q k θ k ∧ θ ,
2
The degree of a smooth map 153
so
Q = A T Q A.
But under this change of basis,
σ → σ = (det A−1 )σ = (det A)−1 σ,
so
n! pf(Q)σ = λn = n! pf(Q )σ = n! pf(A T Q A)(det A)−1 σ,
and therefore
pf(A T Q A) = (det A) pf(Q).
Appendix D
Riemann normal coordinates
D.1 In the specified coordinates the Christoffel symbols vanish. From the defini-
tion of the Riemann tensor,
R i jk = i j,k − i k j, .
Lowering all the indices with the Euclidean metric gives
Ri jk = ij,k − ik j,
1
= (g ji, + gi, j − gj,i ),k − (k ↔ )
2
1
= (gi, jk − gj,ik − gki, j + gk j,i )
2
1
= (gi, jk − g j,ik − gik, j + g jk,i ).
2
D.2 There are n choices for i. By the symmetry of the partial derivatives we must
count the number of multisets of size
3 chosen from n+2
a set of size n. By the
solution to Exercise 2.21a, this is 3 , so a = n 3 .
n+2
154
Riemann normal coordinates 155
m
F.1 First we show that (2) and (3) are equivalent. If dθ i = j=1 A j θ then
i j
because θ k (X ) = 0 if k = 1, . . . , m.
If is involutive then θ i ([X, Y ]) = 0, so a i jk = 0 for j, k = m −n, . . . , n,
and (3) holds. Conversely, if (3) holds then dθ i (X, Y ) = 0, which means
θ i ([X, Y ]) = 0 for i = 1, . . . , m, and this implies [X, Y ] ∈ .
156
Appendix G
The topology of electrical circuits
G.1 Let G be a tree. Pick an edge e with endpoints p and q. If removing e failed
to disconnect G, there would be a path from p to q in G − e. But then adding
e to that path would produce a cycle, a contradiction. Conversely, let G be a
connected graph with the property that removing every edge disconnects G.
Then G cannot have any cycles and must therefore be a tree.
157
Appendix H
Intrinsic and extrinsic curvature
H.1 We have
f X (gY ) = f ∇
∇ X (gY ) = f X (g)Y + f g ∇
X Y
and
so
f X (gY ) − ∇ f X (gY ) = f g(∇
α( f X, gY ) = ∇ X Y − ∇ X Y ) = f gα(X, Y ).
where ‘(X ↔ Y )’ means ‘repeat the previous terms with X and Y inter-
changed’. Taking projections onto the tangent and normal spaces yields the
Gauss and Codazzi-Mainardi equations immediately:
P( R(X, X α(Y, Z ) − ∇
Y )Z ) = R(X, Y )Z + P(∇ Y α(X, Z ))
and
(1 − P)( R(X, Y )Z ) = α(X, ∇Y Z ) − α(Y, ∇ X Z ) − α([X, Y ], Z )
+ (1 − P)(∇X α(Y, Z ) − ∇Y α(X, Z )).
158
Intrinsic and extrinsic curvature 159
By Equation (8.44),
Rabcd = g(R(ec , ed )eb , ea ),
so choosing X = ∂k , Y = ∂ , Z = ∂ j , and W = ∂i gives
Ri jk = bik b j − bi b jk .
where we used the fact that bi j = b ji .
For the Codazzi-Mainardi equation, we first observe that the components
of the second fundamental form are related to those of the shape operator by
the metric function. Specifically, if we choose a coordinate basis and set
S(∂i ) = S k i ∂k ,
then
bi j = II(∂i , ∂ j ) = g(S(∂i ), ∂ j ) = S k i gk j =: S ji .
Second, recall that [∂i , ∂ j ] = 0. Hence, by Equation (H.14) with X = ∂i ,
Y = ∂ j , and Z = ∂k ,
0 = g(∇∂i Sξ (∂ j ) − ∇∂ j Sξ (∂i ), ∂k )
= g(∇∂i (S j ∂ ) − ∇∂ j (S i ∂ ), ∂k )
= g(S j;i ∂ − S i; j ∂ , ∂k )
= gk (S j;i − S i; j )
= Sk j;i − Ski; j
= (Sk j,i − Sm j m ki − Skm m ji ) − (i ↔ j)
= (bk j,i − bm j m ki ) − (i ← j).
H.7 a. The ambient space is Euclidean, so the ambient connection ∇ just reduces
to the ordinary derivative. In particular, if X is a tangent vector field on ,
X n = −X (n k )∂k .
Sn (X ) = −∇
Now let X = σ∗ (∂/∂u). Then
J = IIn (X, X ) = g(Sn (X ), X )
= −X (n k )(σu ) j g(∂k , ∂ j ) = −(n u )k (σu )k = −g(n u , σu ).
But σu is tangent to and n is normal, so
0 = g(n, σu )u = g(n u , σu ) + g(n, σuu ),
and therefore
J = g(n, σuu )
as well. The other terms are obtained similarly.
Intrinsic and extrinsic curvature 161
b. We can do this the simple way or the fancy way. First the simple way. Let
X = σ∗ ∂u and Y = σ∗ ∂v , as before. As X and Y span the tangent space to
,
S(X ) = a X + bY
S(Y ) = cX + dY,
for some a, b, c, and d. Taking inner products gives
J = g(S(X ), X ) = ag(X, X ) + bg(Y, X ) = a E + bF
K = g(S(X ), Y ) = ag(X, Y ) + bg(Y, Y ) = a F + bG
K = g(S(Y ), X ) = cg(X, X ) + dg(Y, X ) = cE + d F
L = g(S(Y ), Y ) = cg(X, Y ) + dg(Y, Y ) = cF + dG,
which we write as
J K a b E F
= .
K L c d F G
Inverting and multiplying gives
−1
a b J K E F
=
c d K L F G
1 J K G −F
=
EG − F2 K L −F E
1 GJ − FK EK − FJ
= .
EG − F2 G K − F L E L − F K
But, relative to the basis {X, Y }, the matrix representing S is just the
transpose of this matrix, by virtue of our conventions (cf., (1.11)).
Alternatively, the fancy way is to observe that (using the symmetry of bi j )
bi j = II(∂i , ∂ j ) = g(S(∂i ), ∂ j ) = S k i gk j ⇒ S k i = g k j b ji ,
which gives the same result as before.
c. This follows immediately from the previous solution by taking determi-
nants:
det S = (det II)(det I )−1 .
d. Differentiate both sides of
g(σu , σu ) = g(σv , σv )
with respect to u to get
g(σuu , σu ) = g(σvu , σv ).
162 Intrinsic and extrinsic curvature
But
0 = g(σu , σv )v = g(σuv , σv ) + g(σu , σvv ),
so combining the two equations we get
g(σu , σuu + σvv ) = 0.
A similar computation shows that
g(σv , σuu + σvv ) = 0
as well, demonstrating that σuu + σvv is normal to . Taking inner products
with n and using the results of Part (a) we get
g(n, σuu + σvv ) = J + L .
On the other hand, from Part (b),
G J + E L − 2F K
g(n, 2λ2 H ) = λ2 tr S = λ2 = J + L.
EG − F2
because E = G = λ2 and F = 0.
e. We have
σu = (−a cosh v sin u, a cosh v cos u, 0),
σuu = (−a cosh v cos u, −a cosh v sin u, 0),
σv = (a sinh v cos u, a sinh v sin u, a),
σvv = (a cosh v cos u, a cosh v sin u, 0).
It follows that
g(σu , σu ) = a 2 cosh2 v,
g(σv , σv ) = a 2 sinh2 v + a 2 = a 2 (1 + sinh2 v) = a 2 cosh2 v,
g(σu , σv ) = 0,
so the parameterization is indeed isothermal. Moreover, σuu + σvv = 0,
so by Part (d), the catenoid is a minimal surface. (It is the only minimal
surface of revolution.)
H.8 We have
σu = (−a sin u, a cos u, 0),
σuu = (−a cos u, −a sin u, 0),
σuv = (0, 0, 0),
σv = (0, 0, 1),
σvv = (0, 0, 0),
Intrinsic and extrinsic curvature 163
and
σu × σv (a cos u, a sin u, 0)
n= = = (cos u, sin u, 0).
σu × σv a
Thus,
E = (σu , σu ) = a 2 ,
F = (σu , σv ) = 0,
G = (σv , σv ) = 1,
J = (n, σuu ) = −a,
K = (n, σuv ) = 0,
L = (n, σvv ) = 0.
The shape operator is
2 −1 GJ − FK GK − FL −1/a 0
S = (E G − F ) = ,
EK − FJ EL − FK 0 0
which is already diagonalized. Therefore the Gaussian curvature (product
of eigenvalues) is 0, while the mean curvature (average of eigenvalues) is
−1/2a.
H.9 We have
σu = (1 − u 2 + v 2 , 2uv, 2u),
σuu = (−2u, 2v, 2),
σuv = (2v, 2u, 0),
σv = (2uv, 1 − v 2 + u 2 , −2v),
and σvv = (2u, −2v, −2).
Also, n = η/η where
î ĵ k̂
η = σu × σv = 1 − u 2 + v 2 2uv 2u .
2uv 1 − v2 + u 2 −2v
Therefore,
η2 = η · η = (−2uα)2 + (2vα)2 + (1 − x 2 )2
= 4(u 2 + v 2 )α 2 + (1 − 2x 2 + x 4 ) = 4x(1 + x)2 + 1 − 2x 2 + x 4
= 4x(1 + 2x + x 2 ) + 1 − 2x 2 + x 4 = 1 + 4x + 6x 2 + 4x 3 + x 4
= (1 + x)4 = α 2 ,
and
η (−2uα, 2vα, 1 − x 2 ) 2u 2v 1 − (u 2 + v 2 )2
n= = = − , , .
η α2 α α α2
Using these results we get the following.
a.
E = (σu , σu ) = (1 − u 2 + v 2 )2 + (2uv)2 + (2u)2
= 1 + u 4 + v 4 − 2u 2 + 2v 2 − 2u 2 v 2 + 4u 2 v 2 + 4u 2
= 1 + u 2 + v 4 + 2u 2 + 2v 2 + 2u 2 v 2 = (1 + u 2 + v 2 )2 .
Similar reasoning gives F = 0 and G = E.
b.
1
J = (n, σuu ) = [(−2uα)(−2u) + (2vα)(2v) + 2(1 − x 2 )]
α 2
1 2
= 2 [4x(1 + x) + 2 − 2x 2 ] = 2 (1 + 2x + x 2 ) = 2.
α α
Similar reasoning gives K = 0 and L = −2.
c. The shape operator is
2 −1 G J − F K GK − FL 2 1 0
S = (E G − F ) = 2 ,
EK − FJ EL − FK α 0 −1
which is already diagonalized. Therefore the principal curvatures are
λ1 = 2/α 2 and λ2 = −2/α 2 . In particular, the mean curvature (average
of the principal curvatures) is zero.
H.10 a. Parameterize the surface in the obvious way by
σ (x, y) = (x, y, f (x, y)).
Then
σx = (1, 0, f x ) and σ y = (0, 1, f y ),
so
E = g(σx , σx ) = 1 + f x2 ,
F = g(σx , σ y ) = f x f y ,
G = g(σ y , σ y ) = 1 + f y2 .
Intrinsic and extrinsic curvature 165
b. We have
ex ey ez
σx × σ y = 1 0 f x = (− f x , − f y , 1),
0 1 fy
and
(− f x , − f y , 1)
n= .
(1 + f x2 + f y2 )
Also,
σx x = (0, 0, f x x ), σx y = (0, 0, f x y ), and σ yy = (0, 0, f yy ),
so
fx x
J = g(n, σx x ) = ,
(1 + f x2 + f y2 )
fx y
K = g(n, σx y ) = ,
(1 + f x2 + f y2 )
f yy
L = g(n, σ yy ) = .
(1 + f x2 + f y2 )
c. The Gaussian curvature is the ratio of the determinants of the second and
first fundamental forms, namely
det II JL − K2
λ1 λ2 = =
det I E G − F32 4
1 f x x f yy − f x2y
=
(1 + f x2 + f y2 )2 (1 + f x2 )(1 + f y2 ) − ( f x f y )2
f x x f yy − f x2y
= .
(1 + f x2 + f y2 )3
d. The mean curvature is the trace of the shape operator, namely
G J + E L − 2F K
λ1 + λ2 = .
EG − F2
This vanishes when the numerator vanishes, so the equation of a minimal
surface is
(1 + f y2 ) f x x + (1 + f x2 ) f yy − 2 f x f y f x y = 0.